Search

Abstract In recent times, AI and UAV have progressed significantly in several applications. This article analyzes applications of UAV with modern green computing in various sectors. It addresses cutting-edge technologies such as green computing, generative AI, future scope, and related concerns in UAV. The research investigates the role of green computing and generative AI in combination with UAVs for navigation, object recognition and tracking, wildlife monitoring, precision agriculture, rescue operations, surveillance, and UAV communication. This study examines how modern computing technologies and UAVs are being applied in agriculture, surveillance, disaster management, and other areas. The ethics of UAV and AI applications, including safety, legal frameworks, and other issues, are thoroughly investigated. This research examines AI-based UAV applications across different disciplines, using open-source data and current advancements for future growth in this domain. This investigation will aid future researchers in their exploration of UAVs using cutting-edge computing technologies.

Abstract Hand gesture recognition based on sparse multichannel surface electromyography (sEMG) still poses a significant challenge to deployment as a muscle–computer interface. Many researchers have been working to develop an sEMG-based hand gesture recognition system. However, the existing system still faces challenges in achieving satisfactory performance due to ineffective feature enhancement, so the prediction is erratic and unstable. To comprehensively tackle these challenges, we introduce a novel approach: a lightweight sEMG-based hand gesture recognition system using a 4-stream deep learning architecture. Each stream strategically combines Temporal Convolutional Network (TCN)-based time-varying features with Convolutional Neural Network (CNN)-based frame-wise features. In the first stream, we harness the power of the TCN module to extract nuanced time-varying temporal features. The second stream integrates a hybrid Long short-term memory (LSTM)-TCN module. This stream extracts temporal features using LSTM and seamlessly enhances them with TCN to effectively capture intricate long-range temporal relations. The third stream adopts a spatio-temporal strategy, merging the CNN and TCN modules. This integration facilitates concurrent comprehension of both spatial and temporal features, enriching the model’s understanding of the underlying dynamics of the data. The fourth stream uses a skip connection mechanism to alleviate potential problems of data loss, ensuring a robust information flow throughout the network and concatenating the 4 stream features, yielding a comprehensive and effective final feature representation. We employ a channel attention-based feature selection module to select the most effective features, aiming to reduce the computational complexity and feed them into the classification module. The proposed model achieves an average accuracy of 94.31% and 98.96% on the Ninapro DB1 and DB9 datasets, respectively. This high-performance accuracy proves the superiority of the proposed model, and its implications extend to enhancing the quality of life for individuals using prosthetic limbs and advancing control systems in the field of robotic human–machine interfaces.

The emergence of generative adversarial networks (GANs) has ignited substantial interest in the domain of synthesizing images from textual descriptions. This approach has demonstrated remarkable versatility and user-friendliness in producing conditioned images, showcasing notable progress in areas like diversity, visual realism, and semantic alignment in recent years. Notwithstanding these developments, the discipline still faces difficulties, such as producing high-resolution pictures with several objects and developing trustworthy evaluation standards that are in line with human vision. The goal of this study is to provide a comprehensive overview of the state of stochastic text-to-image creation models as of right now. It examines how they have changed over the previous five years and suggests a classification system depending on the degree of supervision required. The paper highlights shortcomings, provides a critical evaluation of current approaches for assessing text-to-image synthesizing models, and suggests further study areas. These goals include improving the training of models and designs for architecture, developing more reliable assessment criteria, and fine-tuning datasets. This review, which focuses on text-to-image synthesizing, is a useful addition to earlier surveys on adversarial networks that are generative and offers guidance for future studies on the subject.

Colorectal cancer has become a significant global health challenge and causes millions of fatalities every year. A key factor contributing to its high fatality rate is the absence of timely screening and accurate diagnosis. Early diagnosis is based on the detection and analysis of polyps which are cancer precursors linked to aging and declining health. Artificial intelligence-based computer-aided polyp detection (CAPD) tools are vital for affordable and efficient screening while mitigating the global impact of colorectal cancer. Aimed at improving the polyp detection capabilities, this research presents a real-time framework for detecting polyps in endoscopy frames, employing contrast-limited adaptive histogram equalization (CLAHE) and the proposed polyp detector YOLOv8p. The proposed detector uses multiple convolutional, C2f (Faster Cross Stage Partial Bottleneck with 2 convolutions), and SPPF (Spatial Pyramid Pooling-Fast) blocks. Moreover, a generic suitability assessment metric named YOLO-Score is proposed for overall performance evaluation. The proposed framework demonstrates superior clinical relevance and comparable performance to existing SOTA methods by achieving a mAP50 score of 95.5% and an $F_{1}$ score of 92.6% on the Hyper-Kvasir-Seg dataset. The research underscores the essential role of CAPD tools in improving diagnostic precision and streamlining real-time clinical processes in colorectal cancer screening.

Sensor-based datasets are extensively utilized in human-computer interaction (HCI) and medical applications due to their portability and strong privacy features. Many researchers have developed sensor-based human activity recognition (HAR) systems to increase recognition performance. However, existing systems still face challenges in achieving satisfactory performance due to insufficient time-varying features and gradient explosion issues. To address these challenges, we proposed a multi-stream temporal convolutional network (TCN)-based approach for time-varying feature extraction and feature selection to recognize human activity (HA) from sensor datasets. The proposed model effectively extracts and emphasizes the spatial-temporal features of various human activities based on a 4-stream model. Each stream uses TCN to extract time-varying features and enhances them using an appropriate integration module. The first stream extracts fine-grained temporal features with TCN. The second and third streams integrate TCN features with LSTM, applying pre-integration and post-integration, respectively. The fourth stream uses CNN for spatial features and TCN for enhancing temporal features. The concatenation of the 4-stream features captures complex dependencies, improving the model’s understanding of prolonged activities. In addition, we proposed a modified effective channel attention network (ECA-Net) that assigns higher dimensionality weight to lower dimensionality, enabling the proposed model to learn and recognise human activities effectively despite their complex patterns. Evaluations on the WISDM, PAMAP2, USC-HAD, Opportunity UCI, and UCI-HAR datasets showed accuracy improvements of 1.12%, 1.99%, 1.30%, 5.72%, and 0.38%, respectively, over state-of-the-art systems. The high-performance accuracy of the proposed model demonstrates its superiority, with implications for improving prosthetic limb functionality and advancing robotics human-machine interfaces. Our data preprocessing approach, deep learning model code, and dataset information are available at the following link: https://github.com/musaru/HAR_Sensor.

Colorectal cancer (CC) is a significant public health concern and make it necessary to identify reliable biomarkers and elucidate their molecular and biological mechanisms. This study proposed a system by integrating weighted gene co-expression network analysis (WGCNA) and machine learning-based integrative bioinformatics (ML-IB) analysis to identify key genes for CC. WGCNA was implemented to find a co-expression network of genes and identify important genes by intersecting gene sets obtained using module membership and gene significance criteria across datasets. WGCNA-based significant genes were determined by intersecting important genes between two datasets. ML-IB based approach primarily identified differentially expressed genes (DEGs), then employed support vector machine to determine differentially expressed discriminative genes (DEDGs) and took their common DEDGs across datasets. Protein-protein interaction networks were built and identified hub genes based on the degrees of connectivity and hub module genes using MCODE scores. The ML-IB based significant genes were determined by intersecting hub genes and hub module genes. Four common significant genes were found by intersecting significant genes derived from WGCNA and ML-IB based perspectives. Finally, two genes (AURKA and CCNA2) were determined as key genes for showing strong correlation with survival of CC patients and validated their discriminative capability on an independent test dataset using AUC analysis. The key genes of AURKA and CCNA2 may be used for the early detection of patients with CC. This study will helpful for physicians and doctors to determine and understand the associated the molecular mechanisms and pathway of patients with CC.

Researchers have been developing Hand Gesture Recognition (HGR) systems to enhance natural, efficient, and authentic human-computer interaction, especially benefiting those who rely solely on hand gestures for communication. Despite significant progress, automatic and precise identification of hand gestures remains a considerable challenge in computer vision. Recent studies have focused on specific modalities like RGB images, skeleton data, and spatiotemporal interest points. This paper comprehensively reviews HGR techniques and data modalities from 2014 to 2024, exploring advancements in sensor technology and computer vision. We highlight accomplishments using various modalities, including RGB, Skeleton, Depth, Audio, Electromyography (EMG), Electroencephalography (EEG), and Multimodal approaches and identify areas needing further research. We reviewed over 250 articles from prominent databases, focusing on data collection, data settings, and gesture representation. Our review assesses the efficacy of HGR systems through their recognition accuracy and identifies a gap in research on continuous gesture recognition, indicating the need for improved vision-based gesture systems. The field has experienced steady research progress, including advancements in hand-crafted features and deep learning (DL) techniques. Additionally, we report on the promising developments in HGR methods and the area of multimodal approaches. We hope this survey will serve as a potential guideline for diverse data modality-based HGR research.

Ensuring product quality and integrity is paramount in the rapidly evolving landscape of industrial manufacturing. Although effective to a certain degree, traditional quality control methods often fail to meet the demands for efficiency, accuracy, and adaptability in today’s fast-paced production environments. The advent of Deep Learning (DL) and Computer Vision (CV) technologies has opened new vistas for automated defect detection, promising to revolutionize the way industries approach quality control and inspection. This systematic review focuses on recent advancements in DL and CV applications for automated defect detection in manufacturing processes. It provides a comprehensive overview of state-of-the-art techniques for detecting, classifying, and predicting defects, highlighting the significant strides made in addressing challenges such as varying lighting conditions, complex defect patterns, and the seamless integration of these technologies into existing manufacturing workflows. Through a critical analysis of current methodologies, this study identifies key areas of opportunity, outlines the challenges that persist and suggests directions for future research. This review synthesizes findings from a broad spectrum of industrial applications, offering insights into the potential of DL and CV to enhance quality control mechanisms. By charting the progress and pinpointing the gaps in current practices, this paper aims to serve as a valuable resource for researchers, practitioners, and policymakers seeking to leverage the benefits of DL and CV for improved product management and manufacturing excellence.

The classification of gender from handwriting is a challenging issue that have great attention recently. Most of the exiting works were conducted on gender classification using face image and offline handwritten texts. This study explored an automated system for gender classification from online handwritten patterns. The handwritten samples were collected from 79 (Male: 32 and Female: 47) using pen tablet device. Each subject was asked to perform four tasks such as Zigzag trace (ZigZ-T), Zigzag predict (ZigZ-P), periodic line trace (PL-T), and periodic line predict (PL-P) and repeated its into three times. Thirty-six statistical features were derived from the six raw features, obtained from the Pen tablet device. Following that, we selected the best subset of features by employing Sequential Forward Floating Selection (SFFS)-based algorithm. At the same time, four machine learning (ML)-based algorithms like support vector machine (SVM), random forest, AdaBoost, and Gradient Boosting (GB) were employed for gender classification. We trained these four ML-based algorithms with leave-one-out method and optimized their hyperparameters using Optuna. The experimental results showed that SVM achieved a recognition accuracy of 88.10% for adult ZigZ-T tasks and 90.09% recognition accuracy was obtained by GB-based algorithm for children whose drawing ZigZ-P tasks. our proposed system demonstrates promise in automating gender classification based on handwriting patterns, offering insights into the significant differences between adult and child handwriting. The ability to identify gender accurately from handwriting has broad applications, including security enhancements and personalized service provision.

Attention-deficit/hyperactivity disorder (ADHD) is a neurodevelopmental disorder characterized by the symptoms of inattention, hyperactivity, and impulsivity that significantly affect daily functioning. It is usually 1st diagnosed in childhood and often lasts into adulthood. Various researchers introduced different statistical tools to identify children with ADHD based on phonotype, vision, and image-based data. The main causes of children with ADHD are still unknown and its diagnostic accuracy rate remains low. There is still some scopes to establish valid biomarkers for ADHD and improve the classification accuracy, which will help for children enhance their quality of life. This study aimed to establish and determine potential biomarkers for children with ADHD and then proposed a machine learning (ML)-based ADHD detection system. A multicenter approach was employed to collect Reverse Stroop Task (RST) data, including age, behavioral, and physiological indicators from 72 children with ADHD (aged 6–13 years) and 171 typically developing (TD) children. Moreover, we also collected the signal information from each subject using functional near-infrared spectroscopy (fNIRs) and quantified the change in prefrontal cortex oxygenated hemoglobin during RST. At the same time, we also computed the mean signal for every channel during the last 20 seconds of RST. We selected 70% of the dataset as a training set (ADHD: 51 and TD: 120) and the remaining dataset was used as a testing set (ADHD: 21 and TD: 51). To determine the potential biomarkers for ADHD, we employed independent t-test (p

Recognizing sign language plays a crucial role in improving communication accessibility for the Deaf and hard-of-hearing communities. In Korea, many individuals facing hearing and speech challenges depend on Korean Sign Language (KSL) as their primary means of communication. Many researchers have been working to develop a sign language recognition system for other sign languages. Still, little research has been done for KSL alphabet recognition due to the lack of dataset availability. Moreover, existing KSL recognition systems have faced significant performance limitations due to the ineffectiveness of the features. To overcome the challenges, we newly created a KSL alphabet dataset and introduced an innovative KSL recognition system employing a strategic fusion approach. In the proposed study, we combined joint skeleton-based handcrafted features and pixel-based resnet101 transfer learning features to overcome the limitations of feature effectiveness in traditional systems. Our system consists of two distinct feature extraction streams: the first stream extracts essential handcrafted features, emphasizing capturing hand orientation information within KSL gestures. In the second stream, concurrently, we employed a deep learning-based resnet101 module stream to capture hierarchical representations of the KSL alphabet sign. By combining essential features from the first stream with the hierarchical features from the second stream, we generate multiple levels of fused features with the goal of forming a comprehensive representation of KSL gestures. Finally, we fed the concatenated feature into the deep learning-based classification module for the classification. We conducted extensive experiments with the newly created KSL alphabet dataset, the existing KSL digit and the existing ArSL and ASL benchmark datasets. Our proposed model undeniably shows that our fusion approach substantially improves high-performance accuracy in both cases, which proves the system’s superiority.

Weakly supervised video anomaly detection (WS-VAD) is a crucial research domain in computer vision for the implementation of intelligent surveillance systems. Many researchers have been working to develop WS-VAD systems using various technologies by assessing anomaly scores. However, they are still facing challenges because of lacking effective feature extraction. To mitigate this limitation, we propose a multi-stage deep-learning model for separating abnormal events from normality to extract the hierarchical effective features. In the first stage, we extract two stream features using pre-trained techniques: the first stream employs a ViT-based CLIP module to select top-k features, while the second stream utilizes a CNN-based I3D module integrated into the Temporal Contextual Aggregation (TCA) mechanism. These features are concatenated and fed into the second-stage module, where an Uncertainty-regulated Dual Memory Units (UR-DMU) model is employed to learn representations of regular and abnormal data simultaneously. The UR-DMU integrates global and local structures, leveraging Graph Convolutional Networks (GCN) and Global and Local Multi-Head Self Attention (GL-MHSA) modules to capture video associations. Subsequently, feature reduction is achieved using the multilayer-perceptron (MLP) integration with the Prompt-Enhanced Learning (PEL) module via the knowledge-based prompt. Finally, we employed a classifier module to predict the snippet-level anomaly scores. In the training phase, the based function transfers the snippet-level scores into bag-level predictions for learning high activation in anomalous cases. Our approach integrates these cutting-edge technologies and methodologies, offering a comprehensive solution to video-based anomaly detection. Extensive experiments on ShanghaiTech, XD-Violence, and UCF-Crime datasets validate the superiority of our method over state-of-the-art approaches by a substantial margin. We believe that our model holds significant promise for real-world applications, demonstrating superior performance and efficacy in anomaly detection tasks.

Font style recognition plays a vital role in the field of computer vision, particularly in document and pattern analysis, and image processing. In the industry context, this recognition of font styles holds immense importance for professionals such as graphic designers, front-end developers, and UI-UX developers. In recent times, font style recognition using Computer Vision has made significant progress, especially in English. Very few works have been done for other languages as well. However, the existing models are computationally costly, time-consuming, and not diversified. In this work, we propose a state-of-the-art model to recognize Bangla fonts from images using a quantized Convolutional Autoencoder (Q-CAE) approach. The compressed model takes around 58 KB of memory only which makes it suitable for not only high-end but also low-end computational edge devices. We have also created a synthetic data set consisting of 10 distinct Bangla font styles and a total of 60,000 images for conducting this study as no dedicated dataset is available publicly. Experimental outcomes demonstrate that the proposed method can perform better than existing methods, gaining an overall accuracy of 99.95% without quantization and 99.85% after quantization.

A type of blood malignancies known as leukemia leads to elevated quantities of abnormally formed blood cells and often originates in the bone marrow. An abrupt rise in the quantity of immature blood cells is one of the distinctive characteristics of acute leukemia, that predominantly affects both children and adults. The possibility exists to drastically lower the death rate associated with acute leukemia through early detection and diagnosis. A significant and time-consuming task for the early detection of acute leukemia is microscopic examination of blood cells. In this paper, a two-staged deep learning-based computer-aided diagnosis system regarding microscopic blood smear images is proposed to assist hematologists and improve the diagnosis accuracy of acute leukemia in which for the feature fusion-based ensemble, two deep neural network branches adopting pretrained, fine-tuned EfficientNetB7 and MobileNetV3Large architectures were employed, and feature maps generated from those branches were fused and fed to the second stage of the architecture to achieve the final result. Additional dropout layers and ReLu activation were employed in the architecture to speed up the network, and compound scaling, bottleneck, and fusion architectures enhanced the overall performance. The ALLIDB1, ALLIDB2, and ASH databases were incorporated to evaluate the performances of the proposed method. The experimental findings demonstrated that the proposed approach detected acute leukemia (ALL, AML, Healthy) with an accuracy of 99.3%, F1-score of 99.3%, and AUC score of 0.997. Whereas in detecting acute lymphocytic leukemia (ALL, Healthy) and acute myeloid leukemia (AML, Healthy), the accuracies reached 100.0% and 99.8%, respectively. Thus, we believe that clinics can adopt our proposed architecture for quick and automated diagnosis.

Hand gesture-based Sign Language Recognition (SLR) serves as a crucial communication bridge between hard of hearing and non-deaf individuals. The absence of a universal sign language (SL) leads to diverse nationalities having various cultural SLs, such as Korean, American, and Japanese sign language. Existing SLR systems perform well for their cultural SL but may struggle with other or multi-cultural sign languages (McSL). To address these challenges, this paper introduces a novel end-to-end SLR system called GmTC, designed to translate McSL into equivalent text for enhanced understanding. Here, we employed a Graph and General deep-learning network as two stream modules to extract effective features. In the first stream, produce a graph-based feature by taking advantage of the superpixel values and the graph convolutional network (GCN), aiming to extract distance-based complex relationship features among the superpixel. In the second stream, we extracted long-range and short-range dependency features using attention-based contextual information that passes through multi-stage, multi-head self-attention (MHSA), and CNN modules. Combining these features generates final features that feed into the classification module. Extensive experiments with five culture SL datasets with high-performance accuracy compared to existing state-of-the-art models in individual domains affirming superiority and generalizability.

Parkinson’s disease (PD) is a neurological disorder caused by degeneration of dopaminergic neurons in the midbrain. PD patients mainly suffer from motor symptoms, which significantly impact their daily lives. The diagnostic criteria for PD include the presence of muscle rigidity, tremor, and postural reflex disturbances. The Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) is the standard tool for evaluating PD symptoms, part III of which is dedicated to motor symptoms. That part involves a comprehensive set of specific physical examinations, and physicians assign semi quantitative scores from 0 to 4. However, this approach faces notable challenges, including the requirement for movement-disorder experts proficient in using MDS-UPDRS and the presence of substantial inter rater variability even among experts. Overcoming these challenges requires a quantitative and objective assessment method. Given that the rating of motor symptoms predominantly involves assessing kinematic characteristics, the integration of sensor-based devices and machine learning techniques holds the potential to outperform human experts in symptom evaluations. This study used the Leap Motion optical motion-capture device to quantitatively measure and analyze hand movements while 45 PD patients performed the following 3 tasks from the MDS-UPDRS part III: finger tapping (FT), hand opening and closing (OC), and forearm pronation and supination (PS). Data from these tasks were collected and processed, resulting in the extraction of 31 movement patterns for each task. Additionally, 69 statistical features were extracted from each movement pattern, yielding 2139 features for each task. We subsequently employed a random forest algorithm to select the top 15% of features based on the reduction of Gini impurity. These selected features were subsequently fed into a sequential-forward-floating-selection algorithm, combined with a support vector machine, to identify relevant feature combinations and predict the severity of the motor symptoms. The classification accuracy was 87.0% for FT, 93.2% for OC, and 92.2% for PS. One-way analysis of variance identified 13 features of the OC task that were significantly more discriminative for classifying the movement disability of PD patients ( ${p}$

Generative Adversarial Networks are a class of artificial intelligence algorithms that consist of a generator and a discriminator trained simultaneously through adversarial training. GANs have found crucial applications in various fields, including medical imaging. In healthcare, GANs contribute by generating synthetic medical images, enhancing data quality, and aiding in image segmentation, disease detection, and medical image synthesis. Their importance lies in their ability to generate realistic images, facilitating improved diagnostics, research, and training for medical professionals. Understanding its applications, algorithms, current advancements, and challenges is imperative for further advancement in the medical imaging domain. However, no study explores the recent state-of-the-art development of GANs in medical imaging. To overcome this research gap, in this extensive study, we began by exploring the vast array of applications of GANs in medical imaging, scrutinizing them within recent research. We then dive into the prevalent datasets and pre-processing techniques to enhance comprehension. Subsequently, an in-depth discussion of the GAN algorithms, elucidating their respective strengths and limitations, is provided. After that, we meticulously analyzed the results and experimental details of some recent cutting-edge research to obtain a more comprehensive understanding of the current development of GANs in medical imaging. Lastly, we discussed the diverse challenges encountered and future research directions to mitigate these concerns. This systematic review offers a complete overview of GANs in medical imaging, encompassing their application domains, models, state-of-the-art results analysis, challenges, and research directions, serving as a valuable resource for multidisciplinary studies.

Sign Language Recognition (SLR) represents a revolutionary technology aiming to establish communication between hearing impaired and non-hearing impaired communities, surpassing traditional interpreter-based approaches. Existing efforts in automatic sign recognition predominantly rely on hand skeleton joint information, steering clear of image pixels to address challenges like partial occlusion and redundant backgrounds. Many researchers have been working to develop automatic sign recognition using hand skeleton joint information instead of image pixels to overcome partial occlusion and redundant background problems. However, body motion and facial expression play an essential role in increasing the inner gesture variance in expressing sign language emotion besides hand information for large-scale sign word datasets. Recently, some researchers have been working to develop muti-gesture-based SLR recognition systems, but their performance accuracy and efficiency are unsatisfactory for real-time deployment. Addressing these limitations, we propose a novel approach, a two-stream multistage graph convolution with attention and residual connection (GCAR) designed to extract spatial-temporal contextual information. The multistage GCAR system, incorporating a channel attention module, dynamically enhances attention levels, particularly for non-connected skeleton points during specific events within spatial-temporal features. The methodology involves capturing joint skeleton points and motion, offering a comprehensive understanding of a person’s entire body movement during sign language gestures and feeding this information into two streams. In the first stream, joint key features undergo processing through sep-TCN, graph convolution, deep learning layer, and a channel attention module across multiple stages, generating intricate spatial-temporal features in sign language gestures. Simultaneously, the joint motion is processed in the second stream, mirroring the steps of the first branch. The fusion of these two features yields the final feature vector, which is then fed into the classification module. The model excels in capturing discriminative structural displacements and short-range dependencies by leveraging unified joint features projected onto a high-dimensional space. Owing to the effectiveness of these features, the proposed method achieved significant accuracies: 90.31%, 94.10%, 99.75%, and 34.41%, for the WLASL, PSL, MSL, and ASLLVD large-scale datasets, respectively, with 0.69 million parameters. The high-performance accuracy, coupled with stable computational complexity, demonstrates the superiority of the proposed model. This innovative approach is anticipated to redefine the landscape of sign language recognition, setting a new standard in the field.

Visually impaired people have previously been brought into learning and educational systems through various forms of assistive technology, such as haptic feedback systems. Haptic systems generally need expensive equipment and support from sighted teachers. Moreover, the learning has always been carried out with letters of different alphabets mapped into some tactile pattern. Writing is a big concern for the visually impaired as most official work, like signing, is still carried out by conventional handwriting methods. Most of the existing systems are limited to teaching a single language’s alphabet and basic grammar or may not provide feedback to let the learners know of their learning progress. Therefore, the objectives of this research are to develop an efficient system that includes voice-over guidance to teach writing in multiple alphabets to visually impaired people and to evaluate the performance of the proposed system. As such, a system was developed for teaching multilingual alphabets to visually impaired people with voice instructions. With the aid of a voice-over guide, learners were able to write letters with a stylus on a graphics pad. The progress assessment of the learners is carried out by an image processing algorithm and scored by a machine learning (ML) model. The Random Forest model was used due to its high accuracy (f1-score of 99.8% on test data) among the existing ten different ML algorithms. Finally, the performance and usability of this system were evaluated through an empirical study replicated with 16 participants, including four teachers and twelve visually impaired people. It was found that visually impaired people made fewer attempts to learn handwriting with the proposed system than with the normal handwriting teaching system. 100% of the participants agreed to recommend the system in the future.

The generation of high-quality images from textual descriptions is a challenging task in computer vision and natural language processing. The goal of text-to-image synthesis, a current topic of research, is to produce excellent images from written descriptions. This study proposes a hybrid approach to evaluating a dataset consisting of various text-image pairs by efficiently combining conditional generative adversarial networks (C-GAN), attention mechanisms, and contrastive learning (C-GAN+ATT+CL). We suggest a two-step method to improve image quality that starts by utilizing generative adversarial networks (GANs) with attention mechanisms to create low-resolution images and then contrastive learning to improve. Contrastive learning modules train on a separate dataset of high-resolution pictures; GANs learn on datasets of low-resolution text and image pairs. The Conditional GAN with Attention Mechanism and Contrastive Learning Method provides state-of-the-art performance in terms of image quality, diversity, and visual realism, among the several methods. The results of this study demonstrate that the proposed approach works better than all other methods, achieving an Inception Score (IS) of 35.23, a Fréchet Inception Distance (FID) of 18.2, and an R-Precision of 89.14. Our findings demonstrate that our “C-GAN+ATT+CL” approach significantly improves image quality and diversity and offers exciting paths for further study.

Handwriting is controlled by neurons in the brain’s nervous system, reflecting an individual’s personality and psychology. This unique characteristic can be used for various applications, including user authentication, assessment of neurodegenerative disorders, and classification of handedness, gender, and age groups. Traditional authentication systems require memorization, information leakage, and fingerprints, making them vulnerable to security breaches. The majority of researchers have studied the limitations of image quality, camera frames, and light effects on text and image-dependent performance. Therefore, this paper mainly focused on real-time, text-independent handwriting fine-motor data and proposed an efficient authentication system with low cost using efficient feature extraction and optimal feature selection approaches. This research utilizes two benchmark databases, including the handwriting data of 48 (24+24) participants collected via a sensor-based pen tablet. Each participant wrote the 10 words five times repeatedly, making it a total of 2400 samples. The handwriting classification of the different individuals is in 3 phases: feature extraction, feature selection, and classification. A total of 91 features (statistical, kinematic, spatial, and composite) were extracted from more accurate, real-time numerical handwriting data. The efficient and optimal features have been selected using four feature selection approaches, namely, Pearson’s r correlation, ANOVA-F, Mutual Information Gain, and PCA, among which the ANOVA-F test and PCA perform well for handwriting-extracted data. Then, 14 machine learning (ML) models and 7 deep learning (DL) models were applied to handle the problem of individual classification, with both no- and full-feature-selection scenarios considered. The experimental analysis has been conducted with different angles and perspectives, such as K-Fold cross-validation, testing system efficiency considering 5/10/15/24/48 individuals, and in the case of individual tasks. It shows that ML-based algorithms, namely, CATBOOST (99.07%) with ANOVA-F and DL-based models, namely, BiLSTM (98.31%) with PCA-selected features, provide the highest accuracy with dataset 2, among others that advocate the practicality and reliability of choosing this system for user identification.

Human interaction recognition (HIR) between two people in videos is a critical field in computer vision and pattern recognition, aimed at identifying and understanding human interaction and actions for applications such as healthcare, surveillance, and human–computer interaction. Despite its significance, video-based HIR faces challenges in achieving satisfactory performance due to the complexity of human actions, variations in motion, different viewpoints, and environmental factors. In the study, we proposed a two-stream deep learning-based HIR system to address these challenges and improve the accuracy and reliability of HIR systems. In the process, two streams extract hierarchical features based on the skeleton and RGB information, respectively. In the first stream, we utilised YOLOv8-Pose for human pose extraction, then extracted features with three stacked LSM modules and enhanced them with a dense layer that is considered the final feature of the first stream. In the second stream, we utilised SAM on the input videos, and after filtering the Segment Anything Model (SAM) feature, we employed integrated LSTM and GRU to extract the long-range dependency feature and then enhanced them with a dense layer that was considered the final feature for the second stream module. Here, SAM was utilised for segmented mesh generation, and ImageNet was used for feature extraction from images or meshes, focusing on extracting relevant features from sequential image data. Moreover, we newly created a custom filter function to enhance computational efficiency and eliminate irrelevant keypoints and mesh components from the dataset. We concatenated the two stream features and produced the final feature that fed into the classification module. The extensive experiment with the two benchmark datasets of the proposed model achieved 96.56% and 96.16% accuracy, respectively. The high-performance accuracy of the proposed model proved its superiority.

Abstract Supply chain management relies on accurate backorder prediction for optimizing inventory control, reducing costs, and enhancing customer satisfaction. Traditional machine-learning models struggle with large-scale datasets and complex relationships. This research introduces a novel methodological framework for supply chain backorder prediction, addressing the challenge of collecting large real-world datasets with 90% accuracy. Our proposed model demonstrates remarkable accuracy in predicting backorders on short and imbalanced datasets. We capture intricate patterns and dependencies by leveraging quantum-inspired techniques within the quantum-classical neural network QAmplifyNet. Experimental evaluations on a benchmark dataset establish QAmplifyNet’s superiority over eight classical models, three classically stacked quantum ensembles, five quantum neural networks, and a deep reinforcement learning model. Its ability to handle short, imbalanced datasets makes it ideal for supply chain management. We evaluate seven preprocessing techniques, selecting the best one based on logistic regression’s performance on each preprocessed dataset. The model’s interpretability is enhanced using Explainable artificial intelligence techniques. Practical implications include improved inventory control, reduced backorders, and enhanced operational efficiency. QAmplifyNet also achieved the highest F1-score of 94% for predicting “Not Backorder” and 75% for predicting “backorder,” outperforming all other models. It also exhibited the highest AUC-ROC score of 79.85%, further validating its superior predictive capabilities. QAmplifyNet seamlessly integrates into real-world supply chain management systems, empowering proactive decision-making and efficient resource allocation. Future work involves exploring additional quantum-inspired techniques, expanding the dataset, and investigating other supply chain applications. This research unlocks the potential of quantum computing in supply chain optimization and paves the way for further exploration of quantum-inspired machine learning models in supply chain management. Our framework and QAmplifyNet model offer a breakthrough approach to supply chain backorder prediction, offering superior performance and opening new avenues for leveraging quantum-inspired techniques in supply chain management.

The accuracy of classifying motor imagery (MI) activities is a significant challenge when using brain–computer interfaces (BCIs). BCIs allow people with motor impairments to control external devices directly with their brains using electroencephalogram (EEG) patterns that translate brain activity into control signals. Many researchers have been working to develop MI-based BCI recognition systems using various time-frequency feature extraction and classification approaches. However, the existing systems still face challenges in achieving satisfactory performance due to large amount of non-discriminative and ineffective features. To get around these problems, we suggested a multiband decomposition-based feature extraction and classification method that works well, along with a strong feature selection method for MI tasks. Our method starts by splitting the preprocessed EEG signal into four sub-bands. In each sub-band, we then used a common spatial pattern (CSP) technique to pull out narrowband-oriented useful features, which gives us a high-dimensional feature vector. Subsequently, we utilized an effective feature selection method, Relief-F, which reduces the dimensionality of the final features. Finally, incorporating advanced classification techniques, we classified the final reduced feature vector. To evaluate the proposed model, we used the three different EEG-based MI benchmark datasets, and our proposed model achieved better performance accuracy than existing systems. Our model’s strong points include its ability to effectively reduce feature dimensionality and improve classification accuracy through advanced feature extraction and selection methods.

We investigated the relationship between the prefrontal cortex (PFC) and executive function during a drawing task. Thirty-three participants using pen tablets provided the data for this task. PFC activity was recorded using functional near-infrared spectroscopy (fNIRS) during a simple zig-zag task and a complex periodic line (PL) pattern task. For each task, there was a trace condition and a prediction condition. The Executive Function Questionnaire (EFQ) was used to examine the association between brain-function measurements and executive function during the task. PFC activity was analyzed in the right, middle, and left regions. Oxygenated hemoglobin values measured with fNIRS were converted to z-values and analyzed as a measure of brain activity. Drawing fluency was measured using the line length. In the PL pattern task, the line length was significantly shorter under the prediction condition than under the trace condition. Activity in the right PFC under the prediction condition was significantly higher than that under the trace condition in the PL pattern task, and the score of the EFQ planning subscale was associated with activity in the right PFC. Activity in the right PFC is important for fluent drawing, suggesting that it is also important during drawing activities involving symbols such as letters.

Machine translation (MT) is the process of translating text from one language to another using bilingual data sets and grammatical rules. Recent works in the field of MT have popularized sequence-to-sequence models leveraging neural attention and deep learning. The success of neural attention models is yet to be construed into a robust framework for automated English-to-Bangla translation due to a lack of a comprehensive dataset that encompasses the diverse vocabulary of the Bangla language. In this study, we have proposed an English-to-Bangla MT system using an encoder-decoder attention model using the CCMatrix corpus. Our method shows that this model can outperform traditional SMT and RBMT models with a Bilingual Evaluation Understudy (BLEU) score of 15.68 despite being constrained by the limited vocabulary of the corpus. We hypothesize that this model can be used successfully for state-of-the-art machine translation with a more diverse and accurate dataset. This work can be extended further to incorporate several newer datasets using transfer learning techniques.

Alzheimer’s Disease (AD) poses a significant health risk globally, particularly among the elderly population. Recent studies underscore its prevalence, with over 50% of elderly Japanese facing a lifetime risk of dementia, primarily attributed to AD. As the most prevalent form of dementia, AD gradually erodes brain cells, leading to severe neurological decline. In this scenario, it is important to develop an automatic AD-detection system, and many researchers have been working to develop an AD-detection system by taking advantage of the advancement of deep learning (DL) techniques, which have shown promising results in various domains, including medical image analysis. However, existing approaches for AD detection often suffer from limited performance due to the complexities associated with training hierarchical convolutional neural networks (CNNs). In this paper, we introduce a novel multi-stage deep neural network architecture based on residual functions to address the limitations of existing AD-detection approaches. Inspired by the success of residual networks (ResNets) in image-classification tasks, our proposed system comprises five stages, each explicitly formulated to enhance feature effectiveness while maintaining model depth. Following feature extraction, a deep learning-based feature-selection module is applied to mitigate overfitting, incorporating batch normalization, dropout and fully connected layers. Subsequently, machine learning (ML)-based classification algorithms, including Support Vector Machines (SVM), Random Forest (RF) and SoftMax, are employed for classification tasks. Comprehensive evaluations conducted on three benchmark datasets, namely ADNI1: Complete 1Yr 1.5T, MIRAID and OASIS Kaggle, demonstrate the efficacy of our proposed model. Impressively, our model achieves accuracy rates of 99.47%, 99.10% and 99.70% for ADNI1: Complete 1Yr 1.5T, MIRAID and OASIS datasets, respectively, outperforming existing systems in binary class problems. Our proposed model represents a significant advancement in the AD-analysis domain.

Abstract Hepatocellular carcinoma (HCC) is the most common lethal malignancy of the liver worldwide. Thus, it is important to dig the key genes for uncovering the molecular mechanisms and to improve diagnostic and therapeutic options for HCC. This study aimed to encompass a set of statistical and machine learning computational approaches for identifying the key candidate genes for HCC. Three microarray datasets were used in this work, which were downloaded from the Gene Expression Omnibus Database. At first, normalization and differentially expressed genes (DEGs) identification were performed using limma for each dataset. Then, support vector machine (SVM) was implemented to determine the differentially expressed discriminative genes (DEDGs) from DEGs of each dataset and select overlapping DEDGs genes among identified three sets of DEDGs. Enrichment analysis was performed on common DEDGs using DAVID. A protein-protein interaction (PPI) network was constructed using STRING and the central hub genes were identified depending on the degree, maximum neighborhood component (MNC), maximal clique centrality (MCC), centralities of closeness, and betweenness criteria using CytoHubba. Simultaneously, significant modules were selected using MCODE scores and identified their associated genes from the PPI networks. Moreover, metadata were created by listing all hub genes from previous studies and identified significant meta-hub genes whose occurrence frequency was greater than 3 among previous studies. Finally, six key candidate genes (TOP2A, CDC20, ASPM, PRC1, NUSAP1, and UBE2C) were determined by intersecting shared genes among central hub genes, hub module genes, and significant meta-hub genes. Two independent test datasets (GSE76427 and TCGA-LIHC) were utilized to validate these key candidate genes using the area under the curve. Moreover, the prognostic potential of these six key candidate genes was also evaluated on the TCGA-LIHC cohort using survival analysis.

Attention deficit hyperactivity disorder (ADHD) for children is one of the behavioral disorders that affect the brain’s ability to control attention, impulsivity, and hyperactivity and its prevalence has increased over time. The cure for ADHD is still unknown and only early detection can improve the quality of life for children with ADHD. At the same time, children with ADHD often suffer from various comorbidities like autism spectrum disorder (ASD), major depressive disorder (MDD), etc. Various researchers developed computational tools to detect children with ADHD depending on handwriting text. Handwriting text-based systems are depending on a specific language that causes problems for non-native speakers of that language. Moreover, very few researchers considered other comorbidities such as ASD, MDD, etc., in their studies to detect ADHD for children. In this study, handwriting patterns or drawing is assumed as an aspect to identify/detect ADHD children who have ASD using machine learning (ML)-based approaches. We collected handwriting samples from 29 Japanese children (14 ADHD with coexisting ASD children and 15 healthy children) using a pen tablet. We asked each child to draw two patterns, namely zigzag lines (ZL) and periodic lines (PL) on a pen tablet and repeated them three times. We extracted 30 statistical features from raw datasets and these features were analyzed using sequential forward floating search (SFFS) and selected the best combinations or subsets of features. Finally, these selected features were fed into seven ML-based algorithms for detecting ADHD with coexisting ASD children. These classifiers were trained with leave-one-out cross-validation and evaluated their performances using accuracy, recall, precision, f1-score, and area under the curve (AUC). The experimental results showed that the highest performance scores (accuracy: 93.10%; recall: 90.48%; precision: 95.00%; f1-score: 92.68%; and AUC: 0.930) were achieved by the RF-based classifier for the PL predict task. This study will be helpful and provide evidence of the possibility of classifying ADHD children having ASD and healthy children based on their handwriting patterns.

Sign language recognition is crucial for improving communication accessibility for the hearing impaired community and reducing dependence on human interpreters. Notably, while significant research efforts have been devoted to many prevalent languages, Korean Sign Language (KSL) remains relatively underexplored, particularly concerning dynamic signs and generalizability. The scarcity of KSL datasets has exacerbated this limitation, hindering progress. Furthermore, most KSL research predominantly relies on static image-based datasets for recognition, leading to diminished accuracy and the inability to detect dynamic sign words. Furthermore, most KSL research predominantly relies on static image-based datasets for recognition, leading to diminished accuracy and the inability to detect dynamic sign words. Additionally, existing KSL recognition systems grapple with suboptimal performance accuracy and heightened computational complexity, further emphasizing the existing research gap. To address these formidable challenges, we propose a robust dynamic KSL recognition system that combines a skeleton-based Graph Convolution network with an attention-based neural network, effectively bridging the gap. Our solution employs a two-stream deep learning network to navigate the intricacies of dynamic signs, enhancing accuracy by effectively handling non-connected joint skeleton features. In this system, the first stream meticulously processes 47 pose landmarks using the Graph Convolutional Network (GCN) to extract graph-based features. These features are meticulously refined through a channel attention module and a general CNN, enhancing their temporal context. Concurrently, the second stream focuses on joint motion-based features, employing a similar approach. Subsequently, these distinct features from both streams are harmoniously integrated and channelled through a classification module to achieve precise sign-word recognition. A significant contribution of our work lies in creating a novel KSL video dataset, addressing the scarcity of data in this domain. This dataset comprises comprehensive information, including skeletal data from 47 joint skeleton points and details from both hands, body, and facial expressions. Our dataset aims to fill a critical gap in KSL research and provides a solid foundation for more extensive and inclusive studies in the field. Through this innovative approach, we aim to contribute significantly to the field of KSL recognition, filling the gaps in dynamic sign recognition and bolstering the accessibility of sign language communication within the Korean hearing impaired community and beyond. Our evaluations on a benchmark KSL-77 dataset and our proprietary lab dataset resulted in recognition accuracies of 99.87% and 100%, respectively. These results highlight the superiority of our model in the KSL recognition domain, outperforming existing models in terms of accuracy and computational efficiency.

Osteoporosis, a common skeletal disorder, necessitates the identification of its risk factors to develop effective preventive measures. It is crucial to identify the underlying risk factors and their relationships with the response class attribute. Different machine learning (ML) algorithms and feature selection approaches are used to estimate the risk of osteoporosis. However, ML-based algorithms may struggle to detect risk factors as well as grading of osteoporosis due to different measurement scale of data and their probability distributional assumptions. Violation of these assumptions and results interpretation may be improper in the presence of heteroscedasticity, or unequal variance in data. In this study, we seek to overcome distribution assumption constraints and improve the interpretability of our results by using rigorous statistical approaches, ensuring a robust and trustworthy study of osteoporosis risk variables. The study dataset consists of 40 clinical, lifestyle, and genetic attributes, allowing for a comprehensive analysis of potential risk factors associated with osteoporosis. In the analysis, after confirming the normality assumption using Kolmogorov-Smirnov and Shapiro-Wilk tests, independent t-test assess the factor ALT, FBG, HDL-C, LDL-C, FNT, TL, TLT, and URIC has a substantial impact on the risk of developing osteoporosis. The Mann-Whitney U test for the non-normal FN variable likewise showed a p-value of less than 0.05, indicating that this variable has a significant effect on the likelihood of developing osteoporosis. Based on the chi-square test p-values for the categorical factors, gender, calcium, calcitriol, bisphosphonate, calcitonin, COPD, CAD, and drinking have a severe significant risk of osteoporosis. For developing the predictive Gaussian Process (GPs) model, we proposed two customized integrated GP kernels into the analysis to enhance the modeling of complex relationships within the data. The proposed GP kernel model (modified kernel 2) outperforms the other individual kernels in this experiment and has the best accuracy score of 86.64% and AUC score of 86.63% on osteoporosis data. Moreover, a simulation study is also conducted to robustify the proposed model, the results are improved by different evaluation matrices ranging in accuracy from 0.60-11.41% and AUC from 0.50-11.60%.

The Metaverse, a transformative digital realm, holds immense promise for reshaping industries and human interactions while potentially addressing global challenges and democratizing opportunities. However, it also introduces a spectrum of complexities that demand careful navigation. To establish trustworthiness within the Metaverse ecosystem, gaining a deep understanding of its applications, challenges, and existing solutions is imperative. In this comprehensive survey, we first delve into Metaverse applications, drawing insights from existing literature. Subsequently, we explore the diverse challenges the Metaverse presents, analyzing them through the lens of existing research. We then scrutinize the overall trustworthiness of the Metaverse environment and investigate existing solutions to previously identified challenges through a thorough review and analysis of pertinent literature. Lastly, we discussed future research directions aimed at fostering a trustworthy Metaverse environment. This comprehensive review can provide an overview of the Metaverse, its application domains, challenges, existing solutions and research directions for many multidisciplinary studies.

The reliability of Underwater Wireless Sensor Networks (UWSNs) is measured in terms of energy consumption (EC), end-to-end delay(E2E), and packet delivery ratio (PDR). The adverse effects of a channel may cause data loss. Reducing delay up to the possible extent improves the reliability of the network, also increasing the number of nodes in a particular network increases reliability. Besides, increasing the number of nodes improves reliability but also increases power consumption. In order to overcome these shortcomings, the two routing protocols are proposed in this paper, namely the Delay and Reliability Aware Routing (DRAR) protocol and the Cooperative Delay and Reliability Aware Routing (Co-DRAR) protocol for UWSNs. In the DRAR protocol, the network is divided into two equal regions where two sink nodes(SNs) are positioned at the upper region of the network and two SNs are placed at the mid-region of the network. The protocol chooses the relay node based on residual energy (RE), distance, and Bit Error Rate (BER). These parameters protect the data packets from corruption and also provide a stable path (where nodes remain active for longer periods and do not die quickly). The protocol uses a single link and may get worse sometimes while changing channel circumstances. To address this problem, a cooperative routing scheme is added to the DRAR protocol in order to develop its enhanced version known as the Co-DRAR protocol. The protocol works by allowing the destination to receive multiple copies of data packets in order to decide the quality of packets. The proposed protocols DRAR and Co-DRAR perform routing irrespective of the geographical position of sensor nodes conversely to some conventional routing protocols. This is why our proposed protocols perform better than the well-known protocol i.e. Depth base routing (DBR) in terms of EC, E2E, PDR, dead nodes, packet drop ratio, and number of alive nodes.

Train incidents with animals and even humans have gotten more attention in the past. Every year, thousands of animals are killed on train tracks, causing an imbalance in the ecosystem and significant delays in railway traffic. Similarly, sometimes train accident is prevalent at the rail gates, where motor vehicles are crossing the train line. All that happens due to the lack of detecting the objects correctly on the train line. Since the detection depends on the driver or human, there is a possibility of occasionally making an error in honking the horn at the right moment, leading to the accident. That’s why an automated solution for detecting objects on the train line and promptly notifying the driver is essential. In this research, we proposed a system that detects objects only on the train line. Sometimes, there is an object just beside the train line, which is at a safe distance. To eliminate this kind of wrong detection, we detected the train line first and then detected the objects on the detected train line. Here, we used the Mask R-CNN algorithm to detect the train line and things on the train line. A railway traffic dataset with an input size of $512\times512$ pixels was used to test the proposed methodology, which came up with results with a mean average precision of 0.9375 and a frame rate of 30 frames per second. According to the findings of the experiments, the suggested approach can apply to identify objects on railroads in the real world.

Attention deficit hyperactivity disorder (ADHD) for children is one of the most common neurodevelopmental disorders and its prevalence has increased globally. Children with ADHD are faced with various difficulties, including inattention, impulsivity, and hyperactivity. Therefore, it is important to use an early detection system that is simple, non-invasive, and automated. Children with ADHD also suffer from other coexisting one or more disorders, including major depressive disorder (MDD), autism spectrum disorder (ASD), etc and it creates more challenges to detect ADHD children. Very few researchers considered such kinds of these comorbidities in their studies to detect ADHD for children. In this work, we proposed a deep learning (DL)-based algorithm to identify ADHD children with coexisting ASD. Functional near-infrared spectroscopy (fNIRs) signals from thirteen ADHD children who have coexisting ASD and fifteen typically developing (TD) children were recorded during the drawing of handwriting patterns. We asked each child to draw periodic lines (PL) and zigzag lines (ZL) under the predict and trace condition and repeated them three times. Finally, a hybrid approach was designed by combining convolutional neural networks (CNN) and bidirectional long short-time memory (Bi-LSTM) to determine children with ADHD who have ASD. The experimental results showed that our proposed hybrid approach could determine ADHD children with coexisting ASD with a classification accuracy of 94.0%, a sensitivity of 89.7%, specificity of 97.8%, f1-score of 93.3%, and AUC of 0.938, respectively, for the PL predict task.

Falls represent a significant public health concern, particularly concerning vulnerable populations such as older adults. Accurate detection and classification of falls are critical for timely interventions that can prevent injuries and enhance the quality of life of these individuals. This work proposes a class ensemble approach based on convolutional neural networks and long short-term memory networks for three-class classifications of falling processes (non-fall, pre-fall, and fall) using accelerometer and gyroscope data. The research is conducted on the SisFall and UMAFall datasets, the publicly available dataset of annotated video recordings of falls and non-falls. This approach leverages convolutional neural networks for robust feature extraction from the accelerometer and gyroscope data. In addition, long short-term memory networks model the falling process’s temporal dynamics. The proposed approach has demonstrated state-of-the-art performance in detecting falls, with accuracy rates of 96.45% and 96.12% and precision scores of 98.12% and 97.45% in identifying pre-fall and fall states, respectively.

Stroke is a dangerous medical disorder that occurs when blood flow to the brain is disrupted, resulting in neurological impairment. It is a big worldwide threat with serious health and economic implications. To solve this, researchers are developing automated stroke prediction algorithms, which would allow for early intervention and perhaps save lives. The number of people at risk for stroke is growing as the population ages, making precise and effective prediction systems increasingly critical. wo In a comparison examination with six well-known classifiers, the effectiveness of the proposed ML technique was explored in terms of metrics relating to both generalization capability and prediction accuracy. To give insight into the black-box machine learning models, we also studied two kinds of explainable techniques, namely SHAP and LIME, in this study. SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) are well-established and reliable approaches for explaining model decision-making, particularly in the medical industry. The findings of the experiment revealed that more complicated models outperformed simpler ones, with the top model obtaining almost 91% accuracy and the other models achieving 83-91% accuracy. The proposed framework, which includes global and local explainable methodologies, can aid in standardizing complicated models and gaining insight into their decision-making, which can enhance stroke care and treatment.

Automatic recognition of human emotion has become an interesting topic among brain-computer interface (BCI) researchers. Emotion is one of the most fundamental features of a human subject. With proper analysis of emotion, the inner state of a human subject can be assessed directly. The human brain response can be competently represented by electroencephalography (EEG). The selection of potential features in EEG related to human emotion is a very important task for developing an effective emotion recognition system. In this paper, the discriminative features computed from rhythmic components of EEG are used to recognize human emotional states. The narrowband rhythmic components theta, alpha, beta, and gamma are extracted from multichannel EEG signals using filter bank implementation. The short-time entropy and energy features are extracted from each of the rhythmic components. The spatial filtering has been performed on the entropy-energy space by using common spatial pattern (CSP). Thus obtained spatial features are employed to recognize the emotion states using support vector machine (SVM) classifier. The publicly available two datasets DEAP and SEED are used to evaluate the performance of the proposed method. The experimental results reflect that higher recognition accuracy is obtained by using higher frequency subbands (beta and gamma) than that of the lower frequency subbands (theta and alpha). The combination of features from all subbands has better performance than the features obtained from individual subband signals. The performance of the proposed method outperforms the recently developed algorithms of emotion recognition.

Skin cancer is a prevalent form of malignancy globally, and its early and accurate diagnosis is critical for patient survival. Clinical evaluation of skin lesions is essential, but it faces challenges such as long waiting times and subjective interpretations. Deep learning techniques have been developed to tackle these challenges and assist dermatologists in making more accurate diagnoses. Prompt treatment of skin cancer is vital to prevent its progression and potentially life-threatening consequences. The use of deep learning algorithms can improve the speed and accuracy of diagnosis, leading to earlier detection and treatment. Additionally, it can reduce the workload for healthcare professionals, allowing them to concentrate on more complex cases. The goal of this study was to develop reliable deep learning (DL) prediction models for skin cancer classification; (i) deal with a typical severe class imbalance problem, which arises because the skin-affected patients’ class is significantly smaller than the healthy class; and (ii) interpret the model output to better understand the decision-making mechanism (iii) Propose an End-to-End smart healthcare system through an android application. In a comparison examination with six well-known classifiers, the effectiveness of the proposed DL technique was explored in terms of metrics relating to both generalization capability and classification accuracy. A study used the HAM10000 dataset and an optimized CNN to identify the seven forms of skin cancer. The model was trained using two optimization functions (Adam and RMSprop) and three activation functions (Relu, Swish, and Tanh). Furthermore, an XAI-based skin lesion classification system was developed, incorporating Grad-CAM and Grad-CAM++ to explain the model’s decisions. This system can help doctors make informed skin cancer diagnoses in their early stages, with an 82% classification accuracy and 0.47% loss accuracy.

Plant disease is a significant health concern among all living creatures. Early diagnosis can help farmers take necessary steps to cure the disease and accelerate the production rate efficiently. Our research has been conducted with five most common rice leaf diseases, such as bacterial leaf blight, brown spot, leaf blast,leaf scald, and narrow brown spot, including healthy class, and two categories of betel leaf, such as healthy and unhealthy class. A robust new deep ensemble model, based on InceptionResNetV2, EfficientNetV2L,and Xception, has been proposed, known as PlantDet, in this research. PlantDet solves not only underfitting problems but also leverage nourished performances simultaneously for scarce dataset of the sparse number of different background image dataset. PlantDet integrates efficient data augmentation, preprocessing, Global Average Pooling layer, Dropout mechanism, L2 regularizers, PReLU activation function, Batch Normalization layers, and more Dense layers that make the model more robust compared to all existing models and help to handle underfitting and overfitting problems while maintaining high performance. PlantDet exceeds the previous state-of-art model for the Rice Leaf dataset with an Accuracy of 98.53%, a Precision of 98.50%, a Recall of 98.35%, a F1 of 98.42% and a Specificity of 99.71%. In addition, for the Betel Leaf dataset, PlantDet also surpassed all existing base models, including several robust ensemble models. Finally, Grad-CAM and Score-CAM have been accomplished with the Xception method to explain the model performances particularly to elaborate how the Deep Learning (DL) models works for this complex dataset. Score-CAM slightly outperformed Grad-CAM++ in terms of localizing the predicted area.

Attention deficit hyperactivity disorder (ADHD) is one of the major psychiatric and neurodevelopment disorders worldwide. Electroencephalography (EEG) signal-based approach is very important for the early detection and classification of children with ADHD. However, diagnosing children with ADHD using full EEG channels with all features may lead to computational complexity and overfitting problems. To solve these problems, machine learning (ML)-based ADHD detection was designed by identifying optimal channels and its significant features. In this work, support vector machine and t-test based, two separate approaches were devised to select optimal channels individually and then proposed a hybrid channel selection approach by combining these two channel selection methods in order to select the optimal channels. After that, LASSO logistic regression-based model was used to select the important features from the selected channels. Finally, six ML-based classifiers, like Gaussian process classification (GPC), random forest, k-nearest neighbors, multilayer perceptron, decision tree, and logistic regression were applied for the detection of children with ADHD and evaluated their performances using accuracy and area under the curve (AUC). This study utilized a total of one hundred twenty-one children, with sixty-one children with ADHD, aged 7-12 years, and had nineteen channels. Ten different channels were selected by SVM based and an independent t-test-based approach separately and six overlapping channels were identified from both channel selection methods. Then, we selected twenty-eight features from selected six channels using LASSO. Using only six channels and twenty-eight features, GP-based classifier achieved an accuracy rate of 97.53% and AUC of 0.999. This is an improvement of 3% over previously developed techniques published in the literature. This study illustrated that LASSO with GP-based system performed outstanding performance in distinguishing children with ADHD from healthy children. This proposed system will be helpful to doctors and physicians in order to detect children with ADHD at an early stage and take the necessary steps for the patients to access appropriate healthcare services, receive effective treatment, and be more conscious of maintaining their lives.

The dynamic hand skeleton data have become increasingly attractive to widely studied for the recognition of hand gestures that contain 3D coordinates of hand joints. Many researchers have been working to develop skeleton-based hand gesture recognition systems using various discriminative spatial-temporal attention features by calculating the dependencies between the joints. However, these methods may face difficulties in achieving high performance and generalizability due to their inefficient features. To overcome these challenges, we proposed a Multi-branch attention-based graph and a general deep-learning model to recognize hand gestures by extracting all possible types of skeleton-based features. We used two graph-based neural network channels in our multi-branch architectures and one general neural network channel. In graph-based neural network channels, one channel first uses the spatial attention module and then the temporal attention module to produce the spatial-temporal features. In contrast, we produced temporal-spatial features in the second channel using the reverse sequence of the first branch. The last branch extracts general deep learning-based features using a general deep neural network module. The final feature vector was constructed by concatenating the spatial-temporal, temporal-spatial, and general features and feeding them into the fully connected layer. We included position embedding and mask operation for both spatial and temporal attention modules to track the node’s sequence and reduce the system’s computational complexity. Our model achieved 94.12%, 92.00%, and 97.01% accuracy after evaluation with MSRA, DHG, and SHREC’17 benchmark datasets, respectively. The high-performance accuracy and low computational cost proved that the proposed method outperformed the existing state-of-the-art methods.

Japanese Sign Language (JSL) is vital for communication in Japan’s deaf and hard-of-hearing community. But probably because of the large number of patterns, 46 types, there is a mixture of static and dynamic, and the dynamic ones have been excluded in most studies. Few researchers have been working to develop a dynamic JSL alphabet, and their performance accuracy is unsatisfactory. We proposed a dynamic JSL recognition system using effective feature extraction and feature selection approaches to overcome the challenges. In the procedure, we follow the hand pose estimation, effective feature extraction, and machine learning techniques. We collected a video dataset capturing JSL gestures through standard RGB cameras and employed MediaPipe for hand pose estimation. Four types of features were proposed. The significance of these features is that the same feature generation method can be used regardless of the number of frames or whether the features are dynamic or static. We employed a Random forest (RF) based feature selection approach to select the potential feature. Finally, we fed the reduced features into the kernels-based Support Vector Machine (SVM) algorithm classification. Evaluations conducted on our proprietary newly created dynamic Japanese sign language alphabet dataset and LSA64 dynamic dataset yielded recognition accuracies of 97.20% and 98.40%, respectively. This innovative approach not only addresses the complexities of JSL but also holds the potential to bridge communication gaps, offering effective communication for the deaf and hard-of-hearing, and has broader implications for sign language recognition systems globally.

Dynamic human activity recognition (HAR) is a domain of study that is currently receiving considerable attention within the fields of computer vision and pattern recognition. The growing need for artificial-intelligence (AI)-driven systems to evaluate human behaviour and bolster security underscores the timeliness of this research. Despite the strides made by numerous researchers in developing dynamic HAR frameworks utilizing diverse pre-trained architectures for feature extraction and classification, persisting challenges include suboptimal performance accuracy and the computational intricacies inherent in existing systems. These challenges arise due to the vast video-based datasets and the inherent similarity in the data. To address these challenges, we propose an innovative, dynamic HAR technique employing a deep-learning-based, deep bidirectional long short-term memory (Deep BiLSTM) model facilitated by a pre-trained transfer-learning-based feature-extraction approach. Our approach begins with the utilization of Convolutional Neural Network (CNN) models, specifically MobileNetV2, for extracting deep-level features from video frames. Subsequently, these features are fed into an optimized deep bidirectional long short-term memory (Deep BiLSTM) network to discern dependencies and process data, enabling optimal predictions. During the testing phase, an iterative fine-tuning procedure is introduced to update the high parameters of the trained model, ensuring adaptability to varying scenarios. The proposed model’s efficacy was rigorously evaluated using three benchmark datasets, namely UCF11, UCF Sport, and JHMDB, achieving notable accuracies of 99.20%, 93.3%, and 76.30%, respectively. This high-performance accuracy substantiates the superiority of our proposed model, signaling a promising advancement in the domain of activity recognition.

Abstract Immunoglobulin-A-nephropathy (IgAN) is a kidney disease caused by the accumulation of IgAN deposits in the kidneys, which causes inflammation and damage to the kidney tissues. Various bioinformatics analysis-based approaches are widely used to predict novel candidate genes and pathways associated with IgAN. However, there is still some scope to clearly explore the molecular mechanisms and causes of IgAN development and progression. Therefore, the present study aimed to identify key candidate genes for IgAN using machine learning (ML) and statistics-based bioinformatics models. First, differentially expressed genes (DEGs) were identified using limma, and then enrichment analysis was performed on DEGs using DAVID. Protein-protein interaction (PPI) was constructed using STRING and Cytoscape was used to determine hub genes based on connectivity and hub modules based on MCODE scores and their associated genes from DEGs. Furthermore, ML-based algorithms, namely support vector machine (SVM), least absolute shrinkage and selection operator (LASSO), and partial least square discriminant analysis (PLS-DA) were applied to identify the discriminative genes of IgAN from DEGs. Finally, the key candidate genes (FOS, JUN, EGR1, FOSB, and DUSP1) were identified as overlapping genes among the selected hub genes, hub module genes, and discriminative genes from SVM, LASSO, and PLS-DA, respectively which can be used for the diagnosis and treatment of IgAN.

Gesture-based human-robot interaction has been an important area of research in recent years. The primary aspect for the researchers has always been to create a gesture detection system that is insensitive to lighting and backdrop surroundings. This research proposes a 3D gesture recognition and adaption system based on Kinect for human-robot interaction. The framework comprises the following four modules: pointing gesture recognition, 3D dynamic gesture recognition, gesture adaptation, and robot navigation. The proposed dynamic gesture recognition module employs three distinct classifiers: HMM, Multiclass SVM, and CNN. The adaptation module can adapt to new and unrecognized gestures applying semi-supervised self-adaptation or user consent-based adaptation. A graphical user interface (GUI) is built for training and testing the proposed system on the fly. A simple simulator along with two different robot-navigation algorithms are developed to test robot navigation based on the recognized gestures. The framework is trained and tested through a five-fold cross-validation method with a total of 3,600 gesture instances of ten predefined gestures performed by 24 persons (three age categories: Young, Middle-aged, Adult; each with 1,200 gestures). The proposed system achieves a maximum accuracy score of 95.67% with HMM for the Middle-aged category, 92.59% with SVM for the Middle-aged category, and 89.58% with CNN for the Young category in dynamic gesture recognition. Considering all the three age categories, the system achieves average accuracies of 94.61%, 91.95%, and 88.97% in recognizing dynamic gestures with HMM, SVM, and CNN respectively. Moreover, the system recognizes pointing gestures in real-time.

Human activity recognition is one of the most difficult tasks in computer vision. Due to the lack of time information, detecting human activities from still photos is more difficult than sensor-based or video-based techniques. Recently, various deep learning based solutions are being proposed one after another, and their performance is constantly improving. In this paper, we proposed a convolutional neural architecture by ensembling transfer learning based multi-channel attention networks. Here, four CNN branches were used to make feature fusion based ensembling and in each branch, an attention module was used to extract the contextual information from the feature map produced by existing pre-trained models. Finally, the extracted feature maps from four branches were concatenated and fed to fully connected network to produce the final recognition output. We considered 3 different datasets, Stanford 40 actions, BU-101 and Willow human actions datasets to evaluate our system. Experimental analysis showed that the proposed ensembled convolutional architecture outperformed previous works by a noteworthy margin.

Music genre classification (MGC) is the process of putting genre labels on music by analyzing the sounds or words. With the rapid growth of music data repositories, MGC can be used in a lot of ways to organize and manage music recommendation systems, advertising, and streaming services. But there have been a lot of works on classifying English music using different statistical and machine learning methods, but there hasn’t been much progress in classifying Bengali music. Also, Deep Learning (DL) methods have been used in a few important ways to classify different types of music. The content and uniqueness of Bengali music make it much more interesting. Also, there is still a lot to learn about how to use the DL approach in Bengali music. So, Bengali music genre classification is a pretty new area of research in the field of Deep Learning. In this paper, we developed a unique technique called BMNet-5 to perform a multiclass classification of Bangla music genres such as “Bangla Adhunik,” “Bangla Hip-Hop,” “Bangla Band Music,” “Nazrulgeeti,” “Palligeeti,” and “Rabindra Sangeet.” We show the effectiveness of the suggested technique by extracting features from a dataset of 1742 Bangla music pieces and evaluating the automated classification judgments. The proposed BMNet-5 is based on a neural network designed to predict music genre from audio inputs. Our suggested model outperformed the corresponding previous research with an accuracy of 90.32%. The BMNet-5 model is then tested for performance consistency using K-fold cross validation with varying k values. Finally, we use the suggested model to train the interpretable SHAP model for all the genre of the Bangla music dataset, and the development of an explainable outcome may have a significant advantage.

Gait recognition with wearable sensors is an effective approach to identifying people by recognizing their distinctive walking patterns. Deep learning-based networks have recently emerged as a promising technique in gait recognition, yielding better performance than template matching and traditional machine learning methods. However, most recent studies have focused on improving gait detection accuracy while neglecting model complexity in the deep learning domain, making them unsuitable for low-power wearable devices. Therefore, inference from these models results in latency due to calculation overhead. This study proposes an efficient network suitable for wearable devices without sacrificing prediction performance. We have modified the residual block and accumulated it in shallow convolutional neural networks with five weighted layers only for gait recognition and proved the efficacy of all the architectural components with extensive experiments over publicly available IMU-based datasets: whuGait and OU-ISIR. Our proposed model outperforms all the state-of-the-art methods regarding recognition accuracy and is more than 85 percent efficient on average in terms of model parameters and memory consumption.

The rapid development of the petrochemical industry has caused great harm to the environment. Fine suspended particles with a diameter of less than 2.5 microns cause serious health problems when inhaled air of high concentration of PM2.5. Therefore, an estimate of the concentration of PM2.5 is sought. However, it generally requires expensive instruments installed in an air quality monitoring station and a professional to operate them. In addition to the expensive cost, the instruments require a high maintenance fee and are restricted by geographical location. To eliminate the difficulty, this paper presents a low-cost and effective approach to estimate the concentration of PM2.5 using image processing schemes. The proposed approach consists of four stages. First, images with different concentrations of PM2.5 were taken and the related relative humidity (RH) was collected. Second, an automatically selected region of interest (RoI) was used to extract two features from an image, namely high-frequency information and transmittance by an improved dark channel. Third, the two extracted features, together with the RH measurement, were used to build a support vector regression (SVR) model. Fourth, the SVR model was applied to estimate the concentration of PM2.5, whose performance was then evaluated and compared with four simple regression models and a modified reported SVR method. In the given data set, the proposed method outperforms the comparison methods in terms of $R^{2}$ and root mean squared error. The best performance of our method reaches $R^{2}=0.816$ which is generally satisfactory in related applications.