Search

Citation counts are commonly used to evaluate the scientific impact of a publication on the general premise that more citations probably mean more endorsements. However, two questionable assumptions underpin this idea: a) that all authors contributed equally to the paper; and b) that the endorsement is positive. Obviously, neither of these assumptions hold true. Hence, with this study, we examine two components of citations—their purpose, i.e., the reason for the citation, and polarity, being the author's attitude toward the cited work. Our findings provide a new perspective on the scientific impact of highly-cited publications. Our methodology consists of three steps. Firstly, a pre-trained model composed of a Word2Vec—a well-known word embedding approach—and a convolutional neural network (CNN) is used to identify citation polarity and purpose. Secondly, in a set of highly-cited papers, we compare eight categories of purpose from foundational to critical and three categories of polarity: positive, negative, and neutral. We further explore how different types of papers—those discussing discoveries or those discussing utilitarian topics—influence the evaluation of scientific impact of papers. Finally, we mine and discover the knowledge (e.g. method, concept, tool or data) to explain the actual scientific impact of a highly-cited paper. To demonstrate how combining citation polarity with purpose can provide far greater details of a paper's scientific impact, we undertake a case study with 370 highly-cited journal articles spanning "Biochemistry & Molecular Biology" and "Genetics & Heredity". The results yield valuable insights into the assumption about citation counts as a metric for evaluating scientific impact. [ABSTRACT FROM AUTHOR]

With the establishment of the "double carbon" goal, various industries are actively exploring ways to reduce carbon emissions. Cloud data centers, represented by cloud computing, often have the problem of mismatch between load requests and resource supply, resulting in excessive carbon emissions. Based on this, this paper proposes a complete method for cloud computing carbon emission prediction. Firstly, the convolutional neural network and bidirectional long-term and short-term memory neural network (CNN-BiLSTM) combined model are used to predict the cloud computing load. The real-time prediction power is obtained by real-time prediction load of cloud computing, and then the carbon emission prediction is obtained by power calculation. Develop a dynamic server carbon emission prediction model, so that the server carbon emission can change with the change of CPU utilization, so as to achieve the purpose of low carbon emission reduction. In this paper, Google cluster data is used to predict the load. The experimental results show that the CNN-BiLSTM combined model has good prediction effect. Compared with the multi-layer feed forward neural network model (BP), long short-term memory network model (LSTM), bidirectional long short-term memory network model (BiLSTM), modal decomposition and convolution long time series neural network model (CEEMDAN-ConvLSTM), the MSE index decreased by 52 % , 50 % , 34 % and 45 % respectively. [ABSTRACT FROM AUTHOR]

Facial features extracted from deep convolutional networks are susceptible to background, individual identity and other factors. It interferes with facial expression recognition when mixed with useless features. Considering that different scale features have rich semantic and texture information respectively, this paper takes VGG-16 as the basic network structure and combines multiscale features to obtain richer feature information. In addition, the input feature map elements are enhanced or suppressed by the attention module in order to extract salient features more accurately. The proposed method was validated on two commonly used expression data sets CK+ and RAF-DB, and the recognition rates were 98.77 and 82.83%, respectively. Experimental results show the superiority of this method. [ABSTRACT FROM AUTHOR]

Comic inpainting is essential for language localization and adapting comics into animations, involving the removal of speech boxes, textboxes, and sound effects while maintaining artistic integrity. This paper introduces an automated approach leveraging the Manga inpainting method by Xie et al., adapted for western comics like Tex Willer. Tex Willer comics, known for their fusion of Italian art and American Western imagery, pose unique challenges due to their distinct styles. Customized Manga Restoration and Structural Line Extraction codes are employed, along with automated mask generation using deep CNN-based speech balloon detection and segmentation. Our contributions include the development of a precise model tailored to Tex Willer Comics, resulting in visually coherent inpainting results. The paper follows a structured format, covering State-of-the-Art, methodology, results, and conclusions, aiming to streamline the process of western comic inpainting for professionals in the comics and animation industries. [ABSTRACT FROM AUTHOR]

With the increasing complexity and cost of industrial control systems and the rapid development of information technology, the tolerance of industrial control equipment to performance degradation, productivity decline, and hidden safety hazards is getting lower and lower. Immediate detection of failures of industrial control equipment is of great significance for the safety of industrial control systems and reducing maintenance costs. Faced with these challenges, the traditional fault diagnosis technology based on expert knowledge has been insufficient to meet the requirements of accuracy and real-time fault diagnosis of industrial control systems due to its high cost and low efficiency. Aiming at the problem that the fault diagnosis method in the current industrial control environment is not systematic, this paper proposes a Safety Fault diagnosis system for industrial intelligent control equipment based on DevOps concept and deep CNN. Based on the in-depth analysis of the principles of CNN, this paper improves the model's robustness from the perspective of data set enhancement, fault diagnosis performance, and noise immunity analysis. Finally, it is proved through experiments that the fault diagnosis model can effectively deal with the lack of fault samples by adopting the fault data set enhancement method based on periodic overlapping sampling. Our fault diagnosis system based on DevOps and CNN proposed in this paper has high scalability and can accurately predict fault input. [ABSTRACT FROM AUTHOR]

RNA (Ribonucleic Aci) plays a crucial role in a variety of physiological and pathological processes. Over the past few decades, researchers have designed many types of RNAs and their derivatives and applied them to different fields of biology. Studies have confirmed that RNA molecules play an important role in tumor growth, metastasis, metabolism and immune escape. Based on the research of tumor-associated RNA and the application of related technologies, RNA has been designed and applied in tumor immunotherapy. Machine learning (ML) algorithms can be used to solve tumor type classification predictions with the large-scale data, some with high prediction accuracy and others with limited accuracy. For the early detection of tumors, this paper proposes to use machine learning and RNA technology to establish a tumor type prediction model to enhance the effect of anti-tumor immunotherapy. The model first preprocesses the data. Then, this paper uses an improved principal component analysis algorithm to perform feature dimensionality reduction and data fusion on the RNA expression and DNA methylation data of patients with 32 tumor types. Finally, the data is fed into a tumor type classifier based on the CNN-LSTM model for training and prediction. The experimental results show that the method proposed in this paper has excellent performance in tumor classification, and has important guiding significance for the early diagnosis and treatment of tumor patients, which further indicates that the intelligent ML method can be used for data analysis in the field of tumor type prediction. [ABSTRACT FROM AUTHOR]

In modern deep learning, character recognition in images is a very important field of study due to its has many real life applications. The goal of this paper is to create the state-of-the-art character recognition model using a stacking ensemble of convolution neural networks (CNNs).To develop the proposed ensemble model, we evaluated several CNN models. The models were judged on how well they performed on the Chars74k dataset. The dataset contains 74,103 images divided into 62 classes with labels [A-Z], [a-z], and [0-9]. The accuracy distribution based on the dataset's subgroups (uppercase, lowercase, and digit) is shown in results. The proposed ensemble model achieves state-of-the-art performance with a maximum accuracy of 92.31% on complete dataset, 99.22% on Uppercase alphabets, 98.66% on Lowercase alphabets, 99.77% on Digits, 91.97% on Uppercase+Lowercase alphabets. On the complete and partial datasets, a comparison report between the proposed model and other existing approaches is also displayed. A comparative study of the proposed work and the previous methods is also shown in this paper, in order to demonstrate the effectiveness of the proposed work. [ABSTRACT FROM AUTHOR]

Separation of speech and music plays a vital role in multiple fields related to audio and speech processing. The spectrograms of speech and music show distinct patterns. This serves as the motivation for the differentiation of speech and music signals in an audio segment. The patterns have been further emphasized using Sobel edge kernels, Mel-spectrograms. For the inception of this paper, we have made a dataset from "All India Radio" news archives which is having separate and overlapped speech and music data in different languages. The different input features are extracted from these audio segments and further emphasized before feeding them to the different classifiers for distinguishing speech and music frames. We also compared the different classification algorithms for their varied performance in terms of accuracy. We have found that the convolutional neural network based approach on Mel-spectrograms and MFCC-delta-RNN methods have given a significantly better result compared to other approaches. Further, we wanted to see how these approaches work in the audio data of different languages, hence, we have applied the proposed method in three different languages such as Bengali, Punjabi, and Tamil. We have seen that the performance of the proposed method in all languages is consistent. The paper has also attempted to solve the problem of classifying audio segments with overlapped speech and music regions and achieved a good level of accuracy. [ABSTRACT FROM AUTHOR]

Since the dawn of civilization, handwriting has been one of the most important forms of communication. However, as handwriting differs from person to person, writer identification has become a promising application of pattern recognition to identify the actual writer of a handwritten document. Handwriting can be either online or offline, depending on how it was obtained. Users can write directly on tablets, smartphones, touch screens, PDAs, and other devices using input devices with online handwriting, whereas offline handwriting is done with a pen and paper. With the advent of artificial intelligence, and most importantly deep learning techniques, the development of writer identification systems based on offline handwritten documents has gained a lot of attention. Deep learning models have the capability of automatic feature extraction, which results in increased performance. From the literature survey, it was revealed that least attention has been paid towards the development of deep learning-based writer identification systems for offline Urdu handwritten documents, unlike the English and Arabic scripts. Therefore, in this paper, we proposed an offline Urdu handwritten writer identification system using a deep learning model inspired by the VGG-16 model of CNN. The model was trained and tested on a novel Urdu handwritten dataset contributed by 318 distinct Urdu writers, resulting in an overall training accuracy of 98.71% and a testing accuracy of 99.11%. The results achieved showed that the proposed model outperformed the already existing writer identification techniques. [ABSTRACT FROM AUTHOR]

De-mining operations are of critical importance for humanitarian efforts and safety in conflict-affected regions. In this paper, we address the challenge of enhancing the accuracy and efficiency of mine detection systems. We present an innovative Deep Learning architecture tailored for pulse induction-based Metallic Mine Detectors (MMD), so called DL-MMD. Our methodology leverages deep neural networks to distinguish amongst nine distinct materials with an exceptional validation accuracy of 93.5%. This high level of precision enables us not only to differentiate between anti-personnel mines, without metal plates but also to detect minuscule 0.2-g vertical paper pins in both mineralized soil and non-mineralized environments. Moreover, through comparative analysis, we demonstrate a substantial 3% and 7% improvement (approx.) in accuracy performance compared to the traditional K-Nearest Neighbors and Support Vector Machine classifiers, respectively. The fusion of deep neural networks with the pulse induction-based MMD not only presents a cost-effective solution but also significantly expedites decision-making processes in de-mining operations, ultimately contributing to improved safety and effectiveness in these critical endeavors. [ABSTRACT FROM AUTHOR]

In the realm of prognosticating the remaining useful life (RUL) of pivotal components, such as aircraft engines, a prevalent challenge persists where the available historical life data often proves insufficient. This insufficiency engenders obstacles such as impediments in performance degradation feature extraction, inadequacies in capturing temporal relationships comprehensively, and diminished predictive accuracy. To address this issue, a 1D CNN-GRU prediction model for few-shot conditions is proposed in this paper. In pursuit of more comprehensive data feature extraction and enhanced RUL prognostication precision, the Convolutional Neural Network (CNN) is selected for its capacity to discern high-dimensional features amid the intricate dynamics of the data. Concurrently, the Gated Recurrent Unit (GRU) network is leveraged for its robust capability in extracting temporal features inherent within the data. We combine the two to construct a CNN-GRU hybrid network. Moreover, the integration of data distribution alongside correlation and monotonicity indices is employed to winnow the input of multi-sensor monitoring parameters into the CNN-GRU network. Finally, the engine RULs are predicted by the trained model. In this paper, experiments are conducted on a sub-dataset of the National Aeronautics and Space Administration (NASA) C-MAPSS multi-constraint dataset to validate the effectiveness of the method. Experimental results have demonstrated that this method has high accuracy in RUL prediction tasks, which can powerfully demonstrate its effectiveness. [ABSTRACT FROM AUTHOR]

Detecting arrhythmia through electrocardiogram (ECG) signals is a challenging task, and some methods based on deep neural networks have been proposed to assist in the diagnosis of arrhythmia. However, the performance of existing models in obtaining imbalanced key points of ECG data and integrating long-distance information needs to be improved. In this study, we propose an adaptive arrhythmia classification network (ADLNet) based on deformable convolution and bidirectional LSTM. This network uses the Bi-LSTM gating mechanism to obtain and control the cumulative information of the context, enhance the ability of the model to understand the context, and mine the long-term sequential information in the lead. Considering the imbalance of the key points of the focus of ECG and the gradient disappearance of the deep network, the residual deformable convolution module with deformable receptive field is introduced to better obtain the local features of ECG and enhance the ability of feature representation. In addition, the adaptive fusion module adaptively combines the features mined from the first two with different weight coefficients to improve the performance of the model. The proposed network was evaluated on the 12-lead ECG data set of the 2018 China Physiological Signal Challenge 2018, and the F1 of ADLNet reached 0.861, which is 3% higher than the baseline method. The experimental results show that the performance of the ADLNet proposed in this paper is better than that of the mainstream arrhythmia classification network. This study has demonstrated that the adaptive combination of deformable convolutional models and Bi-LSTM models can achieve excellent ECG classification performance without the assistance of other medical features. [ABSTRACT FROM AUTHOR]

The COVID 19 pandemic is highly contagious disease is wreaking havoc on people's health and well-being around the world. Radiological imaging with chest radiography is one among the key screening procedure. This disease contaminates the respiratory system and impacts the alveoli, which are small air sacs in the lungs. Several artificial intelligence (AI)-based method to detect COVID-19 have been introduced. The recognition of disease patients using features and variation in chest radiography images was demonstrated using this model. In proposed paper presents a model, a deep convolutional neural network (CNN) with ResNet50 configuration, that really is freely-available and accessible to the common people for detecting this infection from chest radiography scans. The introduced model is capable of recognizing coronavirus diseases from CT scan images that identifies the real time condition of covid-19 patients. Furthermore, the database is capable of tracking detected patients and maintaining their database for increasing accuracy of the training model. The proposed model gives approximately 97% accuracy in determining the above-mentioned results related to covid-19 disease by employing the combination of adopted-CNN and ResNet50 algorithms. [ABSTRACT FROM AUTHOR]

The convolutional neural network (CNN) has brought about a drastic change in the field of image processing and pattern recognition. The filters of CNN model correspond to the activation maps that extract features from the input images. Thus, the number of filters and filter size are of significant importance to learning and recognition accuracy of CNN model-based systems such as the biometric-based person authentication system. The present paper proposes to analyze the impact of varying the number of filters of CNN models on the accuracy of the biometric-based single classifiers using human face, fingerprint and iris for person identification, and also biometric-based super-classification using both bagging- and programming-based boosting methods. The present paper gives an insight to the optimal set of filters in CNN model that gives the maximum overall accuracy of the classifier system. [ABSTRACT FROM AUTHOR]

Some types of tumors in people with brain cancer grow so rapidly that their average size doubles in twenty-five days. Precisely determining the type of tumor enables physicians to conduct clinical planning and estimate dosage. However, accurate classification remains a challenging task due to the variable shape, size, and location of the tumors.The major objective of this paper is to detect and classify brain tumors. This paper introduces an effective Convolution Extreme Gradient Boosting model based on enhanced Salp Swarm Optimization (CEXGB-ESSO) for detecting brain tumors, and their types. Initially, the MRI image is fed to bilateral filtering for the purpose of noise removal. Then, the de-noised image is fed to the CEXGB model, where Extreme Gradient Boosting (EXGB) is used, replacing a fully connected layer of CNN to detect and classify brain tumors. It consists of numerous stacked convolutional neural networks (CNN) for efficient automatic learning of features, which avoids overfitting and time-consuming processes. Then, the tumor type is predicted using the EXGB in the last layer, where there is no need to bring the weight values from the fully connected layer. Enhanced Salp Swarm Optimization (ESSO) is utilized to find the optimal hyperparameters of EXGB, which enhance convergence speed and accuracy. Our proposed CEXGB-ESSO model gives high performance in terms of accuracy (99), sensitivity (97.52), precision (98.2), and specificity (97.7).Also, the convergence analysis reveals the efficient optimization process of ESSO, obtaining optimal hyperparameter values around iteration 25. Furthermore, the classification results showcase the CEXGB-ESSO model’s capability to accurately detect and classify brain tumors. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) is widely used in various fields and has a particularly promising application in health monitoring. Human behavior can be represented by various sensor data patterns, and millimeter-wave radar has been widely used in HAR systems in combination with deep learning (DL) algorithms, due to its advantages in accuracy and privacy protection. However, the performance of traditional DL models may degrade due to the complexity of everyday life scenarios. In this paper, we propose a multi-channel parallel CNN and Transformer network algorithm for millimeter-wave radar-based HAR. The proposed algorithm uses a CNN and Transformer for feature extraction of Time-range map and micro-Doppler map, respectively, where CNN extracts feature information from the map and the Transformer extracts temporal information. Finally, the features are fused in the fully connected layer and judged for output. We evaluated the proposed method using a public radar dataset containing six human activities. Experimental results show that the proposed RMPCT-Net achieved an average accuracy of 97.5%. The size of the entire model is 9.5 MB, and the proposed system has good real-time performance, making it suitable for deployment in home monitoring devices. [ABSTRACT FROM AUTHOR]

Medical imaging methods, such as CT scans, MRI scans, X-rays, and ultrasound imaging, are widely used for diagnosis in the healthcare domain. However, these methods are often affected by noise, which can lead to incorrect diagnoses. Radiologists used to rely on visual features observed through various imaging techniques to diagnose diseases in patients, but now, intelligent machines and artificial intelligence offer more accurate and early diagnoses. Over the past few decades, the classical problem of image denoising in computer vision has been extensively studied. This survey paper discusses the various techniques applicable which have tried to remove the noise from medical images. A complete overview of the problem hypothesis is stated in the paper, with an in-depth discussion on types and sources of noise and the evaluation metrics deployed, followed by the discussion and implementation of various filtering and image enhancement techniques. The section is succeeded by a comprehensive literature review conducted on leading and state-of-the-art methods in broadly four domains—frequency domain, filtering, CNN-based, Generative Adversarial Networks (GAN)-based and Transformer-based approaches. The conclusion summarises the findings and proposes the importance of image denoising, focusing on Explainable AI (XAI). [ABSTRACT FROM AUTHOR]

Deep learning models have recently attracted great interest as an effective solution to the challenging problem of human activity recognition (HAR) and its widespread applications in medical rehabilitation and human–computer interaction. However, it is difficult for many existing models to simultaneously extract global temporal features and local spatial features of activity data. This paper proposes a deep learning model, ConvTransformer, based on convolutional neural network (CNN), Transformer, and attention mechanism. The model first uses the convolutional layer to model the local information of the sensor time series signal, then uses Transformer to obtain the temporal correlation of the feature sequence, adds an attention mechanism to highlight essential features, and finally completes the activity recognition through the fully connected layer (FC). We evaluate the model on four public datasets, and the experimental results show that the model achieves 92%, 99%, 97%, and 86% accuracy on the OPPORTUNITY, PAMAP2, SKODA, and USC-HAD datasets, respectively. Compared with the baseline Transformer and existing state-of-the-art methods, the model has higher recognition accuracy and better robustness. In addition, we explore the strengths and limitations of the method proposed in this paper by tuning hyperparameters and optimization strategies. [ABSTRACT FROM AUTHOR]

Purpose: As a critical component of the wind turbine drive train, the bearings are easy to fail under the complex environment of variable working conditions and loads in long-term operation. So it is essential to carry out a study targeting at fault diagnosis on it to improve the safety and reliability of the whole wind turbine operating. Methods: This paper presents a kind of bearing fault diagnosis method for wind turbines based on convolutional neural network (CNN) and sparrow search algorithm (SSA) optimized extreme learning machine (ELM). First, the wavelet time-frequency diagram (WTD) is constructed by using the continuous wavelet transform (CWT) to the original vibrational signal of the wind turbine bearing. Then, the WTD is input into deep learning CNN for extracting features. Finally, the SSA-ELM classifier is constructed by searching the optimal parameters of ELM with SSA, and the extracted features are put into SSA-ELM to identify different fault types. Results: The proposed CWT-CNN-SSA- ELM method is experimentally validated by two bearing datasets and compared with other methods. The result shows that the method has better diagnosis capability. Conclusion: In this paper, a wind turbine bearing fault diagnosis method based on CNN and SSA-ELM is proposed. The approach is able to well extract fault features and classify and identify the bearing data under variable working conditions and time-varying speed with good generalization ability. [ABSTRACT FROM AUTHOR]

The increasingly comprehensive penetration of sports science and technology and the constant change in social culture make competitive sports fierce, and the athletes who participate in high-level sports are more and more equal. Under the current situation, it is difficult for athletes to win in competitions only by training their physical, technical, and tactical abilities under heavy load. Therefore, it is imperative to strengthen the psychological training of athletes in training and before a match. Artificial intelligence, especially machine learning, has recently been widely used in sports, which gives us the power to study athletes' psychological balance before training and competition. This paper uses the Long-Short Term Memory (LSTM) - Convolutional Neural Networks (CNN) model, which combines time-domain and space-domain characteristics to classify the collected electroencephalogram signals of athletes into emotional categories. LSTM network is used to process the information of emotions changing with time, while CNN can mine the correlation between channels and finally fuse the two network outputs to obtain the results of emotional recognition of athletes; hence coaches can put forward targeted intervention strategies to help athletes adjust their psychological state in time. The experimental results show that this method has high classification accuracy and short classification time on Valence and Arousal dimensions and is superior to similar emotional recognition methods based on machine learning and deep learning. It provides dynamic support for the research of psychological balance detection of athletes. [ABSTRACT FROM AUTHOR]

Motor imagery (MI) electroencephalography (EEG) signal classification plays an important role in brain–computer interface (BCI), which gives hope to amputees and disabled people.This paper proposes a number of convolutional neural networks (CNNs) models for EEG MI signal classification, and it also proposes a method for enhancing the classification accuracy by feeding the CNN model with different frequency bands filters, which shows a significant improvement. The proposed models are: (1) Basic model, which is the simplest model with only one layer and no convolution layers. (2) CNN1D where the convolution is carried out along time. (3) CNN2D, where time and sensor channels are presented as 2D signals and convolution is carried out along both of them. (4,5) CNN3D and TimeDist, where the EEG signals are represented as video signals. (6) CNN1D_MF, where MF stands for multiband frequency, which is a modified version of the CNN1D model with multiple frequecy bands input. Physionet (5 classes) and BCI Competition IV-2a (4 classes) datasets were used in the evaluation. The software used for this paper, namely Coleeg, works on Google TM Colaboratory and Python language. The validation accuracy, validation kappa, and time were measured and compared to three existing models, which are: EEGNet, ShallowConvNet, and DeepConvNet. The results show that CNN1D_MF model has the best accuracy results, with 58.0% and 69.2% for Physionet and BCI Competition IV-2a datasets, respectively. Coleeg code is available under an open-source license to be an initiative for collaboration in EEG signal classification. [ABSTRACT FROM AUTHOR]

Recently, CNN (Convolutional Neural Network) and Grad-CAM (Gradient-weighted Class Activation Map) are being applied to various kinds of defect detection and position recognition for industrial products. However, in training process of a CNN model, a large amount of image data are required to acquire a desired generalization ability. In addition, it is not easy for Grad-CAM to clearly identify the defect area which is predicted as the basis of a classification result. Moreover, when they are deployed in an actual production line, two calculation processes for CNN and Grad-CAM have to be sequentially called for defect detection and position recognition, so that the processing time is concerned. In this paper, the authors try to apply YOLOv2 (You Only Look Once) to defect detection and its visualization to process them at once. In general, a YOLOv2 model can be built with less training images; however, a complicated labeling process is required to prepare ground truth data for training. A data set for training a YOLOv2 model has to be composed of image files and the corresponding ground truth data file named gTruth. The gTruth file has names of all the image files and their labeled information, such as label names and box dimensions. Therefore, YOLOv2 requires complex data set augmentation for not only images but also gTruth data. Actually, target products dealt with in this paper are produced with various kinds and small quantity, and also the frequency of occurrence of the defect is infrequent. Moreover, due to the fixed indoor production line, the valid image augmentation to be applied is limited to the horizontal flip. In this paper, a data set augmentation method is proposed to efficiently generate training data for YOLOv2 even in such a production situation and to consequently enhance the performance of defect detection and its visualization. The effectiveness is shown through experiments. [ABSTRACT FROM AUTHOR]

The very low resolution (VLR) issue arises in many face application systems because of the rising demand for surveillance camera-based applications. On the VLR face image, the face recognition algorithms in use are unable to work satisfactorily. Although face super-resolution (SR) techniques can be used to improve the resolution of the images, these VLR face images do not respond well to the existing learning-based face SR techniques. This work suggests a novel method for learning the link between the high-resolution image space and the VLR image space for face SR to solve this issue. The paper aims to address the problem of face image super-resolution and solutions for resolving the issues related to facial images. The major issues related to face image super-resolution during reconstruction end is a fusion of more than one kind of feature which may cause noise in the image, blind spot generation, low perceptual quality, and checkboard issues. The existing models are designed to improvise the perceptual quality of face super-resolution but still fail to generate better perceptual quality of image due to loss during the reconstruction stage. Therefore, this paper proposes a novel progressive face hallucination super-resolution (FHSR) model with a loss-aware upscaling network layer. The upscaling layer is integrated with correlation filters and used a combined loss function. The model is designed as cascading of upscaling modules that are progressive with dense skip connection layers. Loss functions from each upscaling module are evaluated and cascaded together and fed forwards to the next layer to minimize the reconstruction losses. The architecture upscaled the images with 3 cascading modules which magnify 2x,4x, and 8x. The paper also presented the ablation study that assured that the designed FHSR model is better as compared to baseline models. This architecture overcomes the issue of extraction of more features with minimum losses and constructs the high magnification at the reconstruction end. The comparative study shows enhanced performance over state-of-art models. [ABSTRACT FROM AUTHOR]

Cancelable biometrics serves as an effective countermeasure against various template attacks launched by intruders, safeguarding the biometric system. This paper proposes a cancelable approach with a novel feature extraction technique for iris recognition, known as the hybrid architecture of the convolutional neural network (CNN) and GRU (gated recurrent unit). To provide cancelability to the system, the paper makes use of a random projection technique. The proposed method has the best outcome in terms of accurate identification. The method is validated on two Iris datasets IITD and MMU, which show promising results on the equal error rate (EER) and accuracy. The proposed model provides 0.02 and 0.045 EER for IITD and MMU, respectively, and accuracy 0.98 and 0.933%, for IITD and MMU Iris dataset, respectively, which is very high compared to other methodologies. The proposed hybrid architecture is being used for a cancelable biometric system for the first time based on literature review. The efficiency of the proposed method is high when validated on the datasets. [ABSTRACT FROM AUTHOR]

Defect detection in various industrial products ensures product quality and safety. This paper introduces an innovative design, training, and evaluation application employing CNN, CAE, YOLO, FCN, and SVM models, to facilitate defect detection without requiring extensive IT expertise. However, conventional usage of Grad-CAM for visualizing defect regions sometimes includes irrelevant areas unrelated to the target defects. A novel data augmentation technique called random masking is proposed to enhance the visualization of defective regions, leading to more accurate and focused defect detection in various industrial products. This technique is used during training, replacing non-target areas in each image with randomly generated mask patterns. The efficacy of the proposed technique is demonstrated through visualization tests of defective regions using Grad-CAM. Furthermore, an ablation study is conducted to assess the effectiveness of the data augmentation techniques, comparing the performance of Grad-CAM with and without random masking augmentation. We further provide insights into the dataset used and present noteworthy findings from the evaluation, showcasing the contributions of our work in advancing defect detection methodologies. [ABSTRACT FROM AUTHOR]

With the world's interaction moving more and more toward using online social media platforms, the advent of cyberbullying has also raised its head. Multiple forms of cyberbullying exist from the more common text based to images or even videos, and this paper will explore the context of textual comments. Even in the niche area of considering only text-based data, several approaches have already been worked upon such as n-grams, recurrent units, convolutional neural networks (CNNs), gated recurrent unit (GRU) and even a combination of the mentioned architectures. While all of these produce workable results, the main point of contention is that true contextual understanding is quite a complex concept. These methods fail due to two simple reasons: (i) lack of large datasets to properly utilize these architectures and (ii) the fact that understanding context requires some mechanism of remembering history that is only present in the recurrent units. This paper explores some of the recent approaches to the difficulties of contextual understanding and proposes an ALBERT-based fine-tuned model that achieves state-of-the-art results. ALBERT is a transformer-based architecture and thus even at its untrained form provides better contextual understanding than other recurrent units. This coupled with the fact that ALBERT is pre-trained on a large corpus allowing the flexibility to use a smaller dataset for fine-tuning as the pre-trained model already has deep understanding of the complexities of the human language. ALBERT showcases high scores in multiple benchmarks such as the GLUE and SQuAD showing that high levels of contextual understanding are inherently present and thus fine-tuning for the specific case of cyberbullying allows to use this to our advantage. With this approach, we have achieved an F1 score of 95% which surpasses current approaches such as the CNN + wordVec, CNN + GRU and BERT implementations. [ABSTRACT FROM AUTHOR]

The world population is growing very fast and the lifestyle of human beings is changing with time and place. So, there is a need for disease management which includes disease diagnosis, its detection and classification, cure and lastly for future disease prevention. The outermost protective layer of a human body is the skin. Skin not only impacts a person's health but also psychologically impacts one's life. Computer-aided systems are very helpful in skin disease detection and classification and their application is growing rapidly in healthcare. This literature review paper aims to help the researchers to get a synthesized and appropriate information for the same. We have included papers from 2021 to 2023 for the review from the Scopus database. 45 studies are selected for the review of which 32 studies use deep learning techniques, 11 use machine learning techniques and 2 studies use a hybrid approach. The studies are compared on various parameters like models, datasets, and performance metrics. The work also identified some of the challenges like dealing with noise and also explained disease symptoms. [ABSTRACT FROM AUTHOR]

Writing is a commonplace activity that individuals partake in regularly. However, the implications behind it are often overlooked. When we write, various psychological factors come into play as the pen creates letters on the paper. Handwriting analysis has long been a subject of study, attracting researchers from diverse disciplines such as graphology, psychology, paleography, neuroscience, criminology, and computer science. Among the promising applications of handwriting analysis is gender classification, where a system can predict the gender of a writer based on their handwriting style. Since each individual's handwriting is unique, and variations exist between the handwriting of different genders, an automatic gender classification system can exploit these differences to make predictions. This paper presents a deep-learning-based gender classification system specifically designed for Urdu handwriting. The proposed approach utilizes a CNN network trained and tested on a self-created dataset contributed by 200 distinct male and 200 female Urdu writers. Through this method, the gender classification system achieved an impressive overall accuracy of 99.63%. The results obtained demonstrate that our technique for Urdu handwriting-based writer identification surpasses existing approaches. In the future, we intend to explore transfer learning techniques to further advance this field. [ABSTRACT FROM AUTHOR]

Deep learning and the Internet of Things (IoT) are rapidly advancing technologies with wide-ranging applications, notably in areas such as face detection. There is a growing need for sophisticated touchless authentication systems within this domain. While numerous security methods are available, many of them suffer from various shortcomings, including issues like forgotten passwords and the potential transmission of diseases through touch-based authentication techniques. This paper proposes IoT-based intelligent control face detection systems that leverage deep learning (DL) models. These systems are designed to significantly enhance security without relying on manually crafted rules. Face detection plays a pivotal role in safeguarding society, identifying wrongdoers, and bolstering community safety. However, challenges persist, particularly in achieving high accuracy with face recognition control systems in uncontrolled environments and real-time scenarios, such as intersections. In response to these challenges, our paper introduces a novel model that harnesses a combination of deep convolutional neural networks (CNNs) with self-attention mechanisms. Our experimental results demonstrate that this model can rapidly detect whole-scale images in a single forward pass. Notably, our proposed method achieved an outstanding accuracy rate of 99.7%. In comparison to existing state-of-the-art methods, our approach exhibits superior efficiency. This work showcases how the integration of IoT and DL, particularly with the use of a CNN with self-attention, outperforms other CNN-based approaches in terms of both speed and accuracy. [ABSTRACT FROM AUTHOR]

Multiclass classification of brain tumors from magnetic resonance (MR) images is challenging due to high inter-class similarities. To this end, convolution neural networks (CNN) have been widely adopted in recent studies. However, conventional CNN architectures fail to capture the small lesion patterns of brain tumors. To tackle this issue, in this paper, we propose a global transformer network dubbed GT-Net for multiclass brain tumor classification. The GT-Net mainly comprises a global transformer module (GTM), which is introduced on the top of a backbone network. A generalized self-attention block (GSB) is proposed to capture the feature inter-dependencies not only across spatial dimension but also channel dimension, thereby facilitating the extraction of the detailed tumor lesion information while ignoring less important information. Further, multiple GSB heads are used in GTM to leverage global feature dependencies. We evaluate our GT-Net on a benchmark dataset by adopting several backbone networks, and the results demonstrate the effectiveness of GTM. Further, comparison with state-of-the-art methods validates the superiority of our model. [ABSTRACT FROM AUTHOR]

Breast cancer remains a substantial public health challenge, marked by a rising prevalence. Accurate early detection is paramount for effective treatment and improved patient outcomes in breast cancer. The diversity of breast tumors and the complexity of their microenvironment present significant challenges. Establishing a reliable breast calcification and micro-calcification detection approach is an ongoing issue that researchers must continue to investigate. The goal is to develop an effective methodology that contributes to increased patient survival. Therefore, this paper introduces a novel approach for classifying breast calcifications in mammography, aiming to distinguish between benign and malignant cases. Aiming to address these challenges, we proposed our hybrid approach for breast calcification classification in mammogram images. The proposed approach starts with an image pre-processing phase that includes noise reduction and enhancement filters. Afterward, we proposed our hybrid classification architecture. It includes two branches: First, the vision transformer (ViT++) branch for contextual features. Secondly, a CNN branch based on transfer learning techniques for visual features. Using the CBIS-DDSM dataset, the application of our proposed ViT++ architecture reached the maximum accuracy of 96.12%. Further, the application of the VGG16 as a single feature extractor had a much lower accuracy of 61.96%. Meanwhile, the combination of these techniques in the same architecture improved the accuracy to 99.22%. Three different pre-trained feature extractors were applied in the CNN branch: Xception, VGG16, and RegNetX002. However, the best-obtained outcomes were from the combination of the ViT++ and the VGG16. The experimental findings indicate that the proposed strategy for breast calcification detection has the potential to surpass the performance of currently top-ranked methods, particularly in terms of classification accuracy. [ABSTRACT FROM AUTHOR]

Recently, image dehazing methods based on deep learning have achieved good results, but most of them are aimed at synthetic hazy images that have corresponding haze-free images. However, in some practical applications, the acquired images always contain haze in hazy weather, but it is difficult to get the corresponding haze-free images. So the image dehazing for real hazy images is a very important and challenging problem due to the absence of the corresponding haze-free images. In this paper, we propose an unpaired image dehazing method to solve the problem of unpaired real hazy images, which includes two parts: one is the unpaired training module and the other one is the supervised training module. For the unpaired training module, we first use cycle generative adversarial network to generate the dehazed images; then we use perceptual loss to make the features of the dehazed images and the unpaired haze-free images more consistency to improve the image dehazing performance; Finally, we propose to introduce instinct properties of clear images to further improve the image dehazing performance. For the supervised training module, we design a pseudo labeling scheme to train the model by using the dehazed images generated from the unpaired training module as pseudo haze-free images, so that we can conduct supervised training to further improve the performance of haze removal. Besides, we dedicate to collect a real hazy images dataset from real high-speed railway scenarios, named HRHI (High-speed Railway Hazy Image) dataset, which contains real hazy images and unpaired haze-free images. In order to verify the generality of our method, we also collect a real sandstorm images dataset from real high-speed railway scenarios, named HRSI (High-speed Railway Sandstorm Image) dataset. Expensive experimental results on HRHI dataset and the generic hazy image dataset show the effectiveness of our method compared with other state-of-the-art image dehazing methods, and the experimental results on HRSI dataset prove the generality of our method. [ABSTRACT FROM AUTHOR]

The communication barriers experienced by deaf and hard-of-hearing individuals often lead to social isolation and limited access to essential services, underlining a critical need for effective and accessible solutions. Recognizing the unique challenges this community faces—such as the scarcity of sign language interpreters, particularly in remote areas, and the lack of real-time translation tools. This paper proposes the development of a smartphone-runnable sign language recognition model to address the communication problems faced by deaf and hard-of-hearing persons. This proposed model combines Mediapipe hand tracking with particle filtering (PF) to accurately detect and track hand movements, and a convolutional neural network (CNN) and bidirectional long short-term memory based gesture recognition model to model the temporal dynamics of Sign Language gestures. These models use a small number of layers and filters, depthwise separable convolutions, and dropout layers to minimize the computational costs and prevent overfitting, making them suitable for smartphone implementation. This article discusses the existing challenges handled by the deaf and hard-of-hearing community and explains how the proposed model could help overcome these challenges. A MediaPipe + PF model performs feature extraction from the image and data preprocessing. During training, with fewer activation functions and parameters, this proposed model performed better to other CNN with RNN variant models (CNN + LSTM, CNN + GRU) used in the experiments of convergence speed and learning efficiency. [ABSTRACT FROM AUTHOR]

This article explores the application of a new optimal convolutional neural network (CNN) to segment remote sensing. The paper designs a modified version of the firefly algorithm to provide an optimal structure of the CNN. The method is then applied to segment the remote-sensing images. The authors compare the proposed method with several other architectures on two multispectral remote sensing datasets, evaluating model performance and accuracy. The findings demonstrate that the projected CNN method provides the most efficient outcomes and can be a viable option for remote sensing analysis and classification due to their relative ease of training and high accuracy rates. Furthermore, the authors discuss the potential for further requests of CNNs in remote sensing statistics analysis. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks (DCNNs) have been mainly powerful and important artificial intelligence techniques, which are exploited in various computer vision applications, such as facial point detection (FPD), owing their versatility and high performance. The role of DCNNs in this area is pivotal due to their ability to learn hierarchical features, which are essential for recognizing complex patterns. On the other hand, DCNNs have considerable computational complexity due to their topology nature. An FPGA-based SoC design is needed for CNN acceleration due to the rapid development, energy efficiency, low latency and high reconfigurability of FPGAs. Despite their performance benefits, FPGA-based SoCs also come with several limitations like complexity of design. This paper proposes a dynamic partial reconfiguration (DPR) and hybrid architecture for DCNN accelerators. We propose (i) a GPU-based software implementation for DCNN-based FPD, (ii) a CNN-based acceleration and optimization method using the high-level synthesis technique, (iii) a DPR-based hybrid architecture to improve the performance of the suggested approach. To validate our design, four scenarios are put forward. The experimental results prove that the proposed work achieves a better performance in terms of reconfigurability and execution time, hardware cost, and power consumption. [ABSTRACT FROM AUTHOR]

In situations where traditional feedback methods face high complexity and rely on channel sparsity, many deep learning-based CSI compression feedback methods have shown the potential to fully capture the diversity and multiplexing gains of Massive MIMO technology in Frequency Division Duplex (FDD) mode. In order to further enhance the accuracy of obtaining downlink channel state information at the Base Station (BS) side, this paper proposes a novel neural network, SfNet. Based on visual analysis, the convolutional layers in the encoder are designed to extract long-distance time-delay correlations of the same antenna across different subcarriers. The SimAM module is introduced to mitigate clustering effects. In the decoder, a spatial-temporal joint modeling approach is presented, utilizing the Convolutional Block Attention Module (CBAM) module to preserve the original spatial structure features lost during the dimension reduction of the CSI matrix to sequence data. Subsequently, two layers of Transformer multi-head attention are employed to achieve more global time-series modeling.The experimental results indicate that, compared to CLNet, the average complexity of SfNet's encoder is reduced by 3.4%. It meets the demand for lightweight devices on the user side. In terms of accuracy, SfNet exhibits an average improvement of 30.17% compared to TransNet in outdoor environments. [ABSTRACT FROM AUTHOR]

The Internet of Things is a disruptive technology that has changed the face of many industries. On the contrary, the unpresidential growth of IoT has also introduced many cybersecurity challenges. An adversary can exploit a zero-day vulnerability in an IoT to create a botnet of things. An IoT botnet is a group of compromised Internet of Things weaponized to launch cyber attacks. Machine learning and other artificial intelligence techniques are being used to combat the wide range of cyberattacks on the Internet of Things. However, in order to overcome challenges such as early diagnosis, real-time monitoring, and adaptability to different threats, these Machine Learning approaches still require significant feature engineering. In order to identify IoT botnet assaults early on, this paper suggests using a customized convolutional neural network (CCNN) model. The four phases of the model are feature extraction, attack detection, mitigation, and pre-processing. The class imbalance has been improved and the input data pre-processed using the Enhanced Synthetic minority oversampling approach. Furthermore, flow-based features, raw attributes, mean, median, standard deviation, improved entropy, mutual information, and other statistical features are retrieved and regarded as part of the feature set. The CCNN model provides the detection or classification output during the attack detection phase, which operates depending on the features derived from the input data. Additionally, a mitigation process based on entropy has been suggested to locate the attacker node, aiding in the removal of the susceptible attacker IoT node from the network. The compromised IoT node is removed through the entropy-based mitigation method, which establishes the entropy formulation based on the node's activity. The suggested model's specificity is 97.09%, compared to the minimal specificity reached by conventional techniques, including CNN (83.58%), RNN (86.17%), RF (60.46%), SVM (78.50%), and DNN (84.12%) and SMIE (88.42%), respectively. [ABSTRACT FROM AUTHOR]

Convolutional neural network (CNN) models have become very powerful machine learning due to their high classification accuracy especially for applications dealing with images. This paper presents an automatic system for Maghrebian (Algerian, Tunisian and Moroccan) accent recognition using CNN 2D model. The CNN model is fed by images constructed from the Mel frequency cepstral coefficients (MFCC) over time. Accents are recognized independently to utterance and its start time position within the speech signal. The silence part that is presented in speech signals, causes confusion in accent recognition. An SVM classifier is trained to remove the silence basing on 12 MFCC. To show the silence effect, accent recognition is performed with and without silence elimination using 5 folds cross validation technique. Many CNN architectures were studied and evaluated using 150 audio files of 30 speakers from the three Maghrebian accents for 1, 1.5 and 2 s test hearing durations (HD). The classification accuracy (CA) of the accent recognition is improved by about 2% after silence elimination and achieves 99.33% with a hearing duration of 1.5 s using the smallest best CNN model. [ABSTRACT FROM AUTHOR]

Mental stress is a common problem that affects individuals all over the world. Stress reduces human functionality during routine work and may lead to severe health defects. Early detection of stress is important for preventing diseases and other negative health-related consequences of stress. Several neuroimaging techniques have been utilized to assess mental stress, however, due to its ease of use, robustness, and non-invasiveness, electroencephalography (EEG) is commonly used. This paper aims to fill a knowledge gap by reviewing the different EEG-related deep learning algorithms with a focus on Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTMs) for the evaluation of mental stress. The review focuses on data representation, individual deep neural network model architectures, hybrid models, and results amongst others. The contributions of the paper address important issues such as data representation and model architectures. Out of all reviewed papers, 67% used CNN, 9% LSTM, and 24% hybrid models. Based on the reviewed literature, we found that dataset size and different representations contributed to the performance of the proposed networks. Raw EEG data produced classification accuracy around 62% while using spectral and topographical representation produced up to 88%. Nevertheless, the roles of generalizability across different deep learning models and individual differences remain key areas of inquiry. The review encourages the exploration of innovative avenues, such as EEG data image representations concurrently with graph convolutional neural networks (GCN), to mitigate the impact of inter-subject variability. This novel approach not only allows us to harmonize structural nuances within the data but also facilitates the integration of temporal dynamics, thereby enabling a more comprehensive assessment of mental stress levels. [ABSTRACT FROM AUTHOR]

Coronavirus, also known as COVID-19, has spread to several countries around the world. It was announced as a pandemic disease by The World Health Organization (WHO) in 2020 for its devastating impact on humans. With the advancements in computer science algorithms, the detection of this type of virus in the early stages is urgently needed for the fast recovery of patients. In this paper, a study of neutrosophic set significance on deep transfer learning models will be presented. The study will be conducted over a limited COVID-19 x-ray. The study relies on neutrosophic set and theory to convert the medical images from the grayscale spatial domain to the neutrosophic domain. The neutrosophic domain consists of three types of images, and they are the True (T) images, the Indeterminacy (I) images, and the Falsity (F) images. The dataset used in this research has been collected from different sources. The dataset is classified into four classes {COVID-19, normal, pneumonia bacterial, and pneumonia virus}. This study aims to review the effect of neutrosophic sets on deep transfer learning models. The selected deep learning models in this study are Alexnet, Googlenet, and Restnet18. Those models are selected as they have a small number of layers on their architectures. To test the performance of the conversion to the neutrosophic domain, more than 36 trials have been conducted and recorded. A combination of training and testing strategies by splitting the dataset into (90–10%, 80–20%, 70–30) is included in the experiments. Four domains of images are tested, and they are, the original domain, the True (T) domain, the Indeterminacy (I) domain, and the Falsity (F) domain. The four domains with the different training and testing strategies were tested using the selected deep transfer models. According to the experimental results, the Indeterminacy (I) neutrosophic domain achieves the highest accuracy possible with 87.1% in the testing accuracy and performance metrics such as Precision, Recall, and F1 Score. The study concludes that using the neutrosophic set with deep learning models may be an encouraging transition to achieve better testing accuracy, especially with limited COVID-19 datasets. [ABSTRACT FROM AUTHOR]

Prediction of a driver's braking intention enables the advanced driver assistance system (ADAS) to intervene in the braking system as early as possible, which may shorten braking distance and improve driving safety. This paper proposes a novel deep learning model called LSTM-CNN-Attention that combines a long short-term memory (LSTM) neural network, convolutional neural network (CNN), and Attention mechanism for extracting spatiotemporal features of multi-sensor data to improve prediction accuracy. The proposed model inherits both temporal and spatial feature extraction abilities from LSTM and CNN. The LSTM-CNN-Attention model has a parallel architecture, which enhances the feature extraction ability of the model for multi-sensor time series data and improves the prediction accuracy of the driver's braking intention before the braking action. Furthermore, a driving simulator is set up to sample driving data for training and evaluating the proposed method. According to the results of the experiment, the model obtains up to 3.16% higher accuracy than the baseline models such as LSTM, CNN, and bidirectional LTSM (Bi-LSTM). Additionally, the influence of sliding window size and prediction horizon on the performance of the method is investigated. A method of tuning hyperparameters using the genetic algorithm is presented. The results demonstrate that the prediction accuracy increases by about 2% after being optimized by GA. [ABSTRACT FROM AUTHOR]

This research paper presents a review of the use of convolutional neural networks (CNNs) for human crowd analysis. The paper discusses the challenges and limitations of methods and highlights the potential of CNNs in addressing these limitations. This study reveals and provides an in-depth analysis of the different techniques, architectures, and algorithms used in CNNs for human crowd analysis and their respective advantages and limitations. Additionally, the paper discusses the potential applications of CNNs in crowd analysis, including pedestrian detection, crowd counting, and crowd behavior recognition. The review also provides insights into the performance evaluation metrics commonly used in this area and the datasets used for training and testing CNNs. Overall, this review provides a comprehensive overview of the latest developments in the use of CNNs for human crowd analysis, as well as insights into future research directions in this field. [ABSTRACT FROM AUTHOR]

Due to the considerable number of deaths and vehicle accidents caused by a driver's inattention, as reported by WHO, automobile manufacturers are aiming to combine advanced driver assistance systems (ADAS) with artificial intelligence algorithms, particularly deep learning and computer vision techniques. One feature that assists drivers is traffic sign recognition, which is a technique that allows vehicles to detect and recognize road signs placed on the road. This can be achieved by the aid of computer vision and Convolutional Neural Networks (CNN). The main aim of this research is to propose and improve a CNN based-model that can be efficiently and accurately applied for embedded applications, this might be accomplished with the help of several preprocessing algorithms. An improved network model called LeNet-5 has been developed for the classification of traffic signs. Furthermore, the proposed model network is trained using both German Traffic Sign Recognition Benchmark (GTSRB) and extended GTSRB (EGTSRB) datasets. According to the test results, the improved LeNet-5 architecture obtained an accuracy of 99.12% on GTSRB and 99.78% on EGTSRB datasets respectively, which has a positive performance compared to other state-of-the-art papers in terms of accuracy. [ABSTRACT FROM AUTHOR]

Image super-resolution is the process performed to improve the resolution of the images from Low Resolution (LR) to High Resolution (HR). Videoscope images are examples of industrial images that have LR. These videoscope images are enhanced in this paper using wavelet multi-scale Convolutional Neural Networks (CNNs). In this paper, we develop a videoscope super-resolution reconstruction technique based on CNNs and wavelet decomposition. The wavelet decomposition is performed on videoscope images for multi-scale representation. The CNN is trained multiple times to approximate the wavelet multi-scale representations, separately. Thus, multiple CNNs are trained to extract the features of videoscope images in several directions and multi-scale frequency bands, and thus, the HR images can be restored. [ABSTRACT FROM AUTHOR]

For the present, many technical problems exist in the convolution neural network for facial expression recognition, such as the complexity of convolutional face feature extraction, the difficulty in accurately recognizing the subtle feature changes of facial expressions, and the low automatic recognition rate of facial expressions. In this paper, a hybrid attention mechanism based on space and channel—Height Performance Module Implement(HPMI) attention mechanism is proposed to realize automatic facial expression recognition. The addition of this attention mechanism can enhance the weight of key features and make the model focused on the features which are useful for expression classification in the training process. An HPMI module based on spatial and channel-based mixed attention mechanism is embedded in VGG-16 network. This can effectively alleviate the overfitting phenomenon of the network, strengthen the useful information, suppress the useless information, promote the information flow between the key information of the image and the network model. At the same time, it can solve the problem of the inconsistency between the input dimension and the output dimension. The accuracy of the method in this paper is 98.97% and 88.44% on CK + and RAF-DB expression datasets. Experimental comparison shows that by embedding the HPMI module, the model can further enhance the learning of spatial and channel feature weight. Our experiments include 2 datasets for expression recognition and show an average improvement of 3.94% in the accuracy. [ABSTRACT FROM AUTHOR]

Object detection is the most important problem in computer vision tasks. After AlexNet proposed, based on Convolutional Neural Network (CNN) methods have become mainstream in the computer vision field, many researches on neural networks and different transformations of algorithm structures have appeared. In order to achieve fast and accurate detection effects, it is necessary to jump out of the existing CNN framework and has great challenges. Transformer's relatively mature theoretical support and technological development in the field of Natural Language Processing have brought it into the researcher's sight, and it has been proved that Transformer's method can be used for computer vision tasks, and proved that it exceeds the existing CNN method in some tasks. In order to enable more researchers to better understand the development process of object detection methods, existing methods, different frameworks, challenging problems and development trends, paper introduced historical classic methods of object detection used CNN, discusses the highlights, advantages and disadvantages of these algorithms. By consulting a large amount of paper, the paper compared different CNN detection methods and Transformer detection methods. Vertically under fair conditions, 13 different detection methods that have a broad impact on the field and are the most mainstream and promising are selected for comparison. The comparative data gives us confidence in the development of Transformer and the convergence between different methods. It also presents the recent innovative approaches to using Transformer in computer vision tasks. In the end, the challenges, opportunities and future prospects of this field are summarized. [ABSTRACT FROM AUTHOR]

One of the current challenges in deep learning is semantic segmentation. Moreover, human part segmentation is a sub-task in image segmentation, which differs from traditional segmentation to understand the human body's intrinsic connections. Convolutional Neural Network (CNN) has always been a standard feature extraction network in human part segmentation. Recently, the proposed Swin Transformer surpasses CNN for many image applications. However, few articles have explored the performance of Swin Transformer in human part segmentation compared to CNN. In this paper, we make a comparison experiment on this issue, and the experimental results prove that even in the area of human part segmentation and without any additional trick, the Swin Transformer has good results compared with CNN. At the same time, this paper also combines the Edge Perceiving Module (EPM) currently commonly used in CNN with Swin Transformer to prove that Swin Transformer can see the intrinsic connection of segmented parts. This research demonstrates the feasibility of applying Swin Transformer to the part segmentation of images, which is conducive to advancing image segmentation technology in the future. [ABSTRACT FROM AUTHOR]

Stress is an affective human state. Different types of stress have varying effects on human health. Previous stress classifications using tissue oxygen saturation (StO2) have relied on baselines to extract features in a way that limits practical application since the baselines cannot be preacquired. In this paper, we explored the correlation and the classification computability of the baseline-independent facial StO2 across four states involving baseline, emotional stress, high-intensity physical stress and low-intensity physical stress. Three analytical approaches verify that facial StO2 is affection-related and state-specific. Building on this theoretical foundation, we validated the baseline-independent facial StO2 before and after modified data augmentation using three input forms as well as two network structures. We also proposed a dual-stream (Channel stream and Spatial stream) attention network, named as CSnet, to perform baseline-independent stress classification. Experimental results suggest that the proposed method can achieve a unweighted average recall (UAR) of 0.6935 and unweighted F1-score (UF1) of 0.6991, which is higher than traditional image descriptors methods, and is more competitive than the baseline-dependent classification with respect of real applications. [ABSTRACT FROM AUTHOR]

Promoting the development of enterprises is a common task of all countries in the world, and the government has always taken promoting enterprise development as a long-term strategy. With the development for market, as an integral part of the social market economy, enterprises have made great contributions to the rapid of national economy. The enterprises are increasing, but the market competition is becoming increasingly fierce, and the weaknesses of enterprises such as limited scale, lax management and difficult financing are constantly highlighted. The appearance of these problems makes enterprises encounter many bottleneck problems in the process of development. Innovation, as the inexhaustible power of enterprise development, plays a significant role in enterprise growth, it is the key to solving the bottleneck problem of enterprise development. If an enterprise wants to achieve considerable development, it must innovate and enhance its competitiveness through innovation. In this context, how to evaluate enterprise innovation has become an important work. This work proposes an IPSO-ATT-MSCNN network via PSO from swarm optimization algorithm and artificial neural network to evaluate enterprise innovation. First, this work designs a multi-scale convolutional neural network (ATT-MSNN) via attention mechanism. This method uses multi-scale convolution to extract features of different scales, which improves the richness of features. This adds attention mechanisms to enhance useful features and reduce the impact of unwanted features such as noise. Second, in view of the performance degradation caused by the random initialization, this paper uses improved PSO algorithm to optimize the initial parameters. Third, this work proposes a series of strategies for promoting enterprise innovation. Finally, this work carried out a comprehensive and systematic experiment for the designed method, and the experiment verified the superiority of this method. [ABSTRACT FROM AUTHOR]

In the last three decades, a lot of work has been done for building Automatic Speech Recognition (ASR) systems for well-established languages such as English, Chinese, etc. However, for implementing a Large Vocabulary Continuous Speech Recognition (LVCSR) system for low resource languages, the research work is also growing rapidly in corpus-focused areas. Hence, there is a requirement of benchmarking large corpus in case of low-resource language like Bangla such that prejudice results can be avoided and limitations can be handled. In the proposed work, an openly available large-scale Bangla speech corpus provided by Google has been used. The work in this paper proposes a combination of image inspired features with well explored cepstral features to build front end feature extraction phase. It uses integration of Convolutional Neural Network (CNN) and bi-directional Long-short Term Memory (bi-LSTM) with Connectionist Temporal Classification (CTC) loss function to implement backend acoustic model. The experiments employ static and dynamic features of Mel-frequency Cepstral Coefficients (MFCC), Constant Q Cepstral Coefficients (CQCC), and Gammatone Cepstral Coefficients (GTCC) techniques, one by one, with Acoustic Ternary Patterns (ATP) features. The proposed work investigates the effect of these various hybrid front-end approaches with CNN, bi-LSTM and integration of these two models. The novelty of this paper lies in the fact that fusion of ATP with cepstral features improves performance of the proposed low resource language ASR system, where the proposed combination of ATP-dynamic CQCC features with integrated backend acoustic model shows a relative improvement of 10–15% in Word Error Rate (WER) over all other experimented combinations. Further to exploit the noise robust nature of GTCC features, the ATP-dynamic GTCC features with integrated CNN-bi-LSTM back-end model are evaluated in noisy scenario, also. [ABSTRACT FROM AUTHOR]