Search

Deep learning has revolutionized object detection, with YOLO (You Only Look Once) leading in real-time accuracy. However, detecting moving objects in visual streams presents distinct challenges. This paper proposes a refined YOLOv8 object detection model, emphasizing motion-specific detections in varied visual contexts. Through tailored preprocessing and architectural adjustments, we heighten the model’s sensitivity to object movements. Rigorous testing against KITTI, LASIESTA, PESMOD, and MOCS benchmark datasets revealed that the modified YOLOv8 outperforms the state-of-the-art detection models, especially in environments with significant movement. Specifically, our model achieved an accuracy of 90%, a mean Average Precision (mAP) of 90%, and maintained a processing speed of 30 frames per second (FPS), with an Intersection over Union (IoU) score of 80%. This paper offers a detailed insight into object trajectories, proving invaluable in areas like security, traffic management, and film analysis where motion understanding is critical. As the importance of dynamic scene interpretation grows in artificial intelligence and computer vision, the proposed enhanced YOLOv8 detection model highlights the potential of specialized object detection and underscores the significance of our findings in the evolving field of object detection.

The real-time analysis of gas flares is one of the most challenging problems in the operation of various combustion-involving industries, such as oil and gas refineries. Despite the crucial role of gas flares in securing safe plant operation and lowering environmental pollution, they are among the least monitored components of petrochemical plants due to their harsh working environments. Remote sensing techniques are emerging as potential alternatives for conventional sampling-based techniques in visual inspection and performance analysis. This paper presents an in-depth review of significant achievements in Gas flares monitoring over past two decades and highlights ongoing challenges in Artificial Intelligence (AI)-enhanced remote inspection. By reading the content, both industry professionals and academic researchers can gain a comprehensive knowledge of improvements from the integration of AI and remote sensing, understanding the trend of Gas flares monitoring. The paper commences with an analysis of RGB camera-based methods, focusing on how their combination with cutting-edge machine learning and deep learning algorithms can significantly improve the detection, segmentation, and measurement of gas flare systems. It then explores the use of hyper-spectral imaging techniques, including infrared cameras and space-borne satellite sensors, underscoring their potential in remote monitoring and performance analysis. Additionally, the effectiveness of multi-view inspection methods is assessed, highlighting how these approaches enhance the monitoring capabilities. Finally, the paper identifies key research areas that require further attention. It also presents a clearer direction for future progress, emphasizing the importance of continuous research to foster advancements and facilitate broader commercial adoption.

Speech and voice disorders are highly prevalent and a significant concern, particularly among children. These disorders have a notable impact on individual’s personalities, academic performance and overall development. Early detection and intervention of these disorders surely help out and generate an effective outcome. Automatic detection of speech and voice disorders at an early stage plays a crucial role in identifying and addressing these communication challenges because of their efficiency over traditional diagnostic methods. Over time, numerous techniques were developed in the past, but some of them have become less efficient in the present scenario. Deep Learning has surfaced as a recent and significant advancement in detecting speech and voice disorders. This paper provides a systematic review of the utilization of deep learning techniques for the detection of speech and voice disorders. The review encompasses studies published from 2018 to 2023 exploring various architectures of deep learning models for capturing complex patterns in speech data. Each deep learning based speech sound disorder detection technique is discussed with critical appraisal and relevant benchmarks available for evaluation of results. This review holds significance for new researchers who are interested in exploring the field of automatic speech disorder detection as the paper concludes by discussing future directions and potential areas of improvement in automatic speech sound disorder detection using deep learning.

Rare event detection (RED) involves the identification and detection of events characterized by low frequency of occurrences, but of high importance or impact. This paper presents a Systematic Review (SR) of rare event detection across various modalities using Machine Learning (ML) and Deep Learning (DL) techniques. This review comprehensively outlines techniques and methods best suited for rare event detection across various modalities, while also highlighting future research prospects. To the extent of our knowledge, this paper is a pioneering SR dedicated to exploring this specific research domain. This SR identifies the employed methods and techniques, the datasets utilized, and the effectiveness of these methods in detecting rare events. Four modalities concerning RED are reviewed in this SR: video, sound, image, and time series. The corresponding performances for the different ML and DL techniques for RED are discussed comprehensively, together with the associated RED challenges and limitations as well as the directions for future research are highlighted. This SR aims to offer a comprehensive overview of the existing methods in RED, serving as a valuable resource for researchers and practitioners working in the respective field.

This paper aims to develop a defect-background separated generative adversarial network (GAN) using deep learning and GAN to enhance the accuracy of battery exterior defect inspection. In actual battery production lines, the occurrence rates of defects vary by defect type, making it challenging to create a large, uniform defect dataset due to the time required for defect acquisition. This leads to a reduction in the accuracy of battery exterior defect inspection. To construct a large, uniform defect dataset, this paper proposes a defect-background separated GAN based on the principles of GANs. The defect-background separated GAN performs effective defect and background separation learning by referring to the segmentation labeling of defects. Through dataset augmentation using the defect-background separated GAN, the performance quality of newly generated synthetic defect images has improved, and accuracy in battery exterior defect inspection can be enhanced through extensive dataset training. Experimental results show the lowest Fréchet inception distance score among various other methods for the battery exterior defect dataset, and it generates clear synthetic defects perceptible to the human eye. By training defect segmentation on this large, uniform defect dataset, an accuracy of 96.1% and an intersection over union value of 0.71 were achieved. Ultimately, applying this defect inspection network to the actual production line demonstrated a 72% improvement in time efficiency. This demonstrates the stability and robustness of the large, uniform defect dataset generated through the defect-background separated GAN.

It is important in production to achieve accurate counting and density estimation of high-density culture fry under the environmental conditions of aquaculture scenarios in an efficient and accurate manner. However, none of the current methods for fry counting works well under the high-density and high-overlap conditions of real aquaculture scenarios. Therefore, in this paper, we propose a high-density farming fry monitoring network model, Super-Resolution GAN Density Estimate Attention Network (SGDAN), which incorporating an image enhancement algorithm and an attention mechanism, and we create a high-density farming fry dataset (HD-FryDataset) based on the environmental conditions of real aquaculture scenarios. The network model is designed to improve and optimize the targeted subnetworks for several key aspects of high-density fish fry monitoring work. Four subnetworks are included for image optimization, feature extraction, attention, and density map estimation. The experimental results show that the SGDAN network model achieved an average counting accuracy of 97.57% on the high-density culture fry dataset, which was 8.23% and 2.06% higher than those of MCNN and CSRNet, respectively. Additionally, the MAE and RMSE of the model were reduced by 71.9% and 67.3% and by 34.3% and 33.2% compared with those of MCNN and CSRNet, respectively. The model proposed in this paper also has a better ability to generate predictive density maps. The density maps generated by SGDAN have values of the evaluation metrics PSNR and SSIM of 20.33 and 0.933, respectively, which are 3.31 and 0.037 and 2.63 and 0.031 higher than those of MCNN and CSRNet. In general, the network model proposed in this paper outperforms existing network models in two applications: accurate counting of fry and generation of density maps for high-density culture in aquaculture. It also provides a good solution for digitizing the number of fry and visualizing the density of high-density culture in intelligent aquaculture systems.

Vision-based smart agriculture is an important way to improve the efficiency of agricultural production. Labeling images for deep learning in complex field photos is a difficult task. In this paper, an image augmentation method that can help reduce the workload of image labeling based on synthetic images is proposed. The synthetic images consisted of three parts: crop, weed, and soil. The crop and weeds were obtained automatically by Excess Green (ExG) and minimum error threshold segmentation. The data augmentation method was tested on image classification, object detection, and semantic segmentation tasks by Resnet, YOLOV5, and DeeplabV3. The accuracy of the classification model reached 0.99. The IoU of object detection and semantic segmentation were 0.98 and 0.96, respectively. The results showed that the method in this paper was acceptable despite slight overfitting. This method was proposed based on the characteristics of field images, and it was meant for reducing the workload of labeling images.

The rapid advancement of Large Language Models (LLMs) has significantly enhanced the capabilities of text generators. With the potential for misuse escalating, the importance of discerning whether texts are human-authored or generated by LLMs has become paramount. Several preceding studies have ventured to address this challenge by employing binary classifiers to differentiate between human-written and LLM-generated text. Nevertheless, the reliability of these classifiers has been subject to question. Given that consequential decisions may hinge on the outcome of such classification, it is imperative that text source detection is of high caliber. In light of this, the present paper introduces DeepTextMark, a deep learning-driven text watermarking methodology devised for text source identification. By leveraging Word2Vec and Sentence Encoding for watermark insertion, alongside a transformer-based classifier for watermark detection, DeepTextMark epitomizes a blend of blindness, robustness, imperceptibility, and reliability. As elaborated within the paper, these attributes are crucial for universal text source detection, with a particular emphasis in this paper on text produced by LLMs. DeepTextMark offers a viable “add-on” solution to prevailing text generation frameworks, requiring no direct access or alterations to the underlying text generation mechanism. Experimental evaluations underscore the high imperceptibility, elevated detection accuracy, augmented robustness, reliability, and swift execution of DeepTextMark.

Environment perception using camera, radar, and/or lidar sensors has significantly improved in the last few years because of deep learning-based methods. However, a large group of these methods fall into the category of supervised learning, which requires a considerable amount of annotated data. Due to uncertainties in multi-sensor data, automating the data labeling process is extremely challenging; hence, it is performed manually to a large extent. Even though full automation of such a process is difficult, semi-automation can be a significant step to ease this process. However, the available work in this regard is still very limited; hence, in this paper, a novel semi-automatic annotation methodology is developed for labeling RGB camera images and 3D automotive radar point cloud data using a smart infrastructure-based sensor setup. This paper also describes a new method for 3D radar background subtraction to remove clutter and a new object category, GROUP, for radar-based object detection for closely located vulnerable road users. To validate the work, a dataset named INFRA-3DRC is created using this methodology, where 75 % of the labels are automatically generated. In addition, a radar cluster classifier and an image classifier are developed, trained, and tested on this dataset, achieving accuracy of 98.26% and 94.86%, respectively. The dataset and Python scripts are available at https://fraunhoferivi.github.io/INFRA-3DRC-Dataset/.

Steel surface defect detection is an indispensable part of industrial production and processing processes. It helps to reduce production costs, ensure product quality, improve production safety and compliance, and maintain sustainability and competitiveness. To address the low detection accuracy of traditional methods, this paper developed and investigated an improved algorithm based on YOLO for steel surface defect detection: ECA(Efficient Channel Attention) -SimSPPF (Simplified Spatial Pooling - Fast) -SIoU (Scylla Intersection over Union) -Yolov5. First, deformable convolutions were used to replace some conventional convolutions in the model, which expanded the receptive field and improved detection accuracy. Additionally, Efficient Channel Attention was integrated into the model to improve the weight of important information. Then, the SimSPPF was employed in place of the SPP module in the model, reducing computational complexity. Finally, the SIoU loss function was utilized to handle bounding box regression more effectively. The paper conducted different ablation experiments, and the improved ECA-SimSPPF-SIoU-Yolov5 algorithm demonstrated superior detection performance. Using the NEU-DET dataset, the mAP reached 78.8%, which was a 7.1% improvement higher than the original model, while the Recall reached 76.4% and improved by 3.7% compared to the original model. The improved model showed significant improvements in terms of mAP and Recall. Furthermore, the paper conducted multiple comparative experiments, comparing the model with other attention mechanisms and loss functions. The results demonstrated that the improved ECA-SimSPPF-SIoU-Yolov5 algorithm achieved good detection results in terms of mAP and Recall. In the third comparative experiment, the model was compared with YOLOv5 model with different network depths and the latest Yolov8 model, and the improved model also achieved good detection accuracy.

This study is devoted to exploring the strategy of automatic sketch generation and optimization of industrial design based on deep learning. By combining the Generative Adversarial Network (GAN) with the optimization algorithm, this paper proposes an innovative method to realize the automatic generation of high-quality and diverse industrial design sketches. In the experiment, this paper selects SketchyCAD and other public data sets, trains them through deep learning model, and introduces genetic algorithm(GA) and differential evolution algorithm to optimize the parameters. In terms of experimental results, we observed that the quality of generated sketches was significantly improved, and the design sketches generated by the mode (GAN+GA) were more realistic and innovative. The introduction of optimization strategy further improves the generation effect and intelligently adjusts the model parameters to adapt to different design styles. In this paper, the influence of hyperparameter tuning is analyzed in detail, and it is found that the adjustment of learning rate plays a key role in generating quality and diversity. However, the experiment also revealed some challenges and room for improvement. We noticed that the generated results may have the risk of over-fitting in the training process, and with the increase of training times, the diversity gradually decreased. This suggests that more complex model structure and richer data sets are needed to improve the generalization performance. Generally speaking, this study provides new ideas and methods for the integration of deep learning and industrial design. By innovatively combining generation model and optimization algorithm, this research has contributed beneficial research results to the development of industrial design automation. This research is of great significance to promote the intelligence and innovation in the field of industrial design.

With the rapid development of artificial intelligence (AI), especially in machine learning and deep learning technologies, the particle and impurity detection and removal processes employed in many industries have been improved. Particles and impurities of any size, shape and in any condition can be detected using advanced technology in both areas. This paper presents a comprehensive overview of research papers that discuss the application of AI techniques for the detection and removal of particles and impurities. The publications featured in this review were mainly retrieved from the Web of Science (WoS) database, covering the timeframe from 2000 to 2023. This paper also covers the review on the impurity detection and removal specifically in edible bird’s nest (EBN). The aim of this paper is to provide a valuable resource for the future development of AI applications in particle and impurity detection and removal technologies that have not been addressed in this study. Through the review and analysis of AI for particle and impurity detection and removal techniques in recent years, this paper includes the following parts: research trend in particle and impurity detection in general and AI methods in particle and impurity detection, applications of AI in particle and impurity detection in related industries including in EBN and AI applications in particle and impurity removal. This review study will offer advantages to researchers engaged in the field of AI with regards to the detection and removal of particles and impurities.

Partial discharge (PD) pattern recognition approaches are designed to identify the types or severities of the insulation defects within the high voltage equipment, which is vital for evaluating potential harmfulness and making follow-up maintenance plan. In recent years, many advanced machine learning (ML) algorithms have been introduced to this field and achieved remarkable results. As the second one of the two-part papers, we aim to give a comprehensive review regarding the pattern recognition approaches for ultra-high frequency (UHF) PD data in this paper. These methods are grouped into three categories, which are the traditional ML-based PD type recognition, the deep learning-based (DL) PD type recognition, and PD severity assessment. Specifically, for the first topic, feature extraction methods, dimensionality reduction methods and classification methods are reviewed separately. For the second topic, many state-of-the-art DL methods are discussed, including the deep belief network (DBN), deep autoencoder network (DAN), convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), graph convolutional network (GCN), deep ensemble learning (DEL), etc. For the third topic, the relevant algorithms are also divided into the conventional ML-based ones and the DL-based ones, which are studied in detail respectively. Finally, a brief discussion about the application effects of the above technologies is given, and some future directions are suggested. This paper covers almost every aspect of the PD pattern recognition and highlights the latest progress, which can provide valuable references for scholars in this field.

This paper addresses the issue of reliable and automated analog circuit’s structure recognition (ACSR). First, it presents a comprehensive critical review of the related state-of-the-art. Then, essentially, this study proposes and validates a novel approach for realizing a dependable structure recognition system for analog circuits through a comprehensive ontology definition, some transformation tools, scene modeling through graph models, and then, by involving a graph attention neural network (GAT) model. Knowingly, identifying sub-circuits requires, in the brute-force mode, extensive screening of the numerous alternative substructures that can be constructed from the basic elements (i.e. the transistors) in presence, especially w.r.t. the actual connectivity topology amongst them; this is evidently a very complex and challenging endeavor. The relevant state-of-the-art has involved mostly unsupervised learning approaches to tackle this analog circuit’s structure recognition-related challenging task, whereby they have reached a clear limitation of not (or only very hardly) reaching beyond 90% to 93% accuracy/precision. But in this work, we develop, for the first time, a comprehensive supervised-learning approach that demonstrates its clear superiority by enabling the reaching of a recognition accuracy or precision that is reliably in the range beyond 99% (or even 100%) for each subblock of the analog circuit. The supervised clustering approach developed and validated in this study consists of a comprehensive modeling and conceptual pipeline that is implemented around and through a Graph Attention Neural Network model, which ensures a reliable recognition of sub-blocks of analog circuits and their related internal adjacency connectivity topology. Besides the modeling pipeline, this study also develops a set of tools for mapping an analog circuit’s schematics into a graph model. To overcome the limited and unbalanced samples for training the graph neural model, this study proposes a special novel augmentation strategy based on a graph sub-cropping technique. This augmentation technique is embedded in a smart stepwise augmentation protocol that leads at the end, through iterative additional dedicated training steps to a significant progressive increase of the recognition accuracy by the graph neural model until reaching more than 99%, better 100%, for each of the subblocks of the analog circuits as demonstrated in the demonstration case-study presented. Indeed, this paper’s quintessence represents a significant breakthrough in view of the clear outperforming of the currently relevant related state-of-the-art.

With the development of artificial intelligence, there have been many attempts to incorporate artificial intelligence into algorithmic trading. In particular, reinforcement learning, which aims to solve dynamic decision-making problems, is attracting attention because of its high utilization in algorithmic trading. In this paper, we will implement a simple Deep Reinforcement Learning (DRL) trading robot to check the performance of DRL. In addition, we tried to find out how much performance improvement can be achieved by comparing a robot that learned a single stock data with a robot that learned stock data, market index, and interest rate data. This paper aims to develop a stock investment robot using a Proximal Policy Optimization (PPO) reinforcement learning algorithm and analyze the performance of the robot. The first robot used only the stock data of APPL INC, a single stock, as input, and the second robot used stock data of APPL INC and the S&P 500 index together with US interest rate data as input data. Afterward, the stock investment performance of the two robots for APPL INC was comparatively analyzed using the test data.

Video forgery detection is a critical aspect of digital forensics, addressing the challenges posed by the manipulation of video content. This paper presents a novel approach for video forgery detection using Deep Convolutional Neural Networks (DCNN). Leveraging the power of deep learning, our method aims to improve the accuracy and efficiency of object-based forgery detection in advanced video sequences. In the proposed approach, we build upon the foundation of an existing method, which utilizes Convolutional Neural Networks, and introduce innovative modifications to the DCNN architecture. These modifications include data preprocessing, network architecture, and training strategies that enhance the model’s ability to detect tampered objects in video frames. We conduct experiments on the SYSU-OBJFORG dataset, the largest object-based forged video dataset to date, with advanced video encoding standards. Our DCNN-based approach is compared with the existing method, demonstrating superior performance. The results show increased accuracy and robustness in detecting object-based video forgery. This paper not only contributes to the field of video forgery detection but also underscores the potential of deep learning, particularly DCNN, in addressing the evolving challenges of digital video manipulation. The findings open avenues for future research in the localization of forged regions and the application of DCNN in lower bitrate or lower resolution video sequences.

Wood diameter needs to be measured in the process of production, sales and import and export. In order to solve the problem that it is difficult to accurately measure the densely stacked and irregularly arranged vehicle wood manually, this paper proposes a timber segmentation methodology that leverages a Swin Transformer model mechanism to enhance the performance of the target detection model. The method automatically learns and calculates distinct regions in the input image, assigning varying weights to different sizes and shapes of wood. This approach achieves finer detection of densely stacked logs, thereby promoting intelligent inspection and enhancing inspection efficiency.This study optimizes the backbone network by refining its modules and incorporating the operation of the log-space bias module. Additionally, improvements are made to the feature fusion network and loss function to further enhance network performance. The instance segmentation model parameters are also modified, encompassing multi-scale training, an increased number of training samples, improved image input size, and effective data widening techniques, all of which enhance log measurement accuracy and resolve the issue of partially occluded logs.This study conducts multiple control experiments to evaluate various scale metrics, such as mean average precision (mAP), log true detection rate, false detection rate, as well as comparing the root count and volume of logs through prediction. The experiments demonstrate that the mAP of this methodology reaches 0.685, and the true detection rate reaches 0.96 when compared with mainstream neural networks of similar scale, highlighting the advantages of this paper’s approach in wood segmentation detection. The model exhibits a strong detection effect on dense wood, effectively overcoming occlusion challenges, leading to more accurate measurement data. Moreover, the algorithm demonstrates robustness and migration ability, rendering it highly applicable to the task of detecting and segmenting dense wood of all sizes.

This paper presents an innovative approach for personalized modeling of Head-Related Transfer Functions (HRTFs) using deep neural networks (DNNs), with a primary focus on addressing interaural time differences (ITDs), torso and pinna anthropometrics. HRTFs play a pivotal role in spatial hearing applications, such as virtual auditory displays and hearing assistive devices. The core objectives of this study involve mitigating measurement irregularities related to sudden head movements, microphone and sound source positioning, and head orientation. A significant component of this paper introduces a novel ITD correction method designed to reduce irregularities within the HRTF database. This method leverages ITD data, in conjunction with torso and pinna anthropometrics, within a DNN-based model to estimate individual HRTFs for precise spatial directions. Comparisons are made between this approach and traditional head measurement methods. Additionally, this research investigates the sensitivity of HRTFs to minor changes in head position. By analyzing variations in left and right pinna HRTFs at different head tilt angles through numerical simulations, using a 3D model of a human head and torso, the study quantifies spectral distortion in HRTFs. It highlights that spectral distortion generally increases with head motion and sound signal frequency, with a minor decrease observed at 16 kHz. In summary, this paper contributes to the advancement of personalized HRTF modeling, particularly in scenarios involving fluctuations in head position. The findings have the potential to enhance the effectiveness of machine learning algorithms for sound localization, particularly when considering both pinnae HRTFs. This research underscores the importance of these aspects in optimizing spatial hearing experiences.

With the rapid development of information communication and mobile device technologies, smart devices have become increasingly popular, providing convenience to households and enhancing the level of intelligence in daily life. This trend is also driving innovation and progress in various fields, including healthcare, transportation, and industry. However, as technology continues to proliferate, network security concerns have become increasingly prominent, making the protection of digital life and data security an urgent priority. Intrusion detection has always played an important role in the field of network security. Traditional intrusion detection systems predominantly rely on anomaly detection technology to identify potential intrusions by detecting abnormal patterns in network traffic. With technological advancements, machine learning-based methods have emerged as the cornerstone of modern intrusion detection, enabling more precise identification of abnormal behaviors and potential intrusions by learning the patterns of normal network traffic. In response to these challenges, this paper introduces an innovative intrusion detection model that amalgamates the Attention-CNN-BiLSTM (ACBL) and Temporal Convolutional Network (TCN) architectures. The ACBL and TCN models excel in processing spatial and temporal features within network traffic data, respectively. This integration harnesses diverse neural network structures to elevate overall model performance and accuracy. Furthermore, a unique approach inspired by dung beetles’ natural behavior, incorporating Tent mapping-enhanced Dung Beetle Optimization Algorithm (TDBO), is leveraged for both optimizing feature selection parameters and searching for optimal model hyperparameters. The feature selection parameters obtained from TDBO are then combined with the importance ranking from the Random Forest algorithm, ensuring optimal features can be better selected to enhance model performance. This paper introduces a novel intrusion detection model, the TDBO-ACBLT model, and validates its performance using the UNSW-NW15 dataset. TDBO excels in feature selection compared to common algorithms and achieves superior parameter optimization accuracy over Harris’s Hawk Optimization (HHO), Particle Swarm Optimization (PSO), and Dung Beetle Optimization (DBO). The proposed model achieves higher accuracy than prevalent machine learning models.

The Chinese furniture market has a high demand for wood floors. Manual defect detection in wood floors is inefficient and lacks stability in accuracy. It is necessary to conduct research on automatic defect detection in wood floors. To improve the accuracy of detecting small defects in wood floors, this paper proposed a new network based on YOLOv7. The new network is called the cascade center of gravity YOLOv7 (CCG-YOLOv7). This paper designed cascade efficient layer aggregation networks (C-ELAN), streamlined the CBS, replaced the ELAN with the C-ELAN, introduced the rapid supervised attention module to connect the backbone and head layers, and simplified the head layer of the YOLOv7 network. These methods improved the detection accuracy and speed for detecting small defects on wood floor surfaces. The improved network can effectively detect small defects on the wooden board surfaces, including knots, scratches, and mildew. Compared to the original YOLOv7, CCG-YOLOv7 improves precision, recall, and mean average precision by 2.1%, 1.6%, and 1.2%, respectively.

In the aviation industry, safety remains vital, often compromised by pilot errors attributed to factors such as workload, fatigue, stress, and emotional disturbances. To address these challenges, recent research has increasingly leveraged psychophysiological data and machine learning techniques, offering the potential to enhance safety by understanding pilot behavior. This systematic literature review rigorously follows a widely accepted methodology, scrutinizing 80 peer-reviewed studies out of 3352 studies from five key electronic databases. The paper focuses on behavioral aspects, data types, preprocessing techniques, machine learning models, and performance metrics used in existing studies. It reveals that the majority of research disproportionately concentrates on workload and fatigue, leaving behavioral aspects like emotional responses and attention dynamics less explored. Machine learning models such as tree-based and support vector machines are most commonly employed, but the utilization of advanced techniques like deep learning remains limited. Traditional preprocessing techniques dominate the landscape, urging the need for advanced methods. Data imbalance and its impact on model performance is identified as a critical, under-researched area. The review uncovers significant methodological gaps, including the unexplored influence of preprocessing on model efficacy, lack of diversification in data collection environments, and limited focus on model explainability. The paper concludes by advocating for targeted future research to address these gaps, thereby promoting both methodological innovation and a more comprehensive understanding of pilot behavior.

With the escalating global challenges of food security and resource sustainability, innovative solutions like deep learning and computer vision are transforming agricultural practices by enabling data-driven decision-making. This paper provides a focused review of recent advancements in deep learning-enabled computer vision techniques tailored specifically for greenhouse environments. First, deep learning and computer vision fundamentals are briefly introduced. Over 100 studies from 2020 to date are then comprehensively reviewed in which these technologies were applied within greenhouses for growth monitoring, disease detection, yield estimation, and other tasks. The techniques, datasets, models, and overall performance results reported in the literature are analyzed. Tables and figures showcase real-world implementations and results synthesized from current research. Key challenges are also outlined related to aspects like model adaptability, lack of sufficient labeled greenhouse data, computational constraints, the need for multi-modal sensor fusion, and other areas needing further investigation. Future trends and prospects are discussed to provide guidance for researchers exploring computer vision in the niche greenhouse domain. By condensing prior work and elucidating the state-of-the-art, this timely review aims to promote continued progress in smart greenhouse agriculture. The focused analysis, specifically on greenhouse environments, fills a gap compared to previous agricultural surveys. Overall, this paper highlights the immense potential of computer vision and deep learning in driving the emergence of data-driven, smart greenhouse farming worldwide.

Deep convolutional neural networks have made some progress in skin lesion classification and cancer diagnosis, but there are still some problems to be solved, such as the challenge of small inter-class feature differences and large intra-class feature differences, which might limit the classification performance of the model as high-level and low-level features are not properly utilized. This paper proposes a multi-scale skin cancer image identification network using self-interactive attention pyramid and cross-layer bilinear-trilinear pooling(SPCB-Net), which mainly consists of three proposed sub-modules that are the self-interacting attention pyramid (SAP), the across-layer bilinear-trilinear pooling operation and the global average algorithm(GAA). The SPCB-Net is applied to two representative datasets of medical images in dermatology and histopathology (HAM10000 and NCT-CRC-HE-100K) to demonstrate the effectiveness of in the skin lesion classification. SPCB-Net(ResNet101) achieves 97.10% and 99.87% accuracy on HAM10000 and NCT-CRC-HE-100K respectively, which are both achieved performance improvements of 0.4% compared to the state-of-the-art models. In addition, a large number of experiments on HAM10000 show that the interactive attention pyramid(SPA) proposed in this paper is superior to the common attention module, and the method with a cross-layer bilinear-trilinear pooling is superior to the cross-layer trilinear pooling method. SPCB-Net is configured on Vgg19 and ResNet101 to evaluate the effectiveness of our proposed module. The experimental results show that SPCB-Net has shown state-of-the-art performance in the two field of dermatology and histopathology. Therefore, it is not only well qualified for the task of identifying skin cancer image but also has the potential to identify skin cancer by identifying pathological tissue.

The dynamic and unpredictable nature of road traffic necessitates effective accident detection methods for enhancing safety and streamlining traffic management in smart cities. This paper offers a comprehensive exploration study of prevailing accident detection techniques, shedding light on the nuances of other state-of-the-art methodologies while providing a detailed overview of distinct traffic accident types like rear-end collisions, T-bone collisions, and frontal impact accidents. Our novel approach introduces the I3D-CONVLSTM2D model architecture, a lightweight solution tailored explicitly for accident detection in smart city traffic surveillance systems by integrating RGB frames with optical flow information. Empirical analysis of our experimental study underscores the efficacy of our model architecture. The I3D-CONVLSTM2D RGB + Optical-Flow (trainable) model outperformed its counterparts, achieving an impressive 87% Mean Average Precision (MAP). Our findings further elaborate on the challenges posed by data imbalances, particularly when working with a limited number of datasets, road structures, and traffic scenarios. Ultimately, our research illuminates the path towards a sophisticated vision-based accident detection system primed for real-time integration into edge IoT devices within smart urban infrastructures.

Sudden cardiac death (SCD) is one of the main causes of death in athletes during exercise and physical activity. By analyzing the results of the ECG classification, doctors can promptly detect the presence of cardiovascular diseases and implement appropriate treatment measures, thereby reducing disease progression and the occurrence of complications. Previous studies have typically used machine learning methods that often required manual feature selection, which could be subjective, time-consuming, and limited in scope. Additionally, the relationships between features may be overlooked, leading to a decrease in model performance. Therefore, our proposed approach automatically learns and selects relevant features, avoiding the issues associated with manual feature selection. In this paper, we introduce the Grunet deep learning network model, which utilizes convolutional neural networks to automatically extract features from ECG signals, thereby enhancing feature utilization and representation capabilities. Given the time-sensitive nature of ECG signals, we introduce gated recurrent units (GRU) to better capture these features. The gating mechanism of GRU helps manage information flow and facilitates capturing long-term dependencies. In our experiments conducted on the MIT-BIH Arrhythmia Database and the European ST-T Database, our proposed method achieved an overall classification accuracy of 99.47% on the European ST-T Database, with a precision of 98.76% and a recall of 97.92% on the MIT-BIH Arrhythmia Database. The classification accuracy for classes N and Q reached 100%. The experimental results demonstrate that our method outperforms existing techniques, significantly reducing the cost of manual intervention and improving the accuracy of heartbeat classification.

Nowadays, along with the rise of digital human system, 3D animation, intelligent medical and other industries, 3D face reconstruction technology has become a popular research direction in computer vision and computer graphics. Traditional 3D face reconstruction techniques are affected by face expression, occlusion, and ambient light, resulting in poor accuracy and robustness of the reconstructed model, etc. With the rise of deep learning, all of the above problems have been greatly improved. Focusing on 3D face reconstruction techniques based on deep learning, this paper categorizes the existing research works into 3D face reconstruction based on hybrid learning and explicit regression. The first category of research work fits 2D faces to 3D models, which is a pathological process that requires solving the basis vector coefficients of the 3D face statistical model. The second type of research work, instead of Model Fitting, represents 3D faces with multiple data types in the display space and directly regresses 2D faces through deep networks. This review provides the latest advances in single-image-based 3D face reconstruction techniques in recent years, summarizing some commonly used face datasets, evaluation metrics, and applications. Finally, we discuss the main challenges and future trends of the single-image 3D face reconstruction task.

With psoriasis and eczema being the two most common diseases worldwide, achieving automatic diagnosis could be useful for healthcare cyber-physical system. However, creating such an automatic classification system is still challenging since it cannot learn positional and spatial information from unstable training. In this paper, we propose a novel abscissa-ordinate focused network (AOFNet) with active label smoothing for the identification of psoriasis and eczema from images. The AOFNet incorporates the developed abscissa-ordinate focused module that focuses on abscissa-ordinate information and leverages the attention mechanism to enhance the network’s ability to learn positional and spatial details, resulting in improved classification performance. Additionally, the adoption of an active label smoothing approach effectively mitigates the problem of overconfidence and effectively captures the dynamic changes that occur during training, thereby providing an added boost to the overall performance of the network. To evaluate the proposed healthcare cyber-physical system, extensive experiments are conducted on the clinical psoriasis and eczema dataset, and the results demonstrate that the designed system could gain comparable classification performance.

In this paper, we propose a novel opportunistic scheduling-based antenna-user selection (OBAUS) scheme to improve the secrecy performance of cooperative non-orthogonal multiple access (NOMA) systems in the presence of a passive eavesdropper. The proposed OBAUS scheme can be divided into two steps such as relay antenna selection which selects the antenna at the relay to minimize the eavesdropper channel condition, and user selection which selects the received antenna at each user to maximize the main channel gains, respectively. To capture the relation between network parameters and secrecy performance, we derive the closed-form expression for secrecy outage probability (SOP) of cell-center and cell-edge users, respectively. Toward a real-time setting, we design the deep neural network (DNN)-based optimization that predicts each user channel capacity, the optimal values of transmit power, and the power allocation coefficient at the same time. Numerical results show that the proposed scheme can improve the SOP and secrecy throughput performances compared to that of the conventional scheduling scheme, called the random relay-user pair antenna selection (RRUAS) scheme. Besides that, the proposed DNN-based optimization can predict the secrecy performance and find the optimal points. Compared to that of the traditional searching algorithm (2D Golden Section Searching), the proposed DNN-based optimization significantly reduces the execution time as well as reasonable secrecy performance prediction and optimal point finding. Moreover, the impacts of essential network parameters such as transmit power at source and relay, the number of relay’s and user’s antennas, power allocation coefficient, and minimum secrecy data rates on the system secrecy performance are investigated to show the effectiveness of the proposed scheme.

Recent technological breakthroughs enable academics to suggest automating hydroponic farms to increase productivity and reduce workforce requests, hence improving agricultural profitability and output. Traditional deep learning models for detecting plant diseases may pose challenges in terms of their compatibility with embedded systems and only a few research works show their implementation on low-resource Internet of Things (IoT) devices like smartphones or Raspberry Pi. The current models have several shortcomings, including a substantial power bill, high computing power demands, quick energy dissipation, and significant storage space demands because of their intricate structures. Hence, in this paper, an intelligent hydroponic lettuce deficiency classification method is presented by employing deep learning. Initially, collected images are taken to the developed Ensemble Multi-Dilated Adaptive Networks (E-MD-ANet)-based deficiency classification network. In this network, techniques such as Visual Geometry Group 16 (VGG16), Residual Attention Network (RAN), and Convolutional Neural Network (CNN) are adopted for the extraction of the feature stage. Further, the extracted features from these networks are given to the Long Short-Term Memory (LSTM) for the deficiency classification. The parameters presented in the E-MD-ANet network are tuned using the proposed Improvised Honey Badger Algorithm (I-HBA). Thus, the hydroponic lettuce deficiency classified image is attained in the end. Here, the experiments are conducted based on the Lettuce NPK dataset which contains the, Fully Nutritional, Potassium Deficient, and Nitrogen Deficient, and Phosphorus Deficient whereas the total count of images is gathered as 12, 58, 66, and 72. Also, various evaluation parameters are taken for the analysis like precision, NPV, specificity, F1-score, FPR, FDR, accuracy, MCC, FNR, and sensitivity. In the end, the efficacy validation of the suggested task is conducted and contrasted with numerous conventional classification mechanisms to showcase the system’s efficacy. The offered approach shows elevated performance regarding accuracy and specificity rate which attains 96% and 95%. From the result analysis, the offered approach achieves greater performance while combining with the baseline techniques.

Unmanned Aerial Vehicles (UAVs) are advanced technologies that are initially utilized for military apps like border monitoring and reconnaissance in opposed territories. Internet of Things (IoTs) assisted UAV networks suggest the combination of IoT technology with UAVs to generate a networked system that improves the abilities and utility of UAVs for several apps. UAVs’ inherent features namely quick deployment, high dynamicity, low deployment and operational costs, and line of sight communication motivated researchers in the IoT field to assume UAV’s combination into IoT systems near the concept of UAV-assisted IoT systems. However, security concerns with UAVs are evolving as UAV nodes are suitable attractive targets for cyber threats because of extremely developing volumes and poor and weak inbuilt security. Therefore, this paper presents a Modified Marine Predators Algorithm with a Deep Learning-Driven intrusion detection (MMPADL-ID) approach for IoT Assisted UAV Networks. The presented MMPADL-ID technique proposes to identify and classify the presence of intrusions in accomplishing security in IoT-assisted UAV networks. In the MMPADL-ID technique, the feature selection process is performed by the design of MMPA. In addition, the MMPADL-ID technique incorporates the Elman neural network (ENN) model for the recognition and classification of the intrusions. Furthermore, the honey badger algorithm (HBA) can be applied for the hyperparameter tuning of the ENN model and results in improved performance. The simulation value of the MMPADL-ID technique can be tested on benchmark datasets. An extensive comparative outcome reported the better solution of the MMPADL-ID algorithm with existing approaches for various aspects.

Deep neural networks (DNNs) have proven highly effective in various computational tasks, but their success depends largely on access to large datasets with accurate labels. Obtaining such data may be challenging and costly in real-world scenarios. Common alternatives, such as the use of search engines and crowdsourcing, often result in datasets with inaccurately labeled, or “noisy,” data. This noise may significantly reduce the ability of DNNs to generalize and maintain reliability. Traditional methods for learning with noisy labels mitigate this drawback by training DNNs selectively on reliable data, but they often underutilize available data. Although data augmentation techniques are useful, they do not directly solve the noisy label problem and are limited in such contexts. This paper proposes a confidence-guided Mixup named ConfidentMix, which is a data augmentation strategy based on label confidence. Our method dynamically adjusts the intensity of data augmentation according to label confidence, to protect DNNs from the detrimental effects of noisy labels and maximize the learning potential from the most reliable portions of the dataset. ConfidentMix represents a unique blend of label confidence assessment and customized data augmentation, and improves model resilience and generalizability. Our results on standard benchmarks with synthetic noise, such as CIFAR-10 and CIFAR-100, demonstrate the superiority of ConfidentMix in high-noise environments. Furthermore, extensive experiments on Clothing1M and mini-WebVision have confirmed that ConfidentMix surpasses state-of-the-art methods in handling real-world noise.

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.

Hyperspectral remote sensing image analysis employing deep learning (DL) models has consistently demonstrated remarkable performance, owing to their robust nonlinear modeling and end-to-end optimization capabilities. Notably, the capsule neural network (CapsNet) has attracted substantial attention for its proficient feature extraction capabilities. However, it tends to overlook the inherent spatial heterogeneity within patch features. In this paper, we introduce a spatial attention-based deep convolutional capsule network (SA-CapsNet) to enhance CapsNet’s performance in hyperspectral image (HSI) classification. The incorporation of a more potent and light-weight spatial attention mechanism introduces diversity among neighboring pixels. Additionally, we enhance the stability of learned spectral-spatial features by implementing a convolutional capsule layer that extends dynamic routing with 3D convolution. Experimental results conducted on three commonly used hyperspectral datasets demonstrate that SA-CapsNet outperforms conventional and state-of-the-art DL-based HSI classification algorithms in terms of classification accuracy and computational efficiency.

Name Entity Recognition (NER) aims to recognize entities in the engine room domain from unstructured engine room domain text. But in the engine room domain, the entities are diverse and complex, and there is a nesting phenomenon, resulting in a low entity recognition rate. In this paper, a deep learning method incorporating language models is proposed to enhance the entity recognition performance within the engine room. domain. Firstly, the Bidirectional Encoder Representation from Transformers (BERT) language model is employed to train text feature extraction, acquiring a matrix of vector representations at the word level. Secondly, the trained word vectors are fed into the Bidirectional Gated Recurrent Unit (BiGRU) for contextual semantic entity feature extraction. Finally, the global optimal sequence is extracted by combining with the Conditional Random Field (CRF) model to obtain the named entities in the ship cabin semantics. The experimental results show that the proposed algorithm can obtain better F1 values for all three types of entity recognition. Compared with BERT-BiGRU, the overall accuracy of entity identification, recall rate and F1 value are improved by 1.35%, 1.45% and 1.40%, respectively.

Residential energy consumption is rapidly increasing every year due to demographic and behavioral changes, such as the rising population and the adoption of work-from-home post-COVID-19. High energy consumption emits a substantial amount of carbon dioxide and other Greenhouse Gases, contributing to global warming. It becomes crucial to accurately predict residential load. To enable smart home electricity consumption control, as well as efficient generation, planning, and usage, we predict household energy consumption at very short-term, short-term, and medium-term forecast levels using univariate and multivariate time series data. This study assesses the impact of different household units (water heater and air conditioning), areas (kitchen, laundry, office, living room, bathroom, ironing room, teenager room, and parents’ room), and time (i.e., hour, day, and month) on energy consumption. Comparative analysis and numerical experimental results between the most used approaches, Support Vector Regression and Long Short-Term Memory, reveal that the former outperforms the latter across all forecast levels using different datasets. The findings of this paper will be useful to energy companies and household owners in enhancing energy efficiency and earning carbon credits by reducing the emission of carbon dioxide and other Greenhouse Gases.

PD control is a widely used model-free method; however, it often falls short of guaranteeing optimal performance. Optimal model-based control, such as the Linear Quadratic Regulator (LQR), can indeed achieve the desired control performance, but only for known linear systems. In this paper, we present a novel approach for designing optimal PD control for unknown mechanical systems. We utilize a conditional Generative Adversarial Network (GAN) and a Long Short-Term Memory (LSTM) neural network to approximate an optimal PD control. We employ Bayesian inference to generate PD control that can be applied at different operating points. This design mechanism ensures both stability and optimal performance. Finally, we apply this control methodology to lower limb prostheses, and the results demonstrate that the optimal PD control, using GAN and Bayesian inference, outperforms other classical controllers.

The recently proposed learning-to-learn iterative search algorithm (LISA) for multiple-input multiple-output (MIMO) detection can achieve excellent BER performance for low-order modulation signals, such as BPSK/QPSK, but its performance is degraded as the modulation order increases. This limits its application in high data rate transmission. Motivated by this observation, this paper proposes a decomposition-ordering-based LISA (DO-LISA) to improve the accuracy of LISA for higher-order QAM signals. Firstly, the detection of high-order QAM signal is transformed into the detection of QPSK signal to exploit the facility of binary classification. To further exploit the sequential features of a bidirectional long short-term memory (BiLSTM) network, channel ordering is adopted to enhance the detection accuracy. Moreover, to make the proposed method more practical, low-rank decomposition and channel pruning are adopted to accelerate and compress the neural network. Numerical experiments show that the proposed method can achieve quasi-ML BER performance. Compared with the full model, the compact model achieves $2.37\times $ compression ratio and $2.38\times $ acceleration ratio with limited accuracy loss. In addition, the proposed DO-LISA shows its good generalization ability by preserving very good BER performance both in the correlated channel condition and imperfect channel condition.

In this paper we delve into the properties of transformers, attained through self-supervision, in the point cloud domain. Specifically, we evaluate the effectiveness of Masked Autoencoding as a pretraining scheme, and explore Momentum Contrast as an alternative. In our study we investigate the impact of data quantity on the learned features, and uncover similarities in the transformer’s behavior across domains. Through comprehensive visualizations, we observe that the transformer learns to attend to semantically meaningful regions, indicating that pretraining leads to a better understanding of the underlying geometry. Moreover, we examine the finetuning process and its effect on the learned representations. Based on that, we devise an unfreezing strategy which consistently outperforms our baseline without introducing any other modifications to the model or the training pipeline, and achieve state-of-the-art results in the classification task among transformer models.

In the service industry, some operations require preparatory measures and cannot deliver immediate and reliable services. Identifying potential clients in densely populated environments presents a significant challenge for improving service efficiency. Current methodologies used to predict pedestrian trajectories demonstrate suboptimal performance in service-oriented contexts. In response to these challenges, this paper introduces “POI-GAN”, a novel approach tailored to forecasting pedestrian trajectories in service-centric settings. POI-GAN devises an interest point model that conceptualizes service locations within the scene as distinct obstacles. These service locations are subsequently characterized employing a social force model. Additionally, the framework introduces a field of view angle model which filters interactions between dynamic and static objects in the scene to establish plausibility. Subsequently, a generative model is used to produce projected pedestrian trajectories for future time frames. Empirical evaluations highlight the effectiveness of POI-GAN in improving trajectory prediction, particularly in scenarios with multiple interest points in the scene. Notably, POI-GAN exhibits superior performance when measured against analogous methods, as evidenced by the improved Average Displacement Error (ADE) and Final Displacement Error (FDE) metrics. This innovative approach empowers service providers with the capacity to effectively discern potential customers within the scene, ultimately elevating the quality of service delivery.

Interpreting machine learning models is facilitated by the widely employed locally interpretable model-agnostic explanation (LIME) technique. However, when extending LIME to signal data, its credibility falters due to perturbation techniques used to generate local datasets. These techniques disrupt temporal dependencies among features, leading to unrealistic data points and potentially misleading explanations. Additionally, LIME faces instability and local fidelity issues, limiting its suitability for real-world applications. The absence of a dedicated LIME package tailored for interpreting signal data further diminishes comprehensibility, especially when applied to models trained on such data. In this paper, we introduce Signal-based LIME (Sig-LIME) to address these limitations. Sig-LIME leverages a novel data generation technique that captures temporal dependence among features, enhancing credibility and stability. It combines a random forest model and heatmaps to provide illuminating explanations for predictions drawn from electrocardiogram (ECG) signals, improving model transparency. Empirical findings underscore the enhanced interpretability and comprehension of model predictions attained by Sig-LIME compared to baseline LIME. Our quantitative evaluation based on an analysis of variance (ANOVA) framework, reveals a notable improvement in stability with Sig-LIME, evidenced by an f-statistic of 0.0 and p-values of 1, indicating a complete absence of variation between multiple runs. Regarding local fidelity, Sig-LIME surpasses the baseline LIME, exhibiting a lower average Euclidean distance of 0.49 compared to 17.24. Sig-LIME excels in generating data more akin to the original, achieving remarkable stability and significantly enhancing credibility and local fidelity in the explanations it generates.

As Long Short-Term Memory (LSTM) accelerators are increasingly being employed in safety-critical applications with high-reliability demands, protecting them against errors becomes imperative. Traditional protection techniques for LSTMs are either costly, conflict with strict area and power constraints for neural network accelerators, or introduce performance overhead that is untenable for real-time and latency-critical accelerators. In this paper, we propose residue-based Concurrent Error Detection (CED) schemes to detect transient faults and alleviate their impact during LSTM computations. CED units are employed in a coarse-grain or fine-grain fashion, depending on the granularity of the components being protected from the whole LSTM computations to individual processing elements. In pursuit of a more cost-effective strategy, we use fine-grain CEDs in a selective manner based on the resilience characteristics of LSTM. The selections are made spatially for LSTM synaptic weights or temporally based on LSTM time steps. The experimental results show that the proposed residue-based CEDs (i) can achieve nearly complete fault coverage even under extremely large bit error rates, (ii) significantly decrease misprediction rates compared to the unprotected LSTM, and (iii) incur low overhead without compromising performance. Our method is compared with modular redundancy techniques such as DMR and TMR (Double and Triple Modular Redundancy).

The instance segmentation of overlapping cells in smear images of epithelial cells is challenging due to the significant overlap and adhesion between the cells’ translucent cytoplasm. In this paper, an improved single-stage instance segmentation network called VoVNet-BiFPN-SOLO (VB-SOLO) is proposed to address this problem. The model takes SOLOv2 model as its main frame. Firstly, the backbone network uses Efficient Channel Attention (ECA) to optimize the VoVNetv2 network to increase the information interaction across channels and enhance the extraction of cell instance features. Secondly, the bi-directional feature pyramid network (BiFPN) is introduced to connect with the new backbone. BiFPN can achieve the weighted fusion of features with different resolutions from bottom to top and keep more shallow semantic information in the network. Finally, the Convolutional Block Attention Module (CBAM) is added to the mask branch to improve cell segmentation results in feature maps. Experimental results on the publicly available datasets CISD and Cx22 demonstrate the effectiveness of the VB-SOLO model, achieving a DCP of 0.966 and 0.940 and a FNRO of 0.055 and 0.03. Compared to the original SOLOv2 algorithm, the proposed method achieved improvements in DCP of 1.3% and 1.1% respectively. Additionally, comparative tests with multiple instance segmentation networks have shown that the proposed improved network can achieve a better balance between segmentation accuracy and efficiency. The experimental results demonstrate the effectiveness of the proposed network improvements and the potential of single-stage instance segmentation networks in overlapping cell image segmentation.

Brain stroke is a complicated disease that is one of the foremost reasons of long-term debility and mortality. Because of breakthroughs in Deep Learning (DL) and Artificial Intelligence (AI) which enable the automated detection and diagnosis of brain stroke as well as intelligently assisting post-brain stroke patients for rehabilitation, is more favorable than a manual diagnosis. Many publications on automated brain stroke detection, diagnosis, and robotic management using DL and AI approaches are now being published. This review provides a study of the detection, diagnosis of brain stroke and robotic management techniques of post-brain stroke rehabilitation from six different perspectives, namely, brain stroke datasets and modalities of brain stroke data collection, pre-processing approaches, DL-based detection and diagnosis of brain stroke, Al-based intelligent post brain stroke rehabilitation assistant, and performance measures. It also examines the conclusions and the consequences of the findings. There are also three ongoing research challenges in the fields of brain stroke detection and diagnosis, as well as post-brain stroke robotic treatment. For this investigation, 130 key papers from the Scopus, PubMed and Web of Science archives were chosen after a comprehensive screening method. This study gives a comprehensive overview of brain stroke detection and post-brain stroke robotic management strategies that may be useful to the scientist’s community working in the field of automatic brain stroke detection and robotic rehabilitation management.

Facial paralysis refers to the abnormal behavior of facial muscles caused by a disorder of the facial nerve, mainly manifested as facial asymmetry. In recent years, deep learning has found extensive applications in facial paralysis detection research. However, most existing methods are constrained to assessing the severity of facial paralysis, thereby concealing crucial symptoms within black-box models. Compared to the severity of facial paralysis, the symptoms of facial paralysis are of greater significance to both physicians and patients. To address this issue, this paper proposes a facial paralysis symptom detection model based on facial action units (AUs). To enhance the accuracy of AU intensity prediction, a novel Difference Ensemble Method (DEM) is introduced. This method leverages differential information between frames within the same video to improve the accuracy of predictions for the current frame. Building upon the predicted AU intensity sequences for keyframes in a video, an interpretable model for detecting facial paralysis symptoms is designed. This model employs an active means to describe the asymmetry in facial muscle strength and utilizes co-occurrence matrices to detect synkinesis. It is noteworthy that DEM is exclusively trained on a dataset of normal faces but exhibits excellent performance when transferred to a facial paralysis dataset. Additionally, DEM exhibits higher accuracy in predicting AU intensity compared to existing methods. The F1 scores for detecting facial muscle function in the eyebrow, eye, and mouth regions with our proposed model are 80.0%, 79.23%, and 90.91%, respectively. To demonstrate the model’s performance, a synkinesis detection experiment is conducted, further validating its applicability in facial paralysis detection.

While deep neural networks (DNNs) have brought revolutions to many intelligent services and systems, the deployment of high-performing models for real-world applications faces challenges posed by resource constraints and diverse operating environments. While existing methods such as model compression combined with inference accelerators have enhanced the efficiency of deep neural networks, they are not dynamically adaptable to dynamically changing resource conditions since they provide static accuracy-efficiency trade-offs. Further, since they are not aware of performance requirements, such as desired inference latency, they are not able to provide robust and effective performance under unpredictable workloads. This paper introduces a holistic solution to address this challenge, consisting of two key components: adaptive depth neural networks and the Quality of Service (QoS)-aware inference accelerator. The adaptive depth neural networks exhibit the ability to scale computation instantly with minimal impact on accuracy, utilizing a novel architectural pattern and training algorithm. Complementing this, the QoS-aware inference accelerator employs a feedback control loop, adapting network depth dynamically to meet desired inference latency. Experimental results demonstrate that the proposed adaptive depth networks outperform non-adaptive counterparts, achieving up to 38% dynamic acceleration via depth adaption, with a marginal accuracy loss of 1.5%. Furthermore, the QoS-aware inference accelerator successfully controls network depth at runtime, ensuring robust performance in unpredictable environments.

The sports health evaluation has been a technical concern in modern society, and massive mobile big data can be employed as the resource support for this purpose. In this context, this paper presents design of a mobile big data processing-based sports health evaluation system, with adopting the graph neural network model as the algorithm part. First of all, a Zigbee-based mobile big data integration platform is designed for sports health demand. Then, a graph neural network model is developed and embedded into such platform to suggest evaluation results according to collected sprots big data. Inside the whole designed system, the collaborative workflow between business support part and algorithmic analysis part is defined accordingly. Finally, some simulation conditions are set to make assessment for the designed technical framework with introduction of real-world sports big data. The obtained simulation results demonstrate the effectiveness and reliability of the designed evaluation system. The proposal can possess both proper big data processing efficiency and proper system operation performance. Under the designed big data computing framework, a sports big data processing platform has been implemented to meet the technical requirements of upper level applications for data storage, management, and analysis. It provides theoretical support for the development and progress of sports.

Accurate root zone soil moisture (RZSM) estimation is essential for precision irrigation (PI) systems that seek to optimize water use efficiency. Large-scale in-situ sensors for direct measurement are costly, while existing satellites lack depth resolution for direct RZSM data. Hence, in-direct RZSM estimation methods are required. Literature illustrates that RZSM at a location is related to changing soil-water-plant characteristics. Therefore, these characteristics can provide auxiliary information on RZSM changes. By leveraging auxiliary information derived from changing soil-water-plant characteristics, this paper enables indirect RZSM estimation at non-sensor locations, effectively addressing the limitations inherent in direct RZSM measurement techniques initially discussed. Compared to existing methods, deep learning (DL) is most suitable for such data associations as they are auto-tuned to extract relative relationships from diverse big data. Among DL models, sequential models are apt for finding these relationships as all these variables are time-series sequences. The transformer neural network (TNN) is the state-of-the-art DL model for analyzing sequences. However, for in-direct RZSM estimation, the data associations within a location and multiple sensor sites with the target need to be found. Conventional TNN cannot incorporate such simultaneous multi-associations, hence, we develop a new TNN model called the hybrid TNN model, which is able to facilitate the capturing of complicated dependencies through thoughtful feature selection and engineering. First, sensor locations exhibiting analogous downscaled 1-km satellite soil moisture (SM) are identified. Next, a dynamic multilayer perceptron (D-MLP) network discerns highly correlated auxiliary-RZSM data, utilizing both ground-based and downscaled satellite data. Following this, the dual attention module identifies essential multi-associations, leveraging selected sensor and target region information. Finally, the Bayesian layer averages multi-location RZSM using the conditional probability generated based on relative relationships to yield the target location RZSM estimate. Our proposed model shows 13.066% better RZSM estimation compared to popular sequential models. The hybrid TNN RZSM estimates are used to monitor root water depletion tolerance levels for optimal PI schedules, which shows 10.846% water and 10.339% cost-saving on our selected sites. Overall the proposed model effectively demonstrates that a more accurate PI predictive algorithm saves water, improves resource conservation, and reduces irrigation costs.

Electrical Impedance Tomography (EIT) has emerged as a valuable medical imaging modality, which visualizes the conductivity distribution of a subject by performing multi-electrode impedance measurements. EIT finds applications in monitoring lung and cardiac function, brain imaging and the detection of malignant tissues. Its mobility, outstanding temporal resolution and the absence of ionizing radiation make it particularly suitable for repetitive real-time monitoring and diagnostics, especially in radiation-sensitive populations, such as neonates. This paper presents a methodological review of EIT image reconstruction approaches spanning from traditional linear regularization and back-projection to more recent techniques, including deep learning, sparse Bayesian learning and non-linear shape-driven reconstruction. Linear and non-linear reconstruction approaches are distinguished, as well as time, frequency difference and absolute reconstruction ones. The exposition includes a concise elaboration of the methodologies’ mathematical foundations and algorithmic deployment, with particular attention to recent advancements. For each approach, an assessment of its merits and drawbacks is given, providing implementation considerations, imaging performance and relevant applications.

Generative Artificial Intelligence (GAI) has sparked a transformative wave across various domains, including machine learning, healthcare, business, and entertainment, owing to its remarkable ability to generate lifelike data. This comprehensive survey offers a meticulous examination of the privacy and security challenges inherent to GAI. It provides five pivotal perspectives essential for a comprehensive understanding of these intricacies. The paper encompasses discussions on GAI architectures, diverse generative model types, practical applications, and recent advancements within the field. In addition, it highlights current security strategies and proposes sustainable solutions, emphasizing user, developer, institutional, and policymaker involvement.

With the rapid progress of computer vision and deep learning techniques, accurately predicting continuous human motions from very few image inputs and generating high-quality 3D human models has become a cutting-edge research direction in this field. Despite the achievements in 2D to 3D conversion techniques, it is still a great challenge to capture coherent movements from limited image frames and generate texture-rich 3D models. In this paper, we propose a 3D clothed human body generation method based on Inter-Frame Motion Prediction (IFMP for short) of 2D images, which is capable of not only predicting a series of coherent human body motions, but also reconstructing a detailed textured 3D human body model from only two image frames. The method automatically focuses on key parts of the image through action coding and uses a conditional generative adversarial network to generate a series of consecutive intermediate frame images. A depth-aware implicit function representation is combined to map a 3D model from the 2D image, and high quality textures of the human body in a clothed state are obtained by texture mapping and model detail enhancement methods. Finally, the experimental results validate the advantages of the IFMP method in image action coherence prediction, as well as verifying the effectiveness of the human 3D model generated by the method in terms of geometric accuracy and texture quality.