Search

Acoustic holography (AH) is a technique with significant potential in realms, such as biomedicine, industry, and augmented reality. The implementation of acoustic holograms can be realized by a passive approach or active ones. Although the passive approach (by a 3D printer) can achieve high-quality acoustic field generation, it is constrained by high manufacturing costs and static field control. On the contrary, the active approach with a phased transducer array (PTA) as the latest technique stands out since it supports dynamic, flexible, and reconfigurable acoustic field generation. However, current PTA-based AH techniques face the drawback of inferior acoustic field fineness due to the Spatial Bandwidth Product (SBP) limit of PTA, which hinders the application of PTA in precise tasks, such as neural electrodes and microfluidics control. To address this issue, we propose a super-resolution acoustic holography (SRAH) method inspired by the concept of super-resolution in ultrasonic imaging and computer vision, by which we can generate acoustic fields reaching the physical diffraction limit of acoustic waves regardless SBP of PTA. In other words, this method enables high-SBP acoustic field generation with low-SBP PTA. The method is based on self-supervised learning, integrating a generative adversarial network and a physical model of acoustic wave propagation, specifically the linear accumulation method. Both simulation and experimental results demonstrate that the proposed method can generate high-fidelity acoustic fields suitable for intricate tasks with low-SBP PTA. Moreover, the performance of the algorithm improves as the target SBP increases. Therefore, the proposed SRAH method shows great potential for applications requiring elaborate manipulation. [ABSTRACT FROM AUTHOR]

Using the computer vision method, this study proposes an analytical model of visual aesthetics for brand communication and analyzes the effects of visual features (i.e., colors and visual complexity) on brand personality. This study illustrates a four-step procedure correlating computationally coded visual attributes with human ratings of perceived brand personality. This study has important methodological implications for advertising researchers and practitioners. [ABSTRACT FROM AUTHOR]

Brand-related content cocreated by consumers can play a crucial role in brand–consumer interactions and provide brands with valuable insights hidden in vast seas of unstructured data. We propose and evaluate a framework integrating a social network approach and scalable automated content analysis of texts and visuals for studying brand-related communication on social media. To illustrate the proposed approach, we use Twitter content related to two brands: Barclays and Sierra Club. By applying network clustering algorithms we identify different types of organically emerging communities around brands. Cluster-specific diffusion leaders are identified using their in-degree centrality values. To examine the unique characteristics of brand-related content within each cluster, we apply and assess the accuracy of popular off-the-shelf solutions for text and image analysis, also known as application programming interfaces (APIs). Of six sentiment analysis solutions, only one shows acceptable reliability levels. For computer vision APIs, we first identify labels that have unclear or imprecise meaning and calculate accuracy levels, resulting in acceptable accuracy levels for four of the five APIs. We discuss conceptual and practical implications of this integrative approach and of the technological hurdles that these popular automated content analysis applications pose. [ABSTRACT FROM AUTHOR]

This project aims to mitigate the critical decline in bee populations, essential for crop pollination and food security. With a shortage of taxonomic specialists to identify the vast array of bee species, the project's goal is to enhance the monitoring of population changes through an automated classification system. Utilizing a dataset of bee wing images, the project aims to develop a computational pipeline to identify species based on their unique wing vein patterns. This approach not only supports bee conservation efforts but also expands our understanding of complex geometric variations in nature, offering wider applications in biological research. This poster was presented at the UCSB Data Science Capstone showcase in 2024.

The erosion of high-end fashion brands by fast-fashion copycats (e.g., Zara, H&M) has stirred controversies and unceasing legal attempts to copyright fashion designs. Despite the purported negative impact of copycats, the effect of fashion copycats on high-end brands remains empirically unclear. Research on this topic has been impeded by the absence of a modeling framework to quantify fashion and by the lack of consumer-level data on fashion choices. The authors collect data on the posting behaviors of consumers on a fashion-specific social media platform and develop a dynamic structural model with deep learning image analytics to characterize consumers' choices of brands and styles. Results suggest that fast-fashion copycats can both harm high-end brands (a cannibalization effect) and help them (a market expansion effect). The authors also identify both static and dynamic mechanisms that contribute to the market expansion effect: The affordability of mixing copycats with high-end brands boosts the number of high-end items featured in posts by financially constrained consumers (a static mechanism). In addition, good styles from copycats enable users to build their popularity on social media over time, which may increase the users' valuation of high-end brands and reduce the users' future costs via sponsorship opportunities (dynamic mechanisms). The results could inform policy makers about the potential consequences of prohibiting fashion copycats. [ABSTRACT FROM AUTHOR]

Background and purposeMotion management is essential to reduce normal tissue exposure and maintain adequate tumor dose in lung stereotactic body radiation therapy (SBRT). Lung SBRT using an articulated robotic arm allows dynamic tracking during radiation dose delivery. Two stereoscopic X-ray tracking modes are available - fiducial-based and fiducial-free tracking. Although X-ray detection of implanted fiducials is robust, the implantation procedure is invasive and inapplicable to some patients and tumor locations. Fiducial-free tracking relies on tumor contrast, which challenges the existing tracking algorithms for small (e.g., 15 mm) and tumor locations (with/without thoracic anatomy overlapping).ResultsOn X-ray views that conventional methods failed to track the lung tumor, Deep Match achieved robust performance as evidenced by >80 % 3 mm-Hit (detection within 3 mm superior/inferior margin from ground truth) for 70 % of patients and

The increasing use of transportation network companies and delivery services has transformed the utilization of curb space, resulting in a lack of parking and contributing to congestion. No systematic method exists for identifying curb usage patterns, but emerging machine learning technologies and low-tech data sources, such as dashboard cameras mounted on vehicles that routinely travel the area, have the potential of monitoring curb usage. To demonstrate how video data can be used to recognize usage patterns, we conducted a case study on Bancroft Way in Berkeley, CA. The project collected video footage with GPS data from a dashboard camera installed on a shuttle bus that circles the area. We trained a machine learning model to recognize different types of delivery vehicles in the data images, and then used the model to visualize curbside usage trends. The findings include identifying hot spots, analyzing arrival patterns by delivery vehicle type, detecting bus lane blockage, and assessing the impact of parking on traffic flow. The proof-ofconcept study demonstrated that machine learning techniques, when coupled with affordable hardware like a dashboard camera, can reveal curb usage patterns. The data can be used to efficiently manage curb space, facilitate goods movement, improve traffic flow, and enhance safety.

Summary: Anomaly event recognition and identification has a crucial part in several areas, particularly in night vision environments. Conventional techniques of event recognition are hugely based upon data extracted from certain images for classification purposes. This needs users to select suitable features to establish the feature depictions for actual images per definite situations. Manual feature selection is laborious as well as heuristic tasks and the features obtained in this manner generally have worse robustness. Here, a Faster Region‐based Convolutional fused Social Generative Adversarial Network (FRC‐SGAN) is designed for anomaly event recognition in a night vision environment. At the cloud, key frame extraction, pre‐processing, feature extraction, human detection (HD) and anomalous event recognition are carried out. Initially, input video from the database is subjected to perform pre‐processing. The visibility enhancement is utilized for pre‐processing. Thereafter, features like ResNet features, texture features and statistical features are extracted. Then, HD is accomplished by DeepJoint segmentation with chord distance. Finally, anomalous detection is done by FRC‐SGAN that is the incorporation of Fast Regional Convolutional Neural Network (FR‐CNN) and Social Generative Adversarial Network (SGAN). In addition, FRC‐SGAN acquired 90.8% of accuracy, 89.7% of precision, and 89.2% of recall. [ABSTRACT FROM AUTHOR]

Owing to time and scene constraints, a significant number of sectional maps exist in paper form. These maps contain a vast amount of data and hold high information value. However, they often suffer from issues such as annotations, stains, deformation, and missing content during preservation. Traditional processing methods require a large amount of manual image registration, which is extremely inconvenient. In this study, a map image labeling program is designed using OpenCV to prepare a map image dataset, and the U2Net-p algorithm for map segmentation is trained on this dataset. Furthermore, a comprehensive method for automatically merging sectional maps is designed and implemented, which can repair and process sectional maps and seamlessly integrate them into target grids according to map sheet numbering rules. This method has been applied to the production of base maps for natural resource demarcation projects, achieving a stitching accuracy of 96.67% on marked anchor points and considerably improving processing speed. This indicates that our approach has broad application value in the field of automatic stitching and fusion of sectional map images. [ABSTRACT FROM AUTHOR]

This paper explores the application of Visual Question-Answering (VQA) technology, which combines computer vision and natural language processing (NLP), in the medical domain, specifically for analyzing radiology scans. VQA can facilitate medical decision-making and improve patient outcomes by accurately interpreting medical imaging, which requires specialized expertise and time. The paper proposes developing an advanced VQA system for medical datasets using the Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (BLIP) architecture from Salesforce, leveraging deep learning and transfer learning techniques to handle the unique challenges of medical/radiology images. The paper discusses the underlying concepts, methodologies, and results of applying the BLIP architecture and fine-tuning approaches for VQA in the medical domain, highlighting their effectiveness in addressing the complexities of VQA tasks for radiology scans. Inspired by the BLIP architecture from Salesforce, we propose a novel multi-modal fusion approach for medical VQA and evaluating its promising potential. [ABSTRACT FROM AUTHOR]

Ensuring environmental safety and regulatory compliance at Department of Energy (DOE) sites demands an efficient and reliable detection system for low-level nuclear waste (LLW). Unlike existing methods that rely on human effort, this paper explores the integration of computer vision algorithms to automate the identification of such waste across DOE facilities. We evaluate the effectiveness of multiple algorithms in classifying nuclear waste materials and their adaptability to newly emerging LLW. Our research introduces and implements five state-of-the-art computer vision models, each representing a different approach to the problem. Through rigorous experimentation and validation, we evaluate these algorithms based on performance, speed, and adaptability. The results reveal a noteworthy trade-off between detection performance and adaptability. YOLOv7 shows the best performance and requires the highest effort to detect new LLW. Conversely, OWL-ViT has lower performance than YOLOv7 and requires minimal effort to detect new LLW. The inference speed does not strongly correlate with performance or adaptability. These findings offer valuable insights into the strengths and limitations of current computer vision algorithms for LLW detection. Each developed model provides a specialized solution with distinct advantages and disadvantages, empowering DOE stakeholders to select the algorithm that aligns best with their specific needs. [ABSTRACT FROM AUTHOR]

Recent deep learning-based object detection approaches have led to significant progress in multi-object tracking (MOT) algorithms. The current MOT methods mainly focus on pedestrian or vehicle scenes, but basketball sports scenes are usually accompanied by three or more object occlusion problems with similar appearances and high-intensity complex motions, which we call complex multi-object occlusion (CMOO). Here, we propose an online and robust MOT approach, named Basketball-SORT, which focuses on the CMOO problems in basketball videos. To overcome the CMOO problem, instead of using the intersection-over-union-based (IoU-based) approach, we use the trajectories of neighboring frames based on the projected positions of the players. Our method designs the basketball game restriction (BGR) and reacquiring Long-Lost IDs (RLLI) based on the characteristics of basketball scenes, and we also solve the occlusion problem based on the player trajectories and appearance features. Experimental results show that our method achieves a Higher Order Tracking Accuracy (HOTA) score of 63.48% on the basketball fixed video dataset (5v5) and 66.45% on the 3v3 basketball dataset. These results outperform other recent popular approaches on both datasets, demonstrating the robustness of our method across different basketball game formats. Overall, our approach solved the CMOO problem more effectively than recent MOT algorithms. [ABSTRACT FROM AUTHOR]

In modern society, yoga is one of the popular types of exercise. To do this exercise, there is a need for training on how to do accurate yoga postures. This is because performing improper yoga postures results in tiredness and injury. Therefore consuming a guide is crucial. An artificial intelligence (AI) based system aids in recognizing poses of yoga and offers ideas to users. Most of the researchers proposed computerized methods for assessing exercise-related actions like handball strikes, volleyball, jumping, football player positions, races, and other physical events. Automated Yoga Pose Recognition is nothing but a computer vision as well as a machine learning (ML) model that goals to recognize and categorize yoga postures that are performed by people depending on input from images or video data. Leveraging innovative techniques namely image processing and deep learning (DL) is probable to help instructors, practitioners and fitness fans in guaranteeing appropriate pose positions and procedures during yoga sessions. In addition, it helps to improve overall excellence of practice and in injury prevention as well as progression tracking. This study develops an automated yoga pose recognition using Enhanced Chicken Swarm Optimization with deep learning (AYPR-ECSODL) technique. In the pre-processing stage, a Wiener Filter (WF) is applied for noise removal, enhancing image quality by reducing unwanted artefacts. Besides, Dynamic Histogram Equalization (DHE) is employed for contrast enhancement, ensuring that relevant features within the image are highlighted effectively. Besides, the segmentation step is executed through the YOLOv8 model. Feature extraction is accomplished via OpenPose, which extracts essential human body key points that serve as the foundation for further analysis. The classification stage employs a Self-Attention Based Gated Recurrent Unit (SA-GRU) model. To optimize model performance, parameter tuning was performed using the ECSO method. The projected technique was validated on the yoga posture image classification dataset from the Kaggle repository. The experimental values stated that developed method properly categorizes different types of yoga poses such as adho mukha svanasana, anantasana, ardha uttanasana, bakasana, bitilasana, camatkarasana, dhanurasana, gomukhasana, padmasana, purvottanasana, sukhasana, utthita, parsvakonasana, vajrasana, virabhadrasana, and vrischikasana. [ABSTRACT FROM AUTHOR]

Millets, often referred to as "Nutri Cereals," hold significant potential for enhancing future food availability. Over the past 50 years, advancements such as the use of fertilizers, insecticides, and controlled irrigation have markedly increased the productivity of pearl millet crops worldwide. According to the 2020 review report by the Directorate of Millets Development, 693 million hectares of agricultural land are dedicated to pearl millet cultivation. However, managing crop diseases remains a substantial challenge for farmers. Effective crop disease management is crucial for sustainable agriculture, yet traditional algorithms often struggle with accurate detection and computational inefficiency especially in the case of golden cereal termed as "Pearl Millet". Our study introduces a novel approach for disease detection in pearl millet crops using aerial imagery, named Split-Attention Networks with Disentangled Nonlocal and Edge Supervision (SANDNES). This method leverages advanced neural network architectures to enhance disease detection accuracy and efficiency. Our comprehensive numerical evaluations demonstrate that SANDNES outperforms traditional U-Net and U-Net + Nonlocal baseline models Comprehensive evaluations demonstrate SANDNES' exceptional performance, achieving an impressive average F1 score of 89.04%, significantly surpassing other methods in accuracy and efficiency. Specifically, SANDNES achieved notable performance improvements, with F1 scores of 94.88% for healthy crops, 85.96% for honeydew, 82.97% for smut, and 92.69% for ergot. This method leverages split-attention mechanisms for precise feature localization, with Disentangled Nonlocal and Edge Supervision enhancing contextual reasoning and boundary delineation. SANDNES superior computational efficiency and high accuracy make it suitable for real-time precision agriculture applications. [ABSTRACT FROM AUTHOR]

In many sports, player re-identification is crucial for automatic video processing and analysis. However, most of the current studies on player re-identification in multi- or single-view sports videos focus on re-identification in the closed-world setting using labeled image dataset, and player re-identification in the open-world setting for automatic video analysis is not well developed. In this paper, we propose a runner re-identification system that directly processes single-view video to address the open-world setting. In the open-world setting, we cannot use labeled dataset and have to process video directly. The proposed system automatically processes raw video as input to identify runners, and it can identify runners even when they are framed out multiple times. For the automatic processing, we first detect the runners in the video using the pre-trained YOLOv8 and the fine-tuned EfficientNet. We then track the runners using ByteTrack and detect their shoes with the fine-tuned YOLOv8. Finally, we extract the image features of the runners using an unsupervised method with the gated recurrent unit autoencoder and global and local features mixing. To improve the accuracy of runner re-identification, we use shoe images as local image features and dynamic features of running sequence images. We evaluated the system on a running practice video dataset and showed that the proposed method identified runners with higher accuracy than some state-of-the-art models in unsupervised re-identification. We also showed that our proposed local image feature and running dynamic feature were effective for runner re-identification. Our runner re-identification system can be useful for the automatic analysis of running videos. [ABSTRACT FROM AUTHOR]

Breast cancer (BC) is a type of cancer which progresses and spreads from breast tissues and gradually exceeds the entire body; this kind of cancer originates in both sexes. Prompt recognition of this disorder is most significant in this phase, and it is measured by providing patients with the essential treatment so their efficient lifetime can be protected. Scientists and researchers in numerous studies have initiated techniques to identify tumours in early phases. Still, misperception in classifying skeptical lesions can be due to poor image excellence and dissimilar breast density. BC is a primary health concern, requiring constant initial detection and improvement in analysis. BC analysis has made major progress recently with combining multi-modal image modalities. These studies deliver an overview of the segmentation, classification, or grading of numerous cancer types, including BC, by employing conventional machine learning (ML) models over hand-engineered features. Therefore, this study uses multi-modality medical imaging to propose a Computer Vision with Fusion Joint Transfer Learning for Breast Cancer Diagnosis (CVFBJTL-BCD) technique. The presented CVFBJTL-BCD technique utilizes feature fusion and DL models to effectively detect and identify BC diagnoses. The CVFBJTL-BCD technique primarily employs the Gabor filtering (GF) technique for noise removal. Next, the CVFBJTL-BCD technique uses a fusion-based joint transfer learning (TL) process comprising three models, namely DenseNet201, InceptionV3, and MobileNetV2. The stacked autoencoders (SAE) model is implemented to classify BC diagnosis. Finally, the horse herd optimization algorithm (HHOA) model is utilized to select parameters involved in the SAE method optimally. To demonstrate the improved results of the CVFBJTL-BCD methodology, a comprehensive series of experimentations are performed on two benchmark datasets. The comparative analysis of the CVFBJTL-BCD technique portrayed a superior accuracy value of 98.18% and 99.15% over existing methods under Histopathological and Ultrasound datasets. [ABSTRACT FROM AUTHOR]

In this study, we examine the assessment of surface roughness on turned surfaces of Ti 6Al 4V using a computer vision system. We utilize the Dual-Tree Complex Wavelet Transform (DTCWT) to break down the images of the turned surface into sub-images oriented in directions. Three different methods of feature generation have been compared, i.e., the use of Gray-Level Co-Occurrence Matrix (GLCM) and DTCWT-based extraction of second-order statistical features, DTCWT Image fusion, and the use of GLCM for feature extraction, and DTCWT image fusion using Particle Swarm Optimization (PSO) based GLCM features. Principal Component Analysis (PCA) was utilized to identify and select features. The model was developed using a Radial Basis Function Neural Network (RBFNN). Accordingly, six models were designed based on the three feature generation methods, considering all features and features selected using PCA. The RBFNN model, which incorporates DTCWT Image fusion and utilizes PSO with PCA features, achieved a training data prediction accuracy of 100% and a test data prediction accuracy of 99.13%. [ABSTRACT FROM AUTHOR]

Artificial intelligence and machine learning (AI/ML) can be used to automatically analyze large image datasets. One valuable application of this approach is estimation of plant trait data contained within images. Here we review 39 papers that describe the development and/or application of such models for estimation of stomatal traits from epidermal micrographs. In doing so, we hope to provide plant biologists with a foundational understanding of AI/ML and summarize the current capabilities and limitations of published tools. While most models show human-level performance for stomatal density (SD) quantification at superhuman speed, they are often likely to be limited in how broadly they can be applied across phenotypic diversity associated with genetic, environmental, or developmental variation. Other models can make predictions across greater phenotypic diversity and/or additional stomatal/epidermal traits, but require significantly greater time investment to generate ground-truth data. We discuss the challenges and opportunities presented by AI/ML-enabled computer vision analysis, and make recommendations for future work to advance accelerated stomatal phenotyping. [ABSTRACT FROM AUTHOR]

This study contributes to understanding semi-automated ergonomic risk assessments in industrial manufacturing environments, proposing a practical tool for enhancing worker safety and operational efficiency. In the Industry 5.0 era, the human-centric approach in manufacturing is crucial, especially considering the aging workforce and the dynamic nature of the entire modern industrial sector, today integrating digital technology, automation, and sustainable practices to enhance productivity and environmental responsibility. This approach aims to adapt work conditions to individual capabilities, addressing the high incidence of work-related musculoskeletal disorders (MSDs). The traditional, subjective methods of ergonomic assessment are inadequate for dynamic settings, highlighting the need for affordable, automatic tools for continuous monitoring of workers' postures to evaluate ergonomic risks effectively during tasks. To enable this perspective, 2D RGB Motion Capture (MoCap) systems based on computer vision currently seem the technologies of choice, given their low intrusiveness, cost, and implementation effort. However, the reliability and applicability of these systems in the dynamic and varied manufacturing environment remain uncertain. This research benchmarks various literature proposed MoCap tools and examines the viability of MoCap systems for ergonomic risk assessments in Industry 5.0 by exploiting one of the benchmarked semi-automated, low-cost and non-intrusive 2D RGB MoCap system, capable of continuously monitoring and analysing workers' postures. By conducting experiments across varied manufacturing environments, this research evaluates the system's effectiveness in assessing ergonomic risks and its adaptability to different production lines. Results reveal that the accuracy of risk assessments varies by specific environmental conditions and workstation setups. Although these systems are not yet optimized for expert-level risk certification, they offer significant potential for enhancing workplace safety and efficiency by providing continuous posture monitoring. Future improvements could explore advanced computational techniques like machine learning to refine ergonomic assessments further. [ABSTRACT FROM AUTHOR]

This paper presents the design, implementation, and testing of an advanced conveyor belt surface monitoring system, specifically engineered for harsh and complex industrial environments. The system integrates multiple cutting-edge technologies, including programmable logic controllers (PLC), laser scanning, industrial-grade cameras, and deep learning algorithms, particularly YOLOv7, to achieve real-time, high-precision monitoring of conveyor belt conditions. Key innovations include optimized detection location based on failure modes, advanced PLC integration for seamless automation, and intelligent dust-proof features to maintain accuracy in challenging conditions. Through strategic placement of detection devices and multi-mode control strategies (local, remote, and automatic), the system offers unparalleled adaptability and responsiveness. The system leverages robust data management for trend analysis and predictive maintenance, enhancing operational efficiency. The hardware architecture comprises PLC-based control systems, high-resolution industrial cameras, and laser emitters, while the software features a two-tier structure combining human-machine interaction (HMI) with real-time data processing capabilities. Experimental results show that the system is highly effective in detecting common belt defects such as foreign objects, tears, and shallow scratches, ensuring optimal operational efficiency and minimizing downtime. The system's scalability, robust data management, and adaptability to low-light and dusty conditions make it ideal for deployment in large-scale industrial operations, where continuous monitoring and early fault detection are critical to maintaining productivity and safety. [ABSTRACT FROM AUTHOR]

This work applies the computer vision “You only look once” (YOLO) algorithm to extract bounding boxes around dislocations in weak‐beam dark‐field transmission electron microscopy (WBDF TEM) images of semiconductor thin films. A formula is derived to relate the sum of the relative heights of the bounding boxes to the dislocation densities in the films. WBDF TEM images reported in the literature and taken from our α‐Ga2O3 samples are divided into train, evaluation, and test datasets. Different models are trained using the train dataset and evaluated using the evaluation dataset to find the best confidence values, which are used to select the best model based on the performance against the test data set. For α‐Ga2O3 thin films, dislocation density output by this model is on average ≈58% of those estimated by the traditional Ham method. A factor of 4/π may contribute to the systematic underestimation of the model versus the Ham method. [ABSTRACT FROM AUTHOR]

The phenomenon of overweight vehicles severely threatens traffic safety and the service life of transportation infrastructure. Rapid and effective identification of overweight vehicles is of significant importance for maintaining the healthy operation of highways and bridges and ensuring the safety of people's lives and property. With the problems of high cost and low efficiency, the traditional vehicle weighing systems can only meet some of the requirements of different scenarios. The development of artificial intelligence technologies, especially deep learning, has greatly enhanced the accuracy and efficiency of computer vision. To this end, the paper proposes a method using computer vision and deep learning for the non‐contact identification of overweight vehicles. By constructing two deep learning models and combining them with the vehicle vibration model and relevant specifications, the weight and maximum allowable weight of the vehicle are obtained to make a comparison for determining overweight. Experimental verification was performed using a two‐axle vehicle as an illustrative example, and the results demonstrate that the proposed method exhibits excellent feasibility and effectiveness. It shows significant potential in real‐world scenarios, laying a research foundation for practical engineering applications. Additionally, it provides a reference for the governance and decision‐making of overweight issues for relevant authorities. [ABSTRACT FROM AUTHOR]

Robot roller hemming is an advanced assembly process used to achieve high-precision manufacturing of autobody panels. The coupling action between metal panels and non-Newtonian adhesive, complex contour curve, and coverage of the autobody and hemming parameters all have significant influence on the forming quality of the autobody. In traditional robot hemming process, roller pose and trajectory are totally set at the beginning and incapable to adjust according to the hemming quality, making it difficult to further improve the appearance and assembly accuracy. Therefore, taking dimension deviation of the autobody as research object, an optimization model and algorithm of the robot roller hemming process were proposed, and built a robot hemming platform to analyze the optimization model and algorithm. Firstly, the stress-geometry interactions of roller hemming process were considered, and the mapping model between the dimension deviation and the hemming parameter (TCP-RTP) and roller pose was established based on FEM-SPH numerical simulation, Frenet frame of differential geometry, and Cartesian coordinate transformation theory, then the optimized roller pose was obtained through the mapping model. Secondly, the detection of hemming defects and measurement of dimensional deviation were realized based on the machine vision. Finally, the experimental platform was constructed and the experiment showed that the dimension deviation of the closure panel was reduced and then overshoots, and finally stabilized within 0.1 mm, after adopting the optimization algorithm. [ABSTRACT FROM AUTHOR]

Monocular depth estimation is essential in computer vision and robotics applications, including localization, mapping, and 3D object detection. In recent years, supervised learning algorithms that model large amounts of data have been successful in depth estimation. However, obtaining dense ground truth depth labels remains a challenge in supervised training. Therefore, unsupervised methods trained using monocular image sequences have gained wider attention. However, the depth estimation results of most existing models often produce blurred edges. Therefore, we propose various effective improvement strategies to construct a depth estimation network TL-Depth. (1) We propose a tower connection structure that utilizes convolutional processing to facilitate feature fusion, achieve precise semantic classification of pixels, and yield more accurate depth results. (2) We employ a Laplacian-filtering residual to focus on boundary information and enhance detailed results. (3) During the feature extraction stage, multiple pooling excitations are used by embedding them in the convolutional layer. This reduces redundant information while enhancing the network's feature extraction capability. The experimental results on the KITTI dataset and the Make3D dataset demonstrate that this method achieves good results compared to current methods. [ABSTRACT FROM AUTHOR]

BerryPortraits (Phenotyping of Ripening Traits) is open source Python-based image-analysis software that rapidly detects and segments berries and extracts morphometric data on fruit quality traits such as berry color, size, shape, and uniformity. Utilizing the YOLOv8 framework and community-developed, actively-maintained Python libraries such as OpenCV, BerryPortraits software was trained on 512 postharvest images (taken under controlled lighting conditions) of phenotypically diverse cranberry populations (Vaccinium macrocarpon Ait.) from the two largest public cranberry breeding programs in the U.S. The implementation of CIELAB, an intuitive and perceptually uniform color space, enables differentiation between berry color and berry brightness, which are confounded in classic RGB color channel measurements. Furthermore, computer vision enables precise and quantifiable color phenotyping, thus facilitating inclusion of researchers and data analysts with color vision deficiency. BerryPortraits is a phenotyping tool for researchers in plant breeding, plant genetics, horticulture, food science, plant physiology, plant pathology, and related fields. BerryPortraits has strong potential applications for other specialty crops such as blueberry, lingonberry, caneberry, grape, and more. As an open source phenotyping tool based on widely-used python libraries, BerryPortraits allows anyone to use, fork, modify, optimize, and embed this software into other tools or pipelines. [ABSTRACT FROM AUTHOR]

The quantification of centipede populations is one of the key measures in achieving intelligent management of edible centipedes and promoting the upgrade of the rural centipede industry chain. However, current centipede counting techniques still face several challenges, including low detection accuracy, large model size, and difficulty in deployment on mobile devices. These challenges have limited existing network models to the experimental stage, preventing their practical application. To tackle the identified challenges, this study introduces a lightweight centipede detection model (FCM-YOLO), which enhances detection performance while ensuring fast processing and broad applicability. Based on the YOLOv5s framework, this model incorporates the C3FS module, resulting in fewer parameters and increased detection speed. Additionally, it integrates an attention module (CBAM) to suppress irrelevant information and improve target focus, thus enhancing detection accuracy. Furthermore, to enhance the precision of bounding box positioning, this study proposes a new loss function, CMPDIOU, for bounding box loss. Experimental results show that FCM-YOLO, while reducing parameter size, achieves an improved detection accuracy of 97.4% (2.7% higher than YOLOv5s) and reduces floating-point operations (FLOPs) to 11.5G (4.3G lower than YOLOv5s). In summary, this paper provides novel insights into the detection and enumeration of centipedes, contributing to the advancement of intelligent agricultural practices. [ABSTRACT FROM AUTHOR]

Chinese stroke segmentation is a crucial and challenging task for various downstream applications such as font generation, aesthetic evaluation etc. Conventional semantic segmentation techniques typically face difficulties in accurately segmenting strokes, as intersection regions in Chinese characters can belong to multiple strokes simultaneously, and these approaches often lack a holistic understanding of character composition. This paper proposes a character stroke segmentation framework named Stroke‐Seg that integrates with various semantic segmentation architectures, demonstrating adaptability to different backbone networks to tackle the above tasks. A multi‐label output strategy is proposed to effectively classify strokes in intersection areas, overcoming the limitations of traditional semantic segmentation approaches. Additionally, a prior knowledge vector is incorporated into the input layer to provide character‐specific information on stroke composition, enhancing the ability of the framework to precisely identify and segment strokes. The effectiveness of the proposed framework is demonstrated through evaluations of a comprehensive dataset (brush calligraphy stroke segmentation dataset). Evaluations show that the proposed framework significantly improves the capability of semantic segmentation networks, achieving a remarkable improvement of up to 19.9% in the true stroke rate, compared to traditional stroke segmentation techniques. The Stroke‐Seg framework integrated with TransUNet demonstrates its high performance with an impressive 98.2% true stroke rate. Furthermore, the proposed framework combined with FCN still achieves good performance while consuming the least amount of computational resources and memory, demonstrating its potential for lightweight design. [ABSTRACT FROM AUTHOR]

Conventional computed tomography (CT) images often suffer from blurred edges and unclear details. Image super‐resolution methods can significantly enhance CT image quality, thereby improving diagnostic accuracy. To better extract detailed features and enhance the cascading effects of different feature levels, we propose a novel medical image super‐resolution algorithm that integrates discrete wavelet transform and multi‐scale adaptive feature selection. Our approach uses both the low‐resolution image and its high‐frequency component from the frequency domain as network inputs, with the high‐frequency component providing learning supervision, which enhances detail fidelity in reconstruction. Additionally, we introduce a multi‐scale adaptive feature selection module to learn from different layers of CT images and their inter‐layer correlations. Finally, the pixel information is efficiently integrated by a coordinate attention mechanism incorporating the concept of squeeze excitation. Experimental results show that our method outperforms state‐of‐the‐art methods, achieving superior reconstruction at scale factors of 2, 4, and 8, especially at scale factor 8, where it surpasses others by 1.12 in PSNR, 0.0145 in SSIM, and 0.0038 in LPIPS. Visually, our method also delivers more accurate details and better perceptual quality. [ABSTRACT FROM AUTHOR]

In pedestrian detection task, numerous predicted boxes and their corresponding scores are generated and these scores are used to filter these predicted boxes by non‐maximum suppression. This paper analysed the training process of the popular anchor‐based pedestrian detection models (e.g. YOLO and Faster RCNN), and found that the score of the predicted box reflects the overlap between the corresponding anchor and the ground truth, rather than the predicted box itself. Due to the many‐to‐one strategy adopted by anchor‐based methods, multiple predicted boxes could be generated around one predicted box. This study refers to the number of other predicted boxes around the target predicted box as its local density. When a predicted box has a higher local density, it should have a greater overlap with the ground truth. Therefore, this study proposed the fused score by introducing local density into the score. The experiments showed that replacing the score with the fused score can effectively improve the model's detection accuracy. The code and experiments will soon be open‐sourced at https://github.com/zefeichen/FusedScore. [ABSTRACT FROM AUTHOR]

The surface quality of aluminium alloy castings is crucial to quality control. Aiming to address the challenges of limited samples and extensive computation in deep learning‐based surface defect detection for aluminium alloy castings, this paper proposes a surface defect detection method based on data enhancement and the Casting Real‐Time DEtection TRansformer. First, to tackle the issue of small sample sizes and uneven distribution in surface defect data sets of aluminium alloy castings, ECA‐MetaAconC Deep Convolution Generative Adversarial Networks is proposed for generating defects with fewer samples and employ the image augmentation (IMGAUG) library for sample enhancement. Second, building upon the Real‐Time DEtection TRansformer (RT‐DETR), a lightweight partial‐rep convolution is designed to decrease the network's parameter count. Simultaneously, the Deformable attention module and the DRBC3 module are introduced to enhance the neck network, thereby improving the model's capability to capture information and enhancing its detection performance. Compared to RT‐DETR, this method reduces the number of model parameters by 38.7%, increases mAP by 1.5%, and achieves a frame rate that is 1.58 times higher than the original model. The experimental results demonstrate that this method can effectively and accurately detect surface defects in aluminium alloy castings, satisfying industrial requirements. [ABSTRACT FROM AUTHOR]

To effectively address common issues such as cattle being obscured by fences and images prone to colour shifts and high brightness in a ranch setting, this paper proposes an algorithm for counting cows based on multi‐scale perception and image correlation. The algorithm first adjusts the model output scale to enhance cattle detection under current conditions. It incor‐porates efficient Partial Convolution (PConv) to replace 3 × 3 convolutions in the Neck segment of the YOLOv7 network, boost‐ing computational speed and reducing complexity. To streamline feature fusion, Dynamic Head (DyHead) unifies multiple at‐tentional operations in the Neck segment, enhancing efficiency. Additionally, it introduces a novel bounding box similarity metric Minimum Point DioU (MPDIoU) based on minimum point distance, encompassing factors from existing loss functions, while simplifying computations. Experimental results demonstrate the algorithm significantly improves detection, achieving 98.8% accuracy, 99.0% recall, and a 92.1% mAP value. Compared with mainstream SOTA models, Precision increases by 0.4%, Recall by 2.0%, and mAP value by 2.2%. Model size decreases by 23.9%, parameter count by 23.0%, and computational load by 6.1%. the algorithm shows improvements across all indices, meeting the challenge of real‐time cattle counting in ranches under complex conditions. [ABSTRACT FROM AUTHOR]

With the advent of the digital era, millions of facial images are shared online daily, posing severe privacy threats. Generative face anonymization (GFA) methods generate virtual faces to conceal original identities, protecting sensitive information while preserving utility. However, deep learning based user identity linkage (UIL) methods can link similar faces to the same identity and leverage the linked profiles for malicious purposes, including localization and behaviour prediction. These UIL methods pose a significant challenge to the diversity of virtual faces, a challenge that existing GFA methods have not adequately addressed. To address this research gap, we propose Style Diversification‐based Generative Face Anonymization (SD‐GFA), a framework that generates virtual faces with diverse identities and high visual quality. SD‐GFA features an equalized control module to balance input faces and user‐specified keys, a face generation module with a re‐connection strategy for high‐quality synthesis, and a maximum probability simulation module to enhance diversity. Our experiments demonstrate that SD‐GFA effectively mitigates linkage risk by improving the diversity of virtual faces, while also enhancing their utility and visual quality. This study provides a robust solution to enhance the security of anonymized faces shared on the internet. [ABSTRACT FROM AUTHOR]

License plate recognition is crucial in Intelligent Transportation Systems (ITS) for vehicle management, traffic monitoring, and security inspection. In highway scenarios, this task faces challenges such as diversity, blurriness, occlusion, and illumination variation of license plates. This article explores Recurrent Neural Networks based on Connectionist Temporal Classification (RNN‐CTC) for license plate recognition in challenging highway conditions. Four neural network models: ResNet50, ResNeXt, InceptionV3, and SENet, all combined with RNN‐CTC are comparatively evaluated. Furthermore, a novel architecture named ResNet50 Deep Fusion Network using Connectionist Temporal Classification (ResNet50‐DFN‐CTC) is proposed. Comparative and ablation experiments are conducted using the Highway License Plate Dataset of Southeast University (HLPD‐SU). Results demonstrate the superior performance of ResNet50‐DFN‐CTC in challenging highway conditions, achieving 93.158% accuracy with a processing time of 7.91 ms, outperforming other tested models. This research contributes to advancing license plate recognition technology for real‐world highway applications under adverse conditions. [ABSTRACT FROM AUTHOR]

This study evaluates the effectiveness of three loss functions Asymmetric Unified Focal Loss (AUFL), Dice Similarity Coefficient Loss (DSCL), and Cross‐Entropy (CE) for segmenting COVID‐19 lung infections in computed tomography images. Detailed analyses using the intersection over union metric assessed each function's accuracy. AUFL achieved an average Dice Similarity Coefficient (DSC) of 85.18% ± 8.86%, outperforming DSCL 85.18% ± 8.86%, which had the same average DSC but less precise segmentation, and CE, which had an average DSC of 78.31% ± 11.93%. Segmentations using AUFL demonstrated more defined contours and better alignment with actual anatomical structures than those obtained with DSCL and CE. Observations revealed that AUFL‐generated segmentations had more precise boundaries and were more consistent with the expected anatomical regions of lung infections. This study is the first to quantitatively and qualitatively compare the effectiveness of AUFL, DSCL, and CE in segmenting COVID‐19 lung infections, providing concrete evidence of AUFL's superiority in segmentation performance and reliability for clinical applications. The findings underscore the importance of selecting appropriate loss functions to enhance segmentation in medical imaging, highlighting their crucial role in improving image‐based diagnostics and treatment. The study emphasizes the need for ongoing research to optimize these segmentation techniques further. [ABSTRACT FROM AUTHOR]

To address issues such as object‐background confusion and difficulties in multi‐scale object feature extraction in underwater scenarios, this article proposes an underwater organisms detection algorithm based on multi‐scale perception and representation enhancement. The key innovation of the proposed algorithm is the perception improvement of the deep learning model for underwater multi‐scale objects. First, for underwater large‐scale objects, omni‐dimensional dynamic convolution is embedded as an attention mechanism (AM) into the deep network to improve the network's sensitivity to large‐scale underwater objects. For underwater small‐scale objects, an information retention downsampling module is designed to reduce the effects of serious information loss. Then, a contextual transformer as an AM is introduced into shallow networks to strengthen the network's ability to extract features from small objects. The second innovation of the proposed algorithm is an underwater spatial pooling pyramid module which enhances the representation ability of the model. Furthermore, a lightweight decoupled head is designed to eliminate the conflict between classification and localization. The ablation experiment on the URPC dataset shows that the proposed models are effective for underwater object detection. The comparative experiments on the URPC and DUT‐USEG datasets demonstrate that the proposed algorithm achieves an advantage in detection performance compared with the mainstream detection algorithms and underwater detection algorithms. [ABSTRACT FROM AUTHOR]

Deep learning has made significant progress in the field of low‐light image enhancement. However, challenges remain, such as the substantial parameter consumption required for effective image enhancement. Inspired by multi‐scale geometric transformations in image detail enhancement, a novel model called the fixed‐directional filters network is proposed. Fixed‐directional filters network takes the original image as input and employs multiple branches for parallel processing. One branch uses conventional convolutional layers to extract features from the original image, while the other branches apply non‐linear mapping layers based on wavelet transforms. These wavelet transform branches capture the multi‐scale information of the image by combining different directions and convolutional kernels and utilize a trainable custom gamma mapping layer for non‐linear modulation to enhance specific regions of the image. The feature maps processed by each branch are merged through concatenation operations and then passed through convolutional layers to output the enhanced image. Using trainable mapping functions alone to enhance details significantly reduces the reliance on convolutional layers, effectively lowering the model's parameter count to only 13k parameters. Additionally, experiments demonstrate that fixed‐directional filters network significantly improves image quality, particularly in capturing image details and enhancing image contrast. [ABSTRACT FROM AUTHOR]

Melanoma, a highly prevalent and lethal form of skin cancer, has a significant impact globally. The chances of recovery for melanoma patients substantially improve with early detection. Currently, deep learning (DL) methods are gaining popularity in assisting with the early identification of melanoma. Despite their high performance, relying solely on an image classifier undermines the credibility of the application and makes it difficult to understand the rationale behind the model's predictions highlighting the need for Explainable AI (XAI). This study provides a survey on skin cancer identification using DL techniques utilized in studies from 2017 to 2024. Compared to existing survey studies, the authors address the latest related studies covering several public skin cancer image datasets and focusing on segmentation, classification based on convolutional neural networks and vision transformers, and explainability. The analysis and the comparisons of the existing studies will be beneficial for the researchers and developers in this area, to identify the suitable techniques to be used for automated skin cancer image classification. Thereby, the survey findings can be used to implement support applications advancing the skin cancer diagnosis process. [ABSTRACT FROM AUTHOR]

Fisheye cameras have been widely used in autonomous navigation, visual surveillance, and automatic driving. Due to severe geometric distortion, fisheye images cannot be processed effectively by conventional methods. The existing object detection algorithms cannot better detect the small targets or the objects with large distortion in the fisheye images. The size and scene of available fisheye datasets (such as WoodScape and VOC‐360) cannot satisfy the training of robust network models. Herein, the authors propose Swin‐Fisheye, an end‐to‐end object detection algorithm based on Swin Transformer. A feature pyramid module based on deformable convolution (DFPM) is designed to obtain richer contextual information from the multi‐scale feature maps. In addition, a projection transformation algorithm (PTA) is proposed, which can convert rectilinear images into fisheye images more accurately, and then create a fisheye image dataset (COCO‐Fish). The results of extensive experiments conducted on VOC‐360, WoodScape, and COCO‐Fish demonstrate that the proposed algorithm can achieve satisfactory results compared with state‐of‐the‐art methods. [ABSTRACT FROM AUTHOR]

Insulators, as typical equipment for distribution networks, provide good electrical insulation between live conductors and earth. Timely and accurate detection is essential for insulator detection issues. However, as the complexity of neural networks increases, the detection efficiency is often lower. Therefore, this paper proposes a fast insulator positioning and defect detection method. Firstly, for insulator target localization, the SqueezeNet network is improved using ECA attention mechanism. In addition, to address the issue of low defect detection accuracy, a joint algorithm has been proposed. The integration of convolutional variational autoencoder (CVAE) and generative adversarial network (GAN) solve their own shortcomings due to different image focus angles. The target localization accuracy reaches 94.30%, and the defect detection accuracy reaches 89.60%. It solves the problems of difficulty in locating small targets in a large field of view and inaccurate detection due to a small number of abnormal samples. This method has been tried and tested in practical distribution network systems. [ABSTRACT FROM AUTHOR]

Unsupervised anomaly detection, often approached as a one‐class classification problem, is a critical task in computer vision. Knowledge distillation has emerged as a promising technique for enhancing anomaly detection accuracy, especially with the advent of reverse distillation networks that employ encoder–decoder architectures. This study introduces a novel reverse knowledge distillation framework known as RDMS, which incorporates a pretrained teacher encoding module, a multi‐level feature fusion connection module, and a student decoding module consisting of three independent decoders. RDMS is designed to distill distinct features from the teacher encoder, mitigating overfitting issues associated with similar or identical teacher–student structures. The model achieves an average of 99.3%$\%$ image‐level AUROC and 98.34%$\%$ pixel‐level AUROC on the MVTec‐AD dataset and demonstrates state‐of‐the‐art performance on the more challenging BTAD dataset. The RDMS model's high accuracy in anomaly detection and localization underscores the potential of multi‐student reverse distillation to advance unsupervised anomaly detection capabilities. The source code is available at https://github.com/zihengchen777/RDMS [ABSTRACT FROM AUTHOR]

Detecting taxi passengers is crucial for assessing taxi driver behavior, which plays a significant role in regulating the taxi industry. Despite the advancements in deep learning, object detection algorithms have not been extensively applied to this domain. In this article, an innovative taxi passenger detection algorithm is introduced based on YOLOv8, a lightweight and highly accurate method designed to automatically monitor driver behavior and regulate the taxi industry. To address the challenge of deploying complex object detection models on mobile devices, the ghost module is incorporated in place of standard convolutions within the C2f module, thereby making the model more lightweight. Furthermore, the model's performance is enhanced by integrating an improved version of Frequency Channel Attention (FCA), termed Parallel Frequency Channel Attention (PFCA), which boosts detection accuracy with minimal additional parameters and computational overhead. Experimental results on a specific taxi passenger dataset demonstrate that the proposed method significantly outperforms the baseline YOLOv8n model. Specifically, the model reduces the number of parameters and floating point operations by 12.96% and 8.18%, respectively, while achieving increases in mAP50 and mAP50‐95 by 0.27 and 0.73 percentage points, respectively. [ABSTRACT FROM AUTHOR]

Enhancing hazmat truck safety through advanced driving assistance systems (ADAS) relies on both system efficacy and driver reactions. This study investigates the driving behaviors of hazmat truck drivers in response to forward collision warnings (FCWs). Traditional warning triggering methods struggle to capture diverse and immediate driver responses; therefore, our research employs a vision‐based framework for driving data extraction and utilizes the K‐means++ clustering method for response‐based classification. Moreover, we propose an enhanced version of the intelligent driver model (IDM) based on the concept of a virtual vehicle to reproduce hazmat truck drivers' differential behaviors during risky car‐following periods, achieving results that depict improved driving simulations. This model is compared with classic benchmarks, including the IDM, optimal velocity model (OVM), and full velocity difference (FVD) model, demonstrating superior performance in terms of traffic stability and safety in extreme scenarios. Our findings highlight that preaction drivers tend to accelerate before receiving warnings, opting to overtake rather than maintain safe distances. In contrast, calm drivers decelerate in anticipation of the warning, showcasing their awareness of maintaining safety. The analysis reveals that aggressive drivers are predominantly in the 41–45 age group, indicating a higher skill level, while calm drivers are more commonly older, reflecting a trend in cautious driving behaviors. Overall, our research contributes to the development of effective ADAS by considering real‐time driver responses and emphasizes the potential of our model to revolutionize commercial ADAS adoption and enhance road safety for hazmat operations. [ABSTRACT FROM AUTHOR]

In recent years, convolutional neural networks have significantly advanced the field of computer vision by automatically extracting features from image data. CNNs enable the modeling of complex and abstract image features using learnable filters, eliminating the need for manual feature extraction. However, combining feature maps obtained from CNNs with different approaches can lead to more complex and interpretable inferences, thereby enhancing model performance and generalizability. In this study, we propose a new method called RGB angle rotation to effectively obtain feature maps from RGB images. Our method rotates color channels at different angles and uses the angle information between channels to generate new feature maps. We then investigate the effects of integrating models trained with these feature maps into an ensemble architecture. Experimental results on the CIFAR-10 dataset show that using the proposed method in the ensemble model results in performance increases of 9.10 and 8.42% for the B and R channels, respectively, compared to the original model, while the effect of the G channel is very limited. For the CIFAR-100 dataset, the proposed method resulted in a 17.09% improvement in ensemble model performance for the R channel, a 5.06% increase for the B channel, and no significant improvement for the G channel compared to the original model. Additionally, we compared our method with traditional feature extraction methods like scale-invariant feature transform and local binary pattern and observed higher performance. In conclusion, it has been observed that the proposed RGB angle rotation method significantly impacts model performance. [ABSTRACT FROM AUTHOR]

Modern agriculture relies heavily on the precise application of chemicals such as fertilisers, herbicides, and pesticides, which directly affect both crop yield and environmental footprint. Therefore, it is crucial to assess the accuracy of precision sprayers regarding the spatial location of spray deposits. However, there is currently no fully automated evaluation method for this. In this study, we collected a novel dataset from a precision spot spraying system to enable us to classify and detect spray deposits on target weeds and non-target crops. We employed multiple deep convolutional backbones for this task; subsequently, we have proposed a robustness testing methodology for evaluation purposes. We experimented with two novel data augmentation techniques: subtraction and thresholding which enhanced the classification accuracy and robustness of the developed models. On average, across nine different tests and four distinct convolutional neural networks, subtraction improves robustness by 50.83%, and thresholding increases by 42.26% from a baseline. Additionally, we have presented the results from a novel weakly supervised object detection task using our dataset, establishing a baseline Intersection over Union score of 42.78%. Our proposed pipeline includes an explainable artificial intelligence stage and provides insights not only into the spatial location of the spray deposits but also into the specific filtering methods within that spatial location utilised for classification. [ABSTRACT FROM AUTHOR]

The Train Rolling-Stock Examination (TRSE) is a safety examination process that physically examines the bogie parts of a moving train, typically at speeds over 30 km/h. Currently, this inspection process is done manually by railway personnel in many countries to ensure safety and prevent interruptions to rail services. Although many earlier attempts have been made to semi-automate this process through computer-vision models, these models are iterative and still require manual intervention. Consequently, these attempts were unsuitable for real-time implementations. In this work, we propose a detection model by utilizing a deep-learning based classifier that can precisely identify bogie parts in real-time without manual intervention, allowing an increase in the deployability of these inspection systems. We implemented the Anchor-Free Yolov8 (AFYv8) model, which has a decoupled-head module for recognizing bogie parts. Additionally, we incorporated bogie parts tracking with the AFYv8 model to gather information about any missing parts. To test the effectiveness of the AFYv8-model, the bogie videos were captured at three different timestamps and the result shows the increase in the recognition accuracy of TRSE by 10 % compared to the previously developed classifiers. This research has the potential to enhance railway safety and minimize operational interruptions. [ABSTRACT FROM AUTHOR]

The multi-task learning model is a learning paradigm that shares features across multiple tasks, and it has achieved great success in fields such as computer vision and natural language processing etc. Multiple tasks often conflict and even compete with each other, which seriously reduces the model performance of multi task learning. Most existing optimization methods alleviate multi task gradient conflicts by adjusting task weights. However, it is also important to consider the magnitude and direction of task gradients during the training process. The magnitude and direction of task gradients reflect the conflict and dominance among tasks, which can disrupt the training process and cause instability. In this paper, we present a rescaling and balancing approach for tackling conflicting and dominating gradients. The approach employs a projecting conflict strategy to mitigate the influence of conflicting gradients from multiple tasks and utilize rescaling and balancing techniques to mitigate gradient dominance during training. The proposed method comprehensively considers the weighting, magnitude, and directions of gradients from tasks. We conduct a series of ablation experiments and comparative experiments on different multi-task networks to validate the effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

The application of attention mechanisms, especially channel attention, has achieved huge success in the field of computer vision. However, existing methods mainly focus on more sophisticated attention modules for better performance, but ignore global and local contexts in the frequency domain. This work focuses on the channel relationship and proposes a novel architectural unit called Frequency Global and Local (FGL) context block. It adaptively recalibrates global-local channel-wise feature responses by explicitly modeling interdependencies between channels in the frequency domain. The proposed lightweight FGL module is efficient well generalizable across different datasets. Meanwhile, the FGL context block significantly improves the performance of existing convolutional neural networks (CNNs) at a slight computational cost. Our FGL module is extensively evaluated with applications of image classification, object detection, and semantic segmentation with the backbones of ResNets, MobileNetV2, and MobileNeXt. The experimental results indicate that our module is more efficient than its counterparts. Our model is open-sourced at https://github.com/YunDuanFei/FGL. [ABSTRACT FROM AUTHOR]