You Only Look Once Research Articles

Optimizing deep belief network for concrete crack detection via a modified design of ideal gas molecular dynamics

Concrete structures are prone to developing cracks, which can have a negative impact on their overall performance and longevity. It is essential to promptly identify and repair these cracks in order to ensure the structural integrity of the building. The present research concentrates on the development of crack diagnosis algorithms based on vision using an optimized version of Deep Neural Network (DNN). The DNN model employed in the current study is the deep belief network (DBN), while the optimization technique is based on a newly designed variant of the Ideal Gas Molecular Movement (MIGMM). By combining these two components, a highly effective crack detection system is created, capable of achieving higher classification rates. To train the DNN model, an image dataset comprising two classes, namely “no-cracks” and “cracks”, has been utilized. The MIGMM has been applied to the DBN model, involving fine-tuning the network architecture’s weights, substituting the categorization layer with two classes of output (cracks and no-cracks), and augmenting the picture dataset using stochastic angles of rotation. The proposed DBN/MIGMM model achieves exceptional performance, with an accuracy of 90.189%, specificity of 94.502%, precision of 94.586%, recall of 94.529%, and an F1-score of 88.093%, outperforming state-of-the-art methods such as Fully Convolutional Networks (FCN), You Only Look Once (YOLO), CrackSegNet, Convolutional Neural Networks (CNN), and Convolutional Encoder-Decoder Networks (CedNet). The present outcomes prepare a comprehensive superior assessment of the proposed model’s effectiveness in accurately detecting and classifying cracks.

Deep Learning-Based Real-Time Surf Detection Model During Typhoon Events

Surf during typhoon events poses severe threats to coastal infrastructure and public safety. Traditional monitoring approaches, including in situ sensors and numerical simulations, face inherent limitations in capturing surf impacts—sensors are constrained by point-based measurements, while simulations require intensive computational resources for real-time monitoring. Video-based monitoring offers promising potential for continuous surf observation, yet the development of deep learning models for surf detection remains underexplored, primarily due to the lack of high-quality training datasets from typhoon events. To bridge this gap, we propose a lightweight YOLO (You Only Look Once) based framework for real-time surf detection. A novel dataset of 2855 labeled images with surf annotations, collected from five typhoon events at the Chongwu Tide Gauge Station, captures diverse scenarios such as daytime, nighttime, and extreme weather conditions. The proposed YOLOv6n model achieved 99.3% mAP50 at 161.8 FPS, outperforming both other YOLO variants and traditional two-stage detectors in accuracy and computational efficiency. Scaling analysis further revealed that YOLO models with 2–5 M parameters provide an optimal trade-off between accuracy and computational efficiency. These findings demonstrate the effectiveness of YOLO-based video monitoring systems for real-time surf detection, offering a practical and reliable solution for coastal hazard monitoring under extreme weather conditions.

Adapting YOLO11 for Classification of Unlabelled Data : A Semi-Supervised Approach

Computer vision has advanced rapidly, improving object detection and classification. The YOLO (You Only Look Once) model family is widely used because of its real-time processing efficiency and object detection accuracy. However, most YOLO-based methods focus on object detection rather than categorization. Deep learning models sometimes require large amounts of labelled data, which can be expensive and time-consuming, especially in specialized sectors. Traditional classification methods require extensive manual dataset annotation and fully supervised learning. When labelled data is scarce, this reliance is a major issue. Semi-supervised learning (SSL) improves model efficacy by using labelled and unlabeled data. SSL methods to reduce labelling burdens have been studied, however SSL with YOLO-based designs has not. This project uses semi-supervised learning to adapt YOLO11, an advanced deep learning framework, for classification issues. This work aims to improve classification accuracy in environments with little tagged data, reducing manual tagging. This project will test self-training, pseudo-labeling, and consistency regularization in YOLO11-based classification models. This study uses semi-supervised learning to link object detection and categorization in the YOLO framework. The proposed method could be used in autonomous driving, surveillance, and medical imaging, where comprehensive labelling is often impossible. This work will explain YOLO11 classification optimization and validate SSL techniques in deep learning classification challenges. On the Adidas logo dataset, YOLOv11, the latest framework, was used to classify images according to the object box size into large, medium, and small groups, then chose the best group to use it for the next step, which is the object detection, these steps yielded encouraging results, with a MAP train of 0.913 and a 36-minute training time.

Enhancing Deep Learning to Improve Road Safety: An Accident Detection System

This advanced accident detection system significantly improves road safety by enabling timely identification of traffic incidents and rapid emergency response. Current road safety systems often fail to provide accurate and prompt detection of accidents, resulting in delayed emergency services and increased casualties. To address this issue, an advanced system utilizes Convolutional Neural Networks (CNNs) and the You Only Look Once (YOLO) model for real-time accident detection and classification. By analyzing RGB frames and incorporating optical flow data, this system effectively handles dynamic scenarios such as high-speed traffic, varying weather conditions, and complex vehicle movements. Furthermore, a Real-Time Notification feature integrates to automatically alert nearby hospitals and dispatch ambulances immediately upon detecting an accident. Training and testing use the dataset from Kaggle, which includes CCTV footage frames of accidents and non-accidents, ensuring robust performance. This system detects accidents in real- time and promptly sends notifications via email to alert nearby hospitals and dispatch ambulances.

AI-Powered Image-Based Assessment of Pressure Injuries Using You Only Look once (YOLO) Version 8 Models.

Objective: The primary objective of this study is to enhance the detection and staging of pressure injuries using machine learning capabilities for precise image analysis. This study explores the application of the You Only Look Once version 8 (YOLOv8) deep learning model for pressure injury staging. Approach: We prepared a high-quality, publicly available dataset to evaluate different variants of YOLOv8 (YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, and YOLOv8x) and five optimizers (Adam, AdamW, NAdam, RAdam, and stochastic gradient descent) to determine the most effective configuration. We followed a simulation-based research approach, which is an extension of the Consolidated Standards of Reporting Trials (CONSORT) and Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for dataset preparation and algorithm evaluation. Results: YOLOv8s, with the AdamW optimizer and hyperparameter tuning, achieved the best performance metrics, including a mean average precision at intersection over union ≥0.5 of 84.16% and a recall of 82.31%, surpassing previous YOLO-based models in accuracy. The ensemble model incorporating all YOLOv8 variants showed strong performance when applied to unseen images. Innovation: Notably, the YOLOv8s model significantly improved detection for challenging stages such as Stage 2 and achieved accuracy rates of 0.90 for deep tissue injury, 0.91 for Unstageable, and 0.74, 0.76, 0.70, and 0.77 for Stages 1, 2, 3, and 4, respectively. Conclusion: These results demonstrate the effectiveness of YOLOv8s and ensemble models in improving the accuracy and robustness of pressure injury staging, offering a reliable tool for clinical decision-making.

Leveraging Artificial Intelligence and Clinical Laboratory Evidence to Advance Mobile Health Applications in Ophthalmology: Taking the Ocular Surface Disease as a Case Study

ABSTRACTBackgroundThe advent of mobile health (mHealth) applications has fundamentally transformed the healthcare landscape, particularly within the field of ophthalmology, by providing unprecedented opportunities for remote diagnosis, monitoring, and treatment. Ocular surface diseases, including dry eye disease (DED), are the most common eye diseases that can be detected by mHealth applications. However, most remote artificial intelligence (AI) systems for ocular surface disease detection are predominantly based on self‐reported data collected through interviews, which lack the rigor of clinical evidence. These constraints underscore the need to develop robust, evidence‐based AI frameworks that incorporate objective health indicators to improve the reliability and clinical utility of remote health applications.MethodsTwo novel deep learning (DL) models, YoloTR and YoloMBTR, were developed to detect key ocular surface indicators (OSIs), including tear meniscus height (TMH), non‐invasive Keratograph break‐up time (NIKBUT), ocular redness, lipid layer, and trichiasis. Additionally, back propagation neural networks (BPNN) and universal network for image segmentation (U‐Net) were employed for image classification and segmentation of meibomian gland images to predict Demodex mite infections. These models were trained on a large dataset from high‐resolution devices, including Keratograph 5M and various mobile platforms (Huawei, Apple, and Xiaomi).ResultsThe proposed DL models of YoloMBTR and YoloTR outperformed baseline you only look once (YOLO) models (Yolov5n, Yolov6n, and Yolov8n) across multiple performance metrics, including test average precision (AP), validation AP, and overall accuracy. These two models also exhibit superior performance compared to machine plug‐in models in KG5M when benchmarked against the gold standard. Using Python's Matplotlib for visualization and SPSS for statistical analysis, this study introduces an innovative proof‐of‐concept framework leveraging quantitative AI analysis to address critical challenges in ophthalmology. By integrating advanced DL models, the framework offers a robust approach for detecting and quantifying OSIs with a high degree of precision. This methodological advancement bridges the gap between AI‐driven diagnostics and clinical ophthalmology by translating complex ocular data into actionable insights.ConclusionsIntegrating AI with clinical laboratory data holds significant potential for advancing mobile eye health (MeHealth), particularly in detecting OSIs. This study aims to explore this integration, focusing on improving diagnostic accuracy and accessibility. This study demonstrates the potential of AI‐driven tools in ophthalmic diagnostics, paving the way for reliable, evidence‐based solutions in remote patient monitoring and continuous care. The results contribute to the foundation of AI‐powered health systems that can extend beyond ophthalmology, improving healthcare accessibility and patient outcomes across various domains.

Object Detection YOLO Algorithms and Their Industrial Applications: Overview and Comparative Analysis

Deep-learning-based object detection algorithms play a pivotal role in various domains, including face detection, automatic driving, monitoring security, and industrial production. Compared with the traditional object detection algorithms and the two-stage object detection algorithms, the YOLO (You Only Look Once) series improved the detection speed and accuracy. In addition, the YOLO series of object detection algorithms are widely used in the industrial fields due to their real-time and high-precision characteristics. This work summarizes the main versions of YOLO series algorithms as well as their main improving measures. Furthermore, the following is the analysis of the industrial application fields and some application examples of YOLO series algorithms. Furthermore, this work summarizes the general improvement measures for the industrial applications of the YOLO series algorithms. As for the comparison of these algorithms, this work implements the basic tests for the industrial application performance on different datasets. Finally, the development directions and challenges for YOLO series algorithms are pointed out.

AI classification of knee prostheses from plain radiographs and real-world applications

PurposeTotal knee arthroplasty (TKA) is considered the gold standard treatment for end-stage knee osteoarthritis. Common complications associated with TKA include implant loosening and periprosthetic fractures, which often require revision surgery or fixation. Challenges arise when medical records related to the knee prosthesis are lost, making it difficult to plan for revision surgery effectively. This study aims to develop an artificial intelligence (AI) system to classify the types of knee prosthetic implants using plain radiographs.MethodsThis retrospective experimental study includes seven types of knee prostheses commonly used in our hospital. The artificial intelligence (AI) system was trained using YOLO (You Only Look Once) version 9, utilizing a dataset of 3228 post-operative and follow-up knee arthroplasty X-ray images. The plain radiographic images were augmented, resulting in a dataset of 25,800 images. Model parameters were fine-tuned to optimize performance for implant classification.ResultsThe mean age of the patients was 62.8 years. Right knee arthroplasty was performed in 48.3% of cases, while left knee arthroplasty was performed in 51.7%. The images of knee prostheses comprised 50.9% of the dataset from the anteroposterior (AP) view and 49.1% from the lateral view. The AI model demonstrated exceptional performance metrics, achieving precision, recall, and accuracy rates of 100%, with an F1 score of 1. Additionally, the area under the curve (AUC) of the receiver operating characteristic (ROC) curve was calculated to be 100%.ConclusionThis AI model successfully classifies knee prosthetic implants from plain radiographs. This capability serves as a valuable tool for surgeons, enabling precise planning for revision surgeries and periprosthetic fracture fixation surgery, ultimately contributing to improved patient outcomes. The high accuracy achieved by the AI underscores its potential to enhance surgical efficiency and effectiveness in managing knee arthroplasty complications.

Fuzzy control algorithm of cleaning parameters of street sweeper based on road garbage volume grading

The research examines the challenges city street sweepers face, which struggles to adapt cleaning settings based on varying road garbage volume, resulting in inefficient cleaning and high energy consumption. The study proposes a fuzzy control algorithm for adjusting the cleaning parameters of street sweepers based on road garbage volume grading. It starts by utilizing the YOLO (You Only Look Once) v5 deep learning model for target detection and garbage classification on road surfaces. The algorithm calculates road garbage volume grading coefficients based on the coverage and weight of different types of garbage. These coefficients, along with the street sweeper’s traveling speed, are used as input variables to develop a fuzzy control model for optimizing operational parameters such as the street sweeper disc brush and fan speeds. This model aims to achieve adaptive control of cleaning parameters, thereby enhancing the intelligence of street sweepers. Experimental results show that the proposed algorithm can achieve a 29.77% energy saving under the same road garbage conditions compared to the conventional gear control mode.

Lane Detection for Self-Driving Cars

he project focuses on developing a real-time vehicle detection and lane analysis system using computer vision techniques. The system is designed to process video inputs, detect vehicles, analyze lane curvature, and provide safety suggestions to drivers. The application leverages YOLO (You Only Look Once) for vehicle detection, optical flow for speed estimation, and Hough Transform for lane detection. The system is implemented as a Flask-based web application, allowing users to upload video files, process them, and view the results with overlays indicating vehicle distances, lane status, and safety suggestions. The project aims to enhance road safety by providing real-time feedback to drivers about their surroundings. Keywords: Real-time Vehicle Detection, Lane Detection, YOLO, Optical Flow, Hough Transform, Road Safety, Computer Vision, Video Processing, Lane Curvature Analysis, ADAS

A Regret Bound for the AdaMax Algorithm With Image Segmentation Application

ABSTRACTThe AdaMax algorithm provides enhanced convergence properties for stochastic optimization problems. In this paper, we present a regret bound for the AdaMax algorithm, offering a tighter and more refined analysis compared to existing bounds. This theoretical advancement provides deeper insights into the optimization landscape of machine learning algorithms. Specifically, the You Only Look Once (YOLO) framework has become well‐known as an extremely effective object segmentation tool, mostly because of its extraordinary accuracy in real‐time processing, which makes it a preferred option for many computer vision applications. Finally, we used this algorithm for image segmentation.

A Lightweight TA-YOLOv8 Method for the Spot Weld Surface Anomaly Detection of Body in White

The deep learning architecture YOLO (You Only Look Once) has demonstrated its superior visual detection performance in various computer vision tasks and has been widely applied in the field of automatic surface defect detection. In this paper, we propose a lightweight YOLOv8-based method for the quality inspection of car body welding spots. We developed a TA-YOLOv8 network structure which has an improved Task-Aligned (TA) head detection, designed to handle a small sample size, imbalanced positive and negative samples, and high-noise characteristics of Body-in-White welding spot data. By learning with fewer parameters, the model achieves more efficient and accurate classification. Additionally, our algorithm framework can perform anomaly segmentation and classification on our open-world raw datasets obtained from actual production environments. The experimental results show that the lightweight module improves the processing speed by an average of 2.8%, with increases in detection the mAP@50-95 and recall rate of 1.35% and 0.1226, respectively.

A Sustainable Approach to Waste Management of Tyres: Using Artificial Intelligence for Enhanced Accuracy

This study aims to develop a model that can detect and classify tyres, distinguishing between defective and functional ones to aid in quality inspection and support sustainable waste management practices. The study adopted a structured approach to developing an automated system for classifying tyre quality, using ResNet-50 and YOLO (You Only Look Once) for real-time detection. The precision peaks at 0.9552, indicating excellent performance of the model. The study's findings enforce the potential of deep learning models to increase efficiency and safety within the automotive sector, particularly in areas like preventive maintenance and tyre recycling. The quality control and enhancement in waste management practices of tyres within the automotive sector can be achieved by integrating real-time detection and the precise classification of tyres using this model.

Pothole Detection Using Yolo and Computer Vision

Road infrastructure is the infrastructural core and has an important function in transportation and economic growth activities. Nevertheless, potholes have a profound effect on the safety of driving, as they lead to vehicle repairs and operating costs, and on the effectiveness of road systems. State-of-the- art pothole detection methods involve either manual inspection of the road or sensor- based methods, which are inefficient, labor-intensive and expensive. Because of this, there is an increasing demand for automatic intelligent systems to identify potholes automatically and effectively in order to improve road maintenance and roadside safety. This paper introduces an intelligent pothole detection system, using combination of deep learning and real-time video processing for effective pothole detection. The system is based on a state-of-the-art object detection algorithm (YOLO (You Only Look Once) model) to identify potholes in real-time video streams. The YOLO model facilitates real-time high-accuracy pothole detection and localization, which can support automated log and analysis of freeway conditions. The suggested system is an automated, time- saving, economical tool that helps local governments to detect and repair road defects. Keywords—Keywords: computer vision, YOLO, machine learning, CNN, real-time detection, geolocation- based detection, video analysis.

Enhancing Bolt Object Detection via AIGC-Driven Data Augmentation for Automated Construction Inspection

In the engineering domain, the detection of damage in high-strength bolts is critical for ensuring the safe and reliable operation of equipment. Traditional manual inspection methods are not only inefficient but also susceptible to human error. This paper proposes an automated bolt damage identification method leveraging AIGC (Artificial Intelligence Generated Content) technology and object detection algorithms. Specifically, we introduce the application of AIGC in image generation, focusing on the Stable Diffusion model. Given that the quality of bolt images generated directly by the Stable Diffusion model is suboptimal, we employ the LoRA fine-tuning technique to enhance the model, thereby generating a high-quality dataset of bolt images. This dataset is then used to train the YOLO (You Only Look Once) object detection algorithm, demonstrating significant improvements in both accuracy and recall for bolt damage recognition. Experimental results show that the LoRA fine-tuned Stable Diffusion model significantly enhances the performance of the YOLO algorithm, providing an efficient and accurate solution for automated bolt damage detection. Future work will concentrate on further optimizing the model to improve its robustness and real-time performance, thereby better meeting the demands of practical industrial applications.

EGY_PDD: a comprehensive multi-sensor benchmark dataset for accurate pavement distress detection and classification

Abstract Automated detection of pavement distresses using road images remains a research hotspot within the computer vision community. The advent of deep learning has sparked significant interest in enhancing the effectiveness of automated identification and assessment of pavement distresses. Yet, the limited availability of comprehensive ground truth datasets for pavement distresses poses a prominent challenge for training deep learning models. To address this issue, this study introduces the Egyptian Pavement Distress Dataset (EGY_PDD), a publicly available dataset that comprises images of various types of pavement distress, such as cracks, potholes, and rutting, collected from different regions in Egypt. The dataset is annotated with labels that indicate the type of the pavement distress in each image, making it suitable for training and evaluating machine learning models designated for automatic pavement distress detection and classification. The EGY_PDD dataset has some unique features, such as its focus on pavement distress problems commonly found in Egypt and the MENA (Middle East and North Africa) region, which experiences distinct pavement challenges due to specific geographical, climatic, and socioeconomic factors. EGY_PDD aims to create a comprehensive dataset that enables the development of more robust and easily deployable pavement condition assessment systems. The dataset includes annotated 2D images and 3D road scenes captured for the same pavement segments. Both 2D and 3D images are employed for distress detection and classification using deep learning frameworks. While 2D images contribute to these tasks, 3D images provide more precise classification of distress severity and more accurate calculations of density. These enhanced measurements from 3D images are crucial for the automated computation of pavement ratings or the Pavement Condition Index (PCI). The dataset, consisting of 14,612 meticulously annotated 2D images categorized into eleven distinct types of distresses, was evaluated using two iterations of the widely adopted deep learning framework, You Only Look Once (YOLO). The models, trained for no more than 300 epochs, achieved mAP50 and mAP50-95 scores of 0.617 and 0.293, respectively, demonstrating their adequate performance.

Automatic Meniscus Segmentation Using YOLO-Based Deep Learning Models with Ensemble Methods in Knee MRI Images

The meniscus is a C-shaped connective tissue with a cartilage-like structure in the knee joint. This study proposes an innovative method based on You Only Look Once (YOLO) series models and ensemble methods for meniscus segmentation from knee magnetic resonance imaging (MRI) images to improve segmentation performance and evaluate generalization capability. In this study, five different segmentation models were trained, and masks were created from the YOLO series. These masks are combined with pixel-based voting, weighted multiple voting, and dynamic weighted multiple voting optimized by grid search. Tests were conducted on internal and external sets and various metrics. The dynamic weighted multiple voting method optimized with grid search performed the best on both the test set (DSC: 0.8976 ± 0.0071, PPV: 0.8561 ± 0.0121, Sensitivity: 0.9467 ± 0.0077) and the external set (DSC: 0.9004 ± 0.0064, PPV: 0.8876 ± 0.0134, Sensitivity: 0.9200 ± 0.0119). The proposed ensemble methods offer high accuracy, reliability, and generalization capability for meniscus segmentation.

A deep learning based approach for semantic segmentation of small fires from UAV imagery

ABSTRACT In this paper, we propose novel techniques for fire segmentation from unmanned aerial vehicle (UAV) images. (1) We propose the ObjectDetection+CIELAB thresholding technique, which leverages a pre-trained object detector such as YOLO (``you only look once’‘) to generate fire bounding boxes. We then apply thresholding in the CIELAB colour space within these regions to detect fire pixels. This approach significantly improves speed by streamlining the segmentation task into a more focused detection and classification task. (2) We further introduce the SEG-4CHANNEL technique, which generates a fire pixel mask using the ObjectDetection+CIELAB thresholding method. This mask is integrated as a fourth channel into various segmentation networks, allowing models to concentrate on fire detection while minimising background interference. (3) Finally, we explore the AttentionSeg technique, which incorporates an attention module within a segmentation framework (e.g., SegFormer-B5) that utilises all four channels. It combines the advantages of CIELAB colour-space model, convolution neural network (CNN) and transformer. We explore a large design space of various networks and backbones on the FLAME (Fire Luminosity Airborne-based Machine learning Evaluation) segmentation dataset. Our best model, AttentionSeg-B5, segments fire with an intersection-over-union (IoU) score of 84.15% and 91.39% F1-score. The code has been released at https://github.com/CandleLabAI/FireSegmentation.

Conservation in action: Cost-effective UAVs and real-time detection of the globally threatened swamp deer (Rucervus duvaucelii)

Conservation in action: Cost-effective UAVs and real-time detection of the globally threatened swamp deer (Rucervus duvaucelii)

A A YOLO-Based Machine Learning Framework for Detection of Soft Pneumatic Actuator Bending Angles

The bending angle of soft pneumatic actuator (SPA) is a critical parameter influencing their reliability and effectiveness across various applications. Conventional measurement methods are often labour-intensive and impractical for experiments requiring multiple trials, creating a need for efficient, non-invasive techniques. This study proposes a machine learning framework leveraging YOLO (You Only Look Once) models to detect SPA bending angles from image data, eliminating the need for additional hardware. A comprehensive dataset of SPAs under varying actuation pressures, with meticulously labelled bending angles, was curated to train a YOLO-based regression model. The results highlight the model's strong performance, achieving a recall of 99.1%, precision of 70%, and mean average precision (mAP) scores of 86.42% (IoU 0.5) and 84.35% (IoU 0.5–0.95). Low training and validation losses indicate high accuracy in bounding box predictions, object-background differentiation, and object classification. Optimized learning rates ensured efficient parameter updates, achieving convergence without overfitting. The proposed framework demonstrates a robust balance between accuracy, robustness, and efficiency, making it a practical solution for reliable SPA bending angle detection in real-world applications. This study underscores the potential of machine learning-driven techniques to streamline SPA characterization, offering a scalable and non-invasive alternative to traditional methods.