Average Precision Research Articles

Zone Evaluation: Revealing Spatial Bias in Object Detection.

A fundamental limitation of object detectors is that they suffer from "spatial bias", and in particular perform less satisfactorily when detecting objects near image borders. For a long time, there has been a lack of effective ways to measure and identify spatial bias, and little is known about where it comes from and what degree it is. To this end, we present a new zone evaluation protocol, extending from the traditional evaluation to a more generalized one, which measures the detection performance over zones, yielding a series of Zone Precisions (ZPs). For the first time, we provide numerical results, showing that the object detectors perform quite unevenly across the zones. Surprisingly, the detector's performance in the 96% border zone of the image does not reach the AP value (Average Precision, commonly regarded as the average detection performance in the entire image zone). To better understand spatial bias, a series of heuristic experiments are conducted. Our investigation excludes two intuitive conjectures about spatial bias that the object scale and the absolute positions of objects barely influence the spatial bias. We find that the key lies in the human-imperceptible divergence in data patterns between objects in different zones, thus eventually forming a visible performance gap between the zones. With these findings, we finally discuss a future direction for object detection, namely, spatial disequilibrium problem, aiming at pursuing a balanced detection ability over the entire image zone. By broadly evaluating 10 popular object detectors and 5 detection datasets, we shed light on the spatial bias of object detectors. We hope this work could raise a focus on detection robustness.

Margin-enhanced average precision optimization for visible-infrared person re-identification

Average precision-based loss functions are effective for training deep neural networks in ranking tasks and single-modality person re-identification. However, they have difficulty in uniformly ranking cross-modality samples due to high intra-modality similarity. To address this limitation, this paper proposes a Margin-Enhanced Average Precision (MEAP) optimization approach for cross-modality person re-identification. MEAP integrates average precision optimization with inter-class and cross-modality margin parameters. The inter-class margin improves performance by increasing class separation, while the cross-modality margin enhances performance by prioritizing cross-modality positive samples. Additionally, we propose an innovative algorithm to augment visible and infrared person images using horizontal stripes, aiming to bridge the gap between the two modalities by creating diverse and enriched training data. Experiments on two public datasets demonstrate the effectiveness of our approach in comparison to state-of-the-art methods. MEAP and horizontal stripe augmentation significantly improve accuracy and robustness in matching individuals across different modalities. The code is available at: https://github.com/NihatTekeli/meapnet

Precise Localization for Anatomo-Physiological Hallmarks of the Cervical Spine by Using Neural Memory Ordinary Differential Equation.

In the evaluation of cervical spine disorders, precise positioning of anatomo-physiological hallmarks is fundamental for calculating diverse measurement metrics. Despite the fact that deep learning has achieved impressive results in the field of keypoint localization, there are still many limitations when facing medical image. First, these methods often encounter limitations when faced with the inherent variability in cervical spine datasets, arising from imaging factors. Second, predicting keypoints for only 4% of the entire X-ray image surface area poses a significant challenge. To tackle these issues, we propose a deep neural network architecture, NF-DEKR, specifically tailored for predicting keypoints in cervical spine physiological anatomy. Leveraging neural memory ordinary differential equation with its distinctive memory learning separation and convergence to a singular global attractor characteristic, our design effectively mitigates inherent data variability. Simultaneously, we introduce a Multi-Resolution Focus module to preprocess feature maps before entering the disentangled regression branch and the heatmap branch. Employing a differentiated strategy for feature maps of varying scales, this approach yields more accurate predictions of densely localized keypoints. We construct a medical dataset, SCUSpineXray, comprising X-ray images annotated by orthopedic specialists and conduct similar experiments on the publicly available UWSpineCT dataset. Experimental results demonstrate that compared to the baseline DEKR network, our proposed method enhances average precision by 2% to 3%, accompanied by a marginal increase in model parameters and the floating-point operations (FLOPs). The code (https://github.com/Zhxyi/NF-DEKR) is available.

An Ensemble Framework Augmenting Surveillance Cameras for Detecting Intruder Clusters as Potential Mobs

<p>Many developing nations around the world curtail crimes through video surveillance technology, but the crime rate is still high. This is compounded by short-staffed security operatives and a deficiency of security infrastructure to assist security operatives with knowledge-driven decision support systems in the low-resource constraint environment. In a public environment, it is challenging to detect intruder clusters accurately as potential mobs for early warning. Previous research investigated some classical techniques, but their recommendations were insufficient. This research develops a Machine Learning (ML) 3-tiers Ensemble framework, which integrates Gray Level Co-occurrence Matrices (GLCM) principles to enhance the capabilities of surveillance cameras and security operatives to effectively discern and respond to potential mob formations. The University of California San Diego (UCSD) pedestrian datasets that are publicly available were used for the experiments. With an improved overall average precision of 0.98, recall of 0.98, and accuracy of 98.52% on the UCSD dataset, the suggested framework outperforms the widely used methods for the detection of intruder clusters. The reduction in computational time on processors showcases the framework's significant advancements as a promising solution for robust real-time threat assessment applications.</p>

Deep ensemble architectures with heterogeneous approach for an efficient content-based image retrieval

<span lang="EN-US">In the field of digital image processing, content-based image retrieval (CBIR) has become essential for searching images based on visual content characteristics like color, shape, and texture, rather than relying on text-based annotations. To address the increasing demands for efficiency and precision in CBIR systems, we introduce the HybridEnsembleNet methodology. HybridEnsembleNet combines deep learning algorithms with an asymmetric retrieval framework to optimize feature extraction and comparison in extensive image databases. This novel approach, specifically custom-made for CBIR, employs a lightweight query structure skilled at handling large-scale data under resource-constrained environments. The experiments were performed on the ROxford and RParis datasets. The deep learning component of HybridEnsembleNet significantly refines the accuracy of image matching and retrieval. RParis The ROxford dataset, specifically in the medium and hard difficulty benchmarks, demonstrates an enhancement of 5.53% and 10.44%, respectively. Similarly, the RParis dataset, under medium and hard benchmarks, exhibits improvements of 3.01% and 5.83%, showcasing superior performance compared to existing models. By overcoming the traditional limitations of CBIR systems in mean average precision (mAP) metrics, HybridEnsembleNet provides a scalable, efficient, and more accurate solution for retrieving relevant images from vast digital libraries.</span>

Predicting levels of legal case difficulties using machine learning

Lawyers play a crucial role in the courtroom, assisting clients in their defense. Because of their lack of legal expertise, a person or organization facing legal issues requires professional aid. However, we need to know how much money will be spent on paying lawyers. The level of complexity in a case can be used to determine lawyer costs. Therefore, in this research, we propose employing machine learning methodologies, i.e., random forest classifiers and support vector machines (SVM), to determine the level of legal case difficulties. The novelty of this research is applying a machine learning approach in predicting the level of difficulty of legal cases. The data utilized consists of 990 records, which are divided into training and testing data in a 90:10 ratio. The term frequency-inverse document frequency (TF-IDF) approach was then utilized to perform text preprocessing. The text-preprocessing findings are utilized as input in the classification process. According to the research findings, an accuracy value of 85%, a value of weighted average precision is 88%, and a value of weighted average recall is 85%, for support vector machine. Using random forest, we achieve an accuracy value of 89%, a value of weighted average precision is 85.6%, and a value of weighted average recall is 80%.

Enhancing Cardiovascular Disease Prediction Accuracy through an Ensemble Machine Learning Approach

This study explores the efficacy of an ensemble machine learning approach, specifically a Voting Classifier combining Decision Tree, k-Nearest Neighbors, and Gaussian Naive Bayes classifiers, in predicting cardiovascular diseases (CVDs). Utilizing a dataset consisting of 70,000 clinical records, the model was rigorously tested through 5-fold cross-validation, achieving remarkable results with average accuracies, precision, recall, and F1-scores all exceeding 99%. The findings validate the hypothesis that ensemble models, due to their capacity to leverage multiple learning algorithms, provide superior prediction accuracy and reliability compared to single predictor models. This research not only confirms the effectiveness of ensemble methods in medical diagnostics but also highlights their potential to enhance decision-making in clinical settings. Given the model's success in identifying various stages of cardiovascular conditions with high accuracy, it offers significant implications for early intervention and personalized patient management. Future research should aim to validate these results across more diverse populations and explore the integration of additional predictive factors that could refine the model's applicability. This study contributes to the computational health field by demonstrating how advanced machine learning techniques can be effectively applied in predicting health outcomes.

Optimizing Cardiomegaly Detection: A Random Forest Approach to Processed Chest X-ray Imagery

This study explores the application of a Random Forest Classifier for the automated detection of Cardiomegaly from chest X-ray images, utilizing a dataset processed and derived from the NIH Chest X-ray Dataset. Given the crucial need for accurate and timely diagnosis of Cardiomegaly to inform appropriate treatment decisions, this research aims to determine the efficacy of machine learning models in augmenting diagnostic processes. Employing image pre-processing techniques such as Sobel filtering for edge detection and Hu Moments for feature extraction, the study enhances the input features for the model. The performance of the classifier was evaluated using a 5-fold cross-validation approach, yielding results with average accuracy, precision, recall, and F1-scores ranging approximately between 52% and 54%. These findings suggest a moderate level of reliability and consistency, indicating the potential utility of ensemble machine learning methods in medical imaging analysis. However, the variability in performance across different data subsets highlights the challenges and necessitates further optimization. This research contributes to the ongoing discourse on integrating machine learning into clinical settings, demonstrating the potential benefits and current limitations. Future research is recommended to expand the dataset variety, integrate advanced deep learning methodologies, and rigorously test these models in clinical environments. The findings hold significant implications for the development of automated diagnostic tools in healthcare, potentially leading to enhanced diagnostic accuracy and efficiency.

Applying natural language processing to understand symptoms among older adults home healthcare patients with urinary incontinence.

Little is known about the range and frequency of symptoms among older adult home healthcare patients with urinary incontinence, as this information is predominantly contained in clinical notes. Natural language processing can uncover symptom information among older adults with urinary incontinence to promote holistic, equitable care. We conducted a secondary analysis of cross-sectional data collected between January 1, 2015, and December 31, 2017, from the largest HHC agency in the Northeastern United States. We aimed to develop and test a natural language processing algorithm to extract symptom information from clinical notes for older adults with urinary incontinence and analyze differences in symptom documentation by race or ethnicity. Symptoms were identified through expert clinician-driven Delphi survey rounds. We developed a natural language processing algorithm for symptom identification in clinical notes, examined symptom documentation frequencies, and analyzed differences in symptom documentation by race or ethnicity using chi-squared tests and logistic regression models. In total, 39,179 home healthcare episodes containing 1,098,419 clinical notes for 29,981 distinct patients were included. Nearly 40% of the sample represented racially or ethnically minoritized groups (i.e., 18% Black, 14% Hispanic, 7% Asian/Pacific Islander, 0.3% multi-racial, and 0.2% Native American). Based on expert clinician-driven Delphi survey rounds, the following symptoms were identified: anxiety, dizziness, constipation, syncope, tachycardia, urinary frequency/urgency, urinary hesitancy/retention, and vision impairment/blurred vision. The natural language processing algorithm achieved excellent performance (average precision of 0.92). Approximately 29% of home healthcare episodes had symptom information documented. Compared to home healthcare episodes for White patients, home healthcare episodes for Asian/Pacific Islander (odds ratio = 0.74, 95% confidence interval [0.67-0.80], p < 0.001), Black (odds ratio = 0.69, 95% confidence interval [0.64-0.73], p < 0.001), and Hispanic (odds ratio = 0.91, 95% confidence interval [0.85-0.97], p < 0.01) patients were less likely to have any symptoms documented in clinical notes. We found multidimensional symptoms and differences in symptom documentation among a diverse cohort of older adults with urinary incontinence, underscoring the need for comprehensive assessments by clinicians. Future research should apply natural language processing to other data sources and investigate symptom clusters to inform holistic care strategies for diverse populations. Knowledge of symptoms of older adult home healthcare patients with urinary incontinence can facilitate comprehensive assessments, health equity, and improved outcomes.

Cross-shaped windows transformer with self-supervised pretraining for clinically significant prostate cancer detection in bi-parametric MRI.

Bi-parametric magnetic resonance imaging (bpMRI) has demonstrated promising results in prostate cancer (PCa) detection. Vision transformers have achieved competitive performance compared to convolutional neural network (CNN) in deep learning, but they need abundant annotated data for training. Self-supervised learning can effectively leverage unlabeled data to extract useful semantic representations without annotation and its associated costs. This study proposes a novel self-supervised learning framework and a transformer model to enhance PCa detection using prostate bpMRI. We introduce a novel end-to-end Cross-Shaped windows (CSwin) transformer UNet model, CSwin UNet, to detect clinically significant prostate cancer (csPCa) in prostate bpMRI. We also propose a multitask self-supervised learning framework to leverage unlabeled data and improve network generalizability. Using a large prostate bpMRI dataset (PI-CAI) with 1476 patients, we first pretrain CSwin transformer using multitask self-supervised learning to improve data-efficiency and network generalizability. We then finetune using lesion annotations to perform csPCa detection. We also test the network generalization using a separate bpMRI dataset with 158 patients (Prostate158). Five-fold cross validation shows that self-supervised CSwin UNet achieves 0.888±0.010 aread under receiver operating characterstics curve (AUC) and 0.545±0.060 Average Precision (AP) on PI-CAI dataset, significantly outperforming four comparable models (nnFormer, Swin UNETR, DynUNet, Attention UNet, UNet). On model generalizability, self-supervised CSwin UNet achieves 0.79 AUC and 0.45 AP, still outperforming all other comparable methods and demonstrating good generalization to external data. This study proposes CSwin UNet, a new transformer-based model for end-to-end detection of csPCa, enhanced by self-supervised pretraining to enhance network generalizability. We employ an automatic weighted loss (AWL) to unify pretext tasks, improving representation learning. Evaluated on two multi-institutional public datasets, our method surpasses existing methods in detection metrics and demonstrates good generalization to external data.

State of Charge Prediction of Mine-Used LiFePO4 Battery Based on PSO-Catboost

The accurate prediction of battery state of charge (SOC) is one of the critical technologies for the safe operation of a power battery. Aiming at the problem of mine power battery SOC prediction, based on the comparative experiments and analysis of particle swarm optimization (PSO) and Categorical Boosting (Catboost) characteristics, the PSO-Catboost model is proposed to predict the SOC of a power lithium iron phosphate battery. Firstly, the classification model based on Catboost is constructed, and then the particle swarm algorithm is used to optimize the Catboost hyperparameters to build the optimal model. The experiment and comparison show that the optimized model’s prediction accuracy and average precision are superior to other comparative models. Compared with the Catboost model, the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) values of the PSO-Catboost model decreased by 12.4% and 25.4% during charging and decreased by 5.5% and 12.2% during discharging. Finally, the Random Forest (RF) and Extreme Gradient Boosting (XGBoost) models, both ensemble learning models, are selected and compared with PSO-Catboost after being optimized via PSO. The experimental results show that the proposed model has a better performance. In this paper, experiments show that the optimization model can select parameters more intelligently, reduce the error caused by artificial experience to adjust parameters, and have a better theoretical value and practical significance.

Exploring the Effects of Attention Mechanisms on Road Crack Detection and Classification

Pavement cracks are a prevalent form of roadway distress that pose significant safety hazards, necessitating prompt detection and repair. Due to the extensive road network, traditional image processing-based crack detection methods exhibit limitations in recognition accuracy and speed. To address these challenges, this paper proposes an efficient pavement crack detection model based on YOLOv8, termed YOLOv8-ACD (YOLOv8 - Attention for Cracks Detecting), which integrates a global attention mechanism. YOLOv8-ACD enhances detection efficiency and accuracy by focusing on crack identification while filtering out most irrelevant information. We evaluated YOLOv8-ACD on the RDD2022 dataset, and experimental results demonstrate significant improvements in key performance metrics, such as F1-Score and mean Average Precision (mAP), compared to the original YOLOv8 and other mainstream models. The real-time processing capability of this model makes it suitable for practical road inspection and maintenance, effectively reducing the workload of maintenance personnel and enhancing roadway safety.

A Deep Learning-Based Three-Stage Method for Spillway Blockage Detection in Reservoirs

Spillway blockage detection is crucial for flood prevention and disaster reduction in reservoirs. To address the challenge of detecting the spillway blockages under complex environmental conditions such as rain and fog, this study proposed a three-stage spillway blockage detection method based on deep learning. This method involved the removal of the rain and fog interference, the segmentation of the spillway boundary region, and blockage detection. First, a rain and fog interference removal algorithm based on the dark channel prior theory was developed. Next, an improved lightweight DeepLabv3+ semantic segmentation algorithm was adopted to segment the spillway region from the images. Finally, the improved YOLOv7 object detection algorithm was utilized to identify the blockage debris within a segmented spillway area. The experimental results indicated that the proposed method achieved an average precision of 80.32% under normal conditions and 77.77% under complex conditions, representing improvements of 9.93% and 6.65%, respectively, compared to traditional methods. This method significantly enhanced the detection and identification of blockages in complex environments and could provide effective support for intelligent reservoir flood control and disaster reduction.

Laser weld spot detection based on YOLO-weld.

Laser weld point detection is crucial in modern industrial manufacturing, yet it faces challenges such as a limited number of samples, uneven distribution, and diverse, irregular shapes. To address these issues, this paper proposes an innovative model, YOLO-Weld, which achieves lightweight design while enhancing detection accuracy. Firstly, a targeted data augmentation strategy is employed to increase both the quantity and diversity of samples from minority classes. Following this, a Diverse Class Normalization Loss (DCNLoss)function is designed to emphasize the importance of tail data in the model's training. Secondly, the Adaptive Hierarchical Intersection over Union Loss (AHIoU Loss)function is introduced, which assigns varying levels of attention to different Intersections over Union (IoU) samples, with a particular focus on moderate IoU samples, thereby accelerating the bounding box regression process. Finally, a lightweight multi-scale feature processing module, MSBCSPELAN, is proposed to enhance multi-scale feature handling while reducing the number of model parameters. Experimental results indicate that YOLO-Weld significantly improves the accuracy and efficiency of laser weld point detection, with mean Average Precision at 50 ([Formula: see text]) and mean Average Precision at 50:95 ([Formula: see text]) increasing by 15.6% and 15.8%, respectively. Additionally, the model's parameter count is reduced by 0.4M, GFLOPS decreases by 1.1, precision improves by 4.3%, recall rises by 22.2%, and the F1 score increases by 15.1%.

Multi-branch convolutional neural network in building polygonization using remote sensing images

AbstractBuilding extraction and polygonization is important for urban studies, such as urbanization monitoring, urban planning. Remote sensing images, especially in RGB bands, provide sufficient semantic information which is useful for the task of building extraction and polygonization. Deep learning using Convolutional Neural Networks (CNNs) is proven to be successful in many fields, including building extraction from remote sensing images. In this paper, we propose a two-stage method to solve the task of building polygonization from remote sensing images based on deep learning. Firstly, we decompose a 2‑D building footprint model into three basic geometry primitives. Leveraging stacked Multi-Branch Modules (MBMs), we separate the task of building extraction into tasks of predicting the three geometry primitives using our proposed CNN. At the second stage, we propose an efficient enhanced building polygonization and adjustment algorithm to generate the final building polygons. This algorithm is able to handle both building blocks and individual buildings. We evaluate our model on three open datasets. For building blocks, our model achieved average precision of 62.7% and average recall of 73.6% on the CrowdAI mapping challenge dataset, and 13.9% and 24.4% respectively on the Urban Building Classification (UBC) dataset which contains mainly individual buildings. On the Inria aerial image dataset, the proposed method achieved Intersection over Union (IoU) over 71%.

Fish-Finder: A robust small target detection method for aquaculture fish in low-quality underwater images.

Underwater fish object detection serves as a pivotal research direction in marine biology, aquaculture management, and computer vision, yet it poses substantial challenges due to the complexity of underwater environments, occultations, and the small-sized and frequently moving fish in aquaculture. Addressing these challenges, we propose a novel underwater fish object detection algorithm named Fish-Finder. First, we engendered a structure titled "C2fBF," utilizing the dual-path routing attention protocol of BiFormer. The primary objective of this structure is to alleviate the perturbations induced by underwater intricacies during the phase of downsampling in the backbone network, thereby discerning and conserving finer contextual features. Subsequently, we co-opted the RepGFPN method within our neck network-a distinctive approach that adeptly merges high-level semantic constructs with low-level spatial specifics, thus fortifying its multi-scale detection prowess. Then, in an endeavor to diminish the sensitivity toward positional aberrations during the detection of diminutive aquatic creatures, we incorporated a novel bounding box regression loss function, the Wasserstein loss, to the existing CIoU. This innovative function gauges the congruity between the predicted bounding box Gaussian distribution and the reference bounding box Gaussian distribution. Finally, in regard to the dataset, we independently assembled a specific dataset termed "SmallFish." This unique dataset, meticulously designed for the detection of small-scale fish within intricate underwater settings, includes 5000 annotated images of small fish. Experimental results demonstrate that, compared to the state-of-the-art detection methods, our proposed method improves the accuracy by and , and mean average precision (mAP) increases and in public dataset Kaggle-Fish and our SmallFish dataset, respectively.

Deep Learning-Based Student Engagement Classification in Online Learning

Online education has gained significant popularity during the COVID-19 pandemic. The availability of massive open online courses has further strengthened the online learning environment. One of the considerable challenges in online learning environments is to measure the learner’s engagement to meet the educational objectives. This paper addresses the challenge and proposes a convolutional neural network-based architecture that extracts discriminative features from the face (Affective) and the upper body (Behavioral) of the learner and classifies the engagement. The performance of the proposed architecture is evaluated on the publicly available Dataset of Affective States In E-Environments (DAiSEE) and the learning-centered affective state dataset curated from open-source datasets. The experimental results demonstrate that the proposed methodology with affective and behavioral features improves the engagement measurement results. The proposed method outperforms the state-of-the-art in terms of Unweighted Average Recall (UAR), Unweighted Average Precision (UAP), and Unweighted Average F1 (UAF1) with 79.14%, 49.62%, and 55.29% values, respectively. It is threefold more computationally efficient than the previous state-of-the-art method. The proposed method also improves the accuracy of the curated dataset by 2.07% of the prior method.

MSOAR-YOLOv10: Multi-Scale Occluded Apple Detection for Enhanced Harvest Robotics

The accuracy of apple fruit recognition in orchard environments is significantly affected by factors such as occlusion and lighting variations, leading to issues such as missed and false detections. To address these challenges, particularly related to occluded apples, this study proposes an improved apple-detection model, MSOAR-YOLOv10, based on YOLOv10. Firstly, a multi-scale feature fusion network is enhanced by adding a 160 × 160 feature scale layer to the backbone network, which increases the model’s sensitivity to small local features, particularly for occluded fruits. Secondly, the Squeeze-and-Excitation (SE) attention mechanism is integrated into the C2fCIB convolution module of the backbone network to improve the network’s focus on the regions of interest in the input images. Additionally, a Diverse Branch Block (DBB) module is introduced to enhance the performance of the convolutional neural network. Furthermore, a Normalized Wasserstein Distance (NWD) loss function is proposed to effectively reduce missed detections of densely packed and overlapping targets. Experimental results in orchards indicate that the proposed improved YOLOv10 model achieves precision, recall, and mean average precision rates of 89.3%, 89.8%, and 92.8%, respectively, representing increases of 3.1%, 2.2%, and 3.0% compared to the original YOLOv10 model. These results validate that the proposed network significantly enhances apple recognition accuracy in complex orchard environments, particularly improving the operational precision of harvesting robots in real-world conditions.

Enhanced YOLOv8 Ship Detection Empower Unmanned Surface Vehicles for Advanced Maritime Surveillance

The evolution of maritime surveillance is significantly marked by the incorporation of Artificial Intelligence and machine learning into Unmanned Surface Vehicles (USVs). This paper presents an AI approach for detecting and tracking unmanned surface vehicles, specifically leveraging an enhanced version of YOLOv8, fine-tuned for maritime surveillance needs. Deployed on the NVIDIA Jetson TX2 platform, the system features an innovative architecture and perception module optimized for real-time operations and energy efficiency. Demonstrating superior detection accuracy with a mean Average Precision (mAP) of 0.99 and achieving an operational speed of 17.99 FPS, all while maintaining energy consumption at just 5.61 joules. The remarkable balance between accuracy, processing speed, and energy efficiency underscores the potential of this system to significantly advance maritime safety, security, and environmental monitoring.

Development and Application of Small Object Visual Recognition Algorithm in Assisting Safety Management of Tower Cranes

This study presents a novel video-based risk assessment and safety management technique aimed at mitigating the risk of falling objects during tower crane lifting operations. The conventional YOLOv5 algorithm is prone to issues of missed and false detections, particularly when identifying small objects. To address these limitations, the algorithm is enhanced by incorporating an additional small object detection layer, implementing an attention mechanism, and modifying the loss function. The enhanced YOLOv5s model achieved precision and recall rates of 96.00%, with average precision (AP) values of 96.42% at an IoU of 0.5 and 62.02% across the range of IoU values from 0.5 to 0.95. These improvements significantly enhance the model’s capability to accurately detect crane hooks and personnel. Upon identifying the hook within a video frame, its actual height is calculated using an interpolation function derived from the hook’s dimensions. This calculation allows for the precise demarcation of the danger zone by determining the potential impact area of falling objects. The worker’s risk level is assessed using a refined method based on the statistical analysis of past accidents. If the risk level surpasses a predetermined safety threshold, the worker’s detection box is emphasized and flagged as a caution on the monitoring display.