Aug 2026· International journal of pattern recognition and artificial intelligence· 0 citations
TL;DR
A two-stage hand analysis framework that combines an optimized YOLOv5s detector with HRNet-based keypoint estimation and an adapted MSA-CAM module to improve feature representation in cluttered industrial scenes while reducing computational cost is presented.
Abstract
Hand keypoint detection is important for human – computer interaction and industrial process monitoring, but practical deployment on resource-constrained devices still faces challenges such as the trade-off between accuracy and efficiency, limited robustness in dynamic scenes, and sensitivity to occlusion. To address these issues, this paper presents a two-stage hand analysis framework that combines an optimized YOLOv5s detector with HRNet-based keypoint estimation. In the detection stage, the backbone is replaced with InceptionNeXt and an adapted MSA-CAM module is introduced to improve feature representation in cluttered industrial scenes while reducing computational cost. In the pose stage, HRNet is used to estimate 21 hand keypoints from detected hand regions. Experiments on multiple hand datasets show that the proposed detector achieves a favorable balance between accuracy and efficiency. In a discrete workshop packaging scenario, the overall system also supports action-sequence recognition and anomaly detection, achieving 96.3% recognition success in the topview setting and 95.0% in the front-view setting. These results demonstrate the practical value of the proposed framework for real-time industrial hand analysis.
The mixing stage of automotive battery production requires reliable monitoring of raw material types, personnel actions, and correct tool use. However, accurate multi-scale object detection in complex scenes remains challenging because of background interference and the requirements of embedded deployment and real-time operation. This study proposes an enhanced YOLOv5 framework for industrial multi-object detection. The method adopts a dual-stage feature-enhancement strategy designed to improve robustness while limiting parameter count and computational complexity. First, the Convolutional Block Attention Module (CBAM) is embedded in the Backbone and Neck of YOLOv5 to provide multi-granularity feature enhancement and improve the detection of small objects, such as tools. Second, the conventional CIOU loss is replaced with the Focal-EIOU loss function to optimize bounding-box regression, reduce false detections across multiple target scales, and accelerate model convergence. Finally, K-means clustering is applied to target geometric features to generate specialized anchor-box parameters better suited to industrial scenarios. Experimental results on an automotive battery production-site dataset show that the improved model's mAP@0.5 increased by 2.4 % compared with the original model, reaching 95.2 %, while mAP@0.5:0.95 improved by 3.6 %, reaching 78.5 %. The proposed framework provides a lightweight and reliable solution for multi-object detection in complex industrial environments and supports the development of intelligent visual monitoring systems for smart manufacturing.
Y.-X. Li, H. Chen, L. Zhou· Advances in Production Engin...· 0 citations
An Adaptive and Scalable YOLO model named AS-YOLOR (Adaptive and Scalable YOLO for Rotated object detection), based on the YOLOv8 baseline is proposed, providing a solution with strong practical potential for achieving efficient and high-precision detection of small, rotated objects.
Jin Huang, Juntao Shen, Min Wang et al.· Applied Sciences· 0 citations
In human-computer interaction scenarios, gesture recognition technology enables device operators to perform tasks in a more flexible, natural, and immersive manner. However, most existing gesture recognition algorithms are trending toward lightweight architectures. While this development facilitates real-time performance requirements on mobile devices or resource-constrained systems, it also presents several limitations. Traditional lightweight paradigms (such as simple channel reduction, model pruning, or conventional lightweight convolutions) are often achieved at the expense of global feature representation capabilities, which leads to a sharp decline in recognition accuracy when models face multimodal dynamic hand gesture interaction scenarios with complex backgrounds, occlusions, or non-uniform lighting. Addressing the issue of insufficient global modeling capability, this paper proposes a Vision Transformer architecture based on an efficient additive self-attention mechanism and lightweight dilated convolutions. The method comprehensively absorbs the global modeling capability of the self-attention mechanism and the local feature extraction capability of convolutions, successfully achieving global context modeling capabilities comparable to the standard Transformer utilizing a linear computational complexity cost. To address the aforementioned challenges, this paper proposes a Vision Transformer architecture based on an efficient additive self-attention mechanism and lightweight dilated convolution. The proposed approach integrates the global modeling capabilities of self-attention mechanisms with the local feature extraction capabilities of convolutional networks through optimized combination, achieving a balance between resource utilization and recognition efficiency on mobile platforms. Experimental results demonstrate that the proposed method achieves state-of-the-art accuracy rates of 87.55% on the public dataset NVGestures and 97.92% on the multimodal dynamic gesture dataset Briareo. Furthermore, it achieves comparable or even superior accuracy on the more complex NNGestures dataset through both single-modal and multimodal experiments with reduced model parameters, validating the effectiveness of our methodology.
Huiming Wu, Kun Wang· International Conference on...· 0 citations
Strand tensioning is the core process in hybrid tower construction for wind turbines, and accurate detection of ruler scale markings is critical to construction quality. Traditional manual inspection and existing visual recognition techniques suffer from low efficiency, high cost, and excessive computational overhead. To address these challenges, we propose a novel ruler recognition method based on YOLOv11-OBB rotated object detection. The approach adopts the original detection algorithm without any model architecture modifications, and constructs a multi-scene dataset of ruler numerals and spray-painted regions using oriented bounding box annotations. After model training and inference with YOLOv11-OBB, a six-step post-processing pipeline is designed: (1) fitting the ruler straight line via the least squares method, (2) precisely locating the spray-painted box and its positioning point, (3) computing the minor scale value through coordinate calculations, (4) filtering effective major scale values, and (5-6) determining the final scale value based on the relative positional logic between the spray-painted box and major scale boxes. This method achieves end-to-end recognition with a single model, replacing the conventional multi-model pipeline, significantly reducing data annotation effort and computational resource consumption while ensuring accuracy and real-time performance. It can be flexibly deployed on computing devices with varying capabilities at construction sites.
Suo Wang, Haobing Liang, Nana Lu et al.· International Conference on...· 0 citations
With the development of the low-altitude economy, drones have been widely used in traffic monitoring, industrial inspection, and other fields. However, images acquired by drone often suffer from problems such as small object scale, large attitude variations, complex backgrounds, and severe occlusion, which pose serious challenges to the accuracy of existing object detection algorithms. To address these issues, this article proposes a small object detection model for drones based on spatial and frequency domain collaborative enhancement (SFC-DETR). This model introduces a frequency-augmented polarity-aware attention module (FAPAM) that combines polarity-aware attention with adaptive frequency domain enhancement. Through the polarity-aware attention mechanism, it can effectively capture global contextual relationships while maintaining linear computational complexity. Then, an adaptive window frequency domain modulation module is used to refine the features, thereby enhancing the representation of key details. Additionally, this article also designs a dual-domain adaptive feature fusion module (DD-AFFM) that achieves adaptive fusion of cross-layer features in the spatial and frequency domains, fully preserving fine-grained structural information and high-level semantic features. Experimental results on the public VisDrone and TinyPerson datasets demonstrate that our proposed method outperforms state-of-the-art models in small object detection tasks.
Small object detection in industrial scenarios faces challenges including limited pixel coverage, weak feature representation, and background interference. To address these problems, this paper presents an improved YOLOv11 detection model. First, a dual-backbone network architecture is designed to simultaneously capture rich semantic information and spatial details through parallel feature extraction paths. Second, the SimAM parameter-free attention mechanism is integrated into top-level feature fusion to adaptively enhance features relevant to small objects. Finally, the Adaptive Spatial Feature Fusion (ASFF) module is improved with a dual attention mechanism to optimize multi-scale feature fusion and mitigate feature conflicts. On a self-constructed industrial tool dataset, the method achieves an mAP@0.5:0.95 of 0.920, improving upon the baseline YOLOv11n by 5.9 percentage points. For small object detection specifically, mAP_s reaches 0.898, representing a 7.9 percentage point improvement. Experiments on the public VisDrone dataset further validate the generalization capability of the approach. Results demonstrate that the proposed method significantly enhances small object detection performance, providing an effective solution for industrial vision applications.
Chengru Liu, Junqing Yang, Qi-Qi Guo et al.· 2026 8th International Confe...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
