Improved Object Detection Algorithm for Workshop Environments Based on YOLOv5

In the context of workshop-restricted environments, this paper presents an improved lightweight detection algorithm named YOLO-M3C to achieve safety helmet wearing detection on embedded devices in real-time while handling multiple video streams efficiently. YOLO-M3C introduces two key enhancements. Firstly, it replaces the backbone network of YOLOv5 with the more lightweight MobileNetV3 network for feature extraction. This improvement significantly reduces the network's parameter count, resulting in enhanced computational speed and a substantial decrease in computational resource requirements. Secondly, the attention mechanism CBAM is introduced to increase the model's focus on important features, leading to further improvements in detection accuracy. CBAM helps the model better comprehend crucial information in the images, thereby enhancing detection algorithm performance. Experimental results demonstrate that YOLO-M3C achieves increased average precision while shortening the training time by 15%. Additionally, the improved algorithm reduces the parameter count of the YOLOv5 model by 52%, effectively reducing model size and training time, while enhancing detection efficiency. The improved YOLOv5 algorithm can be effectively deployed on embedded devices in workshop environments, enabling efficient real-time processing of multiple video streams and reliable safety helmet wearing detection, thus meeting the real-time detection requirements in workshop settings.