YOLO-Parts: A Lightweight Factory Parts Detection Method

Addressing the current challenges of low accuracy, large model parameters, poor detection real-time performance, and not applicable to mobile devices or edge devices in existing part detection algorithms, this study proposes a factory parts detection method based on the improved lightweight model YOLO-Parts. The method introduces FasterNet with SPPFCSPC structure as the backbone network model of YOLOv5 to construct a lightweight network. Meanwhile, Coordinate attention (CA) mechanism is employed to improve the recognition accuracy of the object in complex environments. In order to improve the localization accuracy of the detection network to the bounding box of the object part, the Alpha-IoU bounding box regression loss function method is introduced. The experimental results show that the YOLO-Parts model achieves mean average precision $(\mathrm{mAP}$) and precision (P) values of 72.7% and 75.0%, respectively, with the model size of 6.6 MB. Compared with the original network model, the improved model has a faster recognition speed, higher localization accuracy, and smaller memory usage, thereby enhancing sorting efficiency while meeting the requirements of precise sorting tasks. Our source code and pre-trained models are available at https://github.com/zhang-dut/yolov5-parts.