Abstract
To enhance the accuracy of identifying parts and goods in automated loading and unloading machines, this study proposes a lightweight detection model, MF-YOLOv10, based on intelligent recognition of goods' shape, color, position, and environmental interference. The algorithm significantly improves the feature extraction and detection capabilities by replacing the traditional IoU loss function with the MPDIoU and introducing the SCSA attention module. These enhancements improve the detection performance of multi-scale targets, enabling the improved YOLOv10 model to achieve precise recognition of goods' shape and quantity. Experimental results demonstrate that the MF-YOLOv10 model achieves accuracy, recall, mAP50, and F1 scores of 92.12%, 84.20%, 92.24%, and 87.98%, respectively, in complex environments. These results represent improvements of 7.11%, 11.29%, 8.51%, and 9.48% over the original YOLOv10 network. Therefore, MF-YOLOv10 exhibits superior detection accuracy and real-time performance in complex working environments, demonstrating significant engineering practicality.