Lightweight metal parts detection algorithm based on improved YOLOv11n

The progress of science and technology and the development of management concept promote the change of warehouse management to improve efficiency, reduce costs and improve service quality. As a key process, the warehouse picking mode has evolved from manual to mechanization to automation and intelligence. Although automation and intelligence improve the efficiency and accuracy of picking, many medium, small and micro enterprises are still in the stage of mechanization or manual picking. For precision parts manufacturers that cannot use automated equipment, people are optimizing picking paths and using voice technology and augmented reality (AR) technologies to improve efficiency. Augmented reality increases worker wellbeing by increasing speed and reducing error rates, but target recognition is still limited by network latency. In order to reduce the number of model parameters, narrow the model size, and ensure that the model can be successfully deployed on mobile devices with limited computing resources. This paper presents a lightweight metal parts detection algorithm based on improved YOLOv11 to improve the warehouse picking efficiency of small and medium enterprises. By constructing a metal parts data set containing real picking scenes, introducing the LSCD-Head detection head, LAMP pruning method and knowledge distillation, the model is lightweight and maintains high accuracy. The experimental results show that while maintaining the accuracy, the number of parameters and the calculation tation are significantly reduced and the speed is improved, which is better than the traditional YOLOv11n model.