HLC-YOLOv8: An algorithm for disordered parts recognition based on improved YOLOv8

Abstract

In order to address the challenge of recognizing parts placed on an assembly line in a disordered manner, a disordered parts recognition algorithm HLC-YOLOv8 based on improved YOLOv8 is proposed. To enhance the accuracy and robustness of image recognition and processing, the HorNet module is introduced into the backbone network. This module is capable of effectively fusing features from different layers, thereby improving the feature extraction capability. Furthermore, to enhance computational efficiency and speed, the LightConv module is employed in the neck network. This module features a simpler structure with a smaller number of parameters, rendering it more efficient than the standard convolutional operation. In Addition, the ConTainer module is integrated into the conventional YOLOv8 architecture, which integrates and understands the contextual information in the image more efficiently, enhances the sensory field of the model, and improves the accuracy of small target recognition. The experimental results on the disordered parts datasets show that the improved model in this paper has better detection performance, and the detection accuracy and speed have been significantly improved to achieve the purpose of real-time identification of disordered parts.