A lightweight bauxite detection and identification algorithm

Abstract

To address the issues of low efficiency and poor accuracy of manual methods in bauxite sorting, this paper proposes a lightweight ore identification and detection algorithm, CL-YOLOv11, based on an improved YOLOv11. This algorithm incorporates a C3k2-GD module integrating GhostModule and dynamic convolution in the backbone network to enhance adaptive feature extraction while reducing redundant parameters. Furthermore, a lightweight shared detail enhancement detection head (LSDECD) is constructed in the detection head, improving the accuracy of capturing small targets and texture edges by introducing detail enhancement convolution and learnable dynamic factors. Experimental results show that compared to the original model, CL-YOLOv11 reduces the number of parameters by approximately 33%, decreases floating-point computation by 23%, and significantly improves inference speed. It achieves a balance between high accuracy and high efficiency while maintaining a high detection accuracy of 93.60%, providing an effective technical solution for automated intelligent bauxite sorting.