A lightweight coal gangue detection method based on multispectral imaging and enhanced YOLOv8n

Abstract

In response to challenges such as low accuracy, slow detection speed, and large model size in traditional coal gangue identification methods, this paper proposes a lightweight coal gangue detection method based on YOLOv8n and Multispectral Imaging (MSI) technology. Initially, MSI technology is employed to collect spectral data of coal and gangue. Three optimal spectral bands are selected based on the accuracy of coal gangue recognition and spectral data correlation, and a pseudo-RGB image is constructed. Subsequently, the FasterNet network is introduced to enhance the YOLOv8n object detection algorithm. A C2f-Faster module is designed to reduce model size and improve detection speed. To address the issue of low coal gangue detection accuracy, an Efficient Channel Attention (ECA) module is introduced after the P3, P4, and P5 layers to dynamically weight the input coal gangue images, obtaining more detailed feature information. Finally, the Neck layer is redesigned, and a Light Bi-directional Feature Pyramid Network (Light-BiFPN) structure is proposed to minimize parameter increase while enhancing recognition accuracy. Experimental results demonstrate a significant improvement in both detection accuracy and speed for the enhanced model, making it suitable for deployment on mobile devices and positively contributing to intelligent underground coal mining.