A Feature Extraction and Detection Method for Multi-Scale Defects on Mineral Surface

This study focuses on the practical engineering application of mineral surface defect detection. To address the complexity and large-scale variations of mineral surface defects, we developed a multi-scale mineral surface defect detection model. By refining the Atrous Spatial Pyramid (ASP) and integrating a pre-trained Xception architecture, we effectively extracted and enhanced multi-scale defect features on mineral surfaces. Subsequently, a Feature Pyramid Network (FPN) was utilized to fuse defect features and locate the multi-scale mineral surface defect. Experimental results demonstrate that the proposed model excels in recognizing multi-scale mineral surface defects, significantly improving the accuracy and efficiency of defect detection.