Surface Defect Detection and Classification Methods for Aluminum Profiles Based on CNN

In the actual production process of aluminum profiles, various such as cracks, peel. Currently, quality inspection still relies on manual visual inspection, resulting in low efficiency and poor distinction between defects. This paper addresses these issues by experimenting with multiple deep learning models and classifiers, finding that the combination of ResNet50 and an SVC classifier performs best, achieving a classification accuracy of 92.84%. Moreover, in defect detection, the proposed method achieves a high accuracy of 98.15%. The model presented here combines Region Proposal Network and ROI Pooling layers to realize efficient and accurate defect localization and classification, offering practical industrial application value for improving the production efficiency and product quality of aluminum profiles.