Improvements for the Recognition Rate of Surface Defects of Aluminum Sheets

In order to improve the recognition rate of surface defects of aluminum sheets, we propose a new feature extraction method in this paper, called NSST-KSR. Non-subsampled Shearlet transform (NSST) method can extract flexibly multiple scales and multiple directions feature information of the image. Kernel spectral regression (KSR) method can quickly remove redundancy and interference information and select important information as features. Combining the advantages of the two methods, the extracted features are more effective and concise, and the classification is easier. The NSST-KSR was tested with samples captured from production lines of aluminum sheets, including five true defect types of point imprints, scratches, dents, roll marks and wrinkles, and three pseudo defect types of lighting variation, oil stains and water marks. The results show that NSST-KSR method can effectively improve the recognition rate of aluminum sheets surface defects.