A Practical Machine-Learning-Based Approach for Leather Automatic Defect Inspection

A Practical Machine-Learning-Based Approach for Leather Automatic Defect Inspection

Leather manual inspection is common in many industries, these methods are low efficiency and cannot be in line with automated manufacturing. In this paper, we propose a leather automated defect inspection (LADI) method based on machine learning and establish a practical LADI system composed of four modules: image acquisition, image preprocessing, image segmentation, and post-processing. The LADI method which forms the image segmentation module is a combination of multi-layer perceptron (MLP) and principal component analysis (PCA), namely MLPPCA. We propose two new algorithms that image preprocessing and post-processing to enhance the image quality and enrich details of the segmentation result. In the result analysis, compare MLPPCA, MLP, KNN, SVMRBF, GMM, show that MLPPCA has strong competitiveness in performance and execution time. The LADI system has been used in a China leather factory, the feedback shows that it combines the advantages of high inspection accuracy and short execution time.