Product Surface Defects Detection Based on Multiple-Kernel Learning Feature Fusion Method

摘要

In order to achieve high accuracy and real-time requirement for product surface defect detection during the process of modern industrial production, a multi-feature fusion method based on Multiple-Kernel -Learning (MKL) is proposed. Firstly, HSV(Hue Saturation and Value) and SIFT(Scale-invariant feature transform) feature are extracted from the real -time acquisition of images, as well as the Multi-scale Equivalent Pattern Local Binary Pattern (MEP-LBP) feature presented in this paper. Secondly, according to the MKL method, three suitable kernel function were selected to train and classify of various defects. At the same time, in the process of detection, the multi-scale sliding window is generated according to the accuracy requirements of different surfaces in the captured images, so as to improve the detection performance. Experiment results show that the proposed method can meet the high accuracy requirements and ensure the real-time demand for industrial production detection.