The machined surface defect detection of improved superpixel segmentation and two-level region aggregation based on machine vision

Abstract

The cutting texture structure of machined surface has a great influence on defect feature information extraction after metal milling. In this paper, a machine vision defect detection method based on improved superpixel segmentation and aggregation is proposed. Firstly, an improved adaptive superpixel segmentation method based on regression prediction model is proposed. This method solves the problem that the segmentation edge of the traditional superpixel method cannot completely fit the edge of the defect target. Then, a two-level aggregation method based on limit learning machine and multi-dimensional characteristic parameters matrix is proposed to aggregate the defective and non-defective regions in the segmented images, respectively. This can eliminate the influence of the cutting texture structure of the machined surface on the defect detection. Further, the geometric feature parameters of the defect regions are extracted. The BP neural network model is constructed to predict the defect type by taking the geometrical feature parameters of the extracted defect regions as the input. Finally, the experimental verification results show that the accuracy of our proposed method for surface defect detection can reach 91.11%. This method can effectively extract more complete defect regions companied with their feature parameters, and realize the defect classification accurately. It provides an important technical support for surface quality detection in industrial automation.