Based On Texture Features Steel Surface Defect Detection Technology Research

This paper based on texture feature of the steel plate surface defects detection, mainly study about the steel plate surface defects detection of related theory and technique, the key discussion is the image of the texture characteristics t; the defects of the steel plate recognition is successful to use the BP neural network method to recognize. The content of the work of this paper are as follows:1. The image preprocessing. Through some experiments the aticle analysis the preprocessing part about gray, image enhancement, image segmentation, in addition, in the part of the edge detection of image segmentation improved Canny operator, with the median filter instead of traditional gaussian filter; With3×3sobel operator instead of traditional2×2in the neighborhood for finite difference for gradient mean amplitude; Using adaptive double threshold to do image segmentation, the improved Canny operator edge detection has better antinoise ability, edge more accurately, also solved the problem of double threshold automatic acquisition.2. Image feature selection and extraction. It discusses the defect image of features about gray feature, texture feature, invariant moment feature; Because there exist a correlation between characteristics, not all of the features on the classification will have obvious function, so we need to extract the characteristics, in other words, reduction the dimension processing. Principal component analysis is one of the most commonly used a linear dimension reduction method, this thesis using this method reduce dimension.3. Design classifier. Design the BP neural network classifier, this thesis use BP neural network based on texture characteristics to classify steel defect images.This thesis using BP neural network of three layers to recognize normal pictures, roller-markers and scratch pictures,the classification rate is86.7%...