Research On Recognition Of Strip Surface Defects Based On Image Processing Method

With the rapid development of science and technology, tip fields such as military project, astronautical technology, auto industry, electronic technology, etc have asked steel production for higher standard to meet with quality requirements and surface requirements. In order to rapid fixed surface defect position in the production process and strengthen the monitoring during the productive process, we usually rely on the surface defect of plate inspection system to realize steel production real-time inspection.which provide the improvement of the quality of steel an important guarantee and vigorously promote steel production enterprise core market competitiveness.Nondestructive testing of steel surface which based on the machine vision inherited and carried forward real-time and reliability requirements, and innovation detection methods; by comparing the differences between image features, easy to distinguish and concluded that judgment, detection of core technology through the programming to realize, easy to update and maintenance, greatly reduce the manual operation consumption and improve production efficiency, and is worth our while to constant innovation and research.This paper combined with machine vision principle structure and steel production practice, according to several common steel plate defects, design a set of testing system which can real-time automatic inspection sheet surface defects,and through a lot of calculation and simulation experiments, suitable for steel plate characteristics of the design parameters, and has initially achieve defect inspection detection positioning and automatic recognition and classification, the main work is as follows:1.To test and verify the reliability of the test results, we select five kind of defect type, the defects in the steel industry is more common, features obvious differences.2.To reduce noise error and camera image acquisition caused by the dark image fuzzy problem, design the multiple ways of filtering noise and image enhancement technique. The method is simple, the algorithm can direct invocation, convenient and quickly.3.For simplified image information, get the needed information, defect edge difference which is the point of different defect, edge detection through edge detection algorithm, inspired by the edge of the image edge gray features.4.With local characteristics instead of object the feature classification principle, we extract the morphological characteristics, gray features and texture features.5.To solve the problem of characteristics value can not be directly separable, design based on the BP neural network classification model, which is of self learning adapt to the characteristics of strong ability.