Research On Surface Detection And Classification Of Cold-Rolled Steel Based On Machine Vision

Nowadays,automobiles,household appliances,military,aerospace and shipbuilding industries are in rapid development,steel materials as the main material,at the same time as one of the pillar industries of the national economy of the metallurgical industry products,their needs are also increasing,steel quality has a direct impact for the follow-up product quality and performance.therefore,study on strip surface quality Inspection for the domestic economy and national security have important significance.In recent years,machine vision with an absolute advantage in non-destructive testing stand out,favored by steel companies and academics,machine vision in the domestic steel market in its initial state,in view of this,we use the key machine vision methods to strip surface defect detection system algorithm-depth study.(1)Set up the test platform of the strip inspection under a simulated factory conditions,design the image acquisition device and the total working process of the system software.Design the algorithm flow.(2)With the strip image processing is proposed based on the global luminance adaptive compensation algorithm,this method is simple,fast operation,after comparison algorithm,this algorithm to determine the brightness compensating advantages of the experimental images.(3)For the image segmentation is proposed binding edge operator,improved Otsu threshold method to achieve regional segmentation,namely the use of edge intensity information to achieve local Otsu threshold segmentation algorithm,threshold selection more appropriately,at the same time,the running time is also a great advantage.(4)Based on the concept of complex scalar quantity,the paper studies the quantitative standard of complexity,and compares the three kinds of complex scalar quantity to verify the universality,validity and real-time performance of the selected standard.The standard gray scale difference and edge segmentation gray value greater than 0 pixels number of double standards applied in the field of defect detection,fast computing speed,reduce the pressure of server,high precision,20 pieces of a defect free image in 125 strip images completely detected.(5)Study the main defects,extract The shape and texture features of the image.By using the Gabor transform to extract the signal processing type texture features,improve the accuracy of classification identification.Using PCA to achieve feature data dimension reduction,the success of reduction from 87 dimensional to 19 dimensional,improved the efficiency of defect classification.Support vector machine is used to realize the classification,and the effectiveness of the texture feature of the new feature Gabor transform is verified by experiments.In this paper,97% of the rapid detection and identification of 94% of the system requirements,the implementation of the strip defect detection system and the promotion of the enterprise has a certain guiding role.