Research On Image Recognition And Defect Detection Based On Machine Vision

In recent years,the increasing demand for automobiles has gradually enhanced the requirements for the production and manufacturing of automobile engines.The application of automatic equipment and automatic detection system has dramatically improved both the automation degree and product quality.However,at present,the engine manufacturers still adopt manual operation in some links during workpiece loading,which has the disadvantages of high labor intensity and low production efficiency.Meanwhile,during the quality inspection of finished products,the manual visual inspection is applied to detect the engine cylinder surface defects,which is likely to cause visual fatigue,and generate high level of false positive rate as well as false negative rate.Such manual work mode has seriously affected the production line efficiency and product quality.In this thesis,the shape identification technology based on machine vision is adopted to automatically identify and locate the workpiece to be loaded,which,by cooperating with the sucker mechanical arm can realize automatic loading.Besides,the machine vision technology is combined with deep learning technology to detect defects on the machined surface of engine cylinder so as to improve the defect detection rate and guarantee product quality.The main research contents can be summarized as follows:(1)By means of taking LED as the lighting solution and regarding monocular vision end-open loop vision system as the fundamental hardware system,a contrast experiment is conducted on the image processing algorithms such as image preprocessing,edge detection and morphological processing.In addition,the deep learning and convolutional neural network are theoretically researched,which lays technical foundation for the subsequent target recognition and localization,as well as the defect detection.(2)Directing at the identification and localization of the workpiece,the image of loaded workpiece is pre-processed firstly;secondly,the edge detection and morphological processing is performed on the image;thirdly,Hough detection is used to identify the shape;fourthly,the target is located and the position solution is carried out to acquire plane coordinate information;fifthly,the height information is obtained via the infrared sensor;finally,according to the palletizing strategy,the coordinate information of the workpiece center point is conveyed to sucker to achieve palletizing.(3)In regard to the shortcoming in cylinder surface defect detection,the defect detection method based on machine vision and deep learning is analyzed.A defect detection solution that combines machine vision technology and deep learning technology is proposed.In this way,when the defect detection rate is insufficient through traditional approaches,this method can defect the image for the second time via the supportive deep learning so as to enhance the defect detection rate.In addition,experiments are designed to verify the effectiveness of this method.(4)Based on MFC(Microsoft Foundation Class Library),an application program for identification,localization and defect detection is developed by integrating the above algorithms into different functional modules.The practical application of this software proves that with convenient operation and stable procedure running,the workpiece identification and localization function can acquire the workpiece position information,and the combined defect detection method can obviously improve the rate of detecting defective cylinders.Therefore,the project expected effect is achieved.