Research On Machine Vision Based Injection Product Defects Detection System

The technique of product inspection based on machine vision has been propelled by the development of computer science and machine vision in most recent years, and it seizes more and more researchers'attention. By snatching the images of the produce and comparing them with standard one, machine vision based inspection can give a real-time evaluation on the quality of the product without contiguity. For these merits, this study focuses on the defects detection of injection molding machine (IMM) product based on machine vision. This research is sponsored by NEU key laboratory of process industry automation fund.Firstly, by researching the process of injection production and analyzing relative literatures, this thesis finishes the hardware and software design of defects detection.system and solves the problems in capturing and processing images.The background and noise in the captured image makes the defect detection much harder. A method combining threshold and image subtraction is proposed to segment the background from the object image entirely. To filtering the noise and enhance the robustness of the segmentation method, a new filter algorithm is presented which provides a better result than the traditional methods.The research on defect detection feature extraction consists of two parts. For detecting shape defects, a rapid-inspecting method is proposed under the condition that the defect's information is reserved entirely. For detecting texture defects, by regroup the texture feature vectors, a new method is presented to enhance the efficiency of classification.At last, a multi-class classification support vector is designed for the multi-class defects detection. The basic structure of IMM products defect detection system based on machine vision is completed by using methods proposed above. The test results show that this system demonstrate a high detecting precision.