Research For Surface Defect Dedection Method Based On Machine Vision

Product quality safety is always the key step of industrial production, which concerns with the sales quota of products in the market. Product with high qualit y could occupy a space in market competition. In modern production process, qualit y of majority of products needs artificial inspection, which is a time-consuming and labor-consuming step. Therefore, the factories need invest ing a lot of energy and financial resource to ensure the product quality safety. With the continuous development of computer technology, extension of its application fields and maturity of machine vision technology, complete automatic defect detection system used for industrial product quality inspection is starting to develop.Aiming at computer-assisted detection for product quality safety, the proper paper proposed three universal defect detection methods based on machine vision avoiding traditional methods’ limits to analyze the image of industrial product surface, judge its defect and control product quality. For regular and homogenously textured surfaces, this paper proposes an unsupervised detection method by using image feature of gray level difference of sub-image. For normal surfaces, this paper proposes two supervised learning methods. After pre-processing the surface image, one method is building the defects count vectors of qualified and defective products by using gray level difference of sub-image and fitting their Logistic regression model to detect the defects. The other method is constructing several features of an image, including average gray level difference of sub-image, color histogram and pixel’s regularity to represent the image. By computing these features of many qualified products’ images, feature learning is conducted and the feature rule of qualified product is acquired to judge whether the test product image exists defect.Afterwards, the software of three defect detection methods and their software framework are released and constructed by using visual studio 2008 on the windows platform based on Open CV. Many images of industrial products are taken to verify the proposed method by the accuracy rate of product detection. Detection rates of three methods are respectively more than 92.7%, 93.0% and 95.0%. The experimental results showed that under the circumstance of maintaining high detection accuracy, the proposed methods could inspect the defects of industrial product effectively.