Research On Surface Defect Detection For The Parts Based On Machine Vision

With the development of the detection technology of industrial parts, the modern industry has made the requirements for detecting technology more and more higher. The traditional detection method has been unable to meet the needs of industrial production. So, a new detection technology based on machine vision appears in time. The detection based on machine vision can meet the requirements of modern industrial production with the advantages of higher accuracy, faster detection speed and higher degree of automation.The premise of this topic with the background of the socialized production, has made gear parts as the research object, they are mainly common in the industrial production. Building the hardware platform for collecting images, using the machine vision technology for defecting the detection. The framework of this paper is divided into four parts:(1) The overall scheme of the system research. Including the hardware design and the software design, the hardware design is mainly the design of image acquisition platform; the software design is mainly the design of detection process and development environment. (2) The image preprocessing of the parts. The collected image denoising processing, according to the shortcoming of the median filtering cannot judge the noise pixels and signal pixels, proposing an improved median filtering algorithm based on the mathematical difference which both in the neighborhood.(3) The research on the defect detection technology. This paper proposes a segmentation algorithm for one dimensional Otsu to extract the defect of the parts surface, which is based on the wavelet transform. With the characteristics of the wavelet multi resolution analysis, reconstruct the image which decomposed by wavelet with the low-frequency, then use the Otsu threshold segmentation method to segment the defects. (4) Identification and classification the defects. Design the classifier based on the theory of BP neural network, using the feature parameters of the defects which mentioned in this paper as the input of the classifier, the different types of defects as the output, finishing the recognition and classification of the defects, The experiment proved that this system meet the demand of industrial detection.