Research Of Surface Defects Detection Of Cambered Workpieces Based On Machine Vision

In national "twelve five" plan, intelligent manufacturing equipment industry is vigorously promoted. And, development of intelligent automation testing is advanced. Machine vision plays a key role in the automation of detection.Large-scale production and zero waste control are required of online automatic detection and machine vision in enterprise. Currently, there is more research of automatic detection in plane workpiece. Automatic detection on cambered workpiece surface is more difficult to achieve because of its variety and certain surfaces.Therefore, this article has practical significance to expand the research of cambered workpiece surface defects detection based on machine vision.The paper has accomplished image acquisition, image preprocessing and image feature extraction. Finally, classifier was builded to complete defect image recognition and classification in time domain and frequency domain.After comparing, the results in the frequency domain had better detection.Here are the main contents in this paper:(1) Achieving image acquisition of cambered workpiece surface: CMOS image sensor and uniform spherical diffuse illumination based on LED. Then, in the VC platform image acquisition was completed.(2) Image preprocessing of cambered workpiece surface: the collected image go through a series of image preprocessing, which was consist of image graying, image filtering, image enhancing, image hough circle detection, image mask operation. The purpose of these is to maintain the integrity and clarity of image information in maxinum extent.(3) Calibration method based on length and are: due to curvature of cambered workpiece, calibration method based on length was used to calculate diameter. When it comes to calculate area of defect, area-based calibration method was chose. It is abetter solution to the influence of curvature on cambered workpiece in feature calculation.(4) Image segmentation and feature extraction of cambered workpiece surface: image segmentation was conducted, thus single defective portion was extracted. Feature value was extracted both in time domain and frequency domain. Geometric characteristic were selected in time domain. While, in frequency domain, mean of gray value and variance were chose after the action on the surface of the workpiece image filtering of 5×8 Gabor filter.(5) Building data dimension reduction and classification: In frequency domain,the dimensionality of data classifier construction: In time domain, binary structure classifier was taken to classify the time-domain characteristics.Dimension of the extracted image feature vector is too high. So principal component analysis method was used for dimensionality reduction of characteristic data. Constructing nearest neighbor classifier based on the mahalanobis distance to complete workpiece classification of 6 levels include defect-free image.