Magnets Surface Defect Inspectiong Method Base On Computer Vision

With the improvement of national economy standard, requirement for high quality,high precision and high reliability products is also increasing. The traditional testingmethods can not meet the demand for quick detection of products. At the same time, thepromotion of computer technology makes it possiable to extensive application of digitalimage processing. Therefore, the detection method of products surface, which is base onmachine vision, has been popular in products quality control. In this paper, we made aresearch into the detection of defects in magnetic, according to the magneticcharacteristics of the magnetic, we proposed the algorithm use for magnetic flawdetection and its application in magnetic-line detection system.For component orientation, we proposes an approach based pyramid edge templatematching technique. By selecting the optimal pyramid metric function, the result meetrobustness requirements and positioning results for nonlinear, linear light, featuremissing, defaced, etc. insensitive. Through choicing the vector as well as improvingsearch termination conditions, sparse matrix drop Vega speed computation, we proposefour optimization search strategy to accelerate the process of matching operation. Wealso design the DXF file analysis tool, the shape parameter can be read directly from theAutocad file, it will be more convinient to adjustment parameter. Experiments show thatthe method can resist nonlinear light, partly obscured, defaced with strong robustness,positioning time will less than50ms.In order to extraction the texture feature and remove the texture of the magnetic.Extraction and filtering in the texture, we compares the two integrated filtering methods,the first one is the frequency domain of the image band filter structure, by removing thevertical direction along the grain direction of the spectrum, we can remove the textureobject airspace; The second method is building anisotropic filter to remove the texture,at the same time to keep edge of defects. Experimental results show that the effect issuperior to the anisotropic filtering in frequency domain filtering has a better results.In order to separate defective area and classification. We firstly extracted grayvalue, standard deviation and entropy from the texture regions, and then the defect area will be characterized by the extracting characterizes. Then we apply morphologicaloperations to the defect area. Secondly, by providing a sample defect image defectlocation, size of the area, circularity, aspect ratio bounding rectangle, composed of fiveconvex-dimensional feature vector of training classifier, linear Fisher classifier is use toclassify the defect. Experimental results show that defect extraction and classificationmeet the standards, measurement and the surface detection time within80ms.Finally, we introduce the implementation of software systems, include MVCsoftware architecture and multi-threaded implementation. According to the algorithmwe mention above and based on the MFC and opencv programming. The results showthat the practical application of the detection time of a picture within150ms, continuousand stable operation time of the software is more than5hours and mistake rate is lessthan2%.