Research On Crack Automatic Detection For Fluorescent Magnetic Image

The methods of crack detection for industrial devices are various.Compared to the price of expensive ultrasonic flaw detector,fluorescent magnetic crack detection is widely used,due to the advantages of the low cost,high sensitivity and fast detection speed.But the traditional surface detection of fluorescent magnetic mainly depends on the manual identification to achieve,this method of detection is low efficiency and low accurate,even lead to false judgment duo to visual fatigue,while the human body is easy to harm the health in the ultraviolet environment for a long time.Therefore,it is of great theoretical and practical value to study the automatic detection method of fluorescence magnetic crack based on digital image processing and pattern recognition.Four key steps of image preprocessing,image segmentation,feature extraction and crack recognition are studied in this thesis.Firstly,the method of the eliminating noise of weighting directed smoothing filter is used to preprocess the image.The aim is to eliminate the image noise and preserve the edge information as much as possible,which provides favorable conditions for gray-gradient co-occurrence matrix maximum entropy image segmentation.The experimental results show that this method has better de-noising effect and better preserving edge information than traditional de-noising and median filtering,thus avoid edge blurring.Secondly,an improved maximum entropy segmentation algorithm based on GA for gray-gradient co-occurrence matrix is gived to solve the crack fracture for the traditional the maximum entropy segmentation of the gray-gradient co-occurrence matrix and improve the segmentation speed.the point-neighborhood search method is proposed to eliminate the noise introduced by the improved maximum entropy segmentation algorithm based on GA for gray-gradient co-occurrence matrix and to improve the segmentation quality of the image.Thirdly,According to the difference of shape,grayscale and gradient information between crack and non-crack images,thesis uses Hu Moment Invariant and HOG descriptor as the input samples of the classifier,and improves the HOG descriptor based on hough transform.Finally,aiming at the shortcomings of man-made selection of penalty factors and kernel parameters in SVM algorithm,thesis gives an improved algorithm of SVM based on PSO for fluorescence particle crack detection,in order to further improve the recognition rate of crack.