Research On Method Of Cutting Tool Wear Detection Based On Image Processing

The rapid development of image processing has provided convenience for cutting tool wear detection in computer numerical control(CNC)machine,making a great contribution in cost reduction as well as ensuring the detection accuracy and reliability.To construct an image processing based automatic cutting tool wear state detection system,the involved technologies including image preprocessing(image denoising,image enhancement),wear region extraction,edge detection,wear state recognition and so on,are needed to be researched.The main work of this thesis is as follows:Firstly,a cutting tool image denoising method based on non-subsampled Shearlet transform(NSST)and fast non-local means(FNLM)filter is discussed.The decision based unsymmetrical trimmed median(DBUTM)filter is applied to eliminate the pepper and salt noise in original image;then the image is decomposed by NSST into low-frequency component and high-frequency components;the FNLM filter and the anistropic diffusion model are used to process low-frequency component and high-frequency components seperately,and the denoised image is reconstructed with these modified frequency components' coefficients.The experimental results demonstrate that the proposed method has a good performance in 4 aspects,namely,subjective visual denoising effect,peak signal to noise ratio(PSNR),structural similarity(SSIM)and running speed.Besides,the proposed method has obvious advantages against Wavelet based threshold shrink method,Contourlet based total invariance and anisotropic diffusion method and Shearlet based standard non-local means filtering method.Then,an enhancement method based on guided filtering optimized Retinex and Contourlet transform is studied.The Contourlet transform is used to decompose the image in multi-scales and multi-directions,resulting in a low-pass subband and a series of band-pass subbands.Process the low-pass subband with Retinex and use guided filtering instead of center-surround function to optimize the approximation of the illumination component;the coefficients of band-pass subbands are adjusted with the nonlinear gain function,so as to suppress the noise and enhance the edges;reconstruct the enhanced image with modified coefficients of low-pass and band-pass subbands.Taking contrast gain,sharpness gain and information entropy as evaluation indexes,the results show that proposed method has a better performance compared with double plateaus histogram equalization method,Contourlet based fuzzy enhancement method and Contourlet based Retinex without optimization method.And then,a wear region extraction method based on 0L gradient minimization model and improved Chan-Vese(CV)model is proposed.Smooth the original image using 0L gradient minimization model and segment the smoothed image with Otsu method to accomplish the initial location of the wear region in cutting tool surface;then get one-pixel contour of wear region with edge following method and take this contour as initial condition for CV model to segment the original image to extract the wear region exactly.Meanwhile,to solve the problem that the runtime is too long because of CV model's low convergence speed,introducing the gradient information as accelerator to optimize the CV model.A large amount of results show that the proposed method has a great performance in cutting tool surface wear region extraction,with a fast speed and an exact segmentation result.Compared with Otsu thresholding method,cross-entropy based thresholding method,Snake model based region extraction method and classic CV model based region segmentation method,the proposed method is much better when taking robustness,speed and accurity into account.Subsequently,an edge detection method based on linear intercept histogram Arimoto entropy and Zernike moments is proposed.Acquire image's neighborhood average grayscale information with Gaussian sliding window and build image's two dimensional histogram of grayscale-average grayscale,reducing this histogram to one dimensiona by using linear intercept method;then threshold the achieved linear intercept histogram according to Arimoto entropy,and map the threshold back to the two dimensional histogram to segment out the target region and extract the pixel-level edge;finally,re-locate the edge points using Zernike moments based edge model to get the sub-pixel-level edge of cutting tool image.A large amount of experiments have been done with cutting tool images,compared with Canny based method,space moment based method,gray moment based method and Zernike moment based method,the results show that proposed method has a better performance in speed and accuracy.Finally,selecting Haar local binary pattern(HLBP)as texture feature operator,a cutting tool wear state recognition method based on HLBP feature of machined surface image and optimized support vector machine(SVM)is proposed.Calculate the image's feature histogram with HLBP feature operator and use uniform pattern reducing histogram's dimensions to get images' s low-dimensional feature vector;then train the SVM with training image samples,meanwhile,to use chaos bee colony to optimize the kernel function parameter as well as the penalty factor;lastly,input the recognizing image samples into the SVM to get the recognition result.Experimental results show that compared with the feature exaction method based on Hough transform,histogram of oriented gradient,Krawtchouk moments and local binary pattern(LBP),taking the texture features extracted by proposed method as the input vectors in SVM will achieve a better performance in recognition.