Research On Super Resolution Reconstruction Method Of Ultrasonic Image Of Weld Defects Based On CNN

Ultrasonic testing and imaging technology is widely used in the detection of weld defects of plate,which can obtain the size and location of defects and other information.Ultrasonic image is a visual presentation of defects,providing a basis for defect detection and overall performance evaluation of materials.However,due to the diffraction limit of ultrasonic wave,the imaging is easy to be interfered by electronic shock and noise,which leads to the distortion of ultrasonic echo signal,affects the imaging quality of ultrasonic image,and is not conducive to the quantitative evaluation of weld defects of plate.Aiming at the problems such as insufficient characterization of defects,low imaging resolution and serious artifacts in ultrasonic images of plate weld defects,this thesis carried out research on the super-resolution reconstruction method of ultrasonic images of weld defects.In this thesis,ultrasonic phased array S-wave fan scanning imaging method is used to detect and image weld defects of sheet metal as the original image of super resolution reconstruction.In order to reduce unnecessary calculation and improve the timeliness of image super-resolution reconstruction,a weld defect recognition and region segmentation model is established in this thesis according to the characteristics of relatively concentrated effective defect information in ultrasonic images.The model uses Convolutional Neural Networks(CNN)to locate the defects and segment the original ultrasonic images,and obtains the regional ultrasonic images with defects.According to the distribution characteristics of ultrasonic image pixels,a new nonlinear interpolation kernel function is constructed.The spatial and frequency domain characteristics of the function are analyzed,and the waveform factor constraint conditions and parameters selection principle are determined.Based on the image of phased array ultrasonic detection equipment,the image super-resolution reconstruction can improve the resolution of the defect image,enhance the contrast between the inside and the edge of the defect,and inhibit artifacts at the same time.In order to further improve the resolution of ultrasonic images and improve the quality of image defect edges,a super resolution reconstruction algorithm of ultrasonic images of weld defects based on CNN was proposed to reconstruct ultrasonic images with super resolution and evaluate them by image super resolution evaluation criteria.The main research work and conclusions of this thesis are as follows:(1)This thesis firstly analyzes and summarizes the research status of ultrasonic detection and imaging technology of plate weld defects at home and abroad.According to the characteristics of ultrasonic image of weld defects,the purpose and significance of studying ultrasonic image super resolution reconstruction are expounded,and the principle,characteristics and development status of image super resolution reconstruction technology are analyzed.(2)The common defect types and forming factors in plate weld were analyzed,the principle of ultrasonic phased array detection of plate weld was introduced,the propagation path of sound beam under S-wave fan scanning mode was deduced,and a fan-shaped imaging model was established.The ultrasonic phased array detection system for weld defects was designed and built.In order to verify the technical feasibility,the weld standard defect test blocks,through-hole standard defect test blocks and crack standard defect test blocks were designed and prepared.The actual detection imaging of the defect test blocks was carried out,and the ultrasonic fan scan image data set of weld defects was constructed.(3)In ultrasonic phased array detection imaging of weld defects,since the defect imaging area only accounts for a part of the whole ultrasonic image,if the original ultrasonic image is directly used for image super-resolution reconstruction,a large amount of background information will be introduced,increasing the complexity of image super-resolution reconstruction.In order to improve the timeliness of super resolution reconstruction algorithm and reduce unnecessary calculation,a CNN based segmentation model of weld defect region was built,the model was trained and optimized,and the test set was used to test the model.The test results show that the model can effectively segment weld defect image region to obtain the image of defect region.(4)According to the distribution characteristics of pixels in defect region images,a new ultrasonic image interpolation reconstruction algorithm based on nonlinear interpolation kernel function is proposed,and a new nonlinear interpolation kernel function is constructed.The spatial and frequency domain characteristics of the function are analyzed,and the constraint conditions of waveform factors and the selection principles of parameters are determined.Taking the kernel function of nonlinear interpolation as the base function,the pixel values of adjacent regions of the insertion points were calculated to obtain the pixel values of the insertion points,and then the interpolation images were reconstructed.The evaluation criterion of image super resolution is established,and the subjective effect of the proposed algorithm is compared and the objective evaluation index is analyzed.The experiment shows that the image resolution is improved.(5)In order to further improve the resolution of ultrasonic images and improve the defect edge structure,a CNN ultrasonic image super-resolution reconstruction algorithm was proposed,the model was trained and optimized,and the image superresolution evaluation criteria were used to evaluate it.According to the statistics of the test results,the algorithm improves the image resolution of the defect area,enhances the contrast between the inside and the edge of the defect,and suppress the artifact.The limitation of 1?2 λ is broken in the pixel size,and the information that the equivalent size of the defect is less than 1/2 of the ultrasonic wavelength is recovered.The limitation of the physical resolution of ultrasonic imaging is broken.Compared with the interpolation algorithm based on nonlinear interpolation kernel function,SRCNN and FSRCNN,the proposed method improves the PSNR of reconstructed images by6.08 dB on average.