Workpiece Crack Segmentation Method Based On CT Image

Under the influence of stress or environment,industrial components tend to appear in fine crack damage.These cracks manifest as linear defects in two-dimensional space,whereas they appear as planar defects in three-dimensional space.The existence of cracks will seriously affect the stability of the workpiece,shorten the life of the workpiece,and sometimes even break the workpiece.Therefore,it is very necessary to detect cracks in industrial parts.As an advanced means of internal detecting technology,industrial computed tomography(ICT)has been widely used in various fields.It is important to extract crack structures from CT image stably and correctly in crack measurement,analysis and evaluation.However,due to the imaging characteristics of the CT system,the imaging results may contain noise and artifacts,which brings great trouble to crack segmentation.Consequently,CT image segmentation is still a challenging problem.In this thesis,typical wheel CT images and rock core CT images in industrial CT are taken as research objects.Based on the analysis of existing segmentation algorithms,the following works are carried out:First,a CT image crack segmentation method based on Frangi filter and support vector machine(SVM)is proposed.For the case which the crack area is narrow and low in contrast,while other areas contain a lot of noise,artifacts and other interferences,this method uses the Frangi filter to improve the contrast of the image,and then uses the SVM algorithm to reduce the region of interest.The performance of the algorithm is improved by shielding the non-crack areas in the image,and the effectiveness of the algorithm is verified.Secondly,a crack segmentation method based on linear feature enhancement is proposed.By analyzing the similarities and differences between cracks,artifacts and noise in the image,an enhancement algorithm based on total variation(TV)model and Frangi filter algorithm is proposed to preprocess the image.This method improves the segmentation accuracy and robustness by enhancing the image quality.The robustness and correctness of the proposed algorithm are verified by comparing with the adaptive threshold segmentation method,level set segmentation method and mathematical morphology segmentation method.Thirdly,combining the advantages of semantic segmentation with the attributes of the studied object in this thesis,a method of crack segmentation in CT image combining UNet with residual attention mechanism is proposed.For the purpose of improving the accuracy of the model,TV-Frangi algorithm was implemented to enhance the image quality of the dataset,and the gray mean value of the workpiece region in the image was calculated to fill the background region.Then,to enhance the feature extraction ability and anti-interference,residual block and attention mechanism are used to succeed the convolution block and jump connection of traditional UNet network.At last,the effectiveness of the network model is verified by several experiments.In order to realize the segmentation of crack structures in CT images,three different solutions are proposed,which have different advantages and disadvantages and are suitable for different application requirements.