| Skin is the only body surface organ in the human body,which is the first barrier to enters the human body in the external harmful substance,so it is easy to suffer from various injuries.Melanoma is a disease that is harmful in skin diseases,not only fast,and the condition has developed rapidly,and the mortality rate is as high as seventy-five percent of skin disease.If melanomas can be identified accurately in the early stage,the survival rate of patients within five years can reach more than ninety-seven percent through simple surgical resection.Therefore,an accurate judgment and recognition of early dermoscopic melanoma images can help thousands of patients get timely help and avoid losing their lives due to delayed treatment.At the same time,in view of the relatively small number of excellent surgeons and the relatively insufficient clinical experience of interns,the virtual skin tissue model is established to assist doctors to better understand the skin structure and improve their professional skills,which can achieve the purpose of timely surgical treatment for patients.The main work and innovation of this paper are as follows:(1)Aiming at the problem of noise in the image,an improved adaptive median filter combined with improved Poisson kernel bilateral filtering method is proposed.Firstly,the image is divided into different image blocks according to the similarity of gray value of image pixels,and then the shape and size of the image are adaptively changed according to the similarity of the gray value.Finally,according to the similarity degree between the pixels in the filtering window and the adjacent pixel blocks,different weight will be put according to the similarity degree,and then the high-frequency noise of the image processing is completed.According to the Poisson distribution function,different distribution functions can be formed according to different pixel gray values.Poisson distribution function is used to replace the Gaussian kernel function in traditional bilateral filtering for image processing,and finally,the image effect after noise processing is improved compared with that obtained by single processing.(2)This paper compares the classification effect of several popular neural network models for melanoma,and finds the ResNet50 convolutional neural network model which is more suitable for melanoma classification.In order to solve the problem of insufficient deep information extracted by ResNet50 convolutional neural network,an improved ResNet50 convolutional neural network model based on multi-scale attention is proposed.In the residual network structure,group normalization is used instead of batch normalization to solve the constraint between the size of batch and the size of computer GPU.In order to solve the problem of insufficient feature extraction,the multi-attention module is introduced into the network model to strengthen the extraction of deep feature information,and the final recognition accuracy was increased by about three percent.(3)At present,most of models are based on double skin,and the contact between dermis and subcutaneous tissue is not considered.At the same time,ABAQUS,a finite element simulation software,does not provide a suitable failure mode for viscoelastic materials.Therefore,this paper proposes to use the defined Mises failed criterion of the Interface subroutine VUMAT of ABAQUS,and use the shared nodes on the surface of dermis and subcutaneous tissue layer to couple and build a three-layer skin model.The mesh distortion control and enhanced hourglass control are used to avoid the mesh distortion in the process of skin cutting.In order to help young doctors better understand the skin structure,an experiments are carried out on the friction coefficient between skin and scalpel,curvature radius,angle and speed of scalpel to realize the process of cutting skin structure with scalpel. |
