Research On Automatic Segmentation And Classification Of Cupping Spot Images Based On Deep Learning

Cupping therapy is widely used in the field of rehabilitation due to its high safety and low adverse reactions.Traditional Chinese Medicine cupping therapy has a long history.The cupping spot produced by cupping therapy is one of the important criteria of cupping effect diagnostication.At present,the method of diagnosing cupping mainly involves manually observing the color,size of ecchymosis and degree of skin depression of cupping etc.characteristics.By combining traditional computer technology to extract the features of cupping spot,the cupping effect can be determined.This method has certain rationality and practicality.However,it can be influenced by subjective factors,with a high error rate and even leading to discrimination failure.Therefore,deep learning technology is used in the research of cupping spot for segmenting and classifying.The specific content and innovation are as follows:(1)Methods for collecting and processing cupping spot images are studied.The image acquisition system is designed to obtain the original image of the cupping spots.The Gaussian filter function is used to preprocess the image of the cupping spots.The cupping spot image dataset is obtained.Implementation methods for color space model,quaternion singular value entropy index model,and convolutional neural network based on cupping spot image processing are provided.(2)The segmentation and evaluation methods for cupping spot images are studied.A Grab Cut algorithm for cupping spot image segmentation is provided,which predicts the pixel distribution of the foreground and background of the cupping spot image using a K-dimensional GMM model.The minimum cut method is used to cut off pixel edges.Segmentation of cupping spot images is achieved.A UNET semantic segmentation model is constructed for segmentation of cupping spot images based on feature extraction,up sampling,and prediction category module.Rich Semantic information of cupping spot images is acquired through encoder decoder structure.Segmentation of cupping spot images is achieved.Evaluation indicators for cupping spot segmentation is designed to analyze the segmentation results of Grab Cut and UNET algorithm.The experimental results show that the UNET cupping spot segmentation method based on deep learning can accurately separate cupping spots,with smooth edge contours,ideal segmentation results.(3)A classification of cupping spot images based on Resnet is proposed.The stimulation category of cupping spot is defined,basing on the principle of air pump and the theory of Traditional Chinese Medicine cupping diagnosis.The model overfitting is reduced by expanding the dataset by flipping,contrast enhancement and other methods.The residual unit structure is introduced to improve the ability to learn data features.The hyperparameter required for model training is setted.The Res Net50 is used as the basic network to build the cupping spot classification method.The experiment proves that the cupping spot classification model proposed in this article performs the best in comparative experiments.The cupping spot classification model has good classification performance for data enhanced cupping spots,with an Acc of over 80% for five different stimulus intensity cupping spots.(4)A cupping spot recognition system is researched and proposed.Based on Python language scripts,Py Qt5 technology,and the Py Torch deep learning framework,the development of a cupping spot recognition system software is carried out.The operational function of the system is analyzed.The operating functions of the system include importing image data,calling segmentation models,and calling classification models.The experimental results show that the system can achieve automatic segmentation and classification of cupping spot images.Reference value is provided for the construction of a real-time detection and recognition system for cupping spots in the future.The above research provides a technical support and application reference for the automatic segmentation and classification of cupping spots.