TCM Constitution Detection Based On Tongue Image Analysis Research And Implementation

The theoretical system of traditional Chinese medicine is relatively complete,but it is difficult to be objective and standardized.It is difficult to combine traditional Chinese medicine with modern medical technology.Therefore,it limits the development of TCM diagnosis and treatment.In order to further develop the traditional Chinese medicine,the state vigorously carries out the work on the integration of traditional Chinese medicine theory and computer technology.The tongue diagnosis of traditional Chinese medicine is intuitive,stable,easy to observe,and has high clinical application value,which has become an important research topic.At present,due to the good development of artificial intelligence technology,it is a very popular direction to apply it to tongue diagnosis of traditional Chinese medicine and make it automatic.There are mainly two aspects to study this direction: tongue image segmentation and tongue image analysis.There are also many researchers in these two aspects of research,but there are still some problems.In the aspect of tongue image segmentation,many people mainly use the previous image segmentation related technology for research,the accuracy and automation of the results are not high.In the aspect of tongue image analysis,this paper mainly studies the crack characteristics in tongue image analysis,but the research is relatively less at present.Only some studies have high time complexity and poor universality.Similarly,there are few system platforms based on the principle of tongue image.Therefore,this paper presents a more accurate tongue image segmentation scheme,and studies a crack recognition method based on Radon transform.Finally,the other tongue features are fused to train the model of TCM Constitution detection,and an experimental platform of TCM constitution classification is implemented combined with the segmentation method in this paper.There are three research points in this paper:(1)In order to achieve more accurate segmentation,this paper proposes fcn-16 s model based on improved skip connection structure.The edge of tongue image is processed by mathematical morphology,which solves the problem of edge burr and makes the edge more smooth.In this way,the tongue can be segmented more accurately,which lays a foundation for the subsequent analysis of tongue image.In the traditional FCN segmentation model,based on the idea of residual network,skip connection structure is improved.Firstly,shallow coding and deep decoding are overlapped,and then convolution is used to learn the detail difference between them,and then they are added to the original deep decoding.Through the comparative experiments,it is found that the improved structure can better combine the encoding and decoding features.Furthermore,the details of segmentation image semantics are supplemented.In this paper,Miou is used to evaluate the segmentation effect.After improving the skip connection structure,the index is 95.40%.It is 1 percentage point higher than before.After optimization with mathematical morphology,the segmentation index is 95.67%,which is higher than that before optimization.(2)In this paper,a tongue crack recognition method based on Radon transform is proposed.In this paper,the tongue image segmentation method is used to segment the tongue image,and then preprocess the tongue image.The processed tongue image is processed by Radon transform.The results of Radon transform are analyzed,and three features affecting the tongue image crack recognition are extracted.A crack recognition model is trained by machine learning method.After comparing the relevant experiments,the model can effectively identify the crack features,and the recognition accuracy is97%,and the recall rate is 96.6%.(3)In this paper,an experimental platform for TCM Constitution Classification Based on tongue features is implemented.Firstly,the constitution is labeled,and then the crack features are extracted by the crack recognition model in this paper,and other tongue features are extracted by the existing methods,and the features are fused to train a TCM constitution classification model.Combining fcn-16 s model based on improved skip connection structure with TCM constitution classification model,an experimental platform of TCM Constitution detection based on tongue image analysis is realized.The user uploads the tongue image to the background,and the background calls the fcn-16 s model of the improved skip connection structure to segment the tongue image,extracts cracks and other features from the segmented tongue image,and then calls the TCM constitution classification model to obtain the constitution information.