Application Research Of Chinese Medicine Tongue Diagnosis Based On Multi-task Deep Learning

Focusing on people’s comprehensive physical condition rather than just a single disease has become an important research direction for the world’s medical research.This is consistent with the idea that TCM(traditional Chinese medicine)treats people as a whole and adjusts the circulation of the human body to achieve the whole health.With the continuous improvement of living standard,people pay more and more attention to physical health,and also pay more attention to the non-invasive and painless detection of diseases.The theory of TCM believes that " the internal situation must be expressed externally ".The human body is an organic whole,and the tongue is connected to the five internal organs through the meridians.Chinese medicine believes that the tongue can understand the patient’s disease and evil,the lungs and phlegm,the disease,the cold and the heat,which plays an important role in the evaluation of the disease and prescription.Tongue diagnosis diagnoses and analyzes pathological conditions by observing the patient’s tongue.Therefore,one of the most obvious advantages of TCM tongue diagnosis is that it is painless and non-invasive.These similar places have provided opportunities for the future development of TCM tongue diagnosis.However,the traditional tongue diagnosis process relies too much on the doctor’s subjective judgment,and is also susceptible to environment,light and other factors at that time.It is irreversible and ambiguous,which is not conducive to the experience and long-term development of TCM tongue diagnosis.Therefore,the realization of the objective and modernization of clinical tongue diagnosis has become an inevitable trend in the development of TCM tongue diagnosis.With the development of computer technology and the popularity of intelligent mobile devices,it has gradually become a new development direction that the use of mobile devices to collect and intelligently analyze tongue images in the natural environment.The processing and analysis of tongue images collected by different devices under natural light has become one of the main research topics in computer tongue diagnosis.It is also the main content of this thesis.In order to realize the accuracy and robustness of the algorithm for tongue image recognition from different sources,this thesis constructs a tongue image dataset by professional instrument and intelligent mobile devices,and then performs color correction and tongue segmentation on the tongue images.Finally,the pre-processed tongue image are recognized of multi-label and multi-attribute.The main research work of this thesis is as follows:Pretreatment of tongue image: it is divided into color correction and tongue division.Firstly,several existing color correction methods are analyzed.On the basis of considering the color distribution of the tongue image,the gray world algorithm based on image entropy is selected.Secondly,according to the characteristics of the tongue image,based on the deep learning algorithm,a two-stage convolutional neural network algorithm is proposed for tongue segmentation.A morphological optimization algorithm is designed to further process the segmentation results,and the robustness of tongue segmentation is realized.Tongue recognition: The tongue images are analyzed to determine that the tongue image recognition belongs to the multi-label identification problem.In this thesis,the method of target detection is introduced to label the data,and the idea of multi-task learning is introduced to design the network,so that the network can obtain multiple attributes of the tongue image at the same time.On this basis,in order to further improve the recognition accuracy,the center loss function is combined in the training process of the network for joint training.Experiments show that the loss function joint training method can further improve the recognition accuracy while ensuring that the algorithm prediction time is constant.