Research On Tongue Image Detection And Segmentation Method Based On Deep Learnin

Objective: Tongue image acquisition and extraction is an important part of intelligent tongue diagnosis.At present,tongue image acquisition generally needs to be carried out in a closed,stable light standard environment,but this is not conducive to the use of mobile devices in natural environment for tongue image acquisition and processing.Therefore,in view of the problems faced by tongue image acquisition and processing in natural environment,two aspects of tongue image positioning detection and semantic segmentation are studied respectively.Effective tongue image data plays a crucial role in subsequent diagnosis.Methods: Aiming at the influence of environmental factors in the process of tongue image detection,an improved YOLO V4-tiny tongue image location detection algorithm is proposed.The GELU activation function is integrated into the model to improve the training speed of the model and reduce the number of model parameters,and then the CA mechanism is integrated into the model to improve the detection accuracy and fault tolerance of the model.Aiming at the problem that the segmentation of tongue image is not fine enough in the process of tongue image segmentation,an improved DeepLabv3+ tongue image segmentation algorithm is proposed.The algorithm combines CA and SE attention mechanisms,and the SP stripe pooling layer is embedded in the ASPP feature extraction module to improve the attention of the model to the tongue image area.The ability of the model to establish long-distance spatial dependencies and dependencies is enhanced.Results: Experimental results show that compared with other classical algorithms,the improved YOLO V4-tiny algorithm has better performance in terms of training speed,tongue detection speed and detection accuracy,and the MAP value reaches 98.89%.Experimental results show that the improved DeepLabv3+ algorithm is superior to other semantic segmentation algorithms in terms of segmentation accuracy and robust performance from the actual segmentation application and the comprehensive analysis of objective indicators.Conclusion: In this paper,the improved YOLO V4-tiny algorithm and the improved DeepLabv3+algorithm are proposed,which can quickly locate and detect the tongue area and accurately segment the tongue area in the complex natural environment,greatly reducing the environmental requirements of tongue image acquisition and processing.