基于Multiscale CNN的齿痕舌识别研究

Research on teeth-marked tongue recognition based on multiscale convolutional neural network

齿痕舌是中医舌诊的一个重要指标。建立的多尺度卷积神经网络由多尺度特征图生成、候选区域搜索、目标区域识别模块组成,实现对轻、中、重3种程度齿痕实例分割与目标区域提取。实验结果表明,齿痕实例分割的精确度高,适用于中医智能舌诊系统应用。

Teeth-marked tongue is an important indicator of tongue diagnosis in traditional Chinese medicine.Multiscale convolutional neural network(CNN)model is built,which includes multiscale feature map generation,region proposal searching and target region recognition.As a result,the re-gion instance segmentation of teeth-marked tongue and target region extraction is obtained with light,medium and heavy tooth marks.Results show that the teeth-marked tongue recognition model a-chieved a higher accuracy,which is suitable for the application of the TCM intelligent tongue diagno-sis system.