Lightweight CNN-SVM Based Hybrid Model for Tongue Image Classification

Tongue diagnosis is a cornerstone of traditional Chinese medicine (TCM) due to its simplicity and significant diagnostic value, making it a widely used tool in clinical practice. The color, thickness, and moisture of the tongue coating are closely linked to various pathological conditions. However, traditional tongue diagnosis is largely based on the subjective experience of TCM practitioners, lacking objective evaluation standards, which limits its integration into modern clinical settings. While deep learning approaches have made progress in image classification, they often suffer from high complexity, significant computational demands, and limited generalizability, reducing their practical utility in clinical practice. To address these challenges, this study introduces a lightweight model for tongue coating image classification, combining Convolutional Neural Networks (CNNs) with Support Vector Machines (SVMs) to efficiently extract and classify tongue coating features. Data preprocessing techniques such as geometric transformations and noise addition were employed to enhance the dataset, improving the model's generalizability. Experimental results demonstrate that the proposed model performs exceptionally well in resource-constrained environments, achieving a classification accuracy of 94%, sensitivity of 92%, and specificity of 95%. The model effectively handles variations in tongue coating images, providing a reliable tool for TCM diagnostics. This research offers a novel method for the objective analysis of tongue coating images and TCM diagnostics, with substantial clinical applicability.