Scholarlink

<p>Tongue coating characteristics provide important diagnostic cues in Traditional Chinese Medicine (TCM). However, coating color can be influenced by food intake or medication, potentially leading to misinterpretation in clinical assessments. Existing deep learning-based methods for tongue coating classification often suffer from limited accuracy due to insufficient modeling of color-sensitive features and heavy reliance on manual hyperparameter tuning. To address these challenges, we propose a task-oriented convolutional neural network framework, termed IPOA-IENet. This framework integrates a color-sensitive, lightweight Ghost-enhanced improved EfficientNet, whose hyperparameters are optimized using an improved Pelican Optimization Algorithm (IPOA). Specifically, a Ghost-enhanced improved EfficientNet (G-IENet) backbone architecture is adopted, within which a color fusion-based efficient channel attention mechanism is embedded to enhance the discrimination of staining-related color features. The IPOA is further employed to automatically optimize the hyperparameters of G-IENet, improving optimization efficiency and robustness. Although population-based hyperparameter optimization incurs additional offline computational overhead, it is well-suited for high-performance computing, whereas the optimized IPOA-IENet supports efficient real-time inference during deployment. Experimental results on a stained tongue coating dataset demonstrate that IPOA-IENet achieves superior classification performance compared with baseline and existing methods across multiple evaluation metrics. Specifically, the proposed method outperforms TCM practitioners by approximately 30% in accuracy, 27% in recall, and 29% in F1-score. Furthermore, cross-dataset validation on the public TonguExpert dataset verifies the strong generalization capabilities of the proposed model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Stained tongue coating images classification via pelican optimization algorithm and convolutional neural network

  • 期刊论文

  • 发表日期：

    2026年4月21日

查看全文
  • Tiannuo Liu,
  • Wenjing Xu,
  • Lin Sun,
  • Na Zhang

摘要

Tongue coating characteristics provide important diagnostic cues in Traditional Chinese Medicine (TCM). However, coating color can be influenced by food intake or medication, potentially leading to misinterpretation in clinical assessments. Existing deep learning-based methods for tongue coating classification often suffer from limited accuracy due to insufficient modeling of color-sensitive features and heavy reliance on manual hyperparameter tuning. To address these challenges, we propose a task-oriented convolutional neural network framework, termed IPOA-IENet. This framework integrates a color-sensitive, lightweight Ghost-enhanced improved EfficientNet, whose hyperparameters are optimized using an improved Pelican Optimization Algorithm (IPOA). Specifically, a Ghost-enhanced improved EfficientNet (G-IENet) backbone architecture is adopted, within which a color fusion-based efficient channel attention mechanism is embedded to enhance the discrimination of staining-related color features. The IPOA is further employed to automatically optimize the hyperparameters of G-IENet, improving optimization efficiency and robustness. Although population-based hyperparameter optimization incurs additional offline computational overhead, it is well-suited for high-performance computing, whereas the optimized IPOA-IENet supports efficient real-time inference during deployment. Experimental results on a stained tongue coating dataset demonstrate that IPOA-IENet achieves superior classification performance compared with baseline and existing methods across multiple evaluation metrics. Specifically, the proposed method outperforms TCM practitioners by approximately 30% in accuracy, 27% in recall, and 29% in F1-score. Furthermore, cross-dataset validation on the public TonguExpert dataset verifies the strong generalization capabilities of the proposed model.