The Institute of Digital Medical, School of Intelligent Medicine, Chengdu University of Traditional Chinese Medicine, Chengdu, China.
Medicine (Baltimore). 2023 Jun 23;102(25):e34085. doi: 10.1097/MD.0000000000034085.
When the similarity of medicinal materials is high and easily confused, the traditional subjective judgment has an impact on the identification results. Use high-dimensional features to identify medicinal materials to ensure the quality of Chinese herbal concoction products and proprietary Chinese medicines.
To study the identification algorithm of traditional Chinese medicinals (TCM) microscopic images based on convolutional neural network (CNN) to improve the objectivity and accuracy of microscopic image identification of TCM powders.
Microscopic image datasets of 4 TCM powders sclereids of Rhizoma Coptidis, Cortex Magnoliae Officinalis, Cortex Phellodendri Chinensis, and Cortex Cinnamomi were constructed, and 400 collected images, as the model training and testing objects, were identified and classified by AlexNet model, VGGNet-16, VGGNet-19, and GoogLeNet model.
The average recognition accuracy in the tested microscopic image of AlexNet model, VGGNet-16, VGGNet-19, and the GoogLeNet model is 93.50%, 95.75%, 95.75%, and 97.50% correspondingly.
The GoogLeNet model has a higher classification accuracy and is the best model to achieve real-time. Applying the CNN to the identification of microscopic images of TCM powders makes the operation of TCM identification simpler and the measurement more accurate while improving repeatability.
目的：研究基于卷积神经网络（CNN）的中药微观图像识别算法，提高中药粉末微观图像识别的客观性和准确性。
方法：构建黄连、厚朴、黄柏、肉桂 4 种中药粉末的微观图像数据集，采集 400 张图像作为模型训练和测试对象，分别采用 AlexNet 模型、VGGNet-16、VGGNet-19 和 GoogLeNet 模型对其进行识别和分类。
结果：在测试的微观图像中，AlexNet 模型、VGGNet-16、VGGNet-19 和 GoogLeNet 模型的平均识别准确率分别为 93.50%、95.75%、95.75%和 97.50%。
结论：GoogLeNet 模型的分类准确率更高，是实现实时性的最佳模型。将 CNN 应用于中药粉末的微观图像识别中，使中药鉴定操作更加简单，测量更加准确，同时提高了重复性。
