丰富多样的中草药资源是我国的宝贵财富，但中草药植物种类繁多、形态各异的特点也为种类辨别带来了困难。考虑到传统中草药识别依赖专业知识，效率低下且不易于批量处理的问题，本文旨在利用深度卷积神经网络实现对中草药植物图像的快速准确识别。

本文基于中国植物图像库（PPBC）与百度图片，通过网络爬虫、哈希去重等步骤自主构建了一个中草药图像数据集。通过向MobileNetV3-Small模型引入ECA（Efficient Channel Attention）模块，使用Focal-loss损失函数替代交叉熵损失函数，以平衡样本，构建了一个轻量且高效的中草药图像识别模型。实验结果表明，改进的模型精度提升约4个百分点，参数量降低了15.4%，内存占用量降低了26%，不仅泛用性更强，而且更加轻量化，适合在资源受限的环境中部署使用。本文还进一步开发了一个中草药图像识别Web系统，该系统能高效准确地识别中草药图像类别并展示专业词条，为用户提供了一种方便快捷的中草药识别和学习工具。词条内容均来自国家中医药管理局中医药名词术语成果转化与规范推广项目，内容准确可靠。本文有利于进一步促进中医药文化的传播和应用。

The wide variety and diverse morphologies of medicinal plants pose challenges to species identification. This study aims to achieve fast and accurate recognition of Chinese herbal medicine plants using deep convolutional neural networks.

This research constructs a Chinese herbal medicine image dataset independently, based on the Plant Photo Bank of China (PPBC) and images from Baidu. By incorporating the Efficient Channel Attention (ECA) module into the MobileNetV3-Small model and replacing the cross-entropy loss function with the Focal-loss function to balance the samples, a lightweight and efficient Chinese herbal medicine image recognition model was developed. Experimental results show that the accuracy of the improved model has increased by approximately 4 percentage points, with a 15.4% reduction in the number of parameters and a 26% reduction in memory usage. Based on the Flask framework, this study has developed a web system for Chinese herbal medicine image recognition on the basis of the improved model, providing users with a convenient and fast tool for identifying Chinese herbal medicine.