| 基于深度学习的中药材饮片识别 |
| 点此下载全文 |
| 引用本文：韩 勇，兰 杰，郭瑞瑶，邢聪颖，黄新杰，霍迎秋.基于深度学习的中药材饮片识别[J].西北农业学报,2023,(11):1859~1868 |
| DOI:10.7606/j.issn.1004-1389.2023.11.020 |
| 摘要点击次数: 21 |
| 全文下载次数: 120 |
|
| 基金项目:国家重点研发计划（2017YFC0403203）；农业农村部农业物联网重点实验室开放课题（2018AIOT-09）；校级创新训练项目（X201910712095）；陕西省重点研发计划（2023-YBNY-080）; 国家自然科学基金(41771315)。 |
|
| 中文摘要:为了解决中药饮片种类繁多、形态相似导致难于快速、准确识别的问题，构建一个包含50种，共计 15 622张图像的中药材饮片数据集，并基于Keras框架建立深度学习模型，模型包含4个稠密块和3个过渡层，每个稠密块和过渡层交替连接。最后利用全局最大池化层将稠密块特征向量化，并加入丢弃法来防止过拟合，优化DenseNet-201网络模型。结果表明该模型在43种中药上的识别率可以达90%以上，最高识别率达95.21%。因此，基于深度学习的方法可以有效解决中药饮片快速、智能识别的问题。 |
| 中文关键词:中药饮片 深度学习 图像识别 机器学习 人工智能 |
| |
| Identification of Chinese Herbal Medicine Slices Based on Deep Learning |
|
|
| Abstract:In order to solve the problem that it is difficult to identify Chinese herbal medicine slices quickly and accurately due to their variety and similarity,a dataset containing 50 kinds of Chinese herbal medicine slices with a total of 15 622 images was constructed,and a deep learning model was established based on Keras framework.The model includes 4 Dense Blocks and 3 Trasition Layers,each Dense Block and Trasition Layer alternately connected.Finally,the global maximum pooling layer was used to vectorize the characteristic value Dense Block,and the discard method was added to prevent over-fitting for optimizing the DenseNet-201 network model.Finally,the global maximum pooling layer was used to feature the Dense Block,and the discard method was added to prevent over-fitting for optimizing the DenseNet-201 network model.The experimental results showed that the recognition rate of the model on 43 kinds of Chinese medicine can reach more than 90%,and the highest recognition rate reaches 95.21%.Therefore,the method can quickly and intelligentlyrecognizethe Chinese herbal medicine slices based on deep learning. |
| keywords:Chinese herbal medicine slices Deep learning Image identification Machine iearning Artificial intelligence |
| 查看全文 查看/发表评论 下载PDF阅读器 |
