摘要
中药饮片是指药材经炮制后可直接用于中医临床或制剂生产的药品,针对中药饮片种类繁多、形状各异、鉴别困难的问题,提出一种改进Res2Net和注意力的中药饮片识别模型BIM-Res2Net50-IECA。首先,在Res2Net的基础上引入双向融合策略,促进不同尺度特征之间的有效交互,获取更精细和丰富的特征信息;其次,使用最大池化改进ECA注意力机制,同时增强全局视角和显著性特征,突出中药饮片重要的特征区域;最后,结合Softmax Loss和Center Loss构造联合损失函数,有效地调节类内以及类间距离,提高分类的准确性。实验表明,基线模型能有效提取多尺度特征,BIM-Res2Net50-IECA在构建的16类中药饮片数据集上的准确率、精确率、召回率和F1-Score分别为94.74%、94.27%、94.83%和94.55%,与先进的Tansformer分类模型相比,具有更低的计算复杂度和更高的准确率,能为中药饮片的智能识别提供有力支持。
Chinese herbal medicine decoction pieces refer to drugs that can be directly used in clinical practice or formulation production after processing.In response to the problems of the wide variety,different shapes,and difficult identification of Chinese herbal medicine decoction pieces,this article proposes an improved Res2Net and attention mechanism Chinese herbal medicine decoction piece recognition model BIM-Res2Net50-IECA.Firstly,a bidirectional fusion strategy is introduced on the basis of Res2Net to promote effective interaction between features of different scales and obtain more refined and rich feature information.Secondly,using max pooling to improve the ECA attention mechanism,while enhancing the global perspective and salient features,highlighting the important feature regions of traditional Chinese medicine decoction pieces.Finally,a joint loss function is constructed by combining Softmax Loss and Center Loss,which effectively adjusts the intra-and inter class distances and improves the accuracy of classification.Experiments have shown that the baseline model proposed in this paper can effectively extract multi-scale features.BIMRes2Net50-IECA achieved accuracy,precision,recall,and F1-Score of 94.74%,94.27%,94.83%,and 94.55%,respectively.Compared with the advanced Tansformer classification model,it has lower computational complexity and higher accuracy,providing strong support for intelligent recognition of traditional Chinese medicine decoction pieces.
作者
谷瑞
宋翠玲
李元昊
Gu Rui;Song Cuiling;Li Yuanhao(Nanjing University,Nanjing 210003,China;Suzhou Industrial Park Institute of Services Outsourcing,Suzhou 215123,China;Shanghai University of Traditional Chinese Medicine,Shanghai 200003,China)
出处
《国外电子测量技术》
2024年第9期130-140,共11页
Foreign Electronic Measurement Technology
基金
2023年江苏省高职院校教师专业带头人高端研修项目(2023TDFX010)资助。
