周苏娟1,2  李嘉涛2  何啟森2  孟江3  刘波1

（1.广东工业大学自动化学院，广东 广州 510006  

2.广东药科大学医药信息工程学院，广东 广州 510006   

3.广东药科大学中药学院，广东 广州 510006）

摘要：针对中药饮片识别采用人工方式存在的耗费人力物力、主观性强，容易造成偏差等问题，提出基于通道注意力机制的中药饮片图像识别方法。首先，构建中药饮片图像数据库；然后，采用改进的AlexNet模型对莪术、姜、橘核和牡丹皮的饮片图像进行识别；最后，针对同一饮片不同炮制品的图像特征差异不明显问题，引入通道注意力机制。对比实验结果表明：基于通道注意力机制的AlexNet模型比AlexNet模型的平均精确度提高了2.18%，识别准确率提高了2.05%；且降低了参数量及FLOPs。

关键词：中药饮片；图像识别；通道注意力机制；AlexNet模型

中图分类号：TP391.4          文献标志码：A            文章编号：1674-2605(2023)05-0010-06

DOI：10.3969/j.issn.1674-2605.2023.05.010

Image Recognition Method for Chinese Herbal Pieces Based on 

Channel Attention Mechanism 

ZHOU Sujuan1,2  LI Jiatao2  HE Qisen2  MENG Jiang3  LIU Bo1

(1.School of Automation, Guangdong University of Technology, Guangzhou 510006, China

2.College of Medical Information Engineering, Guangdong Pharmaceutical University, Guangzhou 510006, China

3.College of Traditional Chinese Medicine, Guangdong Pharmaceutical University, Guangzhou 510006, China)

Abstract: A channel attention mechanism based on image recognition method for Chinese herbal pieces is proposed to address the issues of human and material resources consumption, strong subjectivity, and bias in manual recognition of Chinese herbal pieces. Firstly, construct an image database of Chinese herbal pieces; Then, an improved AlexNet model was used to recognize the sliced images of Zedoary Turmeric, Ginger, Orange Kernel, and Moutan Cortex; Finally, a channel attention mechanism is introduced to address the issue of insignificant differences in the characteristics of different processed products of the same slice. The comparative experimental results show that the AlexNet model based on channel attention mechanism has an accuracy improvement of 2.18% and a classification accuracy improvement of 2.05% compared to the AlexNet model; And it reduces the number of parameters and FLOPs.

Keywords: Chinese herbal pieces; mage recognition; channel attention mechanism; AlexNet model