基于深度学习的中草药植物图像识别方法研究

Research on Image Recognition of Chinese Herbal Medicine Plants Based on Deep Learning

目的:针对长白山野外实习中草药分类识别精度较低问题,对基于深度学习的中草药图像分类识别方法进行研究。方法:首先,通过平移、旋转、缩放和仿射变换对图像样本进行数据增强处理。然后,采用基于蚁群优化的图像分割算法将中草药植物叶片从复杂的图像背景中分离出来。最后,提出一种深度编码解码网络对15种中草药图像进行分类训练,并通过迁移学习解决了数据饥饿问题。结果:选取3752张中草药图像组成测试样本,平均识别精度为99.38%。结论:该方法对中草药植物图像能够表现出较高的识别精度。

Objective:To study the classification and recognition method of Chinese herbal medicine images based on deep learning,thus to solve the problem of low accuracy of classification and recognition of Chinese herbal medicine in field practice in Changbai Mountain.Methods:Firstly,image sample data was enhanced through translation,rotation,scaling,and affine transformation.Then,image segmentation algorithm based on ant colony optimization was adopted to separate Chinese herbal plant leaves from complex image backgrounds.Finally,a deep coding and decoding network was proposed to train 15 kinds of Chinese herbal medicine images,and the data hunger problem was solved through transfer learning.Results:3,752 images of Chinese herbal medicine were selected to form the test sample set.There were 15 kinds of Chinese herbal medicine,with an average recognition accuracy of 99.38%.Conclusion:This method can show high recognition accuracy for Chinese herbal plant images.