基于迁移学习和模型融合的中药材图像识别研究

Research on image recognition of Chinese medicinal materials based on transfer learning and model fusion

目的为了提高中药材图像识别的准确率,针对中药材形状不规则、纹理特征细微、种类繁多等特点,结合深度学习方法开展中药材图像识别研究。方法通过爬虫获取中药材图像并进行数据预处理,构建中药材图像数据集,使用Xception、DenseNet作为主干网络提取中药材特征,通过迁移学习、数据增强、模型融合等方法优化网络结构,并提出一种基于Xception和DenseNet融合的中药材图像识别模型DxFusion。结果通过迁移学习、数据增强和模型融合,DxFusion在60种中药材测试集上分类精度可达99.65%,效果优于已有分类模型。结论基于迁移学习和多模型融合的深度卷积神经网络可以提取图像中不规则的中药材特征,有效提升中药材图像识别的准确率。

Objective In order to improve the accuracy of Chinese medicinal materials image recognition,the research of Chinese medicinal materials image recognition is carried out based on deep learning according to the characteristics of irregular shape,subtle texture characteristics,and various types.Methods Images of Chinese medicinal materials were obtained through crawlers and data preprocessing was performed,image data sets of Chinese medicinal materials were established,Xception and DenseNet was used as the backbone network to extract the characteristics of Chinese medicinal materials,and the network structure was optimized by transfer learning,data augmentation and model fusion.A method called DxFusion for image recognition of Chinese medicinal materials based on the fusion of Xception and DenseNet was proposed.Results Through transfer learning,data augmentation and model fusion,the classification accuracy of DxFusion was reached 99.65% on 60 kinds of Chinese medicinal materials test sets,which was better than the existing classification models.Conclusion The deep convolutional neural network based on transfer learning and multi-model fusion can extract the irregular characteristics of Chinese medicinal materials in the image,and effectively improve the accuracy for the image recognition of Chinese medicinal materials.