| 摘 要: 文章以百部、白芍、枸杞、黄精、姜黄、蒲黄、蛇床子、益母草、郁金和枳壳片共10种中药材为研究对象,使用Python爬虫算法构建数据集,使用随机缩放、随机剪切、水平翻转对数据集进行增强,对AlexNet网络模型进行改进,具体改进内容为在原AlexNet网络模型的第三层(卷积层)之后插入一层,为新的第三层,同时在原AlexNet网络模型中引入岭回归和迁移学习,建立基于改进型AlexNet网络模型的10种中药材的图像识别方法,该模型的平均识别准确率达95.4%;数据集足够大时,可以有效地提高图像识别准确率,该模型也可应用于绝大部分需要识别的中药材类别的场景。 |
| 关键词: 卷积神经网络;AlexNet;中药材;图像识别 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 甘肃省科技计划项目(20CX9NA095 |
|
| Image Recognition of Chinese Medicinal Materials Based on Improved AlexNet |
|
LI Wanhu, WU Lili
|
(College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
894083430@qq.com; wull@gsau.edu.cn
|
| Abstract: In this paper, a total of 10 Chinese herbal medicines, including Radix stemonae, Paeonia lactiflora palls, Wolfberry, Rhizoma polygonati, Turmeric, Cattail pollen, Fructus cnidii, Motherwort, Curcuma turmeric, and Aurantium aurantium, are taken as the research objects, Python crawler algorithm is used to build data sets, and random scaling, random clipping, and horizontal flipping are used to enhance the data sets. AlexNet network model is improved by inserting one layer after the third layer (convolutional layer) of the original AlexNet network model, and it becomes a new third layer. At the same time, ridge regression and transfer learning are introduced into the original AlexNet network model to establish an image recognition method for 10 kinds of Chinese medicinal materials based on the improved AlexNet network model. The average recognition accuracy of this model is 95.4% . When the data set is large enough, it can effectively improve the accuracy of image recognition. The proposed model can also be applied to most of the scenarios where the categories of Chinese herbal medicine need to be identified. |
| Keywords: convolutional neural network; AlexNet; Chinese medicinal materials; image recognition |
