| 基于多层EESP深度学习模型的农作物病虫害识别方法 |
| |
| 引用本文： | 宋余庆,谢熹,刘哲,邹小波.基于多层EESP深度学习模型的农作物病虫害识别方法[J].农业机械学报,2020,51(8):196-202. |
| |
| 作者姓名： | 宋余庆 谢熹 刘哲 邹小波 |
| |
| 作者单位： | 江苏大学计算机科学与通信工程学院,镇江212013;江苏大学食品与生物工程学院,镇江212013 |
| |
| 基金项目： | 中国博士后科学基金项目（2017M611737）和国家自然科学基金面上项目（61772242、61572239） |
| |
| 摘 要： | 为了提取图像高层语义特征、解决各种植物病虫害图像尺寸不相同的问题,提出了多层次增强高效空间金字塔(Extremely efficient spatial pyramid,EESP)卷积深度学习模型。首先,对图像进行预处理;其次,构建多层融合EESP网络模型,该模型通过对每层设置不同的空洞率进行空洞卷积,选择性地提取不同层次的特征信息,通过融合各层信息获得各种农作物病虫害图像的不同特征;最后,通过Softmax分类方法实现农作物病虫害识别。数据集包括10种农作物的61种病虫害类别,迭代训练300次,得到本文方法 Top1分类准确率最高达到了88.4%,且采用三阶EESP模型达到了最佳效果。
|
| 关 键 词： | 农作物病虫害 深度学习 卷积神经网络 空间金字塔结构 |
| 收稿时间： | 2019/11/14 0:00:00 |
Crop Pests and Diseases Recognition Method Based on Multi-level EESP Model |
| |
| SONG Yuqing,XIE Xi,LIU Zhe,ZOU Xiaobo.Crop Pests and Diseases Recognition Method Based on Multi-level EESP Model[J].Transactions of the Chinese Society of Agricultural Machinery,2020,51(8):196-202. |
| |
| Authors: | SONG Yuqing XIE Xi LIU Zhe ZOU Xiaobo |
| |
| Institution: | Jiangsu University |
| |
| Abstract: | With the rapid development of Internet of Things and artificial intelligence, the detection and treatment of crop diseases are gradually developing towards intelligence. Using computer vision methods to identify crop diseases accurately and efficiently was of great significance to ensure the normal growth of crops. In order to extract the high level semantic features of images and solve the problem of different image sizes of various plant diseases and insect pests, a multi level extremely efficient spatial pyramid (EESP) model based on deep learning was proposed. Firstly, the image was preprocessed, and then the proposed model was constructed. In order to extract characteristic information of different scales, the cavity ratio was different in each layer. By integrating the information of each layer, different characteristics of various crop pest images were obtained. Finally, crop pests and diseases were identified through image classification method. The data set included 61 pests and disease categories of 10 crops. After 300 epochs training, the experiments showed that the Top1 accuracy of the proposed model reached 88.4%, which was effectively improved by about 3 percentage points compared with that of traditional methods, and it was found that using the three layer EESP model can obtain the best effect. It had certain practical value and can be applied in actual scenarios. |
| |
| Keywords: | crop pests and diseases deep learning convolution neural network spatial pyramid structure |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
|
