| 基于Inception与Residual组合网络的农作物病虫害识别 |
| |
| 引用本文： | 冯广,孔立斌,石鸣鸣,贺敏慧,何雅萱.基于Inception与Residual组合网络的农作物病虫害识别[J].广东工业大学学报,2020,37(3):17-22. |
| |
| 作者姓名： | 冯广 孔立斌 石鸣鸣 贺敏慧 何雅萱 |
| |
| 作者单位： | 广东工业大学 自动化学院, 广东 广州 510006 |
| |
| 基金项目： | 国家自然科学基金资助项目（70971027） |
| |
| 摘 要： | 针对我国农作物病虫害识别方法中存在的速度慢、主观性强、所需专业知识要求高以及识别成本高等问题，提出一种基于Inception与Residual结构组合的Inception-resnet-v2网络模型的农作物病虫害识别方法，以实现精准高效的农作物病虫害识别。网络使用residual结构，采用跨层连接方式将低层特征与高层特征进行组合学习以增加网络深度。同时加入了Inception结构，既能保持网络结构的稀疏性，又能利用密集矩阵的高计算性能，加快了训练速度。最后通过Softmax分类器进行多分类预测。与传统方法相比，本文方法收敛速度更快，不仅准确率达到96.67%、精确度达到90.77%、召回率达到89.72%，还使病虫害识别的不同类别更加均衡，改善了传统方法中对特定类别识别效果差的问题。
|
| 关 键 词： | 农作物病虫害识别 Inception结构 Residual结构 Inception-resnetl-v2 Softmax分类器 |
| 收稿时间： | 2019-12-17 |
Crop Pest Recognition Based on Inception and Residual Combined Network |
| |
| Feng Guang,Kong Li-bin,Shi Ming-ming,He Min-hui,He Ya-xuan.Crop Pest Recognition Based on Inception and Residual Combined Network[J].Journal of Guangdong University of Technology,2020,37(3):17-22. |
| |
| Authors: | Feng Guang Kong Li-bin Shi Ming-ming He Min-hui He Ya-xuan |
| |
| Affiliation: | School of Automation, Guangdong University of Technology, Guangzhou 510006, China |
| |
| Abstract: | Aiming at the problems of slow speed, subjectivity, high requirement of professional knowledge and high recognition cost in the identification methods of crop pests and diseases in China, a crop pest and disease identification method is proposed based on Inception-resnet-v2 network model combining Inception and Residual structure. To achieve accurate and efficient crop pest identification, the network uses a residual structure to combine low-level features with high-level features to increase network depth. At the same time, the Inception structure is added, which not only maintains the sparsity of the network structure, but also utilizes the high computational performance of the dense matrix to speed up the training. Finally, the multi-class prediction is performed by the Softmax classifier. Compared with the traditional method, the proposed method converges faster, not only with an accuracy rate of 96.67%, but also with an accuracy of 90.77% and a recall rate of 89.72%. It also makes the different categories of pest and disease identification more balanced, improving the specific methods in the traditional methods, by which the recognition effect is poor. |
| |
| Keywords: | crop pest recognition Inception structure Residual structure Inception-resnetl-v2 Softmax classifier |
|
| 点击此处可从《广东工业大学学报》浏览原始摘要信息 |
|
点击此处可从《广东工业大学学报》下载全文 |
|
