| 基于深度学习网络实现番茄病虫害检测与识别 |
| |
| 引用本文： | 王铭慧,张怀清,樊江川,陈帮乾,云挺.基于深度学习网络实现番茄病虫害检测与识别[J].中国农业大学学报,2023,28(11):165-181. |
| |
| 作者姓名： | 王铭慧 张怀清 樊江川 陈帮乾 云挺 |
| |
| 作者单位： | 南京林业大学 信息科学技术学院, 南京 210037;中国林业科学研究院 资源信息研究所, 北京 100091;国家农业信息化工程技术研究中心 数字植物北京市重点实验室, 北京 100097;中国热带农业科学院 橡胶研究所, 海口 571737;南京林业大学 信息科学技术学院, 南京 210037; 南京林业大学 林草学院, 南京 210037 |
| |
| 基金项目： | 国家自然科学基金项目(32371876,32071681)；江苏省自然科学基金面上项目(BK20221337)；江苏省农业自主创新项目(CX(22)3048)；自然资源部国土卫星遥感应用重点实验室开放基金(KLSMNR-G202208)；中国热带农业科学院橡胶研究所省部重点实验室及科学观测实验站开放课题(RRIKLOF202301) |
| |
| 摘 要： | 为利用计算机或人工智能技术协助番茄病虫害防治，以存在病虫害侵害问题的番茄植株图像为研究对象，针对番茄病虫害目标小而密的特点提出基于Swin Transformer的YOLOX目标检测网络，用于精确定位图像中的病虫害目标，并采用基于经典卷积神经网络构建的旋转不变Fisher判别CNN分类网络，以此提高病虫害分类的准确率。结果表明:1)将测试结果与传统的目标检测模型和分类模型作对比，基于Swin Transformer的YOLOX网络在番茄病虫害测试集上的精确度比Faster R-CNN和SSD分别高了7.9%和9.5%，旋转不变Fisher判别CNN对病虫害类别的识别准确率与AlexNet、VGGNet相比分别提升了8.7%和5.2%；2)与基于Transformer的目标检测模型DETR和近年来新兴的图像分类模型Vision Transformer(ViT)在番茄病虫害测试集上的结果相比较，本研究的检测和分类方法也存在优势，病虫害检测精度和分类准确率分别提高了3.9%和4.3%。此外消融试验也证明了本研究方法改进的有效性。总之，本研究所构建的网络在番茄病虫害的目标检测和分类识别方面的性能优于其他网络，有助于提升番茄病虫害的防治效果，对计算机视觉在农业领域的应用具有重要意义。
|
| 关 键 词： | 深度学习 病虫害识别 Swin Transformer Fisher判别 番茄 |
| 收稿时间： | 2023/4/22 0:00:00 |
Detection and identification of tomato diseases and pests based on deep learning networks |
| |
| WANG Minghui,ZHANG Huaiqing,FAN Jiangchuan,CHEN Bangqian,YUN Ting.Detection and identification of tomato diseases and pests based on deep learning networks[J].Journal of China Agricultural University,2023,28(11):165-181. |
| |
| Authors: | WANG Minghui ZHANG Huaiqing FAN Jiangchuan CHEN Bangqian YUN Ting |
| |
| Affiliation: | College of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, China;Institute of Forest Resource Information Techniques, Chinese Academy of Forestry, Beijing 100091, China;Beijing Key Lab of Digital Plant, National Engineering Research Center for Information Technology in Agriculture, Beijing 100097, China;Rubber Research Institute, Chinese Academy of Tropical Agricultural Sciences, Haikou 571737, China; College of Information Science and Technology, Nanjing Forestry University, Nanjing 210037, China; College of Forestry and Grassland, Nanjing Forestry University, Nanjing 210037, China |
| |
| Abstract: | To assist in the prevention and control of tomato diseases and pests using computer or artificial intelligence technologies, taking tomato plant images with diseases and pests invasion problems as the research object, aiming at the characteristics of small and dense tomato diseases and pests targets, a YOLOX object detection network based on Swin Transformer is proposed to accurately locate the diseases and pests targets within the images. Additionally, a rotational-invariant Fisher discriminant CNN classification network based on classical Convolutional Neural Networks is constructed to enhance the accuracy of diseases and pests classification. The results demonstrated that: 1)Compared with the existing object detection and classification models, the YOLOX network based on Swin Transformer achieves 7. 9% and 9. 5% higher precision than Faster R-CNN and SSD on the tomato diseases and pests test set, respectively; The rotational-invariant Fisher discriminant CNN improves the recognition accuracy of the diseases and pests categories by 8. 7% and 5. 2% compared to AlexNet and VGGNet, respectively. 2)Compared with the existing Transformer-based object detection model DETR and the emerging image classification model Vision Transformer(ViT)on the tomato diseases and pests test set, the detection and classification methods proposed in this study also have advantages, which were an increase of 3. 9% in diseases and pests detection precision and 4. 3% higher in classification accuracy. In addition, the ablation trials also verified the effectiveness of the method improvement in this study. In conclusion, the performance of the proposed networks is superior to other networks in both detection and recognition of tomato diseases and pests. The methods strengthened the efficacy of diseases and pests control on tomato plants. It is of great significance in the fields of computer vision techniques applied to agriculture applications. |
| |
| Keywords: | deep learning diseases and pests identification Swin Transformer Fisher discrimination tomato |
|
| 点击此处可从《中国农业大学学报》浏览原始摘要信息 |
|
点击此处可从《中国农业大学学报》下载全文 |
|
