An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images
Accurate and fast locating of diseased plants is critical for the sustainability of forest management. Recent developments in computer vision made by deep learning provide a new way for diseased plant detection from images captured by unmanned aerial vehicles (UAV). In this paper, we developed an an...
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Technoscience Publications
2022-06-01
|
| Series: | Nature Environment and Pollution Technology |
| Subjects: | |
| Online Access: | https://neptjournal.com/upload-images/(53)D-1282.pdf |
| _version_ | 1811243349542174720 |
|---|---|
| author | Dashuang Liang, Wenping Liu, Lei Zhao, Shixiang Zong and Youqing Luo |
| author_facet | Dashuang Liang, Wenping Liu, Lei Zhao, Shixiang Zong and Youqing Luo |
| author_sort | Dashuang Liang, Wenping Liu, Lei Zhao, Shixiang Zong and Youqing Luo |
| collection | DOAJ |
| description | Accurate and fast locating of diseased plants is critical for the sustainability of forest management. Recent developments in computer vision made by deep learning provide a new way for diseased plant detection from images captured by unmanned aerial vehicles (UAV). In this paper, we developed an anchor-free detector, an enhanced CenterNet named as Enhanced CenterNet (ECenterNet) model, which significantly improved the overall accuracy over the original CenterNet model without any increase in the running speed or number of parameters. Compared with the original model, in the newly proposed model improvements had been made in the training stage to increase the accuracy of the detector, while procedures in the test stage remained unchanged. Under the hold-out dataset, the proposed model is trained on 5,281 tiles and tested on 3,842 images, the results showed that the overall detection accuracy of ECenterNet reached 54.7% by COCO Challenge metrics (mean average precision (mAP) @[0.5, 0.95]), while mAP accuracy of the original CenterNet was 49.8%. This research indicates that the proposed deep learning detection model provides a better solution for detecting diseased plants from UAV images with high accuracy and real-time speed. |
| first_indexed | 2024-04-12T14:05:52Z |
| format | Article |
| id | doaj.art-baf2ab19e6244304b9eef9cfe7b4686f |
| institution | Directory Open Access Journal |
| issn | 0972-6268 2395-3454 |
| language | English |
| last_indexed | 2024-04-12T14:05:52Z |
| publishDate | 2022-06-01 |
| publisher | Technoscience Publications |
| record_format | Article |
| series | Nature Environment and Pollution Technology |
| spelling | doaj.art-baf2ab19e6244304b9eef9cfe7b4686f2022-12-22T03:30:05ZengTechnoscience PublicationsNature Environment and Pollution Technology0972-62682395-34542022-06-0121289990810.46488/NEPT.2022.v21i02.053An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle ImagesDashuang Liang, Wenping Liu, Lei Zhao, Shixiang Zong and Youqing LuoAccurate and fast locating of diseased plants is critical for the sustainability of forest management. Recent developments in computer vision made by deep learning provide a new way for diseased plant detection from images captured by unmanned aerial vehicles (UAV). In this paper, we developed an anchor-free detector, an enhanced CenterNet named as Enhanced CenterNet (ECenterNet) model, which significantly improved the overall accuracy over the original CenterNet model without any increase in the running speed or number of parameters. Compared with the original model, in the newly proposed model improvements had been made in the training stage to increase the accuracy of the detector, while procedures in the test stage remained unchanged. Under the hold-out dataset, the proposed model is trained on 5,281 tiles and tested on 3,842 images, the results showed that the overall detection accuracy of ECenterNet reached 54.7% by COCO Challenge metrics (mean average precision (mAP) @[0.5, 0.95]), while mAP accuracy of the original CenterNet was 49.8%. This research indicates that the proposed deep learning detection model provides a better solution for detecting diseased plants from UAV images with high accuracy and real-time speed.https://neptjournal.com/upload-images/(53)D-1282.pdfanchor-free detector, diseased plant detection, convolutional neural network, unmanned aerial vehicles, centernet |
| spellingShingle | Dashuang Liang, Wenping Liu, Lei Zhao, Shixiang Zong and Youqing Luo An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images Nature Environment and Pollution Technology anchor-free detector, diseased plant detection, convolutional neural network, unmanned aerial vehicles, centernet |
| title | An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images |
| title_full | An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images |
| title_fullStr | An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images |
| title_full_unstemmed | An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images |
| title_short | An Improved Convolutional Neural Network for Plant Disease Detection Using Unmanned Aerial Vehicle Images |
| title_sort | improved convolutional neural network for plant disease detection using unmanned aerial vehicle images |
| topic | anchor-free detector, diseased plant detection, convolutional neural network, unmanned aerial vehicles, centernet |
| url | https://neptjournal.com/upload-images/(53)D-1282.pdf |
| work_keys_str_mv | AT dashuangliangwenpingliuleizhaoshixiangzongandyouqingluo animprovedconvolutionalneuralnetworkforplantdiseasedetectionusingunmannedaerialvehicleimages AT dashuangliangwenpingliuleizhaoshixiangzongandyouqingluo improvedconvolutionalneuralnetworkforplantdiseasedetectionusingunmannedaerialvehicleimages |