An effective pest detection method with automatic data augmentation strategy in the agricultural field | |
Qian, Shaowei1,4; Du, Jianming1; Zhou, Jianan3; Xie, Chengjun1; Jiao, Lin1,2; Li, Rui1,4 | |
刊名 | SIGNAL IMAGE AND VIDEO PROCESSING |
2022-06-18 | |
关键词 | Deep learning Pest detection and recognition Data augmentation Computer vision |
ISSN号 | 1863-1703 |
DOI | 10.1007/s11760-022-02261-9 |
通讯作者 | Xie, Chengjun(cjxie@iim.ac.cn) ; Jiao, Lin(linj93@mail.ustc.edu.cn) |
英文摘要 | Currently, computer vision technology has been applied to detect and recognize pests for integrated pest management (IPM). Recent studies have shown that the accuracy of pest detection and recognition has been rapidly improved with the development of deep learning. However, complex backgrounds, various poses, and different scales among insect species in the field will aggravate the difficulty of pest detection. To address the pest detection and recognition problem in wild field, in this paper, we firstly devise a novel automatic data augmentation method to search for the appropriate augmentation strategy adaptively and model data more effectively. Secondly, Res2Net is used as backbone for obtaining richer detailed information of small pest, and a reverse feature fusion layer is introduced into feature pyramid networks (FPN) to learn more details. During network training, the CIoU bounding box regression loss function and cross entropy loss after label smoothing are introduced for accurate localization and recognition of small pests. When testing, the test time augmentation (TTA) strategy is used to further improve pest detection performance and reduce the probability of missing detection by inferring pest images at different scales. We evaluate the performance of our method on the pest dataset including 4 k images and 4 classes (wheat sawfly, wheat aphid, wheat mite and rice planthopper). Our method achieves the pest detection performance of 81.0% mean Average Precision (mAP), which improves 5.7%, 4.0% and 3.1% compared to three state-of-the-art approaches YOLOv4, Faster R-CNN, and Cascade R-CNN detectors, respectively. |
资助项目 | National Natural Science Foundation of China[32171888] |
WOS研究方向 | Engineering ; Imaging Science & Photographic Technology |
语种 | 英语 |
出版者 | SPRINGER LONDON LTD |
WOS记录号 | WOS:000812604900002 |
资助机构 | National Natural Science Foundation of China |
内容类型 | 期刊论文 |
源URL | [http://ir.hfcas.ac.cn:8080/handle/334002/131291] |
专题 | 中国科学院合肥物质科学研究院 |
通讯作者 | Xie, Chengjun; Jiao, Lin |
作者单位 | 1.Chinese Acad Sci, Hefei Inst Phys Sci, Inst Intelligent Machines, Hefei 230031, Peoples R China 2.Anhui Univ, Sch Internet, Hefei 230031, Peoples R China 3.Hunan Univ, Changsha 410000, Peoples R China 4.Univ Sci & Technol China, Hefei 230026, Peoples R China |
推荐引用方式 GB/T 7714 | Qian, Shaowei,Du, Jianming,Zhou, Jianan,et al. An effective pest detection method with automatic data augmentation strategy in the agricultural field[J]. SIGNAL IMAGE AND VIDEO PROCESSING,2022. |
APA | Qian, Shaowei,Du, Jianming,Zhou, Jianan,Xie, Chengjun,Jiao, Lin,&Li, Rui.(2022).An effective pest detection method with automatic data augmentation strategy in the agricultural field.SIGNAL IMAGE AND VIDEO PROCESSING. |
MLA | Qian, Shaowei,et al."An effective pest detection method with automatic data augmentation strategy in the agricultural field".SIGNAL IMAGE AND VIDEO PROCESSING (2022). |
个性服务 |
查看访问统计 |
相关权益政策 |
暂无数据 |
收藏/分享 |
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。
修改评论