YOLO-EP: A detection algorithm to detect eggs of Pomacea canaliculata in rice fields

• Published in: Ecological informatics Vol. 77; p. 102211
• Main Authors: Huang, Yao, He, Jing, Liu, Gang, Li, Dian, Hu, Ruining, Hu, Xiaoxi, Bian, Dingding
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2023
• Subjects: adults; algorithms; Deep learning; Detection and recognition; disease surveillance; eggs; Eggs of Pomacea canaliculata; eyes; food security; pests; Pomacea canaliculata; rice; Small targets; Unmanned aerial vehicles; Deep learning; Unmanned aerial vehicles; Detection and recognition; Eggs of Pomacea canaliculata; Small targets
• ISSN: 1574-9541
• DOI: 10.1016/j.ecoinf.2023.102211

The widespread of Pomacea canaliculata, a new “killer” in rice fields, may threaten the productivity and quality of rice. Therefore, keeping an eye on it is crucial for food security. However, direct monitoring is challenging because adult Pomacea canaliculata often reside underwater. The eggs can be observed since they are frequently attached to dry things. Additionally, because the eggs lack the capacity to move, monitoring the eggs is helpful for the early implementation of control measures. As a result, we suggest a method for monitoring the eggs. We used unmanned aerial vehicles (UAVs) to take close-up pictures of rice fields to capture images of the eggs, and then we applied deep learning algorithms to recognize and identify them. This method takes advantage of the convenience and capacity of UAVs to gather images over a broad region. It is difficult for conventional approaches to provide better results because the eggs of Pomacea canaliculata are tiny targets and only occupy a few pixels in UAV photos. To detect eggs of Pomacea canaliculata in rice fields using UAV images, we propose the YOLO-EP (YOLOv5s for eggs of Pomacea canaliculata) algorithm, which is primarily based on YOLOv5s and is improved for egg targets. To decrease the feature map detail loss, the algorithm uses transposed convolution instead of nearest interpolation upsampling and combines swin transformer and ECA attention algorithms to improve feature extraction. We add a modest target detection layer for eggs and utilize NWD as the model's bounding box loss function. According to the experimental findings, the YOLO-EP algorithm is better than other detection techniques in accuracy and performance, with AP@0.5 of 88.6%, precision of 85.1%, and recall of 82.6%. The improvements were 5.1%, 2.7%, and 3.8% above the base model. Overall, using UAV data and deep learning techniques to detect and identify eggs of Pomacea canaliculata opens up new possibilities for agricultural pest and disease surveillance. •We propose a new method for detecting eggs of Pomacea canaliculata in rice fields.•Introducing swin transformer to improve the model's ability to extract features.•Changing the upsampling method to transpose convolution to reduce the loss of detail.•Applying ECA to the feature fusion part to improve the model's perceptual ability.•Using NWD as the model's bounding box loss function to improve detection accuracy.