Use of YOLOv5 object detection algorithms for insect detection

• Main Author: Oliveira, Lino
• Publication Date: 2022
• Other Authors: Victoriano, Margarida, Alves, Adília, Pereira, J.A.
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10198/29259
• topic: Object detection; YOLOv5; Machine learning; Sustainable agriculture; CIMO-IPB dataset

Climate change affects global temperature and precipitation patterns that influence the intensity and, in some cases, the frequency of extreme environmental events, such as forest fires, hurricanes and storms. These events can be particularly conducive to the increase of plant pests and diseases, which causes significant production losses. So, the early detection of pests is of the main importance to reduce pest losses and implement more safe control management strategies protecting the crop, human health, and the environment (e.g., precision in the pesticide application). Nowadays, pests’ detection and prediction are mainly based on counting insects on attacked organs or in traps by experts, but this is a costly and time-consuming task for extensive and geographically dispersed olive groves. Machine learning algorithms, using image analysis, can be used for autonomous pests’ detection and counting. In the present practical work, YOLOv5 was chosen to detect and count the olive fly adults (Bactrocera oleae Rossi), a key pest of olives. YOLOv5s architecture of YOLO’s algorithm was used to test its efficiency in olive fly detection on a mobile deployment solution. The results obtained were quite satisfactory, and the experimental results obtained have been analyzed and presented, encompassing a set of metrics such as precision, recall, and the mean average precision. This study will be extended for other pests and disease detection in future work. Also, this solution will be integrated into a web-based information and management service (with sensors and e-traps) that remotely detect the presence and severity of pest attacks.