YOLO-based detection of Halyomorpha halys in orchards using RGB cameras and drones

Abstract

This paper explores the utilization of innovative technologies such as RGB cameras, drones, and computer vision algorithms, for monitoring pests in orchards, with a specific focus on detecting the Halyomorpha halys (HH), commonly known as the “brown marmorated stink bug”. The integration of drones and machine learning (ML) into integrated pest management shows promising potential for effectively combating HH infestations. However, challenges arise from relying on vision models solely trained using high-quality images from public datasets. To address this issue, we create an ad hoc dataset of on-site images mainly captured with the help of a drone as well as other devices. We initially conduct an in-depth analysis of the captured images, considering factors such as blurriness and brightness, to possibly improve the performance of the ML algorithms. Afterwards, we undertake the training and evaluation of diverse ML models using distinct approaches within the YOLO framework. We employ a range of metrics to compare their performance and ultimately achieve a satisfactory outcome. Through the optimization of ML models and the correction of image imperfections, we contribute to advancing automated decision-making processes in pest insect monitoring and management, specifically in HH monitoring.