Early detection of Hellula undalis in Brassica oleracea var capitata L. : A real-time drone based approach using YOLOv8

• Development of an automated UAV system for real-time pest monitoring. • Specific application for Hellula undalis detection in cabbage crops. • Novel detection method based on larval leaf damage patterns. • Capability for early diagnosis prior to visual identification of the insect. • Development of a novel dataset annotated to identify early larval damage signs on cabbage leaves. • Scalable solution suitable for diverse greenhouse layouts and geometries. The planet population has been reported to be approximately 8.1 billion people, and it is estimated that by 2050 it will increase to 9.8 billion. This growth has significant implications for farmers in meeting the raising demand for food. Moreover, the Food and Agriculture Organization (FAO) has reported that about 20% to 40 % of global crop population is lost due to pests and diseases. Early detection of weeds, plant diseases, and pests in crops is an essential task to prevent or reduce both crop and financial losses. To address this, emerging technologies are transforming the way agricultural activities are controlled, monitored, and managed. These technologies can be grouped in three main areas: Robotics, Artificial Intelligence (AI) and Internet of Things (IoT). Several works have been developed to detect these issues using visual sensors and AI; however, Real-Time detection still is missing. The main objective of the present work is to develop a Real-Time visual inspection system to identify Hellula undalis pest through the damage observed on the leaves in Brassica oleracea var capitata L. , commonly know as cabbage, using a drone and AI, within greenhouses. The dataset was collected using a drone to capture video, from which 1981 images of leaves with pest damage and an equal number without were obtained. YOLOv8 was used as the Deep Learning (DL) model for training to identify leaves with and without pest damage. Finally, a drone was autonomously navigated inside the greenhouse using SLAM techniques to detect, in Real-Time, cabbages with pest damage. The results showed a detection precision of 60%. The system’s implementation proved effective starting from the second week of cabbage growth for early pest sign identification.