Deep Learning Method for Under-Canopy Tree Detection and Mapping using 2D LiDAR

An accurate and robust solution for under-canopy tree detection and mapping in dense oil palm environments is essential for automating on-ground operation in precision agricultural practices. Although vision-based methods have achieved promising results in various object detection tasks, their performance can be impeded in real-world plantation environments by factors such as occlusion, lighting conditions, and cluttered backgrounds, which pose significant challenges to accurate detection. This paper presents a deep learning approach for under-canopy tree detection and mapping in dense oil palm plantations using 2D LiDAR data. Our method processes 2D LiDAR scans to construct a probability grid representation of the environment and subsequently leverages a deep learning approach to detect, localize, and map the position of the tree while navigating through the plantation. The real-world experimental results have demonstrated the effectiveness of the proposed approach, offering a robust solution for improving operational efficiency in oil palm plantations.