Oil Palm Tree Detection in UAV Imagery Using an Enhanced RetinaNet

Lee, Sheng Siang; Lim, Lam Ghai; Palaiahnakote, Shivakumara; Cheong, Jin Xi; Sow, Serene; Lock, Mun; Nizam, Mohamad; Ayub, Bin; Perak, Darul; Ridzuan; Malaysia

Abstract

21 Accurate inventory management of oil palm trees is crucial for optimizing yield and monitoring 22 the health and growth of plantations. However, detecting and counting oil palm trees, particularly 23 young trees that blend into complex environments, presents significant challenges for deep 24 learning models. While current methods perform well in detecting mature oil palm trees, they often 25 struggle to generalize across the diverse variations found in both young and mature trees. In this 26 study, we propose an enhanced RetinaNet model that incorporates deformable convolutions into 27 the ResNet-50 backbone, deeper feature pyramid layers, and an intersection-over-union-aware 28 branch in a multi-head configuration to improve detection performance. The model was evaluated 29 using a diverse dataset of unmanned aerial vehicle imagery from multiple regions, encompassing 30 oil palm and coconut trees, as well as banana plants. To refine detection, confidence thresholding 31 and non-maximum suppression were applied during inference, filtering out low-confidence 32 predictions and eliminating duplicate detections. Experimental results demonstrate that our 33 method outperforms state-of-the-art models, achieving F1-scores of 0.947 and 0.902 for single-34 and dual-species detection tasks, respectively, surpassing existing approaches by 1.5-6.3%. These 35 findings highlight the model's ability to accurately detect oil palm trees, particularly young ones 36 in complex backgrounds, offering a reliable solution to support sustainable agriculture and 37 improved land management. 38 Keywords. Oil palm tree, convolutional neural network, deep learning, object detection, unmanned 39 aerial vehicle 40

• Journal Article Type: Article
• Acceptance Date: Oct 6, 2024
• Deposit Date: Oct 6, 2024
• Print ISSN: 0168-1699
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed