Deep Learning Approach for Detection of Oil Palm Tree on UAV Images

Abstract

Palm oil is an important economic commodity that acts as the main export of South East Asia countries. The growing conditions and health of the oil palm trees have a direct impact on the yield of the trees and the income of the growers. With thousands of trees planted on a plantation, an effective solution for large-scale oil palm tree detection must be developed to maximise income and management efficiency. However, the detection of oil palm trees in high-density crowded regions is difficult. This study proposed an intelligent method to detect oil palm trees by using UAV images and deep learning. This study focused on the detection of oil palm trees within crowded and overlapping regions. The datasets used in this study are very complex, containing highly crowded regions and various growing statuses of oil palm trees. U-Net deep learning-based segmentation model was employed to extract the regions that contain oil palm trees in the input image. The segmented image was then post-processed to refine the region of interest. The performance of the proposed architecture was compared against DeepLab V3+ and PSP-Net with different hyperparameter settings. The experimental results show that the proposed method achieved an accuracy of 97% in oil palm tree detection.