A Hybrid Pipeline for Date Palm Tree Detection in High-Resolution Satellite Imagery

In this paper, we propose a novel hybrid method that combines image processing techniques with deep learning to tackle the problem of detecting date palm trees from high-resolution satellite imagery. The image processing component of the method is based on simple operations such as combining spectral bands, filtering, and peak detection and is used to produce an initial set of detections. Then, a simple convolutional neural network (CNN) trained with a manually labeled dataset is used to refine the initial set of detections. The method was tested on a dataset containing high-resolution 4-band WorldView-3 imagery of date palm tree farms of varying degrees of complexity in the Al-Ahsa region in Saudi Arabia. Comparisons of the proposed method with a Mask R-CNN show that the proposed method can improve the F1-Score on average by 7.8%, especially in dense and unorganized date palm tree farms commonly found in the Al-Ahsa region. Finally, to demonstrate the applicability of the method, the proposed method was used to estimate the number of palm trees within Al-Ahsa Oasis, the largest self-contained oasis in the world.