Deep-Learning-Based Automated Palm Tree Counting and Geolocation in Large Farms from Aerial Geotagged Images

Adel Ammar,Anis Koubaa,Bilel Benjdira

doi:10.3390/agronomy11081458

Abstract

In this paper, we propose an original deep learning framework for the automated counting and geolocation of palm trees from aerial images using convolutional neural networks. For this purpose, we collected aerial images from two different regions in Saudi Arabia, using two DJI drones, and we built a dataset of around 11,000 instances of palm trees. Then, we applied several recent convolutional neural network models (Faster R-CNN, YOLOv3, YOLOv4, and EfficientDet) to detect palms and other trees, and we conducted a complete comparative evaluation in terms of average precision and inference speed. YOLOv4 and EfficientDet-D5 yielded the best trade-off between accuracy and speed (up to 99% mean average precision and 7.4 FPS). Furthermore, using the geotagged metadata of aerial images, we used photogrammetry concepts and distance corrections to automatically detect the geographical location of detected palm trees. This geolocation technique was tested on two different types of drones (DJI Mavic Pro and Phantom 4 pro) and was assessed to provide an average geolocation accuracy that attains 1.6 m. This GPS tagging allows us to uniquely identify palm trees and count their number from a series of drone images, while correctly dealing with the issue of image overlapping. Moreover, this innovative combination between deep learning object detection and geolocalization can be generalized to any other objects in UAV images.

Highlights

Tree counting and monitoring from aerial images is a challenging problem with many applications such as forest inventory [1], crop estimation [2], irrigation policies [3], and farm management [4]
We developed an algorithm that tags each detected tree with its GPS location by applying photogrammetry concepts to the metadata (EXIF and XMP) extracted from drone images and applying a distance correction based on the ratio between the drone altitude and the estimated average palm height
The gap is especially large for the class of other trees, for which Faster R-convolutional neural networks (CNN) and YOLOv3 perform poorly (AP lower than 0.5)

Summary

Introduction

Tree counting and monitoring from aerial images is a challenging problem with many applications such as forest inventory [1], crop estimation [2], irrigation policies [3], and farm management [4]. It enables farmers and decision makers to conduct real-time monitoring, improve productivity, and participate in ensuring sustainable production and food security. Counting the number of trees in large farms has been a challenging problem for agriculture authorities due to the massive number of trees and the inefficiency and excessive cost of old-style manual counting approaches. The problem becomes even more laborious and tedious when we need to identify the GPS location of trees for governance purposes and for regularly monitoring their condition over time. The inefficiency of traditional methods leads to inconsistent data collection about the number of trees, as reported by agriculture experts.

Methods

Results

Discussion

Conclusion