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

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.

Comparison of Deep Learning Methods for Detecting and Counting Sorghum Heads in UAV Imagery

“…Most of the detected targets were still concentrated in the low confidence region, proving that EfficientDet required high target clarity and was suitable for large target detection, which was consistent with the conclusion of related studies focused on detecting large targets such as crop growing circles [56], ships [57], pedestrians [58], and solid waste garbage [59]. In addition, the BiFPN network with integrated bidirectional cross-scale connectivity and fast normalized fusion was largely inferior to the SSD algorithm of the VGG-16 network and the YOLOv4 algorithm of the SPP+ PANet network in terms of detection efficiency [60]. However, the EfficientDet detected very few wrongly detected high-confidence targets, i.e., among the 1879 sorghum heads detected in Figure 7, there were only 175 wrongly detected negative samples, and the number of negative samples accounted for only 9.31% of the number detected, indicating that this method could have high detection accuracy if sufficiently detailed information could be provided [57,60].…”

“…In addition, the BiFPN network with integrated bidirectional cross-scale connectivity and fast normalized fusion was largely inferior to the SSD algorithm of the VGG-16 network and the YOLOv4 algorithm of the SPP+ PANet network in terms of detection efficiency [60]. However, the EfficientDet detected very few wrongly detected high-confidence targets, i.e., among the 1879 sorghum heads detected in Figure 7, there were only 175 wrongly detected negative samples, and the number of negative samples accounted for only 9.31% of the number detected, indicating that this method could have high detection accuracy if sufficiently detailed information could be provided [57,60]. Meanwhile EfficientDet had a medium number of parameters and the time required for model training was in the middle among the three methods [56].…”

Comparing Interpretation of High-Resolution Aerial Imagery by Humans and Artificial Intelligence to Detect an Invasive Tree Species

“…Therefore, additional training data for the miconia detection DNN, particularly including image training sets with low red contrast and red and green SCRs, would likely improve the robustness of the DNN. Additionally, alternative DNN architectures, such as EfficientDet [103,104], YOLOv5 [105], and Mask R-CNN [72,106,107], should be considered to improve accuracy and inference speed.…”

Identification and Classification of Mechanical Damage During Continuous Harvesting of Root Crops Using Computer Vision Methods

“…In our opinion, such adaptive systems have a serious drawback, they cannot be customized to the individual characteristics of each root crop. In digital agriculture [10,11], computer vision systems are used to quickly detect and count plants [12][13][14][15], to determine their ripeness and diseases [16][17][18][19][20], as part of systems to protect against weeds and pests [21,22], to determine the position of cattle [23]. In recent years publications have shown that the problem of identifying diseased or mechanically damaged fetuses on transportation systems such as conveyor belts, drums, turbines and etc.…”