Palm Tree Detection in Drone Images Using Deep Convolutional Neural Networks: Investigating the Effective Use of YOLO V3
Source of Publication
Advances in Intelligent Systems and Computing
Owing to the agricultural and economic importance to many countries, computer based automated palm-tree detection from aerial images, has been an area of research significance to the computer vision research community worldwide. Most previous approaches have applied traditional machine learning algorithms for palm-tree detection. However, in the recent past, deep neural network based learning has been proven to be a far more superior approach for general object detection and recognition tasks in many application areas. Alongside this technological development lightweight UAVs, e.g. Drones, have been widely accepted as having great practical potential and economic benefit in the surveillance of large areas of land, in significantly higher resolution, as compared to the traditional use of satellite images or more expensive large UAVs. This research presents a novel methodology based on the latest YOLO Version-3 Convolutional Neural Network object detector for detecting palm-trees in drone images captured in a desert area that includes palm-trees of different sizes, resolution, ground spread, degree of overlap, etc. In particular, we discuss the specific training strategy adopted and hyper-parameter optimisations carried out to improve the accuracy from a modest 0.78 to 0.96.
Drones, Object detection, Palm tree detection, Convolution neural networks, Unmanned aerial vehicles
Aburasain, R. Y.; Edirisinghe, E. A.; and Albatay, Ali, "Palm Tree Detection in Drone Images Using Deep Convolutional Neural Networks: Investigating the Effective Use of YOLO V3" (2021). All Works. 4376.
Indexed in Scopus