PalmProbNet is introduced as an innovative approach to detecting palm trees in tropical forests using UAV-derived orthomosaic imagery. The study addresses challenges such as noise, illumination variations, and lack of labeled data. By training models with different-sized image patches and employing deep learning techniques, PalmProbNet achieves remarkable accuracy in identifying palm presence and locality.
The research emphasizes the importance of palm trees as ecological indicators and resources for biodiversity, human livelihoods, and wildlife. The methodology involves feature extraction through transfer learning with ResNet-18, classification using MLPs, and application to landscape orthomosaic images. Results demonstrate high accuracy levels across various model configurations for both small and large patches.
The study highlights the potential of integrating UAV technology with deep learning for efficient palm tree detection in dense forest canopies. Future work aims to refine the heatmap generation process by including edge cases in training samples and exploring segmentation networks for improved localization of individual palms.
To Another Language
from source content
arxiv.org
Deeper Inquiries