Deep learning-based precise detection of shrub crown boundaries using UAS imagery

Authors: LI, JIAWEI - Collaborator; Zhang, Huihui; Barnard, David

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 10/1/2024
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Accurate detection of shrub crown boundaries is essential for ecological monitoring, vegetation mapping, and land management. Recently, advancements in deep learning techniques have revolutionized the field of remote sensing and environmental monitoring. One such method, the convolutional neural network (CNN), has proven highly effective for image segmentation tasks. Unmanned Aerial Systems (UAS) offer high-resolution imagery that, when combined with advanced deep learning techniques, can significantly enhance the precision of vegetation analysis. This study explores the application of U-Net by leveraging multispectral images captured over different periods by unmanned aircraft systems (UAS) to precisely detect individual shrub crown boundaries in a semi-arid shrubland in northeastern CO, USA. The model is trained on UAS images from a target shrubland and validated using ground truth data from field surveys. The results demonstrate that our approach achieves superior accuracy in identifying and outlining shrub crowns compared to traditional image processing techniques and highlight the efficacy of the U-Net architecture in identifying individual shrub covers. This approach provides critical insights into land management practices and reveals the potential of using deep learning for long-term environmental monitoring and vegetation management.