Skip to main content
ARS Home » Southeast Area » Auburn, Alabama » Soil Dynamics Research » Research » Publications at this Location » Publication #425145
Research Project: Agronomic and Engineering Solutions for Conventional and Organic Conservation Agricultural Systems in the Southeastern U.S.

Location: Soil Dynamics Research

Title: Agri-NET UAV: A drone-captured dataset and model exploration for enhanced plant recognition using low-cost drones

Author
item BANSALA, R - The University Of Texas At Dallas
item ANAND, A - University Of Texas
item SUNDARAVADIVEL, P - University Of Texas
item PRABHAKARAN, B - The University Of Texas At Dallas
item Torbert Iii, Henry

Submitted to: Proceedings of SPIE
Publication Type: Proceedings
Publication Acceptance Date: 5/28/2025
Publication Date: 7/4/2025
Citation: Bansala, R., Anand, A., Sundaravadivel, P., Prabhakaran, B., Torbert III, H.A. 2025. Agri-NET UAV: A drone-captured dataset and model exploration for enhanced plant recognition using low-cost drones. Proceedings of SPIE 13458. Real-Time Image Processing and Deep Learning. 2025:1345807. May 28, 2025. https://doi.org/10.1117/12.3053474.
DOI: https://doi.org/10.1117/12.3053474

Interpretive Summary: Unmanned Aerial Vehicles (UAVs), or drones, are becoming vital tools for addressing agricultural challenges. While many plant datasets exist, few are designed explicitly for UAV-based crop imagery. This paper aims to fill that gap by developing a dedicated dataset of drone-captured crop images and evaluating the performance of various machine-learning models for on-device classification and detection. Our dataset offers both single-class and multi-class classification data. We assess traditional Convolutional Neural Networks (CNN) and Vision Foundation Models for image recognition for single-class data. The results are compared with models trained on existing plant datasets. We assess and optimize the State of the Art (SOTA) detection model for the multi-class classification data. This research not only provides a valuable dataset optimized for UAV-based agricultural imaging but also identifies the most effective modeling techniques, offering significant potential to enhance agricultural problem-solving.

Technical Abstract: Unmanned Aerial Vehicles (UAVs), or drones, are becoming vital tools for addressing agricultural challenges. While many plant datasets exist, few are designed explicitly for UAV-based crop imagery. This paper aims to fill that gap by developing a dedicated dataset of drone-captured crop images and evaluating the performance of various machine-learning models for on-device classification and detection. Our dataset offers both single-class and multi-class classification data. We assess traditional Convolutional Neural Networks (CNN) and Vision Foundation Models for image recognition for single-class data. The results are compared with models trained on existing plant datasets. We assess and optimize the State of the Art (SOTA) detection model for the multi-class classification data. This research not only provides a valuable dataset optimized for UAV-based agricultural imaging but also identifies the most effective modeling techniques, offering significant potential to enhance agricultural problem-solving.