Disease Detection in Plants Using UAS and Deep Neural Networks
Abstract
Detecting diseases in plants poses a significant challenge, addressed in this study through the integration of Unmanned Aerial Systems (UAS) and Deep Neural Networks (DNNs). The primary obstacle lies in identifying objects, such as leaves, in computer vision, exacerbated by the scarcity of large labeled datasets necessary for effective neural network training. To tackle this, the research employs a UAS equipped with a 3D image capture system, utilizing Jetson Nano and a ZED camera for streamlined data capture, storage, and subsequent analysis. The study's key objectives include developing a functional 3D image capture system and formulating a unique approach for detecting diseased leaves within the dataset. Utilizing images from the ZED camera, the study utilizes a pre-trained EfficientDet model from TensorFlow, initially trained on nine classes of leaves, to identify plant health. Despite limited labeled data, the model undergoes training using the available collected data and corresponding labeled instances. The proposed approach demonstrates commendable performance in identifying diseased leaves and distinguishing leaf types based on color variations resulting from physiological changes and disease conditions. The results affirm the feasibility and effectiveness of the developed system and detection methodology, contributing significantly to advancing leaf and disease detection through UAS and deep neural networks. This research represents progress in overcoming challenges associated with data scarcity in training these networks for leaf detection, highlighting the potential of 3D image capture systems and pre-trained models in this domain.
Subject Area
Computer science|Artificial intelligence|Systematic biology|Plant sciences
Recommended Citation
Pankaj Mishra,
"Disease Detection in Plants Using UAS and Deep Neural Networks"
(2024).
ETD Collection for Tennessee State University.
Paper AAI31236323.
https://digitalscholarship.tnstate.edu/dissertations/AAI31236323