Intelligent Drought Stress Monitoring on Spatio-Spectral-Temporal Drone based Crop Imagery using Deep Networks

Tejasri N, Rajalakshmi Pachamuthu, Balaji Naik, Uday B Desai

Keywords: Drone based Imagery, Deep learning, drought stress, Spectral data, Convolutional Neural network (CNN), Long short-term memory (LSTM), Maize

TL;DR: A method to monitor drought induced stress in maize using CNN-LSTM based deep network on drone based spatial-spectral time-series image data.

Abstract:

n recent years, high-put crop monitoring methods that inte-
grate drone-based imagery and deep learning have been used
to identify crop health and diseases. However, existing meth-
ods follow manual methods to study drought stress making
it more challenging. To alleviate this problem, we propose a
deep learning-based framework to identify drought-induced
stress in maize using RGB and multispectral data. For this
study, we conducted an experiment to grow maize crop in
controlled conditions of water. A pipeline for pre-processing
UAV-based images and extracting the region of interest from
orthomosaic is explained. We used a variant of convolutional
neural network-long short-term memory (CNN-LSTM) net-
work to learn spatio-spectral-temporal patterns on drone-
captured maize for water stress classification. We employed
fine-tuned versions of pre-trained Alexnet, VGG 19, Resnet-
18, Resnet-50 and Mobilenet V2 models for feature extrac-
tion and the LSTM model for sequence prediction on RGB
data and multispectral data. It can be noted that multispectral
data performed better than RGB data on drone captured data.