CNN Based Water Stress Detection in Chickpea Using UAV Based Hyperspectral Imaging

Abstract

Water is an important agronomic input, which plays a vital role in the health and yield of the crop. Water deficiency results in abiotic stress, early detection of water stress help in recovering the health of the crop. Hyperspectral imaging (HSI) sensors acquire rich spectral information of the objects in hundreds of narrow bands, are capable of identifying the change in canopy water content, which is crucial in predicting irrigation requirements of the crop. Due to the wide field of coverages, short revisiting periods, and high spectral resolutions, Unmanned Aerial Vehicle (UAV) based HSI techniques are suitable in precision agriculture. In this paper, water stress detection in chickpea canopy is presented using hyperspectral (HS) images acquired from UAV. The drought classification was performed in two ways, i. by considering selected water-sensitive bands, and ii. by considering the whole spectral bands of the HS images. A 3D-2D convolutional neural network (CNN) model is used to classify well-watered canopy from water-stressed one, and its performance is compared with that of a Support Vector Machine (SVM) and a 2D+1D CNN model in identifying water stress. We obtained the best classification accuracy of 95.44%, which shows the potential of HSI in successfully detecting water stress in chickpea.