Water Stress Detection in Pearl Millet Canopy with Selected Wavebands using UAV Based Hyperspectral Imaging and Machine Learning

Abstract

The major bottleneck in plant phenotyping is the assessment of thousands of genotypes under field conditions, which can be accelerated through Unmanned Aerial Vehicle (UAV) based sensing. Phenotyping for complex traits such as abiotic stress (drought) adaptation can be explored more precisely through the rich spectral information acquired by Hyperspectral Imaging (HSI) sensors. HSI sensors can identify plant water stress early by observing the changes in canopy reflectance due to drought. This study used a UAV-based HSI sensor in the 400-1000 nm range to identify canopy water stress in the pearl millet crop. Five Machine learning-based Feature Selection (FS) methods were used to identify the top-ranked ten wavebands sensitive to canopy water stress. Wavelengths around 692, 714-716, 763-769, 774-882, 870, and 949 nm were repeatedly selected by two or more FS methods. The Recursive feature elimination method with the Support vector machine (SVM) classifier outperformed the other FS methods in selecting the best bands subset. SVM classifier with linear kernel on the selected bands could classify two water stress levels with 95.38% accuracy and early detect stress with 80.76% accuracy in the pearl millet canopy. This study will benefit the agriculture sector by accelerating crop phenotyping using UAV-based HSI.