Automated detection of herbicide drift effects on crops

Drift can be an undesirable effect of herbicide spraying. Remote detection would be a beneficial step towards effectively monitoring herbicide drift. The goal of this study is to investigate the use of hyperspectral reflectance imagery for remote detection of herbicide drift. Spectral reflectance data for corn and soybean collected with a handheld ASD Spectroradiometer is used in this study. Four different rates of herbicide drift were simulated during the experiment stage, viz. 1/2 (maximum drift), 1/8, 1/32, and 1/64 (minimum drift) as well as no drift. Reflectance values of the 5 best spectral bands were used as features where optimization was conducted using ROC analysis. The features were subjected to linear discriminant analysis to increase drift detection accuracy. Classification was then performed using a maximum-likelihood decision. The methods were tested on the experimental data using cross-validation methods.