Estimation of yield loss of maize with disease stress using UAV multispectrum images

This study obtained multi-spectral imagery of southern corn rust (SCR) with different severity grades using UAV multispectral images, and conducted ground surveys of disease and yield simultaneously. It used correlation analysis methods to select the most suitable remote sensing parameters for disease index of SCR (DISCR), and established spatial distribution maps of SCR and yield for corn with different severity grades based on the selected remote sensing parameters. The results showed that, (1) By fusing the selected original waveband reflectance (RRe), vegetation index (NGI), and texture features (DisRe) as input variables, a Random Forest Regression (RFR) method was used to establish different severity grade SCR monitoring models, with an overall accuracy of 76.67%, Kappa coefficient of 0.68. And an estimation model of corn yield based on remote sensing, with an overall accuracy of 87.54%, Kappa coefficient of 0.71. (2) The yield of corn with disease was validated using actual yield measurements. The relative errors between the estimated and actual yield values for healthy, slight, middle, and severe disease were 5.12%, 6.69%, 8.03%, 9.81%, respectively. (3) The yield losses of slight, middle and severe SCR were 1306.06 kg/hm² , 2485.35kg/hm² and 4771.00 kg/hm² , and the loss rates were 13.27%, 25.25% and 48.48%. The research results could provide theoretical basis and reference for crop yield and yield losses under disease stress.