Estimating plant height of red pepper using unmanned aerial vehicle-based multi spectral imagery

Abstract

This study was conducted to develop the plant height estimation model for pepper using vegetation indicies of multispectral imagery using an Unmanned Aerial Vehicle. There were no models to satisfy the conditions (R2T >0.6, MAPE < 10%), despite significant results in both the multiples linear regressions with Green and Blue bands selected by VIF and the simple linear regressions. The multiple linear regression model using PRI, GRVI, and SAVI selected as VIF satisfied the conditions regardless of the ratio of learning data. The 6:4 ratio model was selected as the best model because its validation performance (R2T = 0.638, RMSET = 2.245 cm, MAPET = 1.183%, R2V = 0.338, RMSEV = 3.980 cm, MAPEV = 4.096%) was better than the others, even though the 7:3 ratio model had a higher R2 value. The standardized regression coefficients of the selected model were PRI, SAVI, and GRVI, in that order. When estimating the plant height of peppers, multiple linear regression was more accurate than simple linear regression.