Regression applied to measure normalized difference vegetation index in soybean images with visible color spaces collected by smartphones

Abstract

The leaf area is an indicator of the plant s health and predicted yield production. The normalized difference vegetation index (NDVI) is the most used measure to evaluate leaf area and health. The study aimed to create a model able to calculate the NDVI from common RGB images collected by smartphones in the field through artificial intelligence techniques. A total of 99 Soybean experimental samples were analyzed by portable equipment GreenSeeker model RT100 from NTech radiometer and image acquired by smartphone positioned upright. NDVI was calculated with radiometer absorbance value. The images were initially preprocessed and then pixel information was submitted to Simple Linear, Multiple Linear, Isotonic, Rhythm Regression, Additive, and Linear Regression of Least Median of Squares models. The models tested achieved between 93,75% and 97,11% of correlation with data collected with a radiometer. The Multiple Linear Regression model that best described the leaf area. The soybean leaf area can be easily evaluated by smartphones with distortion corrections and models adjusted to NDVI.