Generation of Hyperspectral Indices for Non-Invasive Crop Property Analysis Through Genetic Programming

Nowadays, the use of hyperspectral images in the agroindustry is growing because it represents a non-invasive methodology for identifying the internal properties of food products. Hyperspectral indices are common in satellite and aerial images for identifying vegetation, water, etcetera. This document proposes a methodology for generating hyperspectral indices on a minor scale to identify dry matter in crops based on Genetic Programming. We use the correlation coefficients Pearson and Kendall as fitness functions. We use the datasets of four crops: apple, broccoli, leek, and mushroom. The results of the experiments confirm that our proposal can generate simple and understanding hyperspectral indices that correlate with dry matter by combining only two or four hyperspectral bands. Our code is available online.