Genetic programming: on the programming of computers by means of natural selection
Genetic programming: on the programming of computers by means of natural selection
| Hi-index | 0.00 |
Nitrogen is one of the most important chemical intakes to ensure the healthy growth of agricultural crops. However, some environmental concerns emerge (soil and water pollution) when a farmer applies nitrogen in excess. In this study, we propose a new method called GP-SVI to search for the best descriptive model of nitrogen content in a cornfield (Zea mays), thanks to airborne hyperspectral data and ground truth nitrogen measurements. Coupling the output of this descriptive model with variable-rate technologies (VRT) would allow farmers to practice site-specific management ensuring them economical savings and ecological benefits. GP-SVI is a parallel search of the best spectral vegetation index (SVI) describing a crop biophysical variable, derived from Genetic Programming (GP). Compared to statistical regression methods on our datasets, GP-SVI improves results obtained with classical approaches, in term of explained-variance and generalization error. We also show that the spectral bands selected by GP-SVI match those selected by Partial Least Square regression optimized by Genetic Algorithms (GA-PLS) as proposed by Leardi in "Application of genetic algorithm-PLS for feature extraction in spectral data sets", in Journal of Chemometrics.