Prédiction des propriétés du sol à l'aide des Signatures spectrales de sols par Intelligence Artificielle

Abstract #

.

As part of my end-of-studies project at the School of Information Sciences (ESI) and

in order to obtain the state engineer diploma in data and knowledge engineering, I carried out

my end of studies internship at MAScIR in Rabat.

In recent years, VIS-NIR spectroscopy has rapidly emerged and become one of the most

powerful tools for the analysis, control and detection of soil characteristics. The aim of VISNIR spectroscopy is to obtain the spectral signature, which can be used as a fingerprint to

characterize the chemical composition of soil characteristics. This work consists of using the

technology of VIS-NIR spectroscopy, for the processing and manipulation of spectral

databases, also using artificial intelligence.We distinguish two machine learning approaches:

regression and classification which in our project has made the objective of several tests in

order to predict with great precision the characteristics of the soil among others the PH,

nitrogen, Phosphorus P, Potassium K, the organic matter OM and the electrical conductivity

EC. Specifically, this report synthesizes the fruit of our work which aimed to predict the

characteristics of European soils. Before embarking on the realization of the project we

started the documentation relating to the spectroscopy and the application of artificial

intelligence in the field of agriculture as well as the selection of attributes and evaluation of

the models. Finally, the last step presents the realization of the different stages of the

prediction process, from the preparation of data to the implementation of the models. The

approach is tested using a European LUCAS 2015 soil spectral database.

Keywords: Deep learning, machine learning, artificial intelligence, VIS-NIR spectroscopy,

spectral data, spectral signature, soils, agriculture, LUCAS 2015