Machine learning for soil analysis

Abstract

The Pfeiffer circular chromatography (PCC) is a qualitative analysis technique that provides microbiological, mineral, and organic matter of the health status of a soil. This paper addresses the building of a data set conformed by PCC images from different types of soils, based on the following three variables: acidity, electrical conductivity, and soil texture. This work also explores the use of deep learning techniques to automatically extract the characteristics of such images and classify the soils based on their type. It is the first time that learning techniques are applied to this classification problem. Experimental results show an F1-score of 0.7889 in the soil texture classification suggesting a significant relationship between this variable with the PCC. On the other hand, results also show a low correlation of the PCC with acidity and electrical conductivity.