Classifying Soil Type Using Radar Satellite Images

Abstract

The growth of the crop is dependent on soil type, apart from atmospheric and geo-location characteristics. As of now, there is no direct and cost free method to measure soil property or to classify soil type. In this work, we proposed a machine learning model to classify soil type using Sentinel-1 satellite radar images. Further, the developed classifier achieved 72.17% F1-score classifying sandy, free and clayish on a set of 65003 data points collected over one year (from Oct 2018 to Sep 2019) over 14 corn parcels near Ourique, Portugal.