Cognitive Machine Learning Model for Soil Property Prediction and Type Classification on Geo-Spatial Data

Abstract

Advancements in the field of machine learning and deep learning have made the research on the soil property and type classification on the spatial data in the recent years with inexpensive and accurate decision making phenomenon. Machine learning techniques are found to the efficient technique that can map high-dimension spectral data to the functional properties of the soil. Most of the existing research works considered nominal regression methods for the prediction, which results in the correlation issue that leverage the actual prediction output. In this research, we proposed deep learning and machine learning methods for the prediction of the functional properties of the soil such as percent organic carbon, total nitrogen, bulk density, pH, vegetation index, water index, percent sand and clay. Methods used for the prediction are divided into two groups, (i) those that use the properties calculated from the Geo-spatial data for the property prediction (ii) those that use the features extracted from the images for the type prediction. Our method is based on the learning algorithms including XGBoost, Support Vector Machine, Faster-Region Based Convolution Neural Network and Random Forest classifier. The implementation is tested on the Geo-spatial data and nominal soil type data and are observed with promising results over property and soil type prediction.. In soil studies with data modelling, there is a need to consider of fluctuation in relations between soil property and its sorts.