A MACHINE LEARNING BASED APPROACH FOR SOIL PROPERTY ESTIMATION USING SPECTRAL IMAGES

Abstract

Soil is essential to the environmental process and growing crops. An accurate as sessment of soil moisture and chemical characteristics is essential for agricultural and land management decisions to be well-informed. Precision agriculture has advanced significantly because of machine learning. Firstly the paper address the study of hyperspectral images with the help of Riese and Keller’s dataset on Very Near Infrared Rays (VNIR) dataset captured by a Cubert UHD-285 snapshot camera for estimating soil moisture that measures real-time band reflectance values. They proposed a framework of Self Organizing Maps (SOM) for regression to estimate soil moisture. Results indicate that MLP performs better than all the machine learning regression-based techniques and SOM framework. It shows promising results and provides a new and suitable regression method to predict soil moisture from the hyperspectral soil moisture dataset. Results indicate that MLP performs better than all the machine learning regression-based techniques and SOM framework. It shows promising results and provides a suitable regression method to predict soil moisture from the hyperspectral soil moisture dataset. Secondly the paper address the study of multispectral images with the help of the "Land Use/Cover Area Frame Statistical Survey Soil" (LUCAS), a comprehensive and frequent topsoil survey conducted throughout the European Union to get data pertinent to policy about how land management affects soil properties. The data covers 28 European Union States. This work analyses and predicts the chemical properties of the soil of Hungary based on the LUCAS 2015 dataset that includes CaCO3, N, P, K, EC, pH. It estimates them using LUCAS and Landsat 8 satellite images using different regression-based algorithms like GPR, SVR, MLP, AdaBoost, Ridge and compares them. The Lucas survey data points and Landsat 8 satellite images (multispectral) are integrated for forecasting different soil nutrients.