Classification of Soil and Prediction of Total Nitrogen Content Present in Soil by Using Hyperspectral Imaging

Abstract

In this paper, the focus is to design a machine learning model that will classify the type of soil and predict the value of total nitrogen (TN) content of soil for an input soil image based on hyperspectral imaging (HSI) technology. To design the model, a total of 176 soil image samples are collected, where each HSI image consists of 870–1735 nm wavelength. The samples belong to three different types of soil, i.e., alluvial soil, red soil and coastal saline soil. For selection of effective wavelengths in spectrum, successive projections algorithm (SPA) has deployed. Then, from the gray-scale images of soil, different texture features are extracted at the effective wavelengths. Finally, the classification followed by prediction model based on the extracted features is performed using support vector machines (SVM) and partial least squares regression (PLSR), respectively. For achieving better result, first SVM is used to classify the type of soil for an input image. Then, to predict the value of TN content, a PLSR model is applied depending on the type of soil. By using the combination of extracted features and effective wavelengths, an average classification accuracy of 91.2% is achieved. The results indicate that HSI technology can be used for classification and prediction of soil nutrient contents, if the features considered correctly at effective wavelengths from full spectrum of the soil image.