Data Fusion of Soil Spectral Reflectance and Image Features for Predicting Soil Parameters Using Real-time Soil Sensor

Abstract

The objective of this study was to improve the accuracy of predicting soil parameters (moisture content: MC, soil organic matter: SOM) of a real-time soil sensor using additional information from underground soil images. Color features, 2-D fast Fourier transform spectra, auto-correlation function (ACF), and gray level co-occurrence matrix (GLCM) were employed as information from underground soil images. In the experimental field, ‘Contrast’ of GLCM could detect obstacles on the soil surface. The PLS model which had soil reflectance spectra and features of ACF as explanatory variables showed the best result (RMSE=2.04) for MC prediction. On the contrary, the image features did not help improve the performance of the models for SOM prediction in the experimental field.