Soil Classification Via Mid-infrared Spectroscopy

The need for rapid and inexpensive techniques for soil characterization has led to the investigation of modern technologies, and in particular those based on reflectance spectroscopy. While near-infrared has been traditionally used, midinfrared in the 400-4000 cm-1 range is becoming increasingly common due to the specificity of the absorbance bands in this spectral range. The present work discusses two methods based on mid-infrared spectroscopy for soil classification: attenuated total reflectance (ATR) and photoacoustic spectroscopy. The ATR method requires a soil sample close to water saturation, and as a result only the 800-1600 cm-1 interval of the spectrum yields a useful signal. Typical ATR soil spectra consist mostly of several broad bands in the 800-1200 cm-1 region and a calcium carbonate band around 1450 cm-1. By comparison, photoacoustic measurements are conducted with air-dried samples, and the photoacoustic spectra exhibit a larger number of clearly-defined bands. Both methods were tested on data sets containing over 100 samples of various soils commonly used in Israeli agriculture. Data analysis was conducted by wavelet decomposition and neural network classifiers. Very good classification performances were achieved, with correct classification rates of the validation samples typically above 95%. Key-Words: Fourier transforms infrared (FTIR); attenuated total reflectance (ATR); photoacoustic spectroscopy (PAS); wavelets; neural networks