H. Parsiani, E. Mattei, A. Lizarraga, and M. Ramos (Puerto Rico)
Radar imaging, soil moisture, Wideband radar, Neural Network
Non-invasive soil moisture determination has its multiple applications in such areas as agriculture, satellite validation, and climate prediction. The previous soil moisture research successfully conducted by the authors was limited to controlled laboratory experimentations, in terms of the purity of the different soil types and the exact amount of moisture inserted into the soil. In this paper, however, research was extended to open field measurements where the soil type was unknown and contained impurities such as rocks and other elements. The images of the soil were obtained by a monostatic wide-band radar (a ground penetrating radar, GPR), at the ground level, and the individual scan wavelets have been appropriately extracted to reject rocks and undesired elements from the soil. The material characteristics in frequency domain coupled with a two-layer back propagation neural network algorithm (MCFD-NN) was used for the determination of soil moisture. Two days of measurements, three weeks apart, produced the raw GPR data for the training of the MCFD-NN algorithm with the aid of a Theta probe. The total number of images collected on both days was 23, from which 16 were used for the training, and seven were used as unknowns and moisture values at those locations were calculated by the algorithm and compared with the actual Theta probe measurements with very good results.
Important Links:
Go Back
