Near-surface soil water content estimation using UWB-GPR based on selective sparse representation

The information of water content in the near-surface soil is very important for many applications such as hydrology, geology, agriculture and civil engineering. However, the conventional measurement techniques such as gravimetric, neutron probe and frequency/time-domain reflectometry are not very effective, especially for spatial large observation. Ground penetrating radar (GPR) is a prospective non-destructive method for solving this problem. To address the building of automatic measurement systems, this paper proposes the selective sparse representation based signal processing techniques in ultrawideband GPR signature which straightforward estimates the near-surface soil water content levels. The response signal of GPR is processed using matching pursuit with Gabor waveform dictionary for producing ten adaptive atom signals. Those atom signal parameters are processed by feature selection algorithm for selecting and ranking the most relevant features which then are feed to multiclass support vector machine for classifying the soil water content into three levels. The depth of moist soil layer is between 10 cm to 50 cm while the top layer is varied homogeneous dry soil. According to the numerical simulation results, the proposed method is able to estimate the water content within accuracy up to 71% so that it is promising enough to be optimized and developed further for real implementation.