SOIL CHARACTERIZATION USING COMPLEX PERMITTIVITY AND ARTIFICIAL NEURAL NETWORKS

摘要

The complex permittivity of Halton Till, a soil recovered from a landfill site in Ontario, Canada, is measured using a custom developed apparatus in laboratory in the frequency range from 0.3 MHz to 1.3 GHz. The soil is mixed with liquids including distilled water, NaCl, copper and zinc salt solutions, and compacted at various water contents, densities and degrees of saturation. A database consisting of 122 soil specimens is established and artificial neural networks (ANNs) are adopted for data processing. Three ANN models are trained, verified and tested to predict the soil water content, degree of saturation, and dry density. The results show that the three models perform well as judged from statistical analyses. The performance of the networks can be further improved by enhancing the database. The principle and results of this study provide encouraging information for the further development of an in-situ measurement system for characterization of soil subsurface.