Research on loss mesoscopic based on image processing of LS-SVM

Applying LS-SVM theory in quantitative analysis of loess SEM images, bring forward a new method of calculating for soil microstructure parameters, which convert pore zone extraction to pore pixel dot assortment. The method distinguishes loess pore and skeleton by using current pixel and its neighborhood gray level information as well as local textural features educed by their gray level co-occurrence matrix, avail itself of both pixel gray level and spatial information, improving image segmentation accuracy. Q3 loess SEM images before and after seismic subsidence test have been processed by above method in this paper, parameters such as pore numbers, pore areas, surface porosity, pore fractal dimension and pore distribution graph have been achieved. The results show that after test macropores decrease in areas, mesopores and fine pores increase in numbers, seismic subsidence is mainly induced by macropores failure. Based on analysis of SEM images of different magnifications, the proper choice of magnification depends on study object and objective; it is suggested to choose 400-800X to study loess pore characteristics.