Identification of soil layers using probabilistic collaborative representation–based classification with CPTu data

This paper proposes a soil layer identification method using probabilistic collaborative representation–based classification (ProCRC). The representative parameters (cone tip resistance qt, pore pressure u2, and pore pressure ratio Bq) are selected from the CPTu data acquired through numerous engineering sites for normalization. Then, the salient features of the data are extracted based on a sparse automatic encoder, and the optimal solution of the test data was obtained. At that point, the soil layer identification and classification can be carried out by combining the ProCRC algorithm. The proposed method was adopted to identify the soil layers in four different kinds of sites and compared with other existing methods. The classification results were analyzed by regression analysis, and it was observed that the proposed method exhibits higher precision and validity than other methods. The underlying soil stratigraphy including interbeds and mixed layers can be well determined via the ProCRC method with CPTu data.