In geotechnical site investigation, the distance between boreholes often makes the inter-borehole soil layer inferences rely on human experience under conditions of limited in situ tests and sampling. How to conduct intelligent and reliable soil stratigraphic division is a crucial research direction in the current information-oriented development of geotechnical engineering. This paper presents an intelligent soil stratigraphic layer division method by combining Bayesian Compressed Sensing (BCS), Support Vector Machine (SVM) classification, Gaussian Mixture Model (GMM), and Hidden Markov Random Field (HMRF) model. The application flowchart and soil layer division results are presented by taking the Nanjing Yangtze River floodplain ground as an example. The study shows that: BCS can reliably extend the blow count data from standard penetration test (SPT) for subsequent soil layer division; SVM classification can intelligently learn soil boundaries in the two-dimensional space of SPT blow count versus test depth, achieving an initial soil stratigraphic division; based on this, the preliminary optimization of soil layer division can be realized by using the GMM by considering the probability distribution of soil characteristic parameters; finally, the secondary optimization of soil layer division can be realized by using the HMRF model by incorporating spatial correlation constraints (i. e., adjacent points tending to be the same soil type). Combining the four methods can intelligently and automatically divide soil layers, and can significantly improve the accuracy of overall soil layer division and soil layer boundary identification.