In the present study, according to the borehole transient electromagnetic method, the transceiver is placed into the borehole in front of the heading face for the purpose of detection. This can not only avoid the interference of metal support in the roadway, but also detect the hidden disaster-causing water body within a short distance. However, there is an issue of abnormal location in borehole transient electromagnetic method. In order to determine the location of transient electromagnetic anomalies in the borehole, the transient electromagnetic three-dimensional forward modeling is used to analyze the response characteristics of three components. According to the response characteristics of the horizontal component of the abnormal body in different quadrants, a method for calculating the rotation angle of the XOY plane of the abnormal center is proposed, based on the amplitude of the horizontal component and the abnormal quadrant. The resistivity of each depth calculated by the vertical component is regarded as an independent abnormal body, and the depths' mapping relationship with the sampling time is established. Next, based on K-means clustering, the corresponding characteristic value of horizontal component anomaly curve are binary classified, and the resistivity distribution quadrant of the entire data set is automatically divided. Finally, the borehole transient electromagnetic inversion resistivity stereo imaging method is obtained. The algorithm of this paper is used to calculate the stereo imaging of the three-dimensional numerical model, and the results show that the algorithm achieves good results for the small-scale and low resistivity abnormal body in the radial direction of the borehole. In summary, the method of borehole transient electromagnetic stereo imaging based on clustering is an organic combination of geophysics and machine learning, which can provide technical support for advanced detection and fine interpretation of hidden water damage in underground heading faces.