In order to improve the positioning accuracy of single-phase grounding faults in distribution network, a single-phase ground fault area location based on transient zero-sequence current image recognition is proposed. First, the fault simulation is realized by PSCAD to construct the image set required for Convolutional Neural Network(CNN) learning. Then, according to the ambiguity and differentiation characteristics of single-phase ground faults, image preprocessing based on Python programming. The pre-processed byte-form fault samples are trained by the VGGNet11 network structure to obtain a fault region localization model, and the classification effect is visually analyzed. Digital simulation and field tests of a typical 10kV distribution network model show that the proposed method can accurately locate the fault area without being affected by the system grounding mode, fault resistance and initial phase angle. The transient oscilloscope type fault indicator can be used to collect the transient zero sequence current of the line, and the signal acquisition is convenient.