Abstract: When single-phase-to-ground (SPG) faults occur in a small current grounding system, the fault transients appear complex due to the weak fault currents, thereby leading to the poor detection accuracy and reliability of the existing faulty feeder detection. This paper proposes a novel detection method based on the complete waveform recognition of zero-sequence signals. Firstly, a general method for identification of the fault signals based on the complete waveform recognition is introduced. It is pointed out that the fault information can be fully exploited when the raw signals are learned and recognized on the waveform scale. Secondly, both zero-sequence voltage (ZSV) on the bus and the zero-sequence currents (ZSC) of all the feeders are used for recognition, where the waveforms with the distinct characteristics are captured based on the distribution features of the ZSCs, and the detection method based on the complete waveform recognition of fault signals is constructed. Thirdly, an improved fully convolutional network is established for waveform recognition, and the credibility of the recognized fault waveforms is estimated based on the Dice coefficient. The PSCAD simulation and the field test show that the proposed method has a better detection accuracy and strong reliability.