In order to identify the aging state of EPR cable insulation more accurately, a recognition method based on partial discharge image features and deep forest is proposed. In this paper, EPR samples with different aging states are prepared, and a PD test platform is built. The PD spectra of EPR samples with different aging states are obtained through the test, and 19 characteristic parameters are extracted from the PD spectra. The samples with different aging degrees are identified by combining the deep forest network. The results show that: by combining the PD spectra characteristics Token and deep forest network can accurately identify the aging state of cable, and the recognition rate is better than other traditional classification algorithms. The combination of PD image features and deep forest has a good engineering application prospect in cable insulation aging diagnosis