Abstract: In order to overcome the difficulty of insufficient sample size in partial discharge (PD) defect diagnosis, and to solve the defect that multi-source PD cannot be effectively identified by traditional pattern recognition algorithms, an improved SSD (single shot multi-box detector)-based multi-source PD map detection algorithm for gas insulated switchgear (GIS) was proposed. Firstly, the PRPD (phase resolve partial discharge) spectral characteristics of multi-source partial discharge were analyzed, GIS experimental platform was built and four typical partial discharge defects were simulated, sample data were collected, and ACGAN was used for sample expansion. Then, an SSD network model was built, and a spatial and channel attention mechanism was introduced into the feature extraction network. Finally, laboratory data and field data collected from a 220 kV substation were used to verify the effectiveness of the proposed algorithm. The experimental results show that the proposed algorithm can be adopted to effectively detect the multi-source partial discharge features in the PRPD spectrum under complex noise. The average detection accuracy of laboratory data can reach 97.8%, and the average detection accuracy of field data can reach 91.6%.