The multi-source partial discharge PRPD pattern can realize the pattern recognition of multi-source partial discharge types through the target detection algorithm training and identification of shape features. However, when the characteristics of different discharge pattern overlap, the small target is easily blocked by the large target, resulting in false detection and missed detection. Therefore, this paper proposes a multi-source partial discharge PRPD pattern identification algorithm with optimized non-maximum suppression. The Soft-NMS algorithm was introduced to solve the missed detection caused by overlapping targets; GIoU was used to replace the traditional IoU to calculate the similarity between targets and the loss function was optimized; the YOLOv7 network model was further built to identify the PRPD pattern of four typical discharges shape features. After cross-validation between simulation experiments and charged field data, the results prove that the average detection accuracy of the algorithm can reach 98.2% in simulation experiments and 88.4% in field experiments, effectively reducing the false detection rate and successfully identifying the characteristics of multi-source local discharge PRPD pattern when the targets overlap.