Gas-insulated switchgear (GIS) is a key power distribution component in modern power systems. Partial discharge (PD) serves as an early indicator of insulation degradation in GIS, and its timely and accurate detection is essential for ensuring the stable operation of the equipment. To overcome the limitations of traditional detection methods that rely heavily on expert experience, and intelligent models that struggle to balance accuracy with real-time performance, this paper proposes a YOLOv8-based detection approach for GIS partial discharge defects. The proposed model possesses enhanced deep feature extraction capabilities and can adaptively learn discharge characteristics from phase-resolved partial discharge (PRPD) patterns. Four typical defect types were reproduced on a simulated experimental platform, and standardized grayscale image datasets were constructed for model training and testing. Experimental results demonstrate that the proposed method achieves a mean average precision (mAP) of 0.974 on the test set, with AP values of 0.978, 0.982, 0.976, and 0.960 for the four defect categories, respectively. These results confirm that the method enables high-precision defect detection and provides effective technical support for the intelligent operation and maintenance of GIS.