Abstract: Solder joint failures are a common cause of circuit interruptions in aerospace electronic equipment, and artificial intelligence (AI) technology provides a new approach for real-time and accurate fault detection. In this study, an improved defect detection model named YOLO-KSD is proposed based on the YOLOv8 architecture. An adaptive optimization strategy was incorporated, which included a Kernel Warehouse (KW) backbone, a lightweight GSConv + Slim Neck module, and a Dynamic Head framework, to enhance feature extraction efficiency and detection adaptability. Several public solder joint and circuit board defect datasets were referenced and combined with actual PCB images to construct an expanded defect dataset for training and validating the AI model. The results indicated that, under identical hardware conditions, the YOLO-KSD model significantly improved the detection accuracy for PCB solder joint and board defects. The average processing time per frame was reduced by approximately 11.1%. The model effectively handled various defect types, with particularly notable improvements in small-object detection. This study lays a foundation for advancing the detection of solder joint defects in aerospace electronic equipment.