Abstract: With the increasing demand for intelligence in radar fault diagnosis and the development of machine learning technology, machine learning technology has gradually been applied to quickly locate faults in internal components of radar. However, research on radar wiring faults has not yet been published. We propose a YOLOv8s EWD radar network cable wiring defect detection model to address the current situation where the target of network cable wiring defects is relatively small, the target environment is complex and diverse, and defects are difficult to detect. On the basis of the YOLOv8s model, the C2f module was replaced with the HA_C2f module, the head network Conv module was optimized to the DWConv module, and the P2 detection layer was added, which further improved the fine-grained feature recognition ability, enhanced local feature description, and reduced parameter count of the YOLOv8s EWD model compared to YOLOv8s, while improving detection accuracy. In addition, a radar network cable connection defect dataset was established for model training and validation. The experimental results show that the YOLOv8s EWD model has certain improvements in performance, training effectiveness, and robustness.