In response to issues such as missed detection of small defects, low detection efficiency for certain defects, and insufficient overall automation in the automated detection of Quick-Connect Fittings (QCF), this paper proposes an intelligent detection model based on the collaborative use of supervised and unsupervised learning. First, a supervised training approach is conducted using an improved YOLOv8 model (QCF-YOLO), which optimizes the feature map representation capability, significantly improving the detection accuracy of small defects. This model has the advantages of fewer parameters and easy deployment, meeting real-time detection requirements. Secondly, unsupervised secondary training is performed in the cloud using detection results and raw images collected from edge devices. By comparing the detection results of supervised and unsupervised models, the model’s ability to detect unknown defect categories is progressively enhanced, effectively avoiding false detections caused by background interference in unsupervised training. Finally, based on a “cloud-edge-end collaborative architecture”, the model is deployed at the edge. Through the collaborative work of cloud training, edge detection, and terminal data collection, the automation level and detection efficiency of Quick-Connect Fitting defect detection are significantly improved.