Abstract: To tackle the practical engineering challenges of low efficiency, high omission rates, and poor traceability inherent to manual visual inspection of components on automobile welding and assembly lines, this paper investigates an online intelligent detection system based on deep learning. Leveraging the YOLOv5 object detection algorithm as its core, a closed-loop detection framework was developed that integrates a Jetson edge computing platform, industrial cameras, production line PLCs, and a vehicle traceability system. By capturing real-time feature images of components for intelligent recognition and establishing direct linkage with the production control system, the system enables a fully automated "perception-decision-execution" workflow. Field application results demonstrate that the system achieves a mean average precision (mAP@0.5) of 98.2% for detecting key feature points of side inner panels and exterior trim parts, with a single-unit inspection time of less than 3 seconds. It significantly reduces the risk of incorrect or missing assembly at the workstation, thereby effectively preventing production line halts and quality defects stemming from such errors. This study presents a high-precision, low-cost, and easily deployable engineering solution for the online quality inspection of analogous components in automotive manufacturing, offering valuable insights for industry applications.