Abstract: Arc additive manufacturing has attracted extensive attention due to its advantages of high deposition rate, short cycle, and low cost, but it is prone to defects such as cracks, porosity, and non-fusion in the manufacturing process. In this paper, a fully automatic defect identification and optimization compensation method was proposed that combined deep learning, visual sensing, image processing, and six-axis industrial robots for real-time detection and repair of surface defects in the process of arc additive manufacturing. The surface of the workpiece was monitored by visual sensors; the defects were identified by image processing algorithms; the start and end coordinates of the defects were accurately obtained; the spatial distance was calculated, so as to quantify the size of the defects. In addition, the combination of deep learning models and six-axis industrial robots enabled the robots to adjust the motion trajectory in real time based on visual feedback information and accurately compensate for defects of different types and sizes. The results show that the YOLOv7-based model achieves a detection accuracy and recall rate of 94.9% and 91.4%, respectively, for surface defects in sedimentary layers. At the same time, the system can accurately locate defects and quantify their geometric dimensions, enabling feedback to the robotic system for automatic path adjustment and defect compensation.