Abstract: To address the anti-corrosion problem of the waste incineration furnace wall, it is the most effective technical method to deposit a layer of anti-corrosion layer on its surface by surfacing. By considering the savings of welding materials and the anti-corrosion effect, an overlap of approximately 25% between the preceding and subsequent weld beads is considered an ideal overlap ratio. Therefore, in the process of automatic surfacing based on vision, it is an important prerequisite to accurately identify the edge of the previous weld bead. Based on this, a recognition method of weld edge based on the improved YOLOv8 model was proposed. To improve the accuracy of the recognition model of weld bead, an efficient multi-scale attention efficient multi-scale attention (EMA) module for cross-space learning was added. To reduce redundant features and simplify the model, a new C2f-RVB module was constructed to replace the C2f module of the original Backbone part. Welding image recognition experiments and ablation experiments have shown that these improvements show that the proposed recognition method of weld edge based on improved YOLOv8 improves the accuracy of weld edge recognition and simplifies the model size. The surfacing welding experiment shows that the proposed recognition method of weld bead edge has a good effect in the automatic surfacing of the anti-corrosion layer of the waste incinerator wall