Objective To guarantee the precise implementation of refueling overhaul, nuclear power plants conduct core verification and spent fuel pool inventory during fuel loading and unloading to check the code and placement of fuel assemblies. Taking a domestic AP1000 third-generation pressurized water reactor as the research object, underwater inspection videos captured in refueling operations are adopted to realize intelligent identification of assembly codes for on-site auxiliary verification, which helps boost inspection efficiency and build an anti-misloading control system.

Method With on-site videos of core verification and spent fuel pool inventory collected from an operating nuclear power plant as datasets, an intelligent verification system was developed via artificial intelligence and image processing algorithms. The improved EPBC-YOLOv8 was deployed for character region detection, and hybrid attention recognition network (HARNet) was used for fuel assembly code recognition. Based on authentic field videos, sample screening, model training and parameter optimization were completed. The identification results were automatically matched against core design documents, and matching positions were highlighted on the core layout drawing.

Result Experimental tests demonstrate that the proposed system reaches a character recognition accuracy of 93.6% and a spatial coordinate lighting rate of 94.7%.

Conclusion The developed system features sound practicability and effectiveness, supporting the construction of an integrated intelligent platform for core verification and spent fuel pool inventory in nuclear overhaul. Further optimization of target detection and character recognition models can accelerate the engineering application of artificial intelligence in nuclear power operation and maintenance.