Abstract: To address the challenge of non-contact identification of surface micro-damage in 12Cr1MoV main steam pipes under high-temperature and high-pressure service conditions, specimens simulating grain coarsening on the surface of main steam pipes after long-term service were prepared by solution heating. A method based on laser ultrasonic surface wave characteristic parameters was adopted to characterize grain size, and two models were developed to characterize surface grain size using laser ultrasonic velocity and attenuation coefficient. The prediction relative errors and determination coefficient R² of the two models were 2.2%, 0.81, and 22.4%, 0.91, respectively. Combined with a genetic algorithm-optimized backpropagation neural network (GA-BP neural network), a parameter characterization model was established using ultrasonic velocity and attenuation coefficient as input features and surface grain size as the output feature. The results demonstrated that the prediction error and R² of the model were 4.5% and 0.99, respectively. This improved the correlation significance between input and output features in the velocity-based model, reduced the prediction error of the attenuation-based method, and validated the advantages of the GA-BP neural network identification in grain size characterization. This study provides technical support for online monitoring of surface microstructural damage in main steam header pipes under high-temperature and high-pressure environments.