Intelligent Migration Velocity Model Building Based on Image Structure Registration

Velocity model building is a key challenge in geophysical exploration. Many methods often focus on picking single-point velocities, ignoring lateral variations of geologic structures. Full waveform inversion methods rely on low-frequency data and initial velocity models to converge. Artificial intelligence methods, while promising, heavily rely on the quantity of labeled data and often lack physical interpretability. To address these issues, we propose a label-free velocity model building approach based on image structure registration. The method follows the principle of geological pattern consistency: different common offset gather (COG) profiles correspond to the same geological body. By maximizing similarity across COG profiles, the network captures structural differences and updates velocity model (network parameter), achieving label-free velocity model building. A global optimization criterion based on the similarity of COG profiles ensures the velocity model aligns with lateral variations in the seismic profile. Additionally, a novel travel-time activation function imposes hyperbolic constraints on image registration, enhancing physical consistency. Synthetic and field data tests show that velocity models from this method achieve high accuracy and lateral resolution, significantly enhancing the imaging accuracy of laterally varying geological structures.