Intelligent High-Resolution Marine Seismic Survey Using Machine Learning and Adaptive Acquisition Optimization

Conventional high-resolution marine seismic surveys often face challenges in detecting small features in the shallow subsurface because acquisition parameters involve trade-offs among spatial resolution, penetration, signal-to-noise ratio (SNR), and operational efficiency. This study presents an intelligent marine seismic acquisition framework that integrates selectable acquisition hardware with modified ML Real domain adaptation, deep learning reconstruction, quantitative quality-control (QC) evaluation, and a human-supervised adaptive agent. Rather than directly predicting acquisition parameters with the neural network, the deep learning model reconstructs high-resolution seismic images from degraded acquisitions, which are subsequently evaluated using QC metrics including reflector continuity, spectral bandwidth, SNR, and depth penetration. The adaptive agent uses these metrics to rank discrete, hardware-feasible configurations of source type, source energy, dominant frequency, and shot interval and recommends the configuration for the subsequent acquisition cycle. Synthetic benchmark tests demonstrate improved imaging of thin layers and localized subsurface features under the tested conditions. A two-cycle nearshore field experiment further demonstrates the operational feasibility of the workflow, in which QC evaluation of the initial acquisition guided the subsequent acquisition configuration and resulted in improved reflector continuity, penetration, and SNR. The results show that integrating AI-assisted numerical modeling and evaluation can guide acquisition-parameter refinement and reduce reliance on repeated offshore trial-and-error surveys. The proposed framework provides a proof-of-concept for integrating configurable high-resolution seismic hardware with AI-assisted, QC-guided acquisition optimization for marine shallow-subsurface imaging.