A Plume Resolution Enhancement Network-based method for enhancing the resolution of cold seep plume seismic data

Cold seep bubble plumes serve as critical indicators of submarine gas hydrate deposits, yet their seismic imaging resolution is often limited by the dominant frequency of marine seismic sources and signal attenuation, making clear identification challenging. This paper proposes a high-resolution processing method based on a lightweight deep learning model, Plume Resolution Enhancement Network (PRENet), to achieve end-to-end intelligent enhancement of cold seep plume seismic data from low to high resolution. For data construction, the migration patterns and morphological characteristics of plumes were analyzed, and a fixed wavelet frequency difference strategy with a controllable signal-to-noise ratio was employed. Using forward convolutional modeling, a large-scale training dataset of paired low- and high-resolution samples was generated, providing a reliable foundation for model training. For the model, a multi-scale atrous spatial pyramid pooling module, a channel attention mechanism, and a global residual connection were introduced into an encoder–decoder architecture. With an extremely low parameter count, the model effectively extracts multi-scale plume features. A combined loss function dominated by the structural similarity index measure was designed to balance structural fidelity and high-frequency detail recovery. Test results demonstrate that PRENet substantially outperformed the comparison model with an equivalent parameter count across all quantitative metrics on an independent test set. Spectral analysis confirms that the signal frequency band is effectively broadened and a substantial portion of high-frequency components is recovered. Noise robustness, missing data, and frequency difference generalization experiments further validate the model’s favorable robustness and generalization capability. This proof-of-concept study demonstrates the feasibility of enhancing the resolution of cold seep plume seismic data using a lightweight deep learning approach, offering a promising direction for future fine-scale delineation of cold seep plumes in submarine gas hydrate exploration, subject to validation using field data.