Seafloor topography data are fundamental for marine resource development, oceanographic research, and maritime rights protection. However, approximately 75% of the ocean remains unsurveyed for bathymetry. Sole reliance on shipborne measurements is insufficient for constructing a global bathymetric model within a short timeframe; consequently, satellite altimetry-based inversion techniques are essential for filling data gaps. Recent advancements have improved the variety and quality of satellite altimetry gravity data. To leverage the complementary advantages of multiparameter gravity data, we propose a 1D convolutional neural network based on a convolutional attention module, termed the Attention Channel 1D Convolutional Network (AC1D). Results of a case study of the Mariana Trench indicated that the AC1D grid predictions exhibited improved agreement with single-beam depth checkpoints, with standard deviation reductions of 6.32%, 20.79%, and 36.77% and root mean square error reductions of 7.11%, 22.82%, and 50.99% compared with those of parallel linked backpropagation, the gravity–geological method, and a convolutional neural network, respectively. The AC1D grid demonstrated enhanced stability in multibeam bathymetric validation metrics and exhibited better consistency with multibeam bathymetry data and the GEBCO2023 grid. Power spectral density analysis revealed that AC1D effectively captured rich topographic signals when predicting terrain features with wavelengths below 6.33 km.

The recognition and segmentation of seafloor topography play a crucial role in marine science research and engineering applications. However, traditional methods for seafloor topography recognition and segmentation face several issues, such as poor capability in analyzing complex terrains and limited generalization ability. To address these challenges, this study introduces the SG-MKD dataset (Submarine Geomorphology Dataset—Seamounts, Sea Knolls, Submarine Depressions) and proposes YOLO-SG (You Only Look Once—Submarine Geomorphology), an algorithm for seafloor topographic unit recognition and segmentation that leverages a lightweight upsampling operator and attention mechanisms. The SG-MKD dataset provides instance segmentation annotations for three types of seafloor topographic units—seamounts, sea knolls, and submarine depressions—across a total of 419 images. YOLO-SG is an optimized version of the YOLOv8l-Segment model, incorporating a convolutional block attention module in the backbone network to enhance feature extraction. Additionally, it integrates a lightweight, general upsampling operator to create a new feature fusion network, thereby improving the model’s ability to fuse and represent features. Experimental results demonstrate that YOLO-SG significantly outperforms the original YOLOv8l-Segment, with a 14.7% increase in mean average precision. Furthermore, inference experiments conducted across various research areas highlight the model’s strong generalization capability.