Generation of Marine Information Data Based on Generative Adversarial Networks

Oceanic information is crucial for the sustainable development of human society. Effective oceanic information can reliably predict storms, red tides, and coastal zone ratings, playing a vital role in maintaining the stable development of human society. Addressing the issues of scarce oceanic information data and insufficient physical consistency in traditional generative models, this paper proposes an ocean spatiotemporal data augmentation method based on an attention-based generative adversarial network. This method integrates an attention mechanism with a generative adversarial network (GAN) and introduces an Adaptive Spectral Block (ASB) in the generator to achieve efficient modeling of ocean physical laws and capture complex spatiotemporal dynamic features. Specifically, an improved loss function is used to constrain the training process of the generator, ensuring the physical consistency of the generated data. The Adaptive Spectral Block performs weighting and masking operations on frequency-domain signals to enhance the model's ability to express multi-scale features. Meanwhile, the attention mechanism is combined to optimize spatiotemporal correlation modeling. Experiments show that the proposed method achieves an error of less than 3% in statistical features between generated and real data for key ocean parameters such as temperature, salinity, and current velocity. Furthermore, in data-scarce scenarios, the model generates diverse data through noise perturbation and random sampling strategies, significantly improving the generalization capability of climate prediction and ocean monitoring models. This study provides an efficient and reliable data augmentation solution for the field of ocean science.