Accurate observation of nearshore waves is crucial for coastal safety. In this study, the feasibility of extracting wave information from wave video images captured by shore-based cameras using deep learning methods was explored, focusing on inverting nearshore significant wave height (SWH) from instantaneous wave video images. The accuracy of deep learning models in classifying wind wave and swell wave images was investigated, providing reliable classification results for SWH inversion research. A classification network named ResNet-SW for wave types with improved ResNet was proposed. On this basis, the impact of instantaneous wave images, meteorological factors, and oceanographic factors on SWH inversion was evaluated, and an inversion network named Inversion-Net for SWH that integrates multiple factors was proposed. The inversion performance was significantly enhanced by the specialized models for wind wave and swell. Additionally, the inversion accuracy and stability were further enhanced by improving the loss function of Inversion-Net. Ultimately, time series inversion results were synthesized from the outputs of multiple models; the final inversion results yielded a mean absolute error of 0.04 m and a mean absolute percentage error of 8.52%. Despite certain limitations, this method can still serve as a useful alternative for wave observation.

The wave spectrum, as a key statistical feature describing wave energy distribution, is crucial for understanding wave propagation mechanisms and supporting ocean engineering applications. This study, based on ERA5 reanalysis spectrum data, proposes a model combining CNN and xLSTM for rapid gridded wave spectrum prediction over the Bohai and Yellow Seas domain. It uses 2D gridded spectrum data rather than a spectrum at specific points as input and analyzes the impact of various input factors at different time lags on wave development. The results show that incorporating water depth and mean sea level pressure significantly reduces errors. The model performs well across seasons with the seasonal spatial average root mean square error (SARMSE) of spectral energy remaining below 0.040 m²·s and RMSEs for significant wave height (SWH) and mean wave period (MWP) of 0.138 m and 1.331 s, respectively. At individual points, the spectral density bias is near zero, correlation coefficients range from 0.95 to 0.98, and the peak frequency RMSE is between 0.03 and 0.04 Hz. During a typical cold wave event, the model accurately reproduces the energy evolution and peak frequency shift. Buoy observations confirm that the model effectively tracks significant wave height trends under varying conditions. Moreover, applying a frequency-weighted loss function enhances the model’s ability to capture high-frequency spectral components, further improving prediction accuracy. Overall, the proposed method shows strong performance in spectrum prediction and provides a valuable approach for regional wave spectrum modeling.