摘要 : © 2023Tropical cyclones are common natural disasters in coastal areas, and accurately assessing their intensity is crucial for disaster management and prevention. Different bands in multispectral images represent distinct physical... 展开 © 2023Tropical cyclones are common natural disasters in coastal areas, and accurately assessing their intensity is crucial for disaster management and prevention. Different bands in multispectral images represent distinct physical meanings, and effectively utilizing this information can enhance the accuracy of TC intensity estimation. However, current TC intensity estimation methods do not emphasize the use of multispectral information. Meanwhile, dictionary learning techniques have shown considerable potential in image fusion. To this end, we introduce a multispectral convolutional dictionary learning tropical cyclone intensity estimation model (MCDL-TCIENet), which employs convolutional dictionary learning to integrate multispectral information, enhancing the use of satellite data across different bands. MCDL-TCIENet comprises two primary stages: the multispectral convolutional dictionary learning stage and the tropical cyclone intensity estimation stage. In the MCDL stage, we amalgamate spectral information from different bands using a multimodal fusion strategy, built upon the principles of convolutional dictionary learning. Each spectral type is characterized by two dictionaries: one representing the unique features of that spectrum, and the other capturing shared features across different spectra. In the TCIE stage, MCDL-TCIENet is optimized using an attention mechanism for estimating the intensity of tropical cyclones. We validate MCDL-TCIENet using data from 150 tropical cyclones between 2016 and 2021. The results indicate that the overall RMSE for intensity estimation is 8.85 kt, with an MAE of 6.36 kt. Compared to current advanced methods, MCDL-TCIENet demonstrates competitive performance. 收起