Multitask network and multispectral image panchromatic sharpening method based on wave band perception dynamic convolution

The invention relates to a multitask network and multispectral image panchromatic sharpening method based on wave band perception dynamic convolution, and belongs to the field of multispectral image processing. The multi-task network comprises a task general sub-network and a task specific sub-network, wherein the task general sub-network comprises two feature extraction modules, a hierarchical feature fusion module, a Transformer and a reconstruction module; and the feature extraction module and the hierarchical feature fusion module adopt wave band sensing dynamic convolution. According to the multispectral image panchromatic sharpening method, firstly, a multitask network is constructed and trained, and then the trained multitask network is used for reasoning, so that multispectral image panchromatic sharpening is realized. In a multi-task network learning stage, a plurality of simulation data sets are constructed based on a Wald protocol widely applied in the field. And then adjusting the parameters of th