MSAC-Net: 3D Multi-Scale Attention Convolutional Network for Multi-Spectral Imagery Pansharpening

Pansharpening fuses spectral information from the multi-spectral image and spatial panchromatic image, generating super-resolution images with high resolution. In this paper, we proposed a novel 3D multi-scale attention convolutional network (MSAC-Net) based on typical U-Net framework for imagery pansharpening. MSAC-Net is designed via convolution, mechanism replaces skip connection between contraction expansion pathways. Multiple pansharpening layers at pathway are to calculate reconstruction results preserving information. The performance verified IKONOS QuickBird satellites’ datasets, proving that achieves comparable or superior state-of-the-art methods. Additionally, 2D convolution compared, influences of number convolutions in block, weight information, network’s depth analyzed.