Due to various physical and technical limitations, the radiation energy received by different sensors and the amount of data collected vary, and a single sensor cannot simultaneously obtain high spatial and spectral images. Therefore, it is necessary to develop an ideal application-oriented technique for generating multi-spectral image with high spatial resolution. The pan-sharpening method fuses the low spatial resolution multispectral image with the high spatial resolution panchromatic image to obtain a hyperspectral image with rich spatial spectral information. Although significant progress has been made in pan-sharpening methods in recent years, most methods still have two limitations: firstly, limited by network structure and single attention mechanism, global and local features cannot be used simultaneously, resulting in loss of spatial information; secondly, using the Wald protocol to obtain high-resolution multispectral images leads to loss of spectral and detail information. To address these problems, this paper proposes a pan-sharpening framework MAPNet based on multiple attention progressive network. In order to extract more important information, we fully utilize the feature information contained in panchromatic and multispectral images to reduce the interference of redundant information. The low resolution and full resolution phases are closely linked using a progressive pattern. MAPNet trains the ability to extract global information, spectral information and gradient information to reduce the loss of spectrum and detail due to size changes. The multi-attention module combines self-attention, spatial attention and channel attention to achieve multi-modal analysis of global features, local features, spatial features and channel features, thereby further improving MAPNet's ability to retain texture details. The algorithm proposed in this paper is compared with the existing traditional methods BT-H, C-MTF-GLP-CBD, GS, BDSD, PRACS and deep learning methods MUCNN, MDCUN, Band-Aware, PNN and TFNet on the GF-2 dataset. Additionally, this paper records the performance of models with different stages and structures. Objective measurements include RMSE, RASE, SAM, ERGAS, QAVE, SSIM, FSIM, QNR, D_s, D_λ. By combining subjective visual assessment with objective evaluation, the results indicate that MAPNet fusion images retain more spectral and detail information.