基于增强型注意力残差模块的遥感图像全色锐化方法研究

Research on Remote Sensing Image Spatial Resolution Enhancement Method Based on Enhanced Attention Residual Module

摘要:

目前基于深度学习的全色锐化方法往往追求性能而忽略效率,导致模型复杂且计算量大。为此,本文提出一种高效的全色锐化网络,其核心是增强型注意力残差模块(EARM)。该模块通过门控融合机制与双重注意力,有效整合多光谱与全色图像特征,在保持高效率的同时提升重建质量。在多个公开数据集上的实验表明,本方法在客观指标与视觉质量上均优于主流方法,为高效全色锐化提供了新思路。

Abstract:

Current pan-sharpening methods based on deep learning often prioritize performance at the expense of efficiency, resulting in complex models with high computational demands. To address this, this paper proposes an efficient pan-sharpening network centered around the Enhanced Attention Residual Module (EARM). This module effectively integrates multispectral and panchromatic image features through a gated fusion mechanism and dual attention, enhancing reconstruction quality while maintaining high efficiency. Experiments on multiple public datasets demonstrate that this method outperforms mainstream approaches in both objective metrics and visual quality, offering a novel approach to efficient pan-sharpening.