Work

DCCPNet: A Dual-Branch Channel Cross-Concatenation Pan-Sharpening Network for Satellite Remote Sensing Imagery

Year: 2025

Type: article

Abstract: Pan-sharpening performance has been significantly improved by deep learning, yet current networks are limited by single-branch architectures or non-interactive dual-branch designs. To surmount these limitations, a dual-branch channel cross-concatenation pan-sharpening network, i.e., DCCPNet, is proposed, which introduces Dual-branch Channel Crossconcatenation (DCC) blocks to facilitate bidirectional feature exchange between panchromatic and multispectral branches. Within this framework, hierarchical features are extracted through Multi-scale Convolution Residual (MCR) blocks, while complementary spatial and spectral information is adaptively fused in the DCC blocks. Comprehensive evaluations conducted on IKONOS and WorldView-2 datasets demonstrate that the proposed method achieves state-of-the-art performance across 11 quantitative metrics. Furthermore, the proposed method exhibits apparent superiority to other advanced methods in image visual effects, operational efficiency testing, and NDVI application, showcasing its robust competitiveness. The code link is https://github.com/Zeyuchunfeng/DCCPNet.

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

DOI: 10.1109/jstars.2025.3574743