NCGLF2: Network combining global and local features for fusion of multisource remote sensing data

The fusion of multisource remote sensing (RS) data has demonstrated significant potential in target recognition and classification tasks. However, there is limited emphasis on capturing both high- and low-frequency information from these data sources. Additionally, effectively integrating multisource data remains a challenging task, as the absence of redundancy and discriminant information hampers the applications of RS data. In this paper, we propose a fusion network called network combining global and local features (NCGLF2) that integrates global and local features (GLF) extracted from multisource RS data. This approach effectively leverages the capabilities of convolutional neural networks (CNNs) to extract high frequency features while utilizing transformer architecture to replicate low frequency information and remote correlations. Firstly, a scale information aggregation (SIA) module extracts multiscale shallow layer features from the input data sources. Secondly, a structural information learning transformer (SIL-Trans) module captures low frequency features, while an invertible neural network (INN) module learns high frequency information. Finally, a GLF fusion module maximizes the complementary characteristics of multisource RS data and GLF to effectively fuse high- and low-frequency information. Our experimental results with three benchmark datasets indicate that NCGLF2 outperforms existing state-of-the-art approaches in terms of feature representation and compatibility with diverse data types. The code is available at https://github.com/renqi1998/NCGLF2.