Global-Local Feature Cross Fusion Network for Semantic Segmentation of Remote Sensing Images

Accurate semantic segmentation of high-resolution remote sensing images requires both global contextual information for scene understanding and local texture details for precise feature delineation. In this article, we propose a novel global-local feature cross fusion network for semantic segmentation of remote sensing images. This network explores global and local features by InceptionNeXt block and ResBlock. The features are then combined via a channel-wise correlation mechanism to address the limitations of each feature and enhance their complementary strength. Moreover, to diminish intra-class inconsistency and alleviate boundary displacement, FreqFusion is employed to integrate the low-level features derived from the encoder with the high-level features from the decoder. The proposed network is evaluated on Vaihingen and Potsdam benchmarks and the quantitative results show the proposed network can achieve overall accuracy (OA) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{9 1. 1 \%}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{8 9. 8 \%}$</tex>, respectively, which are better than several representative semantic segmentation approaches.