LoG-CAN: local-global Class-aware Network for semantic segmentation of remote sensing images

TL;DR

LoG-CAN is a multi-scale network with global and local class-aware modules designed for semantic segmentation of remote sensing images, effectively handling complex backgrounds and scale variations.

Contribution

The paper introduces LoG-CAN, a novel multi-scale network with global and local class-aware modules, improving segmentation accuracy and efficiency on remote sensing datasets.

Findings

Outperforms state-of-the-art methods on ISPRS datasets

Reduces network parameters and computational cost

Effectively segments objects at multiple scales

Abstract

Remote sensing images are known of having complex backgrounds, high intra-class variance and large variation of scales, which bring challenge to semantic segmentation. We present LoG-CAN, a multi-scale semantic segmentation network with a global class-aware (GCA) module and local class-aware (LCA) modules to remote sensing images. Specifically, the GCA module captures the global representations of class-wise context modeling to circumvent background interference; the LCA modules generate local class representations as intermediate aware elements, indirectly associating pixels with global class representations to reduce variance within a class; and a multi-scale architecture with GCA and LCA modules yields effective segmentation of objects at different scales via cascaded refinement and fusion of features. Through the evaluation on the ISPRS Vaihingen dataset and the ISPRS Potsdam dataset, experimental results indicate that LoG-CAN outperforms the state-of-the-art methods for general semantic segmentation, while significantly reducing network parameters and computation. Code is available at~\href{https://github.com/xwmaxwma/rssegmentation}{https://github.com/xwmaxwma/rssegmentation}.