Skip to content
Conference

MSF-SG-DeepLabV3+ network for land-cover recognition in high-resolution remote sensing images

Oct 2026 · Fourth International Conference on Remote Sensing, Mapping, and Geographic Information Systems (RSMG 2026) · 0 citations

Abstract

To address pronounced target-scale variation, ambiguous inter-class boundaries, and the fragmentation of elongated objects such as roads in land-cover classification using high-resolution remote sensing imagery, this paper proposes a Multi-Scale Fusion Decoder with Semantic Skip Gate (MSF-SG) decoder based on DeepLabV3+. During decoding, a Semantic Skip Gate modulates encoder skip connections at the pixel level according to the current semantic confidence, suppressing redundant textures in high-confidence regions while retaining more structural details in low-confidence and boundary regions. In addition, skip-feature fusion is organized at two scales, H/8 and H/4, to improve the utilization of mid-level structural information and shallow details. Experimental results show that the proposed method achieves mIoU scores of 50.38% and 83.13% on the LoveDA and ISPRS Potsdam datasets, respectively. It effectively improves land-cover classification accuracy and shows improved performance in boundary regions and complex land-cover transitions.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.