Skip to content
Open access

Mamba-LiteDSN: Lightweight Dual-Stream Network for Efficient Remote Sensing Change Detection

2026 · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · Vol 19, pp. 31730-31743 · 0 citations · 38 references

Abstract

Change detection (CD) is a fundamental task in remote sensing. Many existing CD methods primarily focus on detection accuracy at the cost of a large number of parameters and high computational complexity. This article proposes a lightweight dual-stream network with Mamba for efficient remote sensing CD, named Mamba-LiteDSN. We design a dual-stream encoder–decoder architecture, in which a weight-shared MobileNetV2 encoder extracts multiscale features from bitemporal image pair. An enhanced memory visual mamba module integrates linear-complexity state-space modeling to strengthen feature representation, while a semantic-guided aggregation module (SAM) captures global semantic information from multiscale difference features. The decoder employs a channel interaction module to optimize cross-channel feature interactions and suppress pseudochanges, followed by a decoding unit (DU) that progressively restores spatial resolution under the guidance of SAM. Experiments on the public datasets LEVIR-CD and WHU-CD demonstrate that Mamba-LiteDSN achieves competitive CD performance with only 2.86 M parameters and 3.16 G FLOPs. We further construct dataset RLB-CD and evaluate the model on our custom RLB-CD dataset, verifying its effectiveness in large-scale water conservancy scenarios.

Read PDF

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