.png){kind=icon}
Return
C2Mamba: Collaborative context state space model for remote sensing object detection
.png){kind=icon}
PRE
AI
DOI：10.1007/s11431-025-3285-3.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Object detection in remote sensing images (RSIs) is a fundamental and valuable task in the field of Earth observation and computer vision. Typically, RSIs are acquired from a bird’s-eye view, leading to inherent characteristics such as complex backgrounds, random and dense distribution of small objects, and large-scale variations. These characteristics significantly constrain the performance of existing object detectors. To address these issues, this paper proposes a lightweight yet effective collaborative context state space model, C2Mamba, which integrates global context and multi-scale local receptive field information, thereby enabling higher accuracy for geospatial object detection. First, a collaborative multi-context Mamba is developed to capture diverse context information, including local, global, and channel-wise features, in a hierarchical interactive manner. This enables the generation of a comprehensive scene description that adaptively focuses on geospatial objects while suppressing interference from complex backgrounds. Second, a Gaussian large-receptive field perception module is designed to progressively aggregate context information and fine-grained details via increasingly larger receptive fields. This process is enhanced by a Gaussian modulation mechanism, which improves sensitivity to subtle yet discriminative features, thereby effectively addressing scale variations of remote sensing objects, especially for small ones. Experimental results on three public datasets (RSOD, NWPU VHR-10.v2, and DIOR) demonstrate that C2Mamba outperforms state-of-the-art methods in both detection accuracy and computational efficiency.
Keywords:
remote sensing image
object detection
Mamba
deep learning
Journal
IF:
4.9
Papers:
4.9K
Citations:
9.9K
Organization
Cited Papers
.png){kind=icon}
ShareSaveShareSaveSpectral division multiplexing-based region projection for 3D measurement under mutual reflection
PREAI
ShareSave
View more