New Horizons in Remote Sensing: Change State Space Models for More Efficient Change Detection

Remote sensing plays an increasingly important role in the observation and analysis of our planet. From monitoring environmental changes to urban planning, the possibilities arising from the analysis of satellite images are diverse. A central aspect of this is change detection, i.e., the identification of differences between images of the same area at different times. Traditionally, methods such as Convolutional Neural Networks (ConvNets) and Vision Transformers (ViTs) have been used for this purpose. However, these approaches reach their limits when it comes to the efficient processing of large amounts of data and the modeling of long-term dependencies.

A promising new approach in this area are the so-called State Space Models (SSMs), which are characterized by their ability to effectively model temporal dependencies in data. One example of this is the "Vision Mamba" architecture, which has already been successfully used in remote sensing, primarily as a basis for feature extraction. Building on this foundation, researchers are now presenting the Change State Space Model (CSSM), a model specifically designed for change detection.

In contrast to conventional approaches, the CSSM focuses specifically on the relevant changes between two temporally offset images and filters out irrelevant information. This focus on the actually changed features allows for a significant reduction in network parameters, which in turn leads to a considerable increase in computational efficiency. At the same time, the detection performance remains high and the model proves robust against a degradation of the input data.

The effectiveness of the CSSM was evaluated using three established benchmark datasets. The results show that the CSSM outperforms ConvNets, ViTs, and Mamba-based models in terms of detection accuracy, while simultaneously requiring significantly less computational effort. These results underscore the potential of CSSM for future change detection in remote sensing.

Advantages of CSSM at a Glance:

The focus on relevant changes enables more efficient processing of large amounts of data.

The reduced number of network parameters leads to higher computational efficiency.

The model shows high detection performance and robustness against data degradation.

The architecture of the CSSM allows for targeted adaptation to the specific requirements of change detection.

The development of CSSM represents an important step towards more efficient and precise change detection in remote sensing. The ability to analyze large amounts of data quickly and reliably opens up new possibilities for environmental monitoring, urban planning, and many other application areas. With the planned release of the implementation on GitHub, research in this area will be further advanced and the application of CSSM for practical use will be facilitated.

Bibliographie: Ghazaei, E., & Aptoula, E. (2025). Change State Space Models for Remote Sensing Change Detection. arXiv preprint arXiv:2504.11080. https://arxiv.org/abs/2504.11080 https://arxiv.org/html/2504.11080v1 https://huggingface.co/papers/2404.03425 https://github.com/ChenHongruixuan/MambaCD https://www.researchgate.net/publication/381589573_ChangeMamba_Remote_Sensing_Change_Detection_With_Spatiotemporal_State_Space_Model https://www.sciencedirect.com/science/article/pii/S0016003224008457 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5184974 https://github.com/Event-AHU/Mamba_State_Space_Model_Paper_List https://ieeexplore.ieee.org/document/10599227/ https://www.researchgate.net/publication/379732300_ChangeMamba_Remote_Sensing_Change_Detection_with_Spatio-Temporal_State_Space_Model