A Mamba-based Siamese Network for Remote Sensing Change Detection

Jay N. Paranjape,Celso de Melo,Vishal M. Patel

2024-07-09

Abstract:Change detection in remote sensing images is an essential tool for analyzing a region at different times. It finds varied applications in monitoring environmental changes, man-made changes as well as corresponding decision-making and prediction of future trends. Deep learning methods like Convolutional Neural Networks (CNNs) and Transformers have achieved remarkable success in detecting significant changes, given two images at different times. In this paper, we propose a Mamba-based Change Detector (M-CD) that segments out the regions of interest even better. Mamba-based architectures demonstrate linear-time training capabilities and an improved receptive field over transformers. Our experiments on four widely used change detection datasets demonstrate significant improvements over existing state-of-the-art (SOTA) methods. Our code and pre-trained models are available at <a class="link-external link-https" href="https://github.com/JayParanjape/M-CD" rel="external noopener nofollow">this https URL</a>

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper is primarily dedicated to addressing the problem of change detection in remote sensing images. Specifically, the research team proposes a Siamese Network based on the Mamba architecture, named M-CD (Mamba-based Change Detector), for more accurately identifying change areas between remote sensing images taken at different times.

The paper points out that change detection is a fundamental tool in remote sensing image analysis, which is significant for monitoring environmental changes, human-induced alterations, and making corresponding decisions and predicting future trends. Existing deep learning methods, such as Convolutional Neural Networks (CNN) and Transformers, have achieved remarkable success in this field. However, these methods still have some limitations, especially when dealing with various differences between images (such as illumination changes, resolution differences, etc.). Therefore, developing a change detection system that can effectively address these challenges is a complex and challenging task.

To solve the above problems, the authors propose M-CD, a change detection model based on the Mamba architecture. The Mamba architecture was chosen for its linear time training capability and larger receptive field compared to Transformers. The main contributions of M-CD include:

1. Developing a novel M-CD model that can generate accurate change masks based on two images.
2. Designing a difference module to combine features from the image pair and pass them to a mask decoder, which is modified to better understand channel information as well as temporal and spatial information.
3. Evaluating the proposed M-CD method on four widely used public datasets and demonstrating consistent improvements compared to existing state-of-the-art methods.

In summary, this research aims to improve the change detection task, particularly for remote sensing images, by introducing the Mamba architecture, thereby achieving more accurate change area segmentation.