TITLE:
Research on the Application of VMamba Scanning Algorithm in High-Resolution Remote Sensing Change Detection
AUTHORS:
Rui Shi, Zhenchuan Wang
KEYWORDS:
VMamba Algorithm, Remote Sensing Images, Change Detection, Mamba Network
JOURNAL NAME:
Journal of Computer and Communications,
Vol.13 No.4,
April
29,
2025
ABSTRACT: The VMamba (Visual State Space Model) is built upon the Mamba model by stacking Visual State Space (VSS) modules and utilizing the 2D Selective Scan (SS2D) module to extend the original Mamba model’s capability from handling one-dimensional sequences to two-dimensional sequences. This enhancement broadens the application of the Mamba model to visual tasks. Compared to CNNs and Transformers, Mamba retains two significant advantages: long-sequence modeling and linear complexity, making it well-suited for high-resolution image tasks. While previous studies have explored its application in high-resolution remote sensing image processing, challenges such as high computational cost and slow training speed persist. The core issue arises from multiple sequence scans and the merging process after sequence processing, which slows down model training. This paper investigates the sequence scanning process and proposes multiple scanning algorithms. Specifically, we employ a unidirectional sequence scanning algorithm in high-resolution remote sensing change detection to reduce the number of scans in the scanning module, thereby accelerating model training. By evaluating its performance in classification and object detection tasks, we thoroughly test the feature extraction capabilities of these scanning algorithms in the VMamba model. Through comparative experiments in high-resolution remote sensing change detection, we demonstrate that our proposed unidirectional scanning algorithm achieves comparable or even superior performance with higher computational efficiency compared to omnidirectional scanning algorithms. Experimental results further suggest a potential correlation between the SS2D algorithm’s feature extraction capability and its performance in remote sensing change detection. This study provides valuable insights for further research on Mamba-based remote sensing change detection algorithms.