Multi-scale Change-Aware Transformer for Remote Sensing Image Change Detection

HUAN CHEN, Tingfa Xu, Zhenxiang Chen, Peifu Liu, Huiyan Bai, Jianan Li

Abstract: Change detection identifies differences between images captured at different times. Real-world change detection faces challenges from the diverse and intricate nature of change areas, while current datasets and algorithms are often limited to simpler, uniform changes, reducing their effectiveness in practical application. Existing dual-branch methods process images independently, risking the loss of change information due to insufficient early interaction. In contrast, single-stream approaches, though improving early integration, lack efficacy in capturing complex changes. To address these issues, we introduce a novel single-stream architecture, the Multi-scale Change-Aware Transformer (MACT), which features the Dynamic Change-Aware Attention module and the Multi-scale Change-Enhanced Aggregator. The Dynamic Change-Aware Attention module, integrating local self-attention and cross-temporal attention, conducts dynamic iteration on images differences, thereby targeting feature extraction of change areas. The Multi-scale Change-Enhanced Aggregator enables the model to adapt to various scales and complex shapes through local change enhancement and multiscale aggregation strategies. To overcome the limitations of existing datasets regarding the scale diversity and morphological complexity of change areas, we construct the Mining Area Change Detection dataset. The dataset offers a diverse array of change areas that span multiple scales and exhibit complex shapes, providing a robust benchmark for change detection. Extensive experiments demonstrate that the our model outperforms existing methods, especially for irregular and multi-scale changes.

Primary Subject Area: [Experience] Multimedia Applications

Relevance To Conference:

This work significantly contributes to multimedia/multimodal processing by introducing an advanced single-stream architecture, the Multi-scale Change-Aware Transformer (MACT), which is specifically designed to handle the intricacies of change detection in image sequences. The innovative Dynamic Change-Aware Attention (DCAA) module within MACT effectively addresses the challenge of identifying changes by focusing on the dynamic nature of changes across different scales and time intervals. By integrating local self-attention and cross-temporal attention mechanisms, MACT is capable of capturing both the fine-grained details and the broader context of changes, leading to a more accurate and nuanced understanding of the scene dynamics. Furthermore, the Multi-scale Change-Enhanced Aggregator (MCEA) component of MACT refines the model's ability to recognize and integrate changes at various scales, which is crucial for applications that involve complex and heterogeneous multimedia data.
The proposed Mining Area Change Detection (MACD) dataset, with its diverse and morphologically complex change areas, serves as a robust benchmark for evaluating the model's performance, pushing the boundaries of multimedia analysis in terms of scale diversity and morphological complexity.
Overall, this research enhances the field by providing a novel approach and a dataset to change detection that can be generalized to various multimedia processing tasks, offering improved performance and efficiency. It sets a new standard for handling the complexity and richness of multimedia data, particularly in scenarios where precise change detection is essential.

Submission Number: 1989