Change Detection in Synthetic Aperture Radar Images Using a Dual-Domain Network (2104.06699v2)

Abstract: Change detection from synthetic aperture radar (SAR) imagery is a critical yet challenging task. Existing methods mainly focus on feature extraction in spatial domain, and little attention has been paid to frequency domain. Furthermore, in patch-wise feature analysis, some noisy features in the marginal region may be introduced. To tackle the above two challenges, we propose a Dual-Domain Network. Specifically, we take features from the discrete cosine transform domain into consideration and the reshaped DCT coefficients are integrated into the proposed model as the frequency domain branch. Feature representations from both frequency and spatial domain are exploited to alleviate the speckle noise. In addition, we further propose a multi-region convolution module, which emphasizes the central region of each patch. The contextual information and central region features are modeled adaptively. The experimental results on three SAR datasets demonstrate the effectiveness of the proposed model. Our codes are available at https://github.com/summitgao/SAR_CD_DDNet.

• The paper introduces DDNet, a dual-domain network that integrates spatial and DCT features to significantly enhance change detection in SAR images.
• The method uses a multi-region convolution module and reshaped DCT coefficients to suppress speckle noise and highlight relevant image patterns.
• Experimental results on three datasets show reduced false positives and negatives along with improved overall accuracy and kappa coefficient.

Change Detection in Synthetic Aperture Radar Images Using a Dual-Domain Network

Change detection in synthetic aperture radar (SAR) imagery represents a significant challenge and a crucial aspect of remote sensing. The paper by Qu et al. introduces a novel approach to improving change detection accuracy by leveraging both spatial and frequency domain features through a proposed Dual-Domain Network (DDNet). A key innovation of their work is the integration of discrete cosine transform (DCT) domain features into the detection process, traditionally dominated by spatial domain analysis.

Method

The proposed DDNet synergizes spatial and frequency domain features to alleviate the impact of speckle noise commonly found in SAR imagery. The spatial domain analysis employs a multi-region convolution (MRC) module that focuses on the central patch region while maintaining contextual information. This approach aims to minimize the introduction of marginally noisy features, thus highlighting relevant image patterns more effectively.

The frequency domain analysis involves the use of DCT for feature extraction, where reshaped DCT coefficients are utilized as a network branch. This not only compresses representations but also enriches image patterns for improved noise suppression, which is beneficial for change detection tasks. The DDNet hence forms a comprehensive framework, incorporating both domains to enhance detection robustness.

Results

Experimental validation across three SAR datasets—Ottawa, Sulzberger, and Yellow River—demonstrated superior performance via reduced false positives and negatives, and improved scores in overall error, percentage of correct classification, and kappa coefficient. The methodology notably excelled particularly with complex datasets like the Yellow River images, where speckle noise prevalence is substantial.

Implications

The proposed dual-domain approach shows substantial potential for improving the accuracy and reliability of change detection models in SAR imagery. It is particularly impactful in scenarios where traditional spatial-only methods falter due to noise interference. Practically, this can lead to more reliable monitoring and assessment in fields such as environmental observation, urban development tracking, and disaster impact analysis.

Future Directions

Future work may include extending this dual-domain approach to larger scale datasets, testing in different sensor configurations, or integrating additional frequency domain compression techniques to further refine feature extraction processes. Exploring unsupervised deep learning paradigms similar to DDNet introduces promising avenues for enhancing remote sensing capabilities in the detection of changes across temporal image datasets. These developments may further solidify AI's role in advancing SAR imagery analysis.