Super-resolution-based Change Detection Network with Stacked Attention Module for Images with Different Resolutions (2103.00188v1)

Abstract: Change detection, which aims to distinguish surface changes based on bi-temporal images, plays a vital role in ecological protection and urban planning. Since high resolution (HR) images cannot be typically acquired continuously over time, bi-temporal images with different resolutions are often adopted for change detection in practical applications. Traditional subpixel-based methods for change detection using images with different resolutions may lead to substantial error accumulation when HR images are employed; this is because of intraclass heterogeneity and interclass similarity. Therefore, it is necessary to develop a novel method for change detection using images with different resolutions, that is more suitable for HR images. To this end, we propose a super-resolution-based change detection network (SRCDNet) with a stacked attention module. The SRCDNet employs a super resolution (SR) module containing a generator and a discriminator to directly learn SR images through adversarial learning and overcome the resolution difference between bi-temporal images. To enhance the useful information in multi-scale features, a stacked attention module consisting of five convolutional block attention modules (CBAMs) is integrated to the feature extractor. The final change map is obtained through a metric learning-based change decision module, wherein a distance map between bi-temporal features is calculated. The experimental results demonstrate the superiority of the proposed method, which not only outperforms all baselines -with the highest F1 scores of 87.40% on the building change detection dataset and 92.94% on the change detection dataset -but also obtains the best accuracies on experiments performed with images having a 4x and 8x resolution difference. The source code of SRCDNet will be available at https://github.com/liumency/SRCDNet.

• The paper introduces SRCDNet, a deep learning model that leverages GAN-based super-resolution to reconstruct low-resolution remote sensing images for enhanced change detection accuracy.
• The paper employs a stacked attention module integrated with CNN features to improve multi-scale feature extraction and clearly distinguish changed areas.
• Experimental results show SRCDNet outperforms five state-of-the-art methods with F1 scores of 87.40% and 92.94% on building and seasonal datasets.

Super-resolution-based Change Detection Network for Remote Sensing Images

The paper discusses a novel approach to change detection in remote sensing images characterized by varying resolutions, which holds relevance for ecological protection and urban planning. The paper introduces a super-resolution-based change detection network (SRCDNet) that addresses challenges associated with integrating bi-temporal images of different resolutions by leveraging deep learning techniques, notably convolutional neural networks and attention mechanisms.

The proposed methodology emphasizes overcoming the traditional limitations posed by subpixel-based methods that often lead to substantial error accumulation when high resolution (HR) images are involved due to their intraclass heterogeneity and interclass similarity. SRCDNet employs a super-resolution module with adversarial learning (GAN) to generate HR images from low resolution (LR) inputs, facilitating change detection across images with differing resolution scales.

Key Components and Results

1. Super-resolution Module: The network employs a GAN structure, with a generator and discriminator, aiming at reconstructing LR images to the resolution comparable with HR images. This approach is shown to effectively recover rich semantic information that is crucial for precise change detection.
2. Stacked Attention Module: The integration of convolutional block attention modules (CBAMs) within the feature extraction process enhances multi-scale feature hierarchy, leading to more distinguishable feature pairs which are instrumental in achieving higher detection accuracy.
3. Metric Learning: Using a Siamese network structure allows SRCDNet to leverage metric learning for change detection, facilitating a comprehensive analysis of distance maps between features from bi-temporal images. This technique notably improves the separation between changed and unchanged areas, resulting in more accurate change maps.
4. Experimental Validation: The proposed network demonstrates superior performance over five state-of-the-art change detection methods for both building change detection datasets and seasonal image datasets. Notably, SRCDNet achieves F1 scores of 87.40% for the building dataset and 92.94% for the seasonal dataset, outperforming all baseline methods.

Implications and Future Directions

The implications of these findings are multifaceted. Practically, SRCDNet significantly enhances the capacity for accurate urban and ecological change detection, particularly useful in contexts where routine acquisition of HR images is not feasible. Theoretically, this paper reinforces the value of integrating super-resolution techniques with deep learning approaches, encouraging future exploration of adaptive resolution harmonization in remote sensing applications.

Looking forward, advancements in SRCDNet may involve integrating additional spectral information, expanding applications to multispectral or hyperspectral image scenarios that further harness complex data structures. Furthermore, improvements in handling extreme resolution disparities and real-time processing capabilities could make SRCDNet applicable to a broader range of environmental monitoring and planning tasks.

The provision of source code via GitHub further facilitates benchmarking and experimental reproducibility, fostering collaborative improvements and developments in the field of remote sensing change detection. Overall, SRCDNet represents a significant step towards more accurate and scalable solutions for change detection in dynamic environments.