Wavelet-based Bi-dimensional Aggregation Network for SAR Image Change Detection

TL;DR

WBANet tackles SAR image change detection by preserving high-frequency information during attention via $DWT$/$IDWT$ with the Haar wavelet and by fusing spatial and channel dependencies in a Bi-dimensional Aggregation Module. It introduces a Wavelet-based Self-attention Module that achieves lossless down-sampling and an expanded receptive field, plus a BAM that enhances non-linear feature representation. Experiments on three public SAR datasets show state-of-the-art PCC and KC gains over multiple baselines, with ablations confirming the contribution of both WSM and BAM. The approach offers practical improvements for reliable change detection in noisy SAR data and is released as open-source code for reproducibility.

Abstract

Synthetic aperture radar (SAR) image change detection is critical in remote sensing image analysis. Recently, the attention mechanism has been widely used in change detection tasks. However, existing attention mechanisms often employ down-sampling operations such as average pooling on the Key and Value components to enhance computational efficiency. These irreversible operations result in the loss of high-frequency components and other important information. To address this limitation, we develop Wavelet-based Bi-dimensional Aggregation Network (WBANet) for SAR image change detection. We design a wavelet-based self-attention block that includes discrete wavelet transform and inverse discrete wavelet transform operations on Key and Value components. Hence, the feature undergoes downsampling without any loss of information, while simultaneously enhancing local contextual awareness through an expanded receptive field. Additionally, we have incorporated a bi-dimensional aggregation module that boosts the non-linear representation capability by merging spatial and channel information via broadcast mechanism. Experimental results on three SAR datasets demonstrate that our WBANet significantly outperforms contemporary state-of-the-art methods. Specifically, our WBANet achieves 98.33\%, 96.65\%, and 96.62\% of percentage of correct classification (PCC) on the respective datasets, highlighting its superior performance. Source codes are available at \url{https://github.com/summitgao/WBANet}.

Figures (6)

• Figure 1: An overview of the proposed Wavelet-based Bi-dimensional Aggregation Network (WBANet). The WBANet comprises pre-classification module, wavelet-based bi-dimensional aggregation blocks. Each wavelet-based bi-dimensional aggregation block has two critical components: Wavelet-based Self-attention Module and Bi-dimensional Aggregation Module.
• Figure 2: Comparison of the traditional self-attention block in PVT(Pyramid Vision Transformer) and the proposed wavelet-based self-attention module. Linear denotes fully connected layers and $\otimes$ denotes matrix multiplication.
• Figure 3: Illustration of the Bi-dimensional Aggregation Module. It includes two branches: global channel attention and local spatial attention. ⓒ indicates to concatenate in channel axis. $\oplus$ implies element-wise addition and $\odot$ denotes element-wise multiplication.
• Figure 4: Visualized results of different change detection methods on the three datasets: (a) Image captured at $t_1$. (b) Image captured at $t_2$. (c) Ground truth image. (d)-(i) Results by different methods.
• Figure 5: Visualization of the feature representations on the Yellow River dataset. (a) Features before the WSM. (b) Features after the WSM.