Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DDLNet: Boosting Remote Sensing Change Detection with Dual-Domain Learning

Xiaowen Ma, Jiawei Yang, Rui Che, Huanting Zhang, Wei Zhang

TL;DR

The paper addresses RSCD by introducing dual-domain learning that leverages both frequency and spatial information. It presents FEM to capture frequency components via the Discrete Cosine Transform and SRM to recover spatial details through multi-scale feature fusion, integrated within a Siamese backbone and a lightweight decoder. The main contributions are the FEM, SRM, and a hybrid loss that together deliver state-of-the-art accuracy with improved efficiency across three public datasets. This approach enhances change detection boundaries and robustness, enabling more reliable and scalable monitoring in remote sensing applications.

Abstract

Remote sensing change detection (RSCD) aims to identify the changes of interest in a region by analyzing multi-temporal remote sensing images, and has an outstanding value for local development monitoring. Existing RSCD methods are devoted to contextual modeling in the spatial domain to enhance the changes of interest. Despite the satisfactory performance achieved, the lack of knowledge in the frequency domain limits the further improvement of model performance. In this paper, we propose DDLNet, a RSCD network based on dual-domain learning (i.e., frequency and spatial domains). In particular, we design a Frequency-domain Enhancement Module (FEM) to capture frequency components from the input bi-temporal images using Discrete Cosine Transform (DCT) and thus enhance the changes of interest. Besides, we devise a Spatial-domain Recovery Module (SRM) to fuse spatiotemporal features for reconstructing spatial details of change representations. Extensive experiments on three benchmark RSCD datasets demonstrate that the proposed method achieves state-of-the-art performance and reaches a more satisfactory accuracy-efficiency trade-off. Our code is publicly available at https://github.com/xwmaxwma/rschange.

DDLNet: Boosting Remote Sensing Change Detection with Dual-Domain Learning

TL;DR

The paper addresses RSCD by introducing dual-domain learning that leverages both frequency and spatial information. It presents FEM to capture frequency components via the Discrete Cosine Transform and SRM to recover spatial details through multi-scale feature fusion, integrated within a Siamese backbone and a lightweight decoder. The main contributions are the FEM, SRM, and a hybrid loss that together deliver state-of-the-art accuracy with improved efficiency across three public datasets. This approach enhances change detection boundaries and robustness, enabling more reliable and scalable monitoring in remote sensing applications.

Abstract

Remote sensing change detection (RSCD) aims to identify the changes of interest in a region by analyzing multi-temporal remote sensing images, and has an outstanding value for local development monitoring. Existing RSCD methods are devoted to contextual modeling in the spatial domain to enhance the changes of interest. Despite the satisfactory performance achieved, the lack of knowledge in the frequency domain limits the further improvement of model performance. In this paper, we propose DDLNet, a RSCD network based on dual-domain learning (i.e., frequency and spatial domains). In particular, we design a Frequency-domain Enhancement Module (FEM) to capture frequency components from the input bi-temporal images using Discrete Cosine Transform (DCT) and thus enhance the changes of interest. Besides, we devise a Spatial-domain Recovery Module (SRM) to fuse spatiotemporal features for reconstructing spatial details of change representations. Extensive experiments on three benchmark RSCD datasets demonstrate that the proposed method achieves state-of-the-art performance and reaches a more satisfactory accuracy-efficiency trade-off. Our code is publicly available at https://github.com/xwmaxwma/rschange.
Paper Structure (9 sections, 13 equations, 2 figures, 4 tables)

This paper contains 9 sections, 13 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Frequency-domain Enhancement Module
  4. Spatial-domain Recovery Module
  5. Loss Function
  6. Experimental
  7. Datasets and Evaluation Metrics
  8. Implementation Details
  9. Comparison and Analysis

Figures (2)

  • Figure 1: Architecture of the proposed DDLNet, which consists of four components: a Siamese backbone, FEM, SRM and a lightweight change detection header. DS_Conv refers to the depth-wise separable convolution, which is employed to reduce the number of parameters and computational cost while minimizing the compromise in the performance.
  • Figure 2: Comparison of our DDLNet (h) and representative RSCD methods: (d) FC-Siam-Di, (e) LGPNet, (f) SNUNet and (g) USSFC-Net, regarding visualized RSCD results, along with (a) $T_1$, (b) $T_2$ and (c) ground truth, on the WHU and LEVIR-CD test sets. Pixels are colored differently for better visualization (i.e., white for true positive, black for true negative, red for false positive, and green for false negative).