LCD-Net: A Lightweight Remote Sensing Change Detection Network Combining Feature Fusion and Gating Mechanism

Wenyu Liu,Jindong Li,Haoji Wang,Run Tan,Yali Fu,Qichuan Tian

2024-10-14

Abstract:Remote sensing image change detection (RSCD) is crucial for monitoring dynamic surface changes, with applications ranging from environmental monitoring to disaster assessment. While traditional CNN-based methods have improved detection accuracy, they often suffer from high computational complexity and large parameter counts, limiting their use in resource-constrained environments. To address these challenges, we propose a Lightweight remote sensing Change Detection Network (LCD-Net in short) that reduces model size and computational cost while maintaining high detection performance. LCD-Net employs MobileNetV2 as the encoder to efficiently extract features from bitemporal images. A Temporal Interaction and Fusion Module (TIF) enhances the interaction between bitemporal features, improving temporal context awareness. Additionally, the Feature Fusion Module (FFM) aggregates multiscale features to better capture subtle changes while suppressing background noise. The Gated Mechanism Module (GMM) in the decoder further enhances feature learning by dynamically adjusting channel weights, emphasizing key change regions. Experiments on LEVIR-CD+, SYSU, and S2Looking datasets show that LCD-Net achieves competitive performance with just 2.56M parameters and 4.45G FLOPs, making it well-suited for real-time applications in resource-limited settings. The code is available at <a class="link-external link-https" href="https://github.com/WenyuLiu6/LCD-Net" rel="external noopener nofollow">this https URL</a>.

Computer Vision and Pattern Recognition,Artificial Intelligence

What problem does this paper attempt to address?

This paper attempts to solve two main problems:

1. **The problem of high computational complexity and a large number of parameters**:
- Although traditional methods based on Convolutional Neural Networks (CNN) have improved the detection accuracy in remote sensing image change detection, these methods usually have high computational complexity and a large number of parameters. This limits their application in resource - constrained environments, such as common edge - computing devices or mobile platforms.
- To meet this challenge, the author proposes a lightweight remote sensing image change detection network (LCD - Net), aiming to reduce the model size and computational cost while maintaining a high - level detection performance.

2. **The problem of complex background interference and the co - existence of multi - scale changes**:
- Textures and similar structures in complex backgrounds (such as water surface reflection and noise) can mask subtle changes, making it difficult for detection methods to effectively distinguish real changes from background interference.
- In addition, traditional methods are not sensitive enough to changes at different scales, and it is difficult to capture the differences between significant changes and subtle local changes, resulting in a decline in detection accuracy, especially prone to false positives and false negatives in complex scenarios.
- To solve these problems, the author introduces a Feature Fusion Module (FFM) and a Gated Mechanism Module (GMM) to enhance the model's sensitivity to multi - scale changes and the ability to suppress background noise.

### Specific solutions

1. **The lightweight backbone network MobileNetV2**:
- Select MobileNetV2 as an encoder because of its excellent performance in feature extraction tasks and extremely small number of parameters.

2. **The Temporal Interaction and Fusion Module (TIF)**:
- Introduce the TIF module in the encoder. By exchanging channel features, it promotes the information interaction between bi - temporal features, provides more abundant context information, and thus improves the model's ability to perceive changes.

3. **The Feature Fusion Module (FFM)**:
- The FFM module enhances the model's sensitivity to multi - scale changes by integrating multi - scale features, especially being able to better separate background noise and real changes in complex backgrounds.

4. **The Gated Mechanism Module (GMM)**:
- Introduce the GMM module in the decoding layer. It dynamically adjusts the channel weights, emphasizes key change areas, and thus improves the model's accuracy and robustness.

Through these designs, the experimental results of LCD - Net on the LEVIR - CD+, SYSU, and S2Looking datasets show that it can achieve performance comparable to existing advanced models with only 2.56M parameters and 4.45G FLOPs, which is suitable for real - time applications and resource - limited environments.

### Summary

The main contributions of this paper include:
- Proposing a new lightweight remote sensing image change detection model, LCD - Net, with only 2.56M parameters and a computational complexity of 4.45G FLOPs.
- Introducing the TIF module to enhance feature expression and semantic consistency through the sharing and interaction of cross - temporal features.
- Designing the FFM module to deeply integrate multi - scale features and enhance the model's sensitivity to subtle changes.
- Introducing the GMM module to dynamically adjust channel weights, enhance the contribution of important channels, and reduce parameter redundancy.
- Experimentally verifying the effectiveness and superiority of the proposed method, achieving better detection performance than nine state - of - the - art models on three benchmark datasets.