A Robust Multisource Remote Sensing Image Matching Method Utilizing Attention and Feature Enhancement Against Noise Interference

Yuan Li,Dapeng Wu,Yaping Cui,Peng He,Yuan Zhang,Ruyan Wang

2024-10-01

Abstract:Image matching is a fundamental and critical task of multisource remote sensing image applications. However, remote sensing images are susceptible to various noises. Accordingly, how to effectively achieve accurate matching in noise images is a challenging problem. To solve this issue, we propose a robust multisource remote sensing image matching method utilizing attention and feature enhancement against noise interference. In the first stage, we combine deep convolution with the attention mechanism of transformer to perform dense feature extraction, constructing feature descriptors with higher discriminability and robustness. Subsequently, we employ a coarse-to-fine matching strategy to achieve dense matches. In the second stage, we introduce an outlier removal network based on a binary classification mechanism, which can establish effective and geometrically consistent correspondences between images; through weighting for each correspondence, inliers vs. outliers classification are performed, as well as removing outliers from dense matches. Ultimately, we can accomplish more efficient and accurate matches. To validate the performance of the proposed method, we conduct experiments using multisource remote sensing image datasets for comparison with other state-of-the-art methods under different scenarios, including noise-free, additive random noise, and periodic stripe noise. Comparative results indicate that the proposed method has a more well-balanced performance and robustness. The proposed method contributes a valuable reference for solving the difficult problem of noise image matching.

Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the accuracy and robustness of multi - source remote sensing image matching under noise interference. Specifically, due to the influence of various noises (such as sensor thermal noise and electromagnetic interference noise), remote sensing images face the following challenges during the image matching process:

1. **Failure to extract effective features**: Noise blurs the structure and details of the image, interfering with feature detection and extraction, making it difficult for the matching method to identify meaningful feature points in noisy images, thus making it difficult to achieve dense matching.
2. **False matching**: Noise may introduce false features or change the descriptors of real features, causing the matching method to wrongly pair these features with noisy features in another image, resulting in false matching.
3. **Reduced robustness**: Robustness refers to the consistency and accuracy of an image matching method under different conditions (including in the presence of noise). Noise introduces uncertainty and inconsistency, reducing the robustness of the image matching method.

To solve these problems, the author proposes a robust multi - source remote sensing image matching method based on the attention mechanism and feature enhancement to deal with noise interference. The main contributions of this method include:

1. **Combining deep convolution and Transformer attention mechanism for dense feature extraction**: By extracting multi - scale feature information and highlighting local features, more discriminative and robust feature descriptors are constructed, and significant feature points can be extracted even in the presence of a large amount of noise and periodic stripes.
2. **Introducing an outlier removal network based on a binary classification mechanism**: The geometric information of feature points is extracted by a multi - layer perceptron, and global and local information is obtained by context normalization to establish an effective geometrically consistent image correspondence, thereby performing weighted classification on the dense matching results and removing outliers.

To verify the effectiveness of the proposed method, the author conducted experiments using multi - source remote sensing image datasets in different scenarios, including no - noise, additive random noise, and periodic stripe noise. The experimental results show that this method can still maintain a high matching success rate and a good performance balance under high - noise interference, providing a valuable reference for solving the difficult problem of noisy image matching.

In summary, this paper aims to improve the matching accuracy and robustness of multi - source remote sensing images under noise interference conditions by introducing deep learning and the attention mechanism, so as to better meet the needs of practical applications.