机载合成孔径雷达海岸带图像匹配优化仿真

Optimization Simulation of Airborne SAR Coastal Image Matching

机载合成孔径雷达(Synthetic Aperture Radar,SAR)海岸带图像匹配技术是海冰监控、石油泄漏检测和海运监控等应用的核心步骤。通过对匹配技术进行优化能提高匹配精度,也能直接提升上述应用的效果。由于机载SAR海岸带图像中广泛存在乘性噪声的干扰,导致传统匹配算法无法提取准确反映地貌信息的特征,出现匹配结果误差大、精度低的问题。针对这一难点,对传统的SAR图像匹配算法进行了优化,使用SIFT提取特征点的位置和方向信息,将特征点周围的图像信息输入卷积神经网络,提取更为全局的特征以消除乘性噪声的影响,用度量学习模块得到更具有区分性的相似性度量做特征点对应,最后使用RANSAC算法实现配准。仿真结果显示,经过优化的算法可以显著地提高匹配精度,可直接用于后续SAR图像应用中。

Airborne Synthetic Aperture Radar（SAR） coastal image matching technology has important applica- tions in the fields of sea ice monitoring, oil leak detection and sea transportation monitoring. The matching accuracy can be improved by optimizing the existing matching technology, which can directly enhance the effect of these applications. However, there are a lot of multiplicative noises in SAR images, so the effect of traditional image matching algorithm based on gradient feature is often poor. Aiming at this problem, a new SAR image registration method based on convolutional neural network （CNN） is proposed. The SIFT algorithm was used to extract the position and orientation information of feature points, and the patches around the feature points were input to the convolution neural network. The RANSAC algorithm was used in the final step of image registration. The simulation results show that our new algorithm gives better results than the traditional algorithms. Our method is important to SAR image processing, and moreover it can directly enhance the effect of follow -up applications.