基于地基雷达图像的无监督变化检测

地基雷达是近20几年逐渐发展成熟的微波遥感成像技术,目前已广泛应用于滑坡、崩塌等地质灾害的监测中。地基雷达通过干涉测量原理可以监测到目标区域发生的微小形变,然而受人为因素、地质因素、气象因素等影响,导致雷达图像失相干严重,给长期定量化监测带来较大的难度。因此,迫切需要在定量监测的基础上,进一步开展变化检测方面的应用,为长期全面了解监测区域的动态变化提供有效信息。针对上述问题,该文提出了一种基于改进的模糊C均值聚类(FCM)算法对地基雷达图像进行无监督变化检测,该方法首次利用相干系数图和均值对数比值图进行非下采样轮廓波变换(NSCT)和局部能量法得到合成差异图,然后利用主成分分析(PCA)提取合成差异图中每个像素的特征向量,根据地基雷达图像特点对FCM进行改进,通过改进的FCM对每个像素的特征向量进行聚类得到最终的变化检测结果。利用地基雷达LSA对中国西南某省出现的堰塞体的治理过程进行监测,获取监测区域的地基雷达图像,监测过程中受降水等影响监测体出现滑坡,使用该文方法对其进行变化检测,结果表明该文方法更容易进行聚类分割,变化检测结果在保留变化区域的同时噪声点明显减少。

Unsupervised Change Detection Using Ground-based Radar Image

Ground-based radar is a microwave remote sensing imaging technology that has been gradually developed throughout the past 20 years so that it has become mature. At present, it has been widely used in monitoring geological disasters such as landslides and collapses. Ground-based radars can detect micro-variations in target areas through the principle of interferometry. However, due to human factors, geological factors, and meteorological factors, the radar image of the monitored area is incoherent, which makes long-term quantitative monitoring difficult. Therefore, further developing the application of change detection while considering quantitative monitoring is urgent, to provide effective information on long-term changes and comprehensively understand the dynamic changes in the monitored area. To solve the above problems, an unsupervised change detection method using ground-based radar images and based on an improved Fuzzy C-Means clustering (FCM) algorithm is proposed in this paper. In this method, for the first time, the Nonsubsampled Contourlet Transform (NSCT) is performed on the coherence coefficient map and the mean log ratio map to obtain the fusion difference map. Then, principal component analysis is used to extract the feature vectors of each pixel in the fusion difference image. The FCM is improved according to the characteristics of the ground-based radar images. The improved FCM is used to cluster the feature vectors of each pixel to obtain the change detection result. A ground-based radar LSA was used to monitor the treatment process of a dam in southwest China. During the monitoring process, landslides occurred in the monitored area affected by precipitation and other factors. This method is used to detect the change of the radar image before and after the landslide. The results show that the proposed method allows for easier clustering and segmenting, and the change detection results can significantly reduce the noise points while retaining the change area.