基于面向对象的极化雷达影像分类

有效的PolSAR影像分类技术是PolSAR成功应用的基础,然而相比于比较成熟的PolSAR成像技术与系统设计,PolSAR影像分类技术的发展相对滞后,针对PolSAR影像面向对象分类研究中存在的问题,提出了一种新的结合多种目标极化分解、ReliefF-PSO_SVM和集成学习的PolSAR影像面向对象分类方法。该方法首先采用多种方法对PolSAR影像进行目标极化分解;然后将利用不同极化分解方法提取的极化参数组合成一幅多通道影像;接下来对多通道影像进行分割、特征提取;采用ReliefF-PSO_SVM算法进行特征选择,并保留适应度最高的N个特征子集进行分类,每一个特征子集对应一个分类结果;最后利用集成学习技术对各分类结果进行集成。以吉林省长春市部分区域为研究区,Radarsat2影像为数据源,将提出的方法应用于土地利用分类中,取得了较好的分类效果,总体精度和Kappa系数分别达到了85.06%和0.8006。此外,还构建了3种对比方法用于分类,对比结果进一步证明了所提方法在PolSAR影像分类中的优越性。

PolSAR image classification based on object-oriented technology

An effective polarimetric synthetic aperture radar (PolSAR) image classification technology is the basis of the successful application of PolSAR. However, compared with relatively mature PolSAR imaging technology and system design, PolSAR image classification technology lags behind. Aiming at the main problems existing in the research of object-oriented classification of PolSAR images, this paper proposed a new object-oriented classification method, which combines multi-target polarimetric decomposition, ReliefF-PSO_SVM and ensemble learning. First, polarimetric decomposition is implemented for PolSAR image using various methods. Polarimetric parameters extracted using different polarimetric decomposition methods are combined into a multichannel image. Second, the multichannel image is divided into numerous image objects by implementing multi-resolution segmentation. Third, features are extracted from the multichannel image. Fourth, ReliefF-PSO_SVM algorithm is applied for feature selection, and N feature subsets with the highest fitness are retained for classification. Each feature subset corresponds to a classification result. Finally, ensemble learning technology is used to integrate the classification results. The study site is located at the southeastern part of Changchun City, Jilin Province. A RADARSAT-2 Fine Quad-Pol image was selected as the data source for this study. The proposed method was applied to land-use classification, and good classification results were obtained. The overall accuracy was 85.06% and the kappa value was 0.8006. In addition, three other classification methods were performed for comparison. The comparison results further proved the superiority of the proposed method in PolSAR image classification.