基于局部竞争策略的极化SAR图像精细分类

Refined Polarimetric SAR Image Classification Based on Localized Competition

合成孔径雷达(Synthetic Aperture Radar,SAR)成像技术已经成为一种高分辨对地观测的重要手段之一,而极化SAR图像地物分类一直是其中的研究热点。基于复Wishart分布的最大似然(Maximum Likelihood,ML)分类器是最经典的极化SAR图像分类算法之一,但由于地物类型的复杂性、区域的不均匀性等原因使得基于像素的ML-Wishart分类器的分类精度不高。针对这个问题,本文提出了一种基于复Wishart分布的局部最大后验概率(Maximum a Posteriori,MAP)竞争方法,该算法通过计算伪先验概率,并在每个像素的局部窗口中实施MAP分类器,可以提高复杂区域图像的分类精度。该文主要研究了4种基于Wishart分布的分类算法,包括经典复Wishart分类算法、混合复Wishart模型、基于马尔科夫随机场(Markov Random Field,MRF)的混合复Wishart模型和基于局部竞争策略的MAP分类算法。在混合模型建模中,不同于以往的对整幅图像进行建模的模型策略,本文采用对单个类别进行混合建模的策略。实验对比分析了上述4个分类器和SVM分类器在C波段RADARSAT-2多时相的全极化SAR农田数据上的分类效果。实验结果表明,所提出的基于局部竞争策略的分类器对数据的分类结果稳定,具有最高的分类精度,基于混合Wishart的MRF模型分类结果次之。

Polarimetric synthetic aperture radar(SAR)imaging technology has become one of the important means of high-resolution ground observation.Land cover classification for polarimetric SAR image has always been a hot research topic.The complex Wishart distribution-based maximum likelihood(ML)classifier is one of the most classic polarimetric SAR image classification algorithms.However,due to the complexity of feature types,regional heterogeneity,etc.,the classification accuracy of the pixel-based Wishart classifier is not high.To solve this problem,this paper proposes a local maximum a posteriori(MAP)competition method based on the complex Wishart distribution,which can improve the classification accuracy for images with complex terrains by calculating pseudo-prior probabilities and then implementing MAP classifier in a local window of each pixel.This paper mainly studies 4 classification algorithms based on the Wishart distribution,including the classic complex Wishart classification algorithm,mixed complex Wishart model,Markov random field(MRF)-based mixed complex Wishart model and local competitive Wishart classification algorithm.For mixed model modeling,different from the previous modeling strategy that models the whole image,this paper adopts the strategy to model a single class with a mixture of Wishart distributions.The experiments were conducted to compare and analyze the classification results of the above-mentioned four classifiers and the SVM classifier on C-band RADARSAT-2 multi-temporal fully polarimetric SAR data collected over farmlands.Experimental results show that the proposed classifier based on local competition strategy has stable and superior classification performances for the multi-temporal data sets over the other four methods.