论文部分内容阅读
以新疆于田绿洲为研究区,利用四极化PALSAR-2数据进行多种目标极化分解处理,获取相应的极化特征参数。通过目视判读选取噪声较少的11种极化参数作为最佳特征信息对支持向量机分类法进行训练。多种极化分解方法与Wishart分类方法及支持向量机分类法相结合,提取研究区不同程度的盐渍化信息。经过目视判读和实地野外考察,结合Landsat-8陆地成像仪影像对分类结果进行定量分析和验证。由混淆矩阵的计算分析可知,相比Wishart分类方法,支持向量机分类法将分类精度从80.48%提高到88.00%,将Kappa系数从0.73提高到0.83。结果表明,单独的相干分解不能充分挖掘PALSAR-2数据包含的丰富信息,将目标极化分解参数用于特征信息分类处理,可以达到较好的分类效果;利用全极化PALSAR-2数据,结合目标极化分解方法和支持向量机分类法提取盐渍化信息有一定的优势。
Taking Yutian Oasis in Xinjiang as a study area, a variety of polarization decomposition methods were performed using the four-polarization PALSAR-2 data to obtain the corresponding polarization characteristic parameters. Eleven kinds of polarization parameters with less noise were selected as the best feature information through visual interpretation to train SVM classifier. A variety of polarization decomposition methods combined with the Wishart classification method and support vector machine classification method to extract different degrees of salinization information in the study area. After visual interpretation and field investigation, combined with Landsat-8 land imager images quantitative analysis and verification results. Computational analysis of the confusion matrix shows that SVM classification improves the classification accuracy from 80.48% to 88.00% and the Kappa coefficient from 0.73 to 0.83 compared with the Wishart classification method. The results show that the independent coherence decomposition can not fully exploit the abundant information contained in the PALSAR-2 data, and the target polarization decomposition parameters can be used to classify the feature information, thus achieving better classification results. Using the fully polarized PALSAR-2 data, The target polarization decomposition method and support vector machine classification method have some advantages in extracting salinization information.