高分辨率遥感影像的自优化迭代分类方法

发布时间：2018-05-05 19:59

本文选题：高分辨率遥感影像 + 像斑　；参考：《国防科技大学学报》2017年04期

【摘要】：针对高分辨率遥感影像提出了一种面向像斑的自优化迭代分类算法,基于半监督聚类算法获取训练样本,以支持向量机为核心设计了自优化迭代分类器。使用分型网络演化算法获取像斑,并从中选取少量标记样本;结合标记样本,利用半监督模糊C均值算法对像斑进行聚类,并基于密集度筛选得到训练样本;设计了自优化迭代支持向量机分类算法,对所有像斑进行迭代分类直到满足分类要求,并在分类过程中对近邻分类结果进行统计得到高可信度样本以自主优化训练样本集。基于以上方法分别对武汉市Quick Bird和World View影像进行分类实验,分类总精度分别达到94.67%与92%,与基于人工选取训练样本情况下进行分类的分类总精度(82%与82.67%)、常规支持向量机分类总精度(87.33%与88%)、最小二乘支持向量机分类总精度(88%与89.33%)相比,精度有明显提升,分类效果较好。
[Abstract]:Based on semi-supervised clustering algorithm, a self-optimized iterative classification algorithm for high-resolution remote sensing images is proposed. A self-optimized iterative classifier is designed based on support vector machine (SVM). Classification network evolution algorithm is used to obtain image spots, and a small number of labeled samples are selected from them. Combined with label samples, semi-supervised fuzzy C-means algorithm is used to cluster image spots, and training samples are obtained based on density screening. A self-optimization iterative support vector machine classification algorithm is designed to classify all image spots until the classification requirements are satisfied. In the process of classification, the close neighbor classification results are statistically analyzed to obtain highly reliable samples to independently optimize the training sample set. Based on the above methods, the classification experiments of Quick Bird and World View images in Wuhan are carried out respectively. The total accuracy of classification reached 94.67% and 92% respectively, compared with 82% and 82.67 of classification based on artificial training samples, 87.33% and 88% of the total accuracy of classification by conventional support vector machine, and 88% and 89.33 sections of classification by least squares support vector machine, respectively. The accuracy is obviously improved and the classification effect is better.
【作者单位】：武汉大学遥感信息工程学院;
【基金】：国家科技支撑计划资助项目(2014BAL05B07) 高等学校博士学科点专项科研基金资助项目(20130141130003) 测绘遥感信息工程国家重点实验室开放基金资助项目(13R04)
【分类号】：TP751

【相似文献】