Scholarlink

<p>A novel, very fast parallel clustering algorithm is proposed based on a combination of common clustering methods. First, the input images are divided into s equal-sized slices. Each slice is segmented into c clusters using a clustering method to obtain the cluster centers. The collected set of these centers is then grouped into c clusters, from which c new main centers are computed using an arbitrary clustering method. Next, the original data are hard-clustered using these c centers with a correlation-based distance metric. Specifically, we combined the Fuzzy C-Means algorithm and the K-Means algorithm to produce a hybrid classification method. The proposed hybrid method was evaluated on real hyperspectral datasets, including Samsun, Jasper Ridge, Urban, and Pavia. Execution time, overall accuracy, and the kappa coefficient improved significantly compared with Fuzzy C-Means, General Interval Type-2 Fuzzy C-Means, and K-Means. For example, this method reduced the classification time for the Jasper dataset to one-seventh of that required by the compared methods, representing&#xa0;a substantial improvement.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sliced Clustering: A Very Fast Parallel Clustering Algorithm for Remote Sensing Imaging

  • 期刊论文

  • 发表日期：

    2026年5月29日

查看全文
  • Mansoor Zeinali

摘要

A novel, very fast parallel clustering algorithm is proposed based on a combination of common clustering methods. First, the input images are divided into s equal-sized slices. Each slice is segmented into c clusters using a clustering method to obtain the cluster centers. The collected set of these centers is then grouped into c clusters, from which c new main centers are computed using an arbitrary clustering method. Next, the original data are hard-clustered using these c centers with a correlation-based distance metric. Specifically, we combined the Fuzzy C-Means algorithm and the K-Means algorithm to produce a hybrid classification method. The proposed hybrid method was evaluated on real hyperspectral datasets, including Samsun, Jasper Ridge, Urban, and Pavia. Execution time, overall accuracy, and the kappa coefficient improved significantly compared with Fuzzy C-Means, General Interval Type-2 Fuzzy C-Means, and K-Means. For example, this method reduced the classification time for the Jasper dataset to one-seventh of that required by the compared methods, representing a substantial improvement.