- 作者: 杜德銘; 陳進興
- 作者服務機構: 國立中正理工學院電機工程學系; 國立成功大學電機工程學系
- 中文摘要: 在本文中,我們針對超高維頻譜影像的像元級分類,提出一個比傳統式最大相似分類器快數十至數百倍的兩級式分類方法。首先在第一級,我們將負荷因子的觀念融入正規分析技術中,發展出一種有效的頻帶選取方法,它可以計量出每一頻帶的分類能力,讓我們剔除有資訊贅餘的頻帶而保留對分類有用的頻帶。接著在第二級,我們設計了一個運用Winograd法則的遞迴式最大相似分類器,藉以減少分類的時問。
- 英文摘要: Classification for high dimensional remote sensing data generally requires a large set of data samplesand enormous processing time, especially for hyperspectral image data. In this paper, a fast classificationscheme is presented. The first stage of process is to develop a strategy for band selection which is designedbased on the canonical analysis (CA) and the concept of loading factors to weigh bands in accordancewith their energies.The suggested band selection algorithm allows one to predetermine which bands willbe used for data processing so that data dimensionality is greatly reduced. It is then followed by a secondstage using a maximum likelihood (ML) classifier which is recursive and designed based on Winograd'salgorithm to achieve computational efficiency. The experimental results show that the proposed fastclassification scheme reduces the computing time by a factor of 27 to 107 compared to the conventionalone-stage ML classifier.
- 中文關鍵字: canonical analysis; band selection; recursive maximum likelihood classifier; winograd's identity
- 英文關鍵字: --