東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

Development of soft classification a...

Li, Zhe.

FindBook

Google Book

Amazon

博客來

Development of soft classification algorithms for neural network models in the use of remotely sensed image classification.

紀錄類型:	書目-語言資料,印刷品 : Monograph/item
正題名/作者:	Development of soft classification algorithms for neural network models in the use of remotely sensed image classification./
作者:	Li, Zhe.
面頁冊數:	98 p.
附註:	Adviser: J. Ronald Eastman.
Contained By:	Dissertation Abstracts International68-09B.
標題:	Engineering, Environmental. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3282765
ISBN:	9780549254256

Development of soft classification algorithms for neural network models in the use of remotely sensed image classification.
Li, Zhe.

Development of soft classification algorithms for neural network models in the use of remotely sensed image classification. - 98 p.

Adviser: J. Ronald Eastman.

Thesis (Ph.D.)--Clark University, 2007.

In the classification of remotely sensed imagery, soft classifiers provide continuous statements of class membership for each class for the purpose of assessing classification uncertainty or sub-pixel membership. Conventional soft classifiers such as those based on Bayesian and Mahalanobis constructs are parametric and limited by assumptions about the form and distribution of input data. In recent years machine learning algorithms have emerged as effective nonparametric alternatives to conventional parametric algorithms when dealing with complex measurement spaces. Among machine learning techniques, there has been considerable interest in the use of neural networks for the classification of remotely sensed imagery due to their numerous advantages over conventional classifiers such as no assumptions about the form and distribution of input data, non-linear decision boundaries and the capabilities of generalizing inputs and learning of complex patterns. However, soft classification has only been established for the Multi-Layer Perceptron (MLP) neural network. Since Kohonen's Self-Organizing Map (SOM) and the fuzzy ARTMAP neural network show significant promise in the classification of remotely sensed imagery, this study is concerned with the development of non-linear and non-parametric soft classification procedures for these two neural networks. In this dissertation, distance-based (Inverse Minimum Mean Distance and Inverse Minimum Distance measures) and triggering/committing-frequency-based algorithms (Commitment and Typicality measures) were proposed for both the SOM and fuzzy ARTMAP models. It is shown that the Commitment measures are very similar in character to Bayesian posterior probabilities while the Typicality measures correspond closely with Mahalanobis typicality probabilities. However, as implemented, they have the advantage of being non-parametric.

ISBN: 9780549254256Subjects--Topical Terms:

783782
Engineering, Environmental.

Development of soft classification algorithms for neural network models in the use of remotely sensed image classification.
LDR:02817nam 2200277 a 45 001 940011
005 20110517
008 110517s2007 ||||||||||||||||| ||eng d
020 $a 9780549254256
035 $a (UMI)AAI3282765
035 $a AAI3282765
040 $a UMI $c UMI
100 1 $a Li, Zhe. $3 1264118
245 1 0 $a Development of soft classification algorithms for neural network models in the use of remotely sensed image classification.
300 $a 98 p.
500 $a Adviser: J. Ronald Eastman.
500 $a Source: Dissertation Abstracts International, Volume: 68-09, Section: B, page: 5828.
502 $a Thesis (Ph.D.)--Clark University, 2007.
520 $a In the classification of remotely sensed imagery, soft classifiers provide continuous statements of class membership for each class for the purpose of assessing classification uncertainty or sub-pixel membership. Conventional soft classifiers such as those based on Bayesian and Mahalanobis constructs are parametric and limited by assumptions about the form and distribution of input data. In recent years machine learning algorithms have emerged as effective nonparametric alternatives to conventional parametric algorithms when dealing with complex measurement spaces. Among machine learning techniques, there has been considerable interest in the use of neural networks for the classification of remotely sensed imagery due to their numerous advantages over conventional classifiers such as no assumptions about the form and distribution of input data, non-linear decision boundaries and the capabilities of generalizing inputs and learning of complex patterns. However, soft classification has only been established for the Multi-Layer Perceptron (MLP) neural network. Since Kohonen's Self-Organizing Map (SOM) and the fuzzy ARTMAP neural network show significant promise in the classification of remotely sensed imagery, this study is concerned with the development of non-linear and non-parametric soft classification procedures for these two neural networks. In this dissertation, distance-based (Inverse Minimum Mean Distance and Inverse Minimum Distance measures) and triggering/committing-frequency-based algorithms (Commitment and Typicality measures) were proposed for both the SOM and fuzzy ARTMAP models. It is shown that the Commitment measures are very similar in character to Bayesian posterior probabilities while the Typicality measures correspond closely with Mahalanobis typicality probabilities. However, as implemented, they have the advantage of being non-parametric.
590 $a School code: 0048.
650 4 $a Engineering, Environmental. $3 783782
650 4 $a Remote Sensing. $3 1018559
690 $a 0775
690 $a 0799
710 2 $a Clark University. $3 1017690
773 0 $t Dissertation Abstracts International $g 68-09B.
790 $a 0048
790 1 0 $a Eastman, J. Ronald, $e advisor
791 $a Ph.D.
792 $a 2007
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3282765