一個對於高光譜影像資料融合平行的模擬退火波段選擇及特徵抽取方法

A Parallel Simulated Annealing Approach to Band Selection and Feature Extraction for the Data Fusion of Hyperspectral and SAR Imageries

劉進男(Jin-Nan Liu)

指導教授 ： 張陽郎 ； 方志鵬

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2008年)

https://doi.org/10.6841/NTUT.2008.00367

摘要

衛星遙測影像可以大範圍的顯示各種地形的區域概況，藉以了解在各種不同地形地貌的分布情形，如蔗田、海洋、稻田、水塘等。近來衛星感測器持續不斷的更新開發，造成可利用的光譜資訊日益龐大，而在這些龐大的光譜資訊中所包含的雜訊或是錯誤資訊的光譜不在少數，因此如何有效率的挑選正確而有效的光譜資訊進行分類，成為一門重要的課題。
本論文提出一個對於高光譜影像資料融合的平行的模擬退火波段選擇及特徵抽取技術，有效率地對進行高光譜進行特徵抽取再進行分類。本論文應用的方法分為三部份:『平行模擬退火法』(Parallel Simulated Annealing, PSA)、『群集特徵空間/特徵尺度齊一化轉換』(Clustered Eigenspace / Feature scale Uniformity Transformation, CE/FSUT)、『平行的布林函數分類器』（Parallel Positive Boolean Function, PPBF）。利用『平行模擬退火法』、『群集特徵空間/特徵尺度齊一化轉換』將高光譜及不同感測器(譬如合成孔徑雷達- SAR, synthetic aperture radar)所收集的遙測影像，做資料融合(Data Fusion)的群組波段選取，再藉著群集高相關度的高光譜及SAR資料融合波段，將之形成一個波段模組的子集合，最後再採用『平行的布林函數分類器』，針對各測試樣本進行分類動作，最後達到最佳的類別組合。
由實驗數據結果顯示，本論文提出以平行模擬退火法波段選取及特徵抽取方法針對平行的PBF分類器，有效率的挑選出高度相關的高光譜波段，進而提升整體的運算效率。

關鍵字

平行布林函數分類器 ； 平行模擬退火法 ； 模擬退火波段選擇 ； 群集特徵空間/特徵尺度齊一化轉換 ； 資料融合

並列摘要

Satellite remote sensing images can interpret all kinds of large-scale terrain to understand the different topography of the distribution, such as sugarcane, oceans, paddy fields, reservoirs, and so on. Recent advances of satellite sensors technologies continue to update the development resulting in the increase of large spectral information available. The noises contained in these huge information can’t avoid the curse of dimensionality., As a result, how to efficiently select the right and effective spectral information has become important.

This paper presents an alternative promising concept, known as the parallel simulated annealing band selection (PSABS), which adopts a novel parallel approach to the data fusion of remote sensing images of the same scene collected from multiple sources. The applications can be divided into three parts: 1.) a parallel simulated annealing (PSA), 2.) a clustered eigenspace / feature scale uniformity transformation (CE/FSUT), and a parallel positive Boolean function (PPBF). PSA and CE/FSUT are used to select the high-dimensional fused datasets, and cluster the highly related information to a set of modular subspaces. Finally, a PPBF classifer is then applied to these selected band modules to execute the classification.

The effectiveness of the proposed PSABS is evaluated by MODIS/ASTER airborne simulator (MASTER) hyperspectral and SAR images for hyperspectral band selection. The experimental results demonstrated that PSABS can significantly improve the computational loads and provide a more reliable quality of solution compared to the traditional methods.

並列關鍵字

Parallel parallel Positive positive Boolean Function function (PPBF) ； Parallel parallel Ssimulated annealing (PSA) ； clustered eigenspace / feature scale uniformity transformation (CE/FSUT) ； simulated annealing band selection (SABS) ； data fusion

參考文獻

[15] 林鈞傑， 一個對於高光譜影像的二維模擬退火波段選擇方法，碩士論文，國立台北科技大學，台北，2007年6月。

[16] 吳至偉， 一個新的演算法改善中文簽名辨識，碩士論文，國立台北科技大學，台北，2007年7月。

[8] Y. C. Chang, Y.-W. Chang, G.-M. Wu, and S.-W. Wu, "B*-Tree: A New Representation for Non-Slicing Floorplans," Proc. Design Automation Conf., Los Angeles, CA., USA, June 2000, pp. 458-463.

[4] Yang-Lang Chang, Jyh-Perng Fang, Hsuan Ren, Wen-Yew Liang, and Jin-Nan Liu, "A Simulated Annealing Feature Extraction Approach for Hyperspectral Images," Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, IGARSS '07, Barcelona, Spain ,July 23-27,2007, NSC 95-2221-E-027-037, (EI).

[1] Michalewicz, Z., Genetic Algorithms + Data Structures = Evolution Programs,