平行粒子尋優演算法於高光譜影像特徵抽取應用

Band Selection for the Hyperspectral Images Based on Parallel Particle Swarm Optimization Schemes

陳佳輝(Jia-Hui Chen)

指導教授 ： 張陽郎 ； 方志鵬

國立臺北科技大學/電資學院/電機工程系所/碩士(2009年)

https://doi.org/10.6841/NTUT.2009.00165

摘要

隨著影像衛星技術的日新月異以及普及化，高光譜影像的資料複雜度也相對的越來越高，高維度的資料且龐大的資料量會帶來費時又複雜的計算問題，且特徵值的品質也左右著分類器的好壞，因此如何有效的萃取高相關度的特徵值、降低雜訊的影響度，並且提升運算效率是相當重要的。
　　本篇論文提出一套平行粒子尋優演算法(Parallel Particle Swarm Optimization, PPSO)，利用PSO具有的快速收歛特性應用在降低資料維度(Reduction Dimension, RD)，並結合利用空間轉換的觀念，使得PSO搜索空間能夠對應到降維的相關係數陣列(Correlation Coefficient Matrix)解空間，來解決空間型態不同的問題。在參數設定的部份，首先對於粒子尋優演算法裡自身搜尋參數採用線性遞減的方法，用來提升粒子本身在周圍區域搜尋的活動力；第二採用generation-delay的方式來調整PSO內部參數全域值取決率，讓每顆粒子在其範圍內做更完整的搜尋；最後為避免粒子陷入區域性搜尋停滯狀態，加入隨機變數以更新粒子位置與速度，使得PSO能更有效的萃取高相關度的特徵值。
除此之外，為了使PPSO可以有效的逼近最佳解，並且提升整體運算效率，本篇論文結合三種平行機制[11]—「非互動式平行」、「同步交換訊息平行」、「非同步交換訊息平行」。最終由實驗結果顯示，本篇論文提出之PPSO與PSA[12]作一比較，PPSO方法可以找到接近最佳解的機率比PSA高出90%以上。

關鍵字

高光譜(hyperspectral) ； 平行粒子尋優法(PPSO) ； 平行模擬退火法(PSA) ； 相關陣列(CM) ； 降維(RD) ； NIPPSO ； PEPPSO ； APPSO

並列摘要

Hyperspectral image data complexity was increasing with satellite technology improvement and popularization.High-dimensional data and a huge amount of data would be time-consuming and complex calculation.The Bands of the quality affected the accuracy from classification.

How to extract the bands of these high-impact was a very important ,as reduction dimenstion could enhance the operation efficiency and reduce the impact of noise.In this paper ,particle swarm optimization algorithm with fast convergence properties was Collective Intelligence algorithm to optimization problem in recent years.In this study, we proposed parallel particle swarm optimization used in reduction dimensions ,combined with the mapping form search space to problem space to solve different space.

PPSO could be an effective approach of optimal solution and use of generation-delay to determination of updata the global best within parameters of PPSO.So that each particles made complete search within its scope. Adding updata new position and speed of particles in order to avoid stagnation in regional search.Part of the parameter setting for own search used linear decrease.It used to enhance particles own search activity in the surrounding area.Divided into three parallel mechanisms were NIPPSO(non-interation parallel particle swarm optimization) , PEPPSO(periodic exchange parallel particle swarm optimization) and APPSO(asynchronous parallel particle swarm optimization).

In this paper, three parallel test method, non-simultaneous exchange of messages was better than the other two methods.The experimental results show that PPSO and PSA compared reduction dimension, PPSO methods are more effective dimensionality reduction effect.

並列關鍵字

hyperspectral ； parallel particle swarm optimization(PPSO) ； parallel simulated annealing (PSA) ； correlation matrix(CM) ； NIPPSO(non-interation parallel particle swarm optimization) ； PEPPSO(periodic exchange parallel particle swarm optimization) ； APPSO(asynchronous parallel particle swarm optimization)

參考文獻

[13] 劉進男，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，碩士論文，國立台北科技大學，台北，民國96年。

[1] C. Lee and D. A. Landgrebe, “Analyzing high-dimensional multispectral data” IEEE Trans. Geosci. Remote Sensing 31, No. 4, 1993, pp. 792–800.

[2] Yang-Lang Chang, Chin-Chuan Han, Kuo-Chin Fan, K.S. Chen, Chia-Tang Chen and Jeng-Horng Chang, "Greedy Modular Eigenspaces and Positive Boolean Function for Supervised Hyperspectral Image Classification" Optical Engineering, Vol. 42, Issue 9, September 2003, pp. 2576-2587.

[3] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, “Optimization by Simulated Annealing” Science 220, No. 4598, 1983, pp. 671-680.

[4] D. L. Hall and J. Llinas, “An introduction to multisensor data fusion” Proc. IEEE, Vol. 85, Jan. 1997, pp. 6–23.