基於GPU之粒子群優法應用於高光譜影像波段選取

Particle Swarm Optimization for Hyperspectral Band Selection Using GPU

王旭(Hsu Wang)

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(2012年)

https://doi.org/10.6841/NTUT.2012.00616

摘要

隨著衛星遙測技術近年來廣泛的發展，高光譜影像的波段數與資料量也越趨龐大，使計算複雜度大幅提高，波段中也可能包含雜訊或是錯誤的資訊導致分類正確率降低。因此，在高光譜影像處理中，進行波段選取降低資料複雜度並萃取具代表性的波段，是不可或缺的一個步驟。
過去已有學者以粒子群優法(Particle Swarm Optimization, PSO)進行高光譜影像的波段選取。利用PSO演算法將高光譜影像的相關係數矩陣(Correlation Coefficient Matrix)聚合成一組群聚模組特徵空間，挑選出具代表性的波段，達到降維(Reduction Dimension, RD)的效果。然而處理波段數較多的高光譜影像時，依然需耗費大量時間。因此本論文應用CUDA（Compute Unified Device Architecture）技術實現平行架構的粒子群優法，利用圖形處理器(Graphics Processing Units, GPU)進行加速，更進一步提升高光譜影像波段選取的整體運算速度。
本文採用鰲鼓溼地的 MASTER 遙測影像以及 Northwest Tippecanoe County 的AVIRIS 遙測影像為實驗圖資。最後由實驗結果可以得知，本文所提出的平行粒子群優法波段選取能夠迅速、有效地挑選出有價值的波段，並透過分類器得到良好的分類效果。

關鍵字

粒子群優法 ； 波段選取 ； 相關係數矩陣 ； 降維 ； CUDA

並列摘要

In recent years, the satellite image technologies have greatly advanced remote sensing community, resulting in the increased number of bands acquired by hyperspectral sensors. The band selection of hyperspectral imagery can reduce the dimensions which can avoid the Hughes phenomena. Therefore, the band selection of hyperspectral imagery has become very important. A band selection algorithm based on particle swarm optimization (PSO) is proposed in this paper. By using the PSO algorithm, the highly correlated bands of hyperspectral imagery can be grouped into the same modules which can extract the most useful information of hyperspectral bands in each module and can further reduce the dimensionality. However the PSO band selection is a time-consuming procedure when the number of hyperspectral bands is huge. Consequently this paper proposes a parallel PSO (PPSO) band selection based on modern graphics processing unit (GPU) architecture using NVIDIA compute unified device architecture (CUDA) technology. It can improve the computational speed of PSO band selection processes. The natural parallelism of proposed PPSO is in the face that each particle can be regarded as an independent agent. Parallel computation benefits the algorithm by providing each agent with one of the parallel processors. The intrinsic parallel characteristics embedded in PPSO can be therefore suitable for a parallel implementation. The effectiveness of the proposed PPSO is evaluated by AVIRIS hyperspectral images. The experimental results demonstrated that the proposed PPSO band selection not only can improve the computational speed but also can offer a satisfactory classification performance.

並列關鍵字

Particle Swarm Optimization (PSO), ； band selection ； correlation coefficient matrix ； dimension reduction ； Compute Unified Device Architecture (CUDA)

參考文獻

[13] 徐斌峰，一個維度優先權方法應用於粒子群優法在高維度影像特徵抽取，碩士論文，國立台北科技大學電機工程研究所，台北，2011。

[2] G. F. Hughes, "On the mean accuracy of statistical pattern recognizers, " IEEE Transactions on Information Theory, vol. 14 no. 1, 1968, pp. 55–63.

[3] 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, no. 9, 2003, pp. 2576-2587.

[4] Yang-LangChang, Chin-Chuan Han, Bormin Huang, Wen-Yen Chang, Benediktsson, J.A.,Lena Chang, "A Parallel Simulated Annealing Approach to Band Selection for High-Dimensional Remote Sensing Images," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Vol. 4, 2011, pp. 579 - 590 .

[6] WeiPang, Kang-Ping Wang, Chun-Guang Zhou, "Modified Particle Swarm Optimization Based On Space Transformation For Solving Traveling Salesman Problem" 2004, pp. 2342-2346.