透過您的圖書館登入
IP:44.200.101.170
  • 學位論文

以GPU實現最鄰近特徵內切圓演算法應用於高光譜影像分類

GPU Acceleration of Incenter-Based Nearest Feature Space Approach to Hyperspectral Image Classification

指導教授 : 張陽郎
共同指導教授 : 方志鵬

摘要


近年「最鄰近特徵空間分類演算法」(Nearest Feature Space, NFS) 被多方應用於以多源遙測資料為目的的分類問題上,通過計算測試樣本到特徵面(Feature Surface , FS)的最短距離來進行分類,相較於其他最鄰近特徵方法,特徵面包含較多的類別資訊,可提高分類正確率,但NFS在類別樣本分布過於接近或交疊時,容易分類錯誤,所以本論文提出一個新的方法,以特徵面為基礎,引入三角形內心(Incenters)概念,稱之為「最鄰近特徵內切圓演算法」(Incenter-based Nearest Feature Space, INFS),利用計算特徵面內心與測試樣本的最短距離來進行分類。因內心只須由同特徵面的三個訓練樣本共同計算而得,可降低單一訓練樣本過於接近其他類別分布所帶給特徵面的影響,本論文最後由實驗結果證明當類別樣本分布交疊時INFS可獲得比NFS更高的分類正確率,且因INFS內心的計算只考慮特徵面的訓練樣本,並不會受測試樣本影響,與NFS相比可大幅減少運算時間完成分類過程。 由於資料量龐大,配合INFS演算法,以一般電腦運算相當耗時,因此,本論文以實現 INFS平行化為主軸,透過高速計算圖形處理器(Graphic Process Unit, GPU)所提供之統一計算架構(Compute Unified Device Architecture, CUDA),實現訓練樣本為基底的平行化,在NTC圖資實驗下,GPU可達到38倍的加速。

並列摘要


In this paper a novel technique based on nearest feature space (NFS), called as incenter-based nearest feature space (INFS), are proposed for supervised hyperspectral image classification. Although NFS has high classification accuracy, multiple overlapping training samples might cause classification errors. To overcome this problem, we borrow the concept from the incircle of a triangle, which is tangent to its three sides and can form a nearest feature space (NFS). In addition, an incenter can only be calculated by the same class of three training samples, which can largely reduce the impact from feature space of single training sample. Furthermore, in order to further speed up the computation performance, this paper proposes a parallel computing version of INFS, namely parallel INFS (PINFS), using a modern graphics processing unit (GPU) architecture with NVIDIA’s compute unified device architecture (CUDA) technology to improve the computational speed of PINFS processes. Experimental results demonstrate the proposed INFS approach is suitable for land cover classification in earth remote sensing. It can achieve the better performance than NFS classifier when the class sample distribution overlaps. Through the computation of GPU by CUDA, we can also gain better speedup.

參考文獻


[15] 王旭,基於GPU之粒子群優法應用於高光譜影像波段選取,碩士論文,國立臺北科技大學電機工程研究所,臺北,2012。
[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,國立臺北科技大學電機工程研究所,臺北,2011。
[14] 傅義翔,以GPU實現最鄰近特徵向量空間演算法應用於高光譜影像分類,碩士論文,國立臺北科技大學電機工程研究所,臺北,2012。
[1] Richard O. Duda, Peter E. Hart and David G. Stork, Pattern Classification, 2nd Ed. Wiley-Interscience, pp. 114-116, 119-124, 2001.
[6] T. Hastie, R. Tibshirani, "Discriminant Adaptive Nearest Neighbor

延伸閱讀