正規化最小平方法為基礎的階層式合作共同進化演算法及其於模糊類神經網路設計和影像對準的應用

進化型演算法經常被使用在訓練模糊類神經網路參數方面，主要是因為該方法有並行搜尋的技術。不過目前此類型的方法有著無法拓展到多數量的訓練參數以及低效率的調整模糊法則問題，所以本篇論文提出了正規化最小平方法為基礎的階層式合作共同進化演算法來改善上述問題。使用正規化最小平方法的主要效用為減少訓練參數的數量，而在階層式合作共同進化演算法方面，兩層級進化法被提出能夠有效地進化模糊規則以及使得網路的參數及其架構能夠被分別被區域性及全域性的進化，因此以正規化最小平方法為基礎的階層式合作共同進化演算法有著參數學習及架構學習的優點，並且進化完成的網路可以被應用到現實世界的實例。第一個應用為二維影像對準問題，本論文所提出的演算法則可用來建立一個以合作式模糊類神經網路為基礎的二維影像對準系統，該系統利用多級模糊神經網路來解決單級模糊神經網路在仿射參數的大範圍應用的困難。第二個應用為三維影像對準問題，採用本論文所提的學習演算法可建立以模糊類神經網路為基礎的粗糙到細緻的三維影像對準系統，該系統改善傳統的主成份分析對準法的高粗糙對準誤差的缺點，在細緻對準階段成功改善了遞迴式最近點法的繁重計算的問題。這些證據可以被發現在實驗結果中來表示本論文提出的二維及三維影像對準系統，相較於其他一些典型的影像對準系統，本文的方法有較佳的性能。

關鍵字

正規化最小平方法；階層式合作進化型演算法；兩層級進化；參數學習；架構學習

並列摘要

Evolutionary algorithms are very popular in training parameters of neural fuzzy network due to their parallel search techniques. However, current methods have problems of not scaling well to a large number of training parameters and adjusting fuzzy rules inefficiently. In this dissertation, a regularized least squares based hierarchical cooperative coevolutionary algorithm (RGLS-HCCA) is proposed to improve above problems. The major utility of RGLS is to reduce the number of learning parameters. In HCCA, two-level evolution is proposed to evolve fuzzy rules efficiently and make the parameters and structure of a network be evolved locally and globally, respectively. Thus, RGLS-HCCA has advantages of parameter learning and structure learning, and the evolved network can be applied to the real world applications. The first application is a 2D image alignment problem. The proposed RGLS-HCCA is used to construct a cooperative neural fuzzy network (CNFN)-based 2D image alignment system, which utilizes the multi-stage of neural fuzzy networks to solve problems that one-stage of neural network have difficulty in applying a large range of affine parameters. The second application is a 3D image alignment problem. The use of RGLS-HCCA can construct a neural fuzzy network (NFN)-based coarse-to-fine 3D image alignment system, which solve the problem of the high alignment error caused by principle component analysis (PCA) and heavy computational cost caused by iterative closest point (ICP). The evidence can be found in experimental results demonstrate the superior performance of the proposed 2D and 3D surface alignment system over typical systems.

並列關鍵字

regularized least squares ； hierarchical cooperative coevolutionary algorithm ； two-level evolution ； parameter learning ； structure learning

參考文獻

[1] L. X. Wang and J. M. Mendel, “Generating fuzzy rules by learning from examples,” IEEE Trans. Systems Man Cybern., vol. 22, no. 6, pp. 1414-1427, 1992.

[3] C. J. Lin and C. T. Lin, “An ART-based fuzzy adaptive learning control network,” IEEE Trans. Fuzzy Systems, vol. 5, no. 4, pp. 477-496, 1997.

[4] T. Takagi and M. Sugeno, “Fuzzy identification of systems and its applications to modeling and control,” IEEE Trans. Systems Man Cybern., vol. 15, pp. 116-132, 1985.

[5] C. T. Lin and C. S. G. Lee, Neural Fuzzy Systems: A Neuro-Fuzzy Synergism to Intelligent System, Prentice-Hall, Englewood Cliffs, NJ, May 1996.

[6] C. F. Juang and C. T. Lin, “An on-line self-constructing neural fuzzy inference network and its applications,” IEEE Trans. Fuzzy Systems, vol. 6, no. 1, pp. 12-31, 1998.

國際替代計量

正規化最小平方法為基礎的階層式合作共同進化演算法及其於模糊類神經網路設計和影像對準的應用

全文下載

主題瀏覽