基於L-模糊測度之Choquet積分於多重辨識器融合之理論與應用

The Theorems and Applications of Combining the Multiple Classification Systems Based on Choquet Integral with Respect to L-Measure

林文質(Lin Wen-Chih)

指導教授 ： 劉湘川

亞洲大學/資訊學院/資訊工程學系碩士班/博士(2010年)

摘要

在進行分類問題上時，支撐向量機辨識器在不同領域上解決問題的能力一直表現的很傑出與穩健。然而在進行分類過程時，往往單一辨識器在訓練過程中常常會發生無法選取適當的初始訓練點或是不良的搜尋方式，最後陷入局部區域解。多重辨識器系統則是為了改進單一辨識器的缺點而衍生而出的，其優勢是在於擷取單一辨識器之間有用資訊，進而改進單一辨識器在辨識效能上的不足。在進行多重辨識器最後決策時，其結合或是融合的方式也將會大大的影響最後的辨識效能。因此本研究將利用支撐向量機辨識器建立出多重辨識器，在假設辨識器之間有關聯性下，利用基於L模糊測度之Choquet模糊積分融合演算法減低辨識器之間關連性的影響，並在模糊積分融合演算法下提出較靈活的模糊密度函數。由實驗結果可以發現基於L模糊測度之Choquet模糊積分融合演算法搭配提出的模糊密度函數能得到較佳的辨識效能。

關鍵字

支撐向量機辨識器糊 ； 多重辨識器系統 ； Choquet模糊積分 ； L模測度

並列摘要

In classification issues, support vector machine (SVM) has an excellent ability to solve the problems. However, the single classifier often gets in the local solution, when the classifier can’t build by a suitable training set or has the poor statistical estimation in training process. Multiple classification system is proposed to overcome these problems from the single classifier. The advantage of multiple classification system is it can gain the more effective classification information from each single classifier, and improve the classification performance via this information. The final step of multiple classification system is the combination or fusion of multiple classifiers, and chooses a proper method to combine or fuse the multiple classification system to make the final decision is the most important task. In this thesis, we try to build a multiple classification system via SVM classifiers, and assume that there are correlations between these classifiers, and applying the Choquet fuzzy integral fusion algorithm with respect to L-measure with a more sensitive fuzzy density we proposed to decrease the influences of the interaction between the classifiers. Experiment results show the Choquet fuzzy integral algorithm with respect to L-measure with the fuzzy density we proposed obtains the advancement in terms of the performance of classification.

並列關鍵字

support vector machine ； multiple classification system ； Choquet fuzzy integral ； L-measure

參考文獻

劉湘川，“基於P測度之改進模糊測度及其模糊積分”，測驗統計年刊，第十四輯上期，2006年，頁1-15。

A. P. Dempster (1967). Upper and lower probabilities induced by multi-valued mapping. Annals of Mathematical Statistics, 38, 325-339.

A. Verikas and A. Lipnickas (2002). Fusing Neural Networks Through Space Partitioning and Fuzzy Integration. Neural Processing Letters, 16(1), 53-65.

B. Schölkopf, C. Burges, and V. Vapnik (1995). Extracting support data for a given task. Proceedings on First International Conference on Knowledge Discovery & Data Mining, AAAI Press, Menlo Park, CA.

B.-C. Kuo, C.-H. Pai, T.-W. Sheu, and G.-S. Chen (2004). Hyperspectral Data Classification using Classifier Overproduction and Fusion Strategies. Proceedings of IEEE International Geoscience and Remote Sensing Symposium, 5, 2937-2940.