Fuzzy Fusion Method for Combining Small Number of Classifiers in Hyperspectral Image Classification

For hyperspectral image classification problem, the random subspace method has been shown that is a good approach to overcome the small sample problem, and the machinery of it is to randomly select a batch of subspaces to train different classifiers and then get the final decision by using the majority vote method. Theoretically, more classifiers we train, more stable and more accurate result we obtain. However, it shows the bad outcome when using small number of classifiers. In this paper, a fuzzy measure has been applied into the fusion process as a new evaluation to combine classifiers to try to improve the performance in the situation of less classifier. From the experiment results, it displays that this fuzzy measure has effectively progressed in the classification accuracy.