Experiments on Individual Classifiers and on Fusion of a Set of Classifiers

In the last decades many classification methods and fusers have been developed. Considerable gains have been achieved in the classification performance by fusing and combining different classifiers. We experiment a new method for ship infrared imagery recognition based on the fusion of individual results in order to obtain a more reliable decision [1]. To optimize the results of every class of ship, we implemented individual classifiers using Dempster-Shafer(DS) method for each class i.e. an individual classifier returns if the ship belongs to the class or not. We compare the result of the DS classifier with the results of the individual classifier. The improvement recognition varies between 3% to 20% for a class. We then experiment a new method based on a fusion of a set of classifiers [2]. The objective of a good fuser is to perform at least as good as the best classifier in any situation. For this purpose, we consider three classifiers: DS classifier, Bayes classifier and nearest neighbor classifier and one fuser: feedforward neural network fuser. We compare the results of the best classifier with the results of the fusion of a combination of classifiers. The fuser gives a performance equal or superior to the best classifier.