Comparison of four methods for deciding objective weights of features for classifying stored-grain insects based on extension theory

The extension theory can be used to solve the problem of recognition of the stored-grain insects that have many different features with overlapping attributes. It is necessary to determine the objective weights of the features in the classification based on extension theory, avoiding the human factor from the traditional method of expert evaluation. Two methods for constructing the fuzzy decision matrix and the mean decision matrix were put forward. Four methods for deciding objective weights of the features, namely, the methods of maximizing deviations, standard variance between-class, criteria importance through inter-criteria correlation and the entropy, were used to classify the grain-stored insects. These four methods were applied to the two extension-theory classifiers for the stored-grain insects based on two methods for constructing standard and extensional matter-element matrices. The nine species of the stored-grain insects in the grain-storage bin were automatically recognized by the classifier based on the extension theory. The results showed that the maximizing deviation method for objective weights of features based on fuzzy decision matrix was the optimum scheme, and the correct identification ratio reached 93%.