Pest Identification Based On Multiple Classifier System

Pest identification technology is crucial to the grain storage. Traditional identification methods with single classifier are not competent under the complicate environment. The multiple classifier system(MCS) is considered to be an effective method, but there should exist the diversity between member classifiers. In order to improve identification accuracy, the MCS based on diversity measure is built in this paper. Our MCS uses the BP and nearest neighbor( NN) classifiers on different features for generating different member classifiers. Among member classifiers, those with larger diversities are selected to construct the MCS. In the final, the voting method or the weighted average method is used for fusing the member classifiers outputs. Experiments show that classification performance of MCS performs well when dealing with the problem of pest identification.