A new evidence classification algorithm for target recognition in underwater acoustic research

The Evidence k-nearest neighbor classification algorithm has been widely used in the field of noise source identification. In this traditional method, there are some problems in defining the weight of evidence function and the rule of combination. In order to effectively overcome these deficiencies and improve the recognition accuracy, a new evidence K-nearest neighbor recognition algorithm (NEK-NN) based on Dezert-Smarandache theory (DSmT) is presented in this paper. In this new method, the basic belief assignments (bba) are determined by using the feature similarity between the object and its K-nearest neighbors in each class of the training underwater acoustic targets, and then the K bba are discounted according to the distance of the K-nearest neighbors. Finally the discounted bba are combined by using DSmT rule, and the mean of these combined results in each training class is used for recognition of the object. Many tests were performed using experiments based on underwater acoustic data sets in order to verify the effectiveness of NEK-NN with respect to the other methods. The experimental results indicate that NEK-NN has effectively improved the recognition accuracy.