Feature Selection of Scale Target Recognition by Underwater Acoustic Homing Weapons Based on Random Forest

When underwater acoustic homing weapons identify underwater scale targets, it is necessary to extract different dimensional features from underwater target echoes and combine the features to form a complementary feature set to improve the recognition accuracy. However, due to the different application scenarios of different features, introducing all features leads to a high dimension of the feature set and may contain redundant information among each other, which will increase the difficulty of recognition. In the active recognition problem of underwater acoustic homing weapons, the feature set has a high dimension and needs to be selected. To solve these problems, a feature selection algorithm based on random forest(RF)was proposed in this paper. At the same time, to solve the problem of small amounts and unbalanced types of active echo data of underwater acoustic homing weapons, the synthetic minority oversampling technique was adopted in the feature domain. The feature subsets selected by the proposed method were put into the classifiers for testing by using real data. The results show that the proposed method can obtain better feature subsets and effectively improve the recognition accuracy.