Underwater low-speed small target trajectory feature selection using support vector data description and recursive feature elimination

For underwater low-speed small targets, the poor performance of trajectory feature and information redundancy lead to the degradation of the classification and recognition performance. This paper proposes an improved trajectory feature selection method based on the improved support vector data description (ISVDD) and the recursive feature elimination (RFE), abbreviated as ISVDD-RFE in abbreviation. First, to solve the problem of small sample size and class imbalance in small target classification, the ISVDD is employed, with trajectory features selected through step-by-step recursive elimination. Then, to enhance the trajectory feature selection capability of SVDD-RFE, improvements are made to the recursion process from three aspects: recursion efficiency, feature correlation, and robustness. Finally, to overcome the inherent limitation of SVDD in lacking overall information, the selected trajectory features are evaluated from the perspective of feature distinguishability and feature situation, which improves the overall classification performance. Experimental results show that the proposed method improves the precision of frogman targets from 93.8% to 94.9%, and the recall from 84.7% to 91.1%. For unmanned underwater vehicle (UUV) targets, the precision increases from 89.0% to 94.7%, and the recall rises from 83.1% to 85.2%. The average classification accuracy of small targets is improved from 87.7% to 91.5%. Under the small sample size and class imbalance, the proposed method outperforms traditional methods.