Abstract: The classification of underwater small targets such as frogmen and unmanned underwater vehicles (UUVs) is a typical small-sample-size and class-imbalanced problem. To solve this problem, a joint classification method is proposed which uses the tracking trajectory features, the support vector data description (SVDD) and the support vector machine (SVM). The proposed method transforms the underwater small target recognition problem to be a classification one, where only three classes including “frogman”, “UUV” and “others” are considered. It makes use of the feature differences produced by the moving processes of different targets, and design a joint classifier called SVDD-SVM to realize classification. Specifically, to solve the small-sample-size problem, it chooses SVDD and SVM having a small number of parameters and requiring small-size training data as the basic unit of the classifier. To solve the class-imbalanced problem, it uses two one-class SVDDs and one two-class SVM to construct SVDD-SVM. The two SVDDs are placed in parallel, and the SVM is placed right after them. Meanwhile, a voting mechanism is designed to obtain the final outputted results of SVDD-SVM. A small-size set of real data is used to verify the classification performance of the proposed SVDD-SVM. The proposed SVDD-SVM method has an averaged recall rate of 86% and an averaged precision rate of 87% for frogmen, an averaged recall rate of 85% and an averaged precision rate of 86% for UUVs, showing a higher accuracy and more robust performance than the traditional methods considered in this paper.