Underwater acoustic object recognition is becoming attractive given the critical information available. However, this comes at the expense of large-scale annotated data, which is expensive to collect and annotate. This paper proposes a semi-supervised learning approach of SE_RseNet_Decoder to recognizing insufficient sample underwater acoustic targets. Given this goal, we introduce the SE_RseNet_Decoder network containing supervised and unsupervised modules. Firstly, we leverage the supervised module to recognize the labeled signals and reduce the dimensional feature extraction of unlabeled samples. Then, the unsupervised network is designed as an auxiliary network to optimize the supervised network, which uses low-dimensional features to restore high-dimensional features of unlabeled samples to enhance the classification ability of the supervised network. We especially introduce ReLU activation function to connect the supervised and unsupervised modules that can help find a balanced relationship between classification and regression tasks for recognizing underwater acoustic signals. Extensive experiments on multiple benchmark datasets demonstrate the superiority of our framework showing that the proposed approach achieves the best recognition accuracy compared with the other approaches with few samples. Moreover, the experimental results can demonstrate the optimal combination of variables for the recognition effect of the proposed method under multiple variables.