As a significant study of sonar signal processing, underwater target recognition based on deep learning has become the research hotspot. Feature extraction is the pivotal step in target recognition. However, due to the complicated background noise, the extracted feature by the traditional algorithm has a poor ability to express data. In this paper, we propose a cascaded neural network for underwater acoustic target recognition via multimodal feature fusion. Firstly, we extract the Mel frequency cepstral coefficient (MFCC) and the Mel spectrogram feature of underwater acoustic signals to form the input for the network. Secondly, to prevent the network overfitting and improve the robustness of the fusion feature against noise, a set of label transformations (noise addition, waveform stretching, etc.) is applied to expand the original data. Finally, we integrate a long short-term memory (LSTM) network and a one-dimensional convolutional neural network (1D-CNN) as the cascaded neural network. Experimental results provide empirical evidence for the superior performance of our proposed model, as compared with the existing deep learning modelsl and single mode.