Abstract: A feature attention enhancement network (FEAN) is proposed for underwater acoustic classification. Based on the line spectrum information and modulation information of radiated noise signals from underwater acoustic targets, the FEAN employs a learnable attention module that performs linear adaptive filtering on features according to the classification task, retaining information relevant to the classification task while filtering out interference. Additionally, it optimizes time-frequency spectrogram features by utilizing the coherence of line spectrum information in the time domain and the correlation of modulation information in the frequency domain. The experimental results on the DeepShip and Shipsear public datasets show that compared with the recognition models with the traditional short-time Fourier transform (STFT) and Fbank features as input, the FEAN has achieved a 4.5% and 1.2% improvement in recognition accuracy on the two datasets, respectively.