Abstract To address the challenge of radar emitter signal individual recognition, where single-dimensional radar fingerprint features are susceptible to noise interference and neural networks exhibit low recognition accuracy in low signal-to-noise ratio (SNR) environments, this study introduces a residual neural network predicated on a two-branch feature fusion. This approach amalgamates two-dimensional time-frequency domain features with one-dimensional intermediate-frequency signal features. Unlike existing algorithms, our proposed method integrates the knowledge gleaned from features learned across different dimensions. This integration enables the neural network to utilize features from multiple dimensions for recognition, thereby mitigating the impact of noise on a single feature. Experimental results demonstrate that our proposed algorithm outperforms others under various low SNRs, achieving an average recognition rate of 88.39%.