Abstract: Automatic modulation recognition (AMR) is an important link in fields such as communication recognition, electronic reconnaissance, and interference detection. In response to the problem of difficulty in accurately modulating and identifying interference sources in low signal-to-noise ratio environments, this paper constructs a modulation recognition model based on attention mechanism, sequential convolution-based attention model (SCAM), for modulating recognition of raw I/Q sequence signals. By introducing the attention mechanism in a one-dimensional convolutional neural network(CNN) model, SCAM can effectively extract feature information from the raw I/Q sequence signals under low signal-to-noise ratio conditions, and jointly extract multi-domain feature information through feature fusion, using the fused features for modulation recognition, thereby improving the accuracy of interference source type recognition. Through modulation recognition experiments under different signal-to-noise ratio environments on the open-source dataset RML2016.10a, compared with traditional convolutional neural network models, SCAM achieves higher modulation recognition accuracy.