To solve the problem that the magnetic acoustic emission (MAE) signal is easily interfered by noise, a magnetic acoustic emission signal feature recognition method based on improved northern goshawk optimization-variational mode decomposition (INGO-VMD) and dispersion entropy is proposed. An experimental platform is established to collect the MAE signal of Q345 steel under stress state from 0 to 450 MPa. The INGO algorithm optimized variational mode decomposition method was employed to determine the number of modes (k) and the penalty factor \(\left( \alpha \right)\) , resulting in the decomposition of MAE signals into ten modal components. Dispersion entropy was then calculated for these components as feature vectors, which were organized into a feature vector matrix for classification using support vector machines. The proposed method achieved a recognition rate of 95.8333%, demonstrating its effectiveness.