Target user EEG identification for two-class motor imagery is crucial In the field of brain computer interface (BCI). We propose a lightweight method combining transfer learning (TL) and wavelet packet transform (WPT). After common average reference preprocessing, WPT decomposes 0-3.5 s EEG from C3/Cz into three layers, reconstructs ERD-related coefficients, and extracts variance and energy mean as features. Using BCI Competition III dataset IVa and a TL classifier, our method achieves 91.8% average accuracy, outperforming the top two competition methods. It is simple, effective, and practical for multi-user motor imagery recognition, promoting robust BCI operations.