In recent years, radar-based gesture recognition technology has been widely used in industry and life, and more complex application scenarios also put forward higher requirements on the accuracy and robustness of gesture recognition algorithms. A high-precision gesture recognition algorithm based on millimeter-wave radar is desgined. By comparing the existing classification algorithms, a Convolutional Neural Network-Long Short Term Memory(CNN-LSTM) deep learning algorithm model is constructed for gesture recognition. At the same time, the Blackman window is employed to suppress the problem of spectrum leakage in gesture signal processing, and efficient clutter suppression and data enhancement is achieved through the combining of wavelet threshold and dynamic zero-padding algorithm. The actual measurement results show that the proposed gesture recognition algorithm achieves a correct classification rate of 97.29%, and can maintain a good recognition accuracy rate under different distances and angles with very good robustness.