To enhance the classification accuracy of lower limb movements, this paper introduces a hybrid recognition model based on surface electromyography (sEMG) that combines convolutional neural networks (CNN) with long short-term memory networks (LSTM). Initially, sEMG data were collected from 20 subjects performing four types of gait movements: ascending stairs, descending stairs, walking, and squatting. Subsequently, the collected sEMG data underwent preprocessing, and both time domain and frequency domain features were extracted to serve as inputs for the machine learning recognition model. The CNN-LSTM model was then constructed for lower limb action recognition and compared against the performances of CNN, LSTM, and SVM models. The results demonstrate that the CNN-LSTM model outperforms the CNN, LSTM, and SVM models by 2.16%, 8.34%, and 11.16% in accuracy, respectively, thereby proving its superior classification performance. This model provides an effective solution for enhancing lower limb motor functions, offering significant benefits for rehabilitation medical equipment and power assist devices.