The gait cycle of lower limb movement in the human body is composed of the stance phase and the swing phase. However, existing methods for recognizing the gait intention of human movement usually extract features from the swing phase, with less research on the stance phase, and most of them are limited to discrete ground contact states, ignoring the continuous detailed information of the stance phase. Therefore, this paper proposes an intelligent lower limb prosthesis intention recognition method driven by the powered plantarflexion phase data. Considering that the adaptive posture adjustment of the human body to terrain changes begins in the powered plantarflexion phase at the end of the stance phase, which serves as a transitional phase connecting the stance phase and the swing phase, and participates in energy release during movement, this method extracts complete powered plantarflexion phase data and combines it with the early swing phase data to define the target data time window. Haar wavelet transform is used to represent the hidden non-stationary signal features in the stance phase, and the number of wavelet decomposition layers is adaptively determined based on the energy change of plantar push-off. Feature vectors are constructed using the corresponding wavelet coefficients, and support vector machines are used for classification. Experimental results show that the recognition rate of the proposed method can reach 99.21% in five steady-state modes of the self-collected dataset and 97.65% in 13 comprehensive motion modes, which is 1.69% and 2.53% higher than the benchmark method, respectively. Utilizing the data from the plantar power phase to assist the intention recognition task improves the recognition rate and robustness of the model.