Zero-velocity interval detection is a typical method to inhibit the error accumulation for pedestrian navigation systems based on MIMU (Micro-inertial Measurement Unit). The traditional threshold adjustment method based on condition judgment has poor robustness to different movement patterns, and it is hard to realize automatic adjustment and precise navigation in the multi-movement state. In this paper, we proposed an algorithm of movement pattern recognition and zero-velocity correction, which is based on deep learning. Collect sensor data series under various movement patterns as training samples. Based on LSTM (Long Short-term Memory) neural network, we train the movement pattern recognition and threshold adaptive model. Combined with the gait recognition and monitoring algorithm, the model adjusts zero-velocity detection thresholds adaptively in the multi-movement state. We evaluate the performance of our system on public datasets and with real-world experiments, and compare the results with other algorithms. Experiments showed that the algorithm detects the zero-velocity interval adaptively, and improves the ability to adapt the movement and environment for pedestrian navigation based on MIMU in complex environments, thereby improving the precision of pedestrian navigation.