The invention discloses an activity recognition method based on stack own coding. The method comprises the following steps: acquiring data points of a user in daily activities to form time sequence data, wherein each data point comprises a signal value and corresponding acquisition time; pre-processing the time sequence data to obtain a plurality of time sequence data frames; solving a global model with a stack own coding neural network method according to the time sequence data frames obtained after pre-processing, determining a depth feature vector corresponding to each time sequence data frame with the global model, and combining the depth feature vectors corresponding to all the time sequence data frames into a depth feature matrix; and training a recognition model on the basis of a plurality of depth feature matrixes of which the activities are marked correctly in order to perform activity recognition. According to the activity recognition method, the global network is constructed through combination of the stack own coding method and a depth feature learning method for solving the depth feature vectors, so that the recognition speed and recognition accuracy are favorably increased.