Mobile terminal double-decoupling network long tail activity identification method

The invention discloses a mobile terminal double-decoupling network long tail activity identification method, which specifically comprises the following steps of collecting action signals, and obtaining a training set; inputting the training set into a double-decoupling network model, and realizing long-tail activity identification through a mobile wearable identification device; wherein the double-decoupling network model is constructed by a neural network and is decoupled in a training stage and a testing stage respectively. According to the method, through decoupling in a training stage, namely separation of feature extraction and activity classification, a multi-branch topological structure is adopted to enhance the feature extraction capability of the model, and meanwhile, weight fine tuning in a classifier stage can effectively improve the long-tail distribution recognition performance of the model; according to the invention, the decoupling of the training stage and the testing stage significantly redu