A behavior recognition method based on deep learning

Abstract

The invention discloses a behavior recognition method based on deep learning. The method is characterized by dividing the behavior recognition into a training stage and a recognition stage, at the training phase, preprocessing the triaxial acceleration data acquired by an acceleration sensor and extracting a resultant acceleration value; obtaining a root-mean-square value of quadratic sum of three-axis acceleration of xyz, performing STFT on a resultant acceleration value to extract a relationship between time and frequency in data, and inputting an energy spectrum of an STFT sequence and a behavior tag corresponding to each collected data as a training set of a behavior recognition model; and enabling the behavior recognition model to adopt a DRNN based on an LSTM unit, training the initial model by using training set data, and selecting the training model with the highest test set classification accuracy to be applied to the recognition stage. Compared with an existing scheme, the method is higher in recognition accuracy, lower in power consumption and suitable for working at an intelligent terminal with limited resources.

Description

technical field [0001] The invention relates to a behavior recognition method based on deep learning, which belongs to the field of behavior recognition. Background technique [0002] At present, the existing behavior recognition based on machine learning often needs to manually extract a large number of feature values, resulting in the lack of objectivity of the recognition model in the selection of feature values ​​and the increase of system energy consumption. The application of deep learning algorithms can avoid the process of manually extracting feature values, and the recognition model can independently find features from the input data, so as to predict the results and ensure the accuracy of the system. The convenience and efficiency of deep learning algorithms make their application in the field of behavior recognition a research hotspot. [0003] In the method of behavior recognition using deep learning, it is an important aspect of research to improve the accuracy...

AI Technical Summary

Problems solved by technology

[0002] At present, the existing behavior recognition based on machine learning often needs to manually extract a large number of feature values, which leads to the lack of objectivity of the recognition model in the selection of feature values ​​and the increase of system energy consumption.

Images