Activity classification model construction method and system based on Wi-Fi signals and transfer learning

Abstract

The invention discloses an activity classification model construction method and system based on Wi-Fi signals and transfer learning. The method comprises the steps of preprocessing Wi-Fi channel state information corresponding to user activities through operations such as Butterworth filtering, singular value decomposition and phase correction; training a deep convolutional feature extraction network in a data triple pairing training mode in a source domain, and extracting activity features in the pre-processed Wi-FiCSI fragments in the target domain by using a feature extraction network model through transfer learning; and realizing a cross-domain activity classification task by combining the extracted activity features with an SVM classifier. According to the invention, a transfer learning mode is adopted; a new cross-domain activity classification method based on Wi-Fi signals is established; aiming at the problem that the model application of the existing activity classification method fails after the environment is changed, a new cross-domain activity classification framework is provided, the method can realize an activity classification task in a target domain based on a small number of labeled samples, and the existing cross-domain activity classification effect is improved.

Description

technical field [0001] The present invention relates to the field of activity classification and neural network technology, in particular to an activity classification method based on Wi-Fi signals and transfer learning, which is mainly used to realize user activity classification under cross-domain conditions. Background technique [0002] Human activity classification has important research value and significance in the fields of smart home, human-computer interaction and intelligent security. In smart homes, we can intelligently adjust indoor temperature, light, humidity, etc. for users through the perception results of user activities. For families with the elderly, it can also provide intelligent care for the elderly. In the field of human-computer interaction, the perception of user behavior is the basis for understanding and realizing user needs. In the field of intelligent security, through the mining of user behavior characteristics, the user's identity can be eff...

AI Technical Summary

Problems solved by technology

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide an activity classification method based on Wi-Fi signals and transfer learning, which solves the low recognition accuracy and the need for a large number of labels in the cross-domain activity classification task in the prior art. For the problem of samples, an effective model training method based on paired data is proposed to solve the problem that the existing deep learning model training methods are not suitable for cross-domain activity classification.

Images