Wireless cross-domain action recognition method based on semi-supervised learning

Abstract

The invention provides a wireless cross-domain action recognition method based on semi-supervised learning, and the method comprises a step of extracting the amplitude information of each subcarrier from the CSI data of a WiFi signal for action recognition according to the principle that a human body action affects a wireless signal; in order to solve the problem that a large amount of annotationdata is not easy to obtain, the method only annotates a real action label for a small amount of action samples, then calculates the similarity between a label-free sample and a labeled sample througha DTW algorithm, and attaches a pseudo label to the label-free sample, thereby expanding a training sample set. In order to improve the generalization of an action recognition model and realize actionrecognition of different positions or different persons, the invention provides a classification clustering comprehensive model, establishes an SOM network to cluster action samples, and finally classifies the samples in combination with classification and clustering results. According to the method, the dependence of the supervised learning model on a large number of labeled samples is reduced,only a small number of labeled samples are combined with unlabeled samples to train the classification model, high generalization is achieved, and cross-domain action recognition can be achieved.

Description

technical field [0001] The invention relates to an action recognition method based on wireless signals, in particular to a cross-domain action recognition method based on commercial WiFi using meta-learning and semi-supervised learning. Background technique [0002] Action recognition is of great significance in people's daily life. With the rapid development of the Internet of Things, pervasive computing, graphics and images, artificial intelligence and other fields, it has become possible to accurately identify human actions through technical means and has gradually been widely used. Monitoring people's daily behaviors through motion recognition can effectively ensure people's health and safety in their daily lives. Currently commonly used motion recognition technologies are mainly based on video surveillance or wearable devices with built-in sensors. The video surveillance method needs to work normally under the conditions of light and no occlusion, and it is difficult ...

AI Technical Summary

Problems solved by technology

This is a supervised learning method, and its disadvantages are: (1) a large amount of labeled sample data is required to train the recognition model, and the collection and labeling of labeled sample data are expensive; (2) the classification of a single classifier Poor performance and generalization performance, when recognizing actions from different domains (such as different locations or different people), the classification ability is greatly reduced

Images