A Depth Migration Indoor Localization Method Based on Parameter Prediction

Abstract

The invention belongs to a method for accurate positioning in a complex indoor environment, and in particular relates to an indoor positioning method for deep migration based on parameter prediction. The present invention first uses labeled source domain data to pre-train a source network positioning model with good performance, and then learns a method from source network parameters to target network with the assistance of unlabeled target domain data by fixing the pre-trained source network parameters. The transformation matrix of the parameters, and finally use the transformation matrix and the source network parameters to calculate the target domain network parameters. The invention overcomes the drawbacks of common domain adaptation techniques that focus too much on domain invariant features and ignores domain differences, and can make the feature distribution of the target domain as close as possible to the source domain feature distribution, thereby ensuring that the target domain network can adapt to the new environment while being able to Part of the data processing capability of the source domain network is reserved. The invention is a high-precision positioning method that can well adapt to complex indoor environments.

Description

technical field [0001] The invention belongs to the technical field of indoor positioning, and in particular relates to an indoor positioning method for deep migration based on parameter prediction. Background technique [0002] The popularization of mobile devices and the development of wireless communication technology have promoted the development of a series of services based on mobile terminals, among which location-based services have greatly changed people's lifestyles. The provision of services has been widely used in scenarios such as pedestrian navigation, advertisement push, and asset security management. As the basis of location-based services, positioning technology has received extensive attention from researchers. [0003] Among many indoor positioning technologies, WiFi-based positioning technology has become one of the most promising positioning technologies due to its advantages of low cost, high real-time performance, and convenient use. The positioning ...

AI Technical Summary

Problems solved by technology

However, both methods use the same network to extract data features in different fields. The disadvantage of this feature extraction method is that it focuses too much on the commonality of the two fields and ignores the unique characteristics of their respective fields.

In addition, since the features extracted by the same network are the common parts of the two domains, it can only effectively constrain the similarity of the common features. It is obviously not sufficient to narrow the domain differences in this way, and when the amount of data in a certain domain is large Small or large domain differences will also lead to a sharp drop in the positioning performance of the model

Based on the above reasons, it is difficult for such methods to achieve accurate positioning in complex indoor positioning environments.

Images