Research On Human Imaging Technology Based On Generative Adversarial Network In WiFi Environmen

Applications for imaging tasks can be found in a variety of domains,including intrusion detection and picture identification.However,there are still some issues with the current imaging techniques.The imaging technique based on camera equipment runs the risk of privacy leakage,while the imaging technique based on radar-based professional equipment is expensive and difficult to use on a broad scale.Researchers have discovered that WiFi signals will send messages while carrying information about the human body in the scene as the Internet of Things develops.As a result,research on WiFi signals has been developed,including WiFi human imaging,WiFi behavior identification,and WiFi fall detection.These studies provide technological support for the area of human-computer interaction and intelligent services with being low cost,widely applicable,and non-invasive.Existing WiFi-based research,however,still faces significant difficulties.For instance,different WiFi data formats that are gathered by various devices result in the lack of adaptability of various datasets.Additionally,incorrect limb imag ing is still a concern with current WiFi imaging.Through the use of generative adversarial networks,this thesis investigates WiFi channel state data related to dataset extension and wireless photography.The following are the specific research contents:1.This thesis proposes a data augmentation approach based on principal component analysis and Wasserstein generative adversarial network to address the issues of inadequate data volume of existing WiFi datasets and low adaptability of datasets in various scenarios.Before extracting useful features from the WiFi signal,this method preprocesses the WiFi signal using filtering,noise reduction,and principal component analysis.After that,the preprocessed data is enhanced using the Wasserstein Generative Ad versarial Network.Finally,the public dataset is used to perform experimental verification.The results demonstrate that the strategy put forward in this thesis can successfully increase the size of the WiFi action recognition dataset,and the extended d ataset’s recognition accuracy has increased to over97%.2.This thesis suggests a wireless imaging solution based on d eep convolutional generative adversarial networks to address the issue of blurring body contours in human body mask images created using WiFi generative models.This technique uses preprocessed WiFi data and video to create a heterogeneous dataset,which is then used to train a deep convolutional generative confrontation network to create a model for creating human masks.The method suggested in this thesis reduces the time to generate the image by 60% while improving the accuracy of the intersection and union ratio of the mask image by 3%compared to the current wireless imaging models.