Zhao Zijian ; Chen Tingwei ; Cai Zhijie ; Li Xiaoyang ; Li Hang ; Chen Qimei ; Zhu Guangxu

Abstract

In recent years, Wi-Fi sensing has garnered significant attention due to itsnumerous benefits, such as privacy protection, low cost, and penetrationability. Extensive research has been conducted in this field, focusing on areassuch as gesture recognition, people identification, and fall detection.However, many data-driven methods encounter challenges related to domain shift,where the model fails to perform well in environments different from thetraining data. One major factor contributing to this issue is the limitedavailability of Wi-Fi sensing datasets, which makes models learn excessiveirrelevant information and over-fit to the training set. Unfortunately,collecting large-scale Wi-Fi sensing datasets across diverse scenarios is achallenging task. To address this problem, we propose CrossFi, a siamesenetwork-based approach that excels in both in-domain scenario and cross-domainscenario, including few-shot, zero-shot scenarios, and even works in few-shotnew-class scenario where testing set contains new categories. The corecomponent of CrossFi is a sample-similarity calculation network called CSi-Net,which improves the structure of the siamese network by using an attentionmechanism to capture similarity information, instead of simply calculating thedistance or cosine similarity. Based on it, we develop an extra Weight-Net thatcan generate a template for each class, so that our CrossFi can work indifferent scenarios. Experimental results demonstrate that our CrossFi achievesstate-of-the-art performance across various scenarios. In gesture recognitiontask, our CrossFi achieves an accuracy of 98.17% in in-domain scenario, 91.72%in one-shot cross-domain scenario, 64.81% in zero-shot cross-domain scenario,and 84.75% in one-shot new-class scenario. The code for our model is publiclyavailable at https://github.com/RS2002/CrossFi.