Research And Implementation Of Sparse Wi-Fi Fingerprint Enhancement And Loacation Based On Capsule Network

With the increasing level of technology and the rapid development of mobile smart devices,smartphones and their various application services have greatly facilitated people’s daily lives.The demand for location services from users is also growing.Among them,due to their wide deployment and low cost,Wi-Fi access points have attracted more and more researchers to engage in algorithm research for indoor positioning based on Wi-Fi.High-density Wi-Fi data plays a crucial role in improving the accuracy of positioning algorithms.However,due to the limitations of wireless scanning frequency by smart terminal systems,it is challenging to achieve high-frequency collection of Wi-Fi data.Therefore,this paper proposes a method for enhancing sparse Wi-Fi fingerprint data and then performing positioning using capsule networks.Firstly,this paper proposes a sparse fingerprint enhancement algorithm,which uses low-rank matrix completion and piecewise lowdegree interpolation to expand the low-frequency collected Wi-Fi fingerprint data.Before enhancing the original sparse fingerprint data using low-rank matrix completion,a low-rank matrix is constructed by adding blank rows row by row,and matrix decomposition is performed to recover the low-rank matrix.Since there is certain regularity in the propagation of signal data,this paper uses the piecewise low-degree interpolation method to construct function curves and obtain unscanned fingerprint data on the curves.The two fingerprint data enhancement methods proposed in this paper can approximately double the density of the original sparse fingerprint data,which is then used to train the subsequent positioning model.Secondly,when positioning the expanded dense fingerprint,this paper proposes a positioning algorithm based on capsule networks.This algorithm uses convolutional layers for feature extraction,then achieves data transformation between capsule layers through dynamic routing algorithms,and finally predicts the positioning results through fully connected layers.Capsule networks can not only learn global and local features simultaneously but also capture spatial features and positional relationships of data better,effectively improving the accuracy of Wi-Fi positioning and demonstrating good performance.Experimental results demonstrate that the proposed sparse fingerprint data enhancement algorithm effectively expands sparse WiFi fingerprint data,and the obtained dense fingerprint data is used in a deep learning positioning algorithm model based on capsule networks.This significantly expands the training data and has a positive effect on training the model parameters.Furthermore,extensive comparative experiments reveal that the proposed positioning algorithm model exhibits excellent positioning performance,outperforming common indoor positioning technologies and traditional Wi-Fi indoor positioning algorithms based on machine learning.