Wireless indoor localization using convolutional neural network

This paper provides a wireless localization method using convolutional neural networks (CNN) and based on WIFI fingerprinting data. We pre-processed WIFI fingerprinting data from Jaume I University public database and used it to train our CNN model. A WIFI fingerprint is composed of a series of received signals' strength (RSS) released by many access points (APs). We changed the RSSs' value at the edge of the signal disappearance. This operation can help the CNN model extract the features of fingerprints and have a better performance. After this operation, a CNN model can have about 21.2% lower loss and 23.5% higher accuracy than usual. Changing edge RSS is shown to be a good way to improve the performance of indoor localization.