Deep Belief Network for Fingerprinting-Based RFID Indoor Localization

摘要

Indoor positioning based on fingerprinting has recently attracted a lot of interest due to its high accuracy. In this paper, we investigate fingerprinting-based radio-frequency identification (RFID) indoor localization algorithm using deep learning. First, the original and normalized received signal strength indication (RSSI) data are extracted from the RFID reader and reference tags. For the offline stage, we design a deep belief network (DBN) with four hidden layers to deeply learn the characteristics of normalized RSSI data, in which the hidden layers are stacked by restricted Boltzmann machine (RBM). We use greedy learning algorithm to train the weights of DBN layer-by-layer, and calculate the parameters of each layer by contrastive divergence with one step iteration (CD-1) algorithm. In the online stage, we collect test tags data for location estimation. In addition, we propose an improved DBN algorithm to achieve accurate location estimation through similarity comparison. Experiments show that the proposed RFID tag localization method can locate targets with high accuracy in complex indoor environment and outperform several fingerprinting-based localization schemes.