Intra-Spacecraft RFID Localization

In this paper we explore two machine learning approaches to improve RFID tag localization in the highly reflective environment imposed by the International Space Station. We propose P-RFIDNet (Passive RFID Net), a neural network with a ResNet50 (He, et al., 2015) [1] backbone for localizing passive RFID tags in high multipath environments with fixed antennas. Furthermore, we show how transfer learning can be used to generalize P-RFIDNet to new RFID environments with limited training data. In addition to P-RFIDNet, we present REALMRFC, a random forest (Breiman, 2001) [2] model with feature engineering performed by an RFID localization expert. We benchmark P-RFIDNet and REALMRFC using data from the RFID Enabled Autonomous Logistics Management (REALM) RFID system on International Space Station (ISS).