Box Size Estimation Using ANNs in UHF RFID Gates from Interrogation Process Features

Different physical parameters of boxes containing batches of RFID-tagged items can be estimated by making smart use of standard information available during interrogation, such as the number of tags identified, the number of collisions, the average power measured in the slots, and so forth. This process can be used for many purposes, such as adding a measuring capacity to a gate without extra-hardware, checking whether the boxes fit their manifest, detecting unusually distributed boxes that require different processing, etc. In this paper, these features are used as input information in a supervised learning model based on Artificial Neural Networks (ANNs), which outputs the box size estimation among a set of possible box size candidates. This model has been trained using standard sizes of boxes, which are fed into a UHF RFID gate simulator that introduces random perturbations in the orientation, the position, and the number of tags in the boxes. The accuracy of the model is about 90% for an 8 candidate boxes setup. Thus, the model can be considered suitable to be further investigated.