Towards Machine Learning Based Fingerprinting of Ultrasonic Sensors

Fingerprinting is now recognized as an approach that identifies or categorizes devices using distinct data characteristics. This paper introduces a sensor fingerprinting model and, through experiments, demonstrates that unique error patterns resulting from manufacturing imperfections can be used to accurately identify sensors. Using various machine learning algorithms such as a random forest classifier, multilayer perceptron, and soft decision tree, accuracies of 87%, 85.5%, and 89.2% respectively were achieved. These results highlight the significant potential for reliable sensor fingerprinting applications.