As the number of embedded devices in commercial markets increase, it becomes increasingly important to secure these devices from malicious actors. In this paper, we demonstrate various ways of non-intrusively extracting data from embedded devices through debugging interfaces and using this data with machine learning and deep learning models to detect anomalies in embedded devices. The results exhibit that deep learning models have great potential in embedded cybersecurity, with Convolutional Neural Network and Long Short-Term Memory network models scoring over 95% accuracy on the testing set of our data. This study not only contributes to the ongoing cybersecurity discussion for embedded devices but also creates a foundation for further advancements in the field.
