Supervised Analysis Dictionary Learning: Application in Consumer Electronics Appliance Classification

The objective of this paper is to estimate if an electrical appliance is 'ON' based on their common mode electromagnetic (CM EMI) emissions. The assumption being that, a user by knowing the state of the appliance can make an informed decision whether to keep it running or switch it off to save power. Here, state estimation of a single appliance is formulated as a classification problem. A new technique called analysis dictionary learning is proposed to generate features from CM EMI. The proposed method outperforms feature extraction based on deep learning techniques as well as a state-of-the-art information theoretic feature extraction technique based on Conditional Likelihood Maximization.