Transform Based Subspace Interpolation for Unsupervised Domain Adaptation Applied to Machine Inspection

In most practical application scenarios, several factors may introduce domain discrepancy between the train (source domain) and test (target domain) data. Domain adaptation techniques address this domain shift to ensure reliable inferencing. However, limited data and the unavailability of annotations of the target domain data pose an additional challenge for the adaptation task. Unlike divergence and adversarial learning-based techniques that are data-hungry, subspace modeling-based techniques are found more suitable for learning representations from limited data. This work presents a novel subspace interpolation-based method via transform learning for unsupervised domain adaptation. Transform learning framework has been used for subspace modeling that provides superior performance in terms of accuracy, computational complexity, and improved convergence over dictionaries. They model the subspace that links the source and target domain data and generates domain invariant features for cross-domain analysis. The potential of the proposed method is demonstrated using the challenging scenario of adaptation between different but related machines using two public datasets. Experimental results show the effectiveness of the proposed method compared to the state-of-the-art methods for machine diagnosis.