In this paper, a new detection approach based on sparse decomposition in terms of a union of learned subspaces is presented. It uses a dictionary that can be interpreted as a bank of matched subspaces. This improves the performance of signal detection, as it is a generalization for detectors and also exploits sparsity in its decision rule. The proposed detector shows a new trade-off for designing a suitable detector. We demonstrate the high efficiency of our method in the case of voice activity detection in speech processing.
Union of subspaces model, sparse representation, signal detection, dictionary learning