THIS COMMUNICATION IS CANCELLED: Intelligent system for data quality assurance in extensive urban noise monitoring networks
Résumé
Monitoring urban noise by means of extensive measurement networks is nowadays becoming affordable. However, a trade-off between quality and number of sensor units is still needed. Detection of breakdowns, drifts, or critical outliers is a crucial aspect. The simplest models, based on heuristic rules derived by prior knowledge, are able to detect breakdowns mainly. However, detecting every possible kind of malfunction only based on prior knowledge is neither theoretically possible nor practically feasible. The introduction of an intelligent learning system is thus advisable. In this paper, a neural network approach is followed. The proposed method is based on the synergy between different submodels that apply both supervised and unsupervised learning strategies. In particular, the submodel that uses a self-organizing map (SOM) is thoroughly studied. It implements an unsupervised learning algorithm, based on a series of psychoacoustic features encoding spectro-temporal irregularities, and currently used in auditory scene analysis models. The developed methodology has been applied to cheap sensors placed near reference microphones to check the validity of the quality assessment.
Origine : Fichiers éditeurs autorisés sur une archive ouverte