A Data-driven Fusion and Its Application to Acoustic Vehicle Classification

Most traditional information fusion methods depend on system models, where certain simplification will be introduced. However, with increased complexity of applications, these models tend to be inadequate and show bias to the real situation. In some cases, precise models are just impossible to build up. Aiming at this problem, two data-driven information fusion methods are presented in this paper. By combining a data-driven feature set with a model-based feature set, the performance of information fusion is improved due to a compensation deficiency for model-based approaches. The proposed method is then applied to acoustic vehicle classification, and better classification performance is achieved, which shows the feasibility and advantages to introduce data-driven ideas into information fusion.