Vehicle recognition in acoustic sensor networks via sparse representation

This paper presents a new approach to recognize the types of moving vehicles in a distributed, wireless sensor network, based on sparse signal representation. Through a sparse representation computed by l₁-minimization, we propose a general classification algorithm for acoustic object recognition. This algorithm first uses Mel frequency cepstral coefficients to extract the acoustic features of vehicles and then constructs an over-complete dictionary by the training feature sets and represents the test feature sets sparsely. The objective vehicles will be recognized by solving the optimization problem of sparse representation. The experiments are based on the data sets of military vehicles collected by DARPA, and the results show that the proposed algorithm model gives good performance on vehicle recognition. Compared with other classification algorithms, the method improves the precision of recognition.