Hybrid neural network based on novel audio feature for vehicle type identification

Abstract

Due to the audio information of different types of vehicle models are distinct, the vehicle information can be identified by the audio signal of vehicle accurately. In real life, in order to determine the type of vehicle, we do not need to obtain the visual information of vehicles and just need to obtain the audio information. In this paper, we extract and stitching different features from different aspects: Mel frequency cepstrum coefficients in perceptual characteristics, pitch class profile in psychoacoustic characteristics and short-term energy in acoustic characteristics. In addition, we improve the neural networks classifier by fusing the LSTM unit into the convolutional neural networks. At last, we put the novel feature to the hybrid neural networks to recognize different vehicles. The results suggest the novel feature we proposed in this paper can increase the recognition rate by 7%; destroying the training data randomly by superimposing different kinds of noise can improve the anti-noise ability in our identification system; and LSTM has great advantages in modeling time series, adding LSTM to the networks can improve the recognition rate of 3.39%.

The AI Mechanic: Acoustic Vehicle Characterization Neural Networks

“…Researchers [77] classify accelerating vehicles using audio and apply random noise and pitch shift to expand their dataset. Other work [14] focuses on vehicle type identification using MFCCs and other audio features, these authors apply city noise to augment captured signals. There have been additional applications of augmentation to acoustic vehicle DL related tasks [7,13,53].…”

Fast environmental sound classification based on resource adaptive convolutional neural network

“…Piczak 4 first proposed the use of 2-D CNN to learn Log-Mel spectrogram features, which has significantly improved ESC performance compared with traditional machine learning algorithms such as KNN and SVM. Chen et al 5 accurately identified the audio signal of the vehicle by fusing the LSTM unit into the convolutional neural network. Boddapati et al 6 uses AlexNet 7 and GoogLeNet 8 to classify the environmental sound features extracted from the spectrum.…”

C-AVDI: Compressive Measurement-Based Acoustic Vehicle Detection and Identification

“…Mel-frequency cepstral coefficients (MFCCs) are used in conjunction with ML or deep learning (DL) as features in a number of existing AVDI systems: in [8] they are used with a modified MLP, in [5] they are extracted from a specific high energy audio region and used with an ANN and knearest neighbors (KNN) classifier, and in [9] they are used in a feature set containing the pitch class profile (PCP) and short-term energy (STE) of vehicle audio signals in a hybrid convolutional neural network (CNN) containing a long shortterm memory (LSTM) layer.…”

“…[5] MFCC KNN /ANN [6] DFT SVM [7] STFT SVM [8] MFCC MLP [9] MFCC CNN /PCP/STE /LSTM [10] Mod-PCEN SNN [11] GCC RANSAC [12] GCC RANSAC [13] DWT LR architecture in which information acquired at sub-Nyquist rates from different frequency bands is used in a range of applications without prior signal reconstruction.…”

“…Various low-cost, low-complexity VDI techniques have been proposed, with acoustic vehicle detection and identification (AVDI) in particular being the subject of wide-ranging research [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15]. The low price and simple installation process of AVDI systems make them an attractive alternative to other more expensive and difficult-to-install VDI systems, such as video camera-, radar-, or induction loop coilbased systems.…”