Real-Time Vehicle Sound Detection System Based on Depthwise Separable Convolution Neural Network and Spectrogram Augmentation

This paper proposes a lightweight model combined with data augmentation for vehicle detection in an intelligent sensor system. Vehicle can be considered as binary classification problem, or non-vehicle. Deep neural networks have shown high accuracy audio classification, and convolution are widely used feature extraction classification. However, the performance of deep is highly dependent on availability large quantities training data. Recordings such tracked vehicles limited, techniques applied to improve overall accuracy. In our case, spectrogram mel before extracting Mel-scale Frequency Cepstral Coefficients (MFCC) features robustness Then depthwise separable CNN network compression migrated hardware platform The proposed approach evaluated dataset recorded field using systems microphones. final frame-level achieved was 94.64% test recordings 34% parameters were reduced after compression.