A Gunshot Recognition Method Based on Multi-Scale Spectrum Shift Module

In view of the issues such as larger network model and lower recognition accuracy current gunshot networks, a neural based on multi-scale spectrum shift module is proposed in this paper to fully mine relevant information among spectrums. This employs architecture densely connected convolutional uses branch realize interaction information. replaces under-sampling operation spectrums, realizes globalized feature extraction spectrum, avoids loss during process, further improves quality map. Experiments were conducted NIJ Grant 2016-DN-BX-0183 dataset YouTube gunshots that have been open public, both whose classification reached 83.2% 95.1%, respectively, with size being controlled at around 16 MB. The experimental results indicate that, compared other existing methods for network, can time-frequency better effectively, has higher recognition.