DeepSqueak: a deep learning-based system for detection and analysis of ultrasonic vocalizations

摘要

Rapid detection of USVs using Faster RCNN. Shown is the Faster RCNN-based detection architecture. a An example waveform from a 1 s section recording containing mouse ultrasonic vocalizations. b A filtered sonogram produced from the audio segment displayed in (a). c A representation of the Faster-RCNN region proposal network determining sections of the sonogram to pass to the classification network. d The classification network receives input from the region proposal network, performs a series of convolutions and filters, and ultimately classifies the section as a call or background noise. Scale bar: Y = 20 kHz, X = 20 ms