Wild animal weak supervision sound recognition method based on deep learning

Abstract

A wildlife weak supervision sound recognition method based on deep learning relates to the field of wildlife sound recognition, and comprises the following steps: training a sound recognition model: obtaining a video set containing wildlife images and sounds; extracting an image frame sequence in the video and a corresponding audio; identifying the image frame by using a deep learning network to obtain a category attribute tag; converting the audio time domain data into frequency domain data; creating a training database; performing voice recognition model training by utilizing the category attribute labels and the frequency domain data; voice recognition model reasoning: acquiring a video set only containing voices of wild animals; extracting audio data in the video; converting the audio time domain data into frequency domain data; and recognizing the frequency domain data by using the voice recognition model to obtain a final category attribute tag. According to the method, the sample labeling cost is reduced, the deep learning method is adopted, the feature template does not need to be manually screened, and the accuracy and the recognition efficiency are improved.

Description

technical field [0001] The invention relates to the technical field of wild animal voice recognition, in particular to a method for wild animal weakly supervised voice recognition based on deep learning. Background technique [0002] In the field of wildlife protection, in order to meet the needs of wildlife statistics, it is often necessary to obtain wild animal-related videos, images and sound clips by setting up wild animal video surveillance cameras, etc., and identify the captured wild animals through certain technical means type. [0003] At present, the research content of sound recognition technology is mainly divided into the following two aspects: one is the acquisition and separation of sound, such as the use of professional recording equipment to capture sound, signal processing and separation of audio signals, etc. The second is the classification and recognition of sounds, such as using traditional pattern recognition methods for audio recognition. Especially...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a wild animal weakly supervised sound recognition method based on deep learning to solve the problems existing in the existing wild animal sound recognition methods, reduce the human, financial and material input of the existing sound labeling samples, and pass the depth The method of learning, intelligent classification and recognition of the sounds of wild animals obtained

Images