Live pig cough identification method based on convolutional neural network

Abstract

The invention discloses a live pig cough identification method based on a convolutional neural network, and belongs to the field of voice signal processing. Respiratory diseases are one of the most common diseases affecting the pig breeding industry, which seriously restricts the development of the pig industry, cough is one of the obvious symptoms of respiratory diseases in live pigs, early warning of respiratory diseases can be achieved by monitoring and identifying the cough, thereby effectively reducing the spread of the diseases and the use of antibiotic drugs. For the problem that the identification precision is low in an existing live pig cough identification method, especially obvious for intensive large pig houses, the invention provides the method based on a convolutional neuralnetwork based on Ale*Net to recognize the cough, wherein a voice spectrum of voice signals serves as network input, the network can automatically extract the deep features of the voice spectrum, compared with a conventional voice recognition method, the method can effectively improve the cough recognition accuracy and overall recognition accuracy.

Description

technical field [0001] The invention belongs to the field of speech signal processing, and in particular relates to a method for recognizing pig cough sounds based on a convolutional neural network. Background technique [0002] Respiratory diseases are one of the most common diseases that affect the pig breeding industry. Respiratory diseases can cause pigs to have difficulty breathing, repeated body temperature, coughing, red ears, some foaming at the mouth, and increased eye and nose secretions. It is easy to infect, and mild cases will cause pigs to lose weight, and severe cases will cause pigs to die, which will bring huge losses to farmers. Therefore, early warning of respiratory diseases is of great significance. One of the more obvious symptoms of respiratory diseases is coughing. Early warning of respiratory diseases can be realized by monitoring and identifying cough sounds, thereby effectively reducing the spread of diseases and the use of antibiotics. [0003] A...

AI Technical Summary

Problems solved by technology

[0003] At present, pig cough sound recognition methods mainly include dynamic time warping (DTW), vector quantization (VQ), fuzzy C-means clustering (FCM), hidden Markov process (HMM), artificial neural network (ANN) and various algorithms. These algorithms need to manually extract the sound signal feature vector as input for signal classification and recognition. Commonly used sound signal feature vectors include power spectral density (PSD), Mel cepstral coefficient (MFCC), linear predictive cepstrum These eigenvectors have a good degree of discrimination between some human voice signals, but they do not have a high degree of discrimination for certain sounds in the pig house, such as coughing, screaming and gnawing metal. The classification accuracy is not high when used as a feature vector

These commonly used cough sound recognition methods have a good classification and recognition effect in obtaining pig cough sounds in a laboratory environment, but in intensive large-scale farms and complex breeding environments, due to relatively large noise interference, the recognition effect is poor

Images