Bird sound recognition method based on multi-feature fusion and combination model

Abstract

The invention discloses a bird sound recognition method based on multi-feature fusion and a combination model. The method comprises the steps: carrying out the preprocessing of a read original bird sound audio, and carrying out the pre-emphasis and framing windowing; extracting four characteristics of the Mel-frequency cepstrum coefficient, the Mel-filtered energy coefficient, the short-time zero-crossing rate and the short-time spectrum centroid of the bird sound, respectively normalizing the four characteristics, and longitudinally splicing the four characteristics to form a fusion characteristic; drawing an STFT speech spectrogram; respectively inputting the fusion characteristic and the drawn STFT speech spectrogram into two constructed CNN models based on an Inception module for training, splicing probability arrays output by the two models to form a characteristic array after training is completed, taking the characteristic array as input of an ANN model for training, and loading optimal parameters of the three models after training is completed; and inputting any to-be-detected bird sound audio into the three models loaded with the optimal parameters to obtain a bird sound recognition and classification result. According to the invention, the characteristic difference between different bird sounds can be improved, and the bird sound recognition accuracy is increased.

Description

technical field [0001] The invention relates to a bird voice recognition method based on multi-feature fusion and combination models, and belongs to the technical field of bird voice classification and recognition. Background technique [0002] Birds are an important part of the natural ecosystem. Because they are very sensitive to changes in the ecological environment in which they live and are easy to be observed and studied, the monitoring and identification of birds is helpful to the monitoring of the ecological environment. The development of protection work is of great significance. Bird monitoring is an important field of research at home and abroad. Traditional bird monitoring mainly depends on the differences in the morphological characteristics of birds. In terms of hearing, birdsong also contains unique features, and has a wide range, Stability, low interference and other advantages, so the research on bird voice recognition is particularly important. [0003] T...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem that the extracted feature is single in the existing bird voice recognition method and the representative bird voice characteristics are insufficient, the present invention provides a bird voice recognition method based on multi-feature fusion and combination model, which uses fusion features instead of single features , so that the feature differences between different bird voices are greater and easier to be distinguished, and the combination of three neural network models is used for recognition, which improves the accuracy of bird voice recognition

Images