Power grid harmful bird chirp recognition method based on VGGish transfer learning network

Abstract

The invention discloses a power grid harmful bird chirp recognition method based on a VGGish transfer learning network. The method comprises the following steps: firstly, establishing a power grid harmful bird species audio library according to bird species information of historical bird-related faults and a survey result of bird species around a power grid, then carrying out pre-processing such as framing, windowing, deep learning noise reduction and clipping on a chirp signal, calculating a chirp signal spectrogram, mapping the chirp signal spectrogram to a 64-order Mel filter bank to obtain a Mel spectrogram, and finally, and taking the Mel spectrogram as the input of the network. In order to solve the problem that a traditional chirp recognition model is weak in generalization ability due to the fact that the number of samples is insufficient, a transfer learning method is adopted, 128-dimensional chirp VGGish features are extracted through a VGGish network pre-trained on an Audio Set data set, dimensionality reduction is conducted on the features through a principal component analysis method, and finally the transfer features are recognized through a classification network. According to the method, different bird species can be effectively identified, and the accurate prevention and control of the bird interference fault of the power grid can be realized.

Description

technical field [0001] The invention relates to the field of power transmission lines, in particular to a method for identifying the song of harmful bird species in a power grid based on a VGGish transfer learning network. Background technique [0002] There are many kinds of birds that are often active around the power grid. Different birds have different habits, so the types of faults caused are also different. Bird-related faults mainly include four types: bird droppings, bird nests, bird pecks and bird body short-circuiting. Types of. In order to ensure effective prevention and control of power grid fault tripping caused by bird activities, it is necessary to take corresponding prevention and control measures according to different types of birds and bird-related fault types. The knowledge of the birds in the surrounding activities is extremely scarce, which makes it difficult to realize the precise prevention and control of bird-related faults. Therefore, it is necessa...

AI Technical Summary

Problems solved by technology

Due to the low dimensionality of traditional feature parameters, the ability to express bird song features is insufficient, resulting in traditional bird song recognition algorithms that can only identify a small number of bird species

With the development of computer vision technology, the visualization of bird song signals is realized by converting audio signals into time-spectrograms, and the time-spectrogram is used as a feature combined with convolutional neural networks to identify bird songs. The convolutional neural network model for recognition requires a large number of bird sound samples, and it is difficult to obtain bird song signals of endangered bird species in the power grid, resulting in unsatisfactory recognition results

Images