Recognition Of Bird Species Based On Song Processing And Machine Learning For Faulty Birds On Transmission Lines

According to the statistics of power grid related departments on transmission line failures,the activity of the birds is the main cause of failure.At present,various bird-proof devices have been widely used on transmission lines,but the main reasons for failing to contain bird-related failures are the diversity of bird-related failure types and blind installation of bird-proof devices,which do not have targeted measures.In response to such problems,it is necessary to carry out intelligent identification of relevant bird species on transmission lines.In this article,traditional manual feature extraction and deep learning are used to classify and identify bird species related to bird-related faults.The main research contents are as follows:(1)According to the statistical results of the power grid,construct a sample set of the singing signal of bird-related faults that harm bird species,perform preprocessing such as framed,windowed,and pre-emphasis on the singing signal,and use the improved spectral subtraction method of multi-window spectrum estimation to analyze the singing signal Perform denoising,and perform endpoint detection through the dual-threshold method.(2)For the pre-processed song signal,extract the Mel cepstrum coefficient and power spectral density value as the feature vector,and then use the Gaussian mixture model and the random forest model as the classifier to detect 16 bird species related to bird failure,and then perform classification and identification.(3)Build a database of song spectrograms of 30 bird species related to bird-related failures,migrate the AlexNet model based on model migration,share the model parameters pre-trained on the ImageNet dataset,and then fine-tune the network model to extract language the characteristics of the spectrogram and classification and identification of 30 bird species.