Bird individual identifying method based on multi-feature fusion transfer learning

Abstract

The invention discloses a bird individual identifying method based on multi-feature fusion transfer learning. The bird individual identifying method comprises the steps of S1, preprocessing an already-known twittering signal; S2, framing and windowing the preprocessed twittering signal, and using the obtained twittering signal with the fixed frame length to generate a sonogram through a linear frequency modulation wavelet; S3, processing a deep convolutional neural network; S4, inputting the sonogram generated in S2 to the neural network in S3 to obtain feature vectors at different layers, andfusing the vectors to obtain final feature vectors; S5, inputting the final feature vectors to a support vector machine, and obtaining an identifying model after training is conducted; S6, detectingthe performance of the obtained identifying model, and continuously improving the model to obtain a final identifying model; S7, inputting the to-be-tested twittering signal to the final identifying model in S6 to identity the types and quantity of birds after processing the to-be-tested twittering signal according to S1, S2 and S4.

Description

technical field [0001] The invention relates to the technical field of bird recognition, in particular to a bird individual recognition method based on multi-feature fusion migration learning. Background technique [0002] Human activities and changes in nature have direct or indirect impacts on the living environment of birds, which may pose a threat to bird diversity. The number of birds is an indispensable ecological variable in the study of community population dynamics. Selecting it as a quantitative evaluation of habitat suitability can provide a basis for formulating reasonable wetland protection and management plans. Therefore, bird identification is of great significance to the protection of birds and their living environment. [0003] For example, the oriental reed warbler prefers to nest in well-growing reeds. It has particularly strict requirements on the environment of the nest area during the breeding period and is very sensitive to changes in the external env...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a bird recognition method based on multi-feature fusion transfer learning to solve the problems of single research object, poor calculation efficiency and high requirements for field samples in the prior art when recognizing birds

Images