Birdsong Signal Analysis Research For Species And Behavior Classification

Ecological civilization construction has become one of the national key tasks in recent years,as the decrease of biodiversity and the deterioration of ecological environment.In order to monitor the changes of biodiversity and ecological environment,ecological monitoring is of great significance in the ecological civilization construction.With the development of science and technology and the appearance of various audio sensors,acoustic monitoring methods have been widely used in the field of ecological monitoring.Birds,as an environmental indicator,can reflect the changes of ecological environment and biodiversity,and their vocalizations are an important part of acoustic monitoring data.Therefore,the research on bird vocalizations is of great significance for the acoustic monitoring of ecological environment.Bird species and behavior classification based on vocalizations is one of the current research hotspots,and the related research results not only helps to study the dynamic change and diversity of bird species,but also can be used to research the biological rhythm,living habits and social relations of particular species,which is beneficial to rare species monitoring and habitats protection.Relying on the National Natural Science Foundation of China(61401203),this thesis focuses on the analysis and research of bird vocalizations for species and behavior classification.Firstly,the characteristics of chirping signal are analyzed and the analysis framework of bird vocalizations is introduced.The research focus of this paper is the species and behaviors classification based on bird vocalizations.This paper proposes an automated method for classification based on fusing spectral pattern and texture feature.The spectral pattern and texture features were extracted for each event to characterize bird vocalizations.Then,the ReliefF algorithm was further employed for feature ranking and selection,helping to reduce redundancy and computational demands,and the fused feature set was obtained.Finally,a support vector machine(SVM)algorithm was used for species or behaviors classification.Besides,this paper tested and analyzed the influence of three adjustable parameters(nearest neighbor k,iteration number m and contribution rate)on classification performance.To the different objects,bird species and behaviors databases were established.Bird species databases include North American birds dataset and European birds dataset,each comprised of field recordings of 11 bird species widely spread in North America or Europe.The bird behaviors database is composed of field recordings of four behaviors of Grus leucogeranus in the Jiangxi Nanjishan Wetland Nature Reserve.Compared with features within a category or simple concatenation of a set of features,the method proposed in this thesis based on fusing spectral pattern and texture feature has outstanding and stable classification performance.The thesis has laid a good foundation for the establishment of an automatic analysis system for birds acoustic monitoring and has a good application value in the field of ecological civilization construction.