170 likes | 347 Views
Frog classification using machine learning techniques. Chenn-Jung Huang a* , Yi-Ju Yang b , Dian-Xiu Yang a , You-Jia Chen a a Department of Computer and Information Science b Institute of Ecology and Environmental Education
E N D
Frog classification using machine learning techniques Chenn-Jung Huang a*, Yi-Ju Yang b, Dian-Xiu Yang a, You-Jia Chen a aDepartmentof Computer and Information Science bInstituteof Ecology and Environmental Education Expert Systems with Applications 36 (2009) 3737–3743, ELSEVIER Presenter Chia-Cheng Chen
Outline • Introduction • Architecture of on-line frog sound identification system • Experimental results • Conclusion and feature work
Introduction • An automatic frog sound identification system is developed in this work. • Three features, spectral centroid, signal bandwidth and threshold-crossing rate, are extracted to serve as the parameters forthe frog sound classification.
Architecture of on-line frog sound identification system(Cont.)
Architecture of on-line frog sound identification system(Cont.) • Signal preprocessing • Resampled at 8 kHz frequency and saved as 8-bit mono format • Normalized to the same level
Architecture of on-line frog sound identification system(Cont.) • Syllable segmentation • Amplitude matrix S(a, t), initially n=1 • Find an and tn, such that S(an, tn)=max{S(|a|, t)} • If |an| <=athreshold, stop the segmentation process. The athreshold is the empirical threshold. • Store the amplitude trajectories corresponding to the nth syllable in function An(τ), where τ=tn-ɛ,…,tn,…,tn+ɛ and is the empirical threshold of the syllable.
Architecture of on-line frog sound identification system(Cont.) • Feature extraction • Spectral centroid • Signal bandwidth
Architecture of on-line frog sound identification system(Cont.) • Threshold-crossing rate
Architecture of on-line frog sound identification system(Cont.) • Classification • kth nearest neighboring (KNN) • Support vector machines (SVM)
Architecture of on-line frog sound identification system(Cont.) • kth nearest neighboring (KNN) • The kNN method is a simple yet effective method for classification in the areas of pattern recognition, machine learning, datamining, and information retrieval.
Architecture of on-line frog sound identification system(Cont.) • Support vector machines (SVM) Lagrangian Multiplier Method:
Architecture of on-line frog sound identification system(Cont.)
Architecture of on-line frog sound identification system(Cont.)
Conclusion and feature work • An automatic frog sound identification system is proposed inthis work to provide the public to consult online. • The sound samples are first properly segmented into syllables.