Bird flight call discrimination using machine learning

Abstract

The development of a software system which can detect and identify the flight calls of migrating birds is reported. The system first produces a spectrogram using a DFT. Calls are detected in the spectrogram using an adhoc combination of local peak-finding and a connectedness measure. Attributes are extracted both globally from the call and from a window moved incrementally through the call. Decision trees are then used to determine the bird species. These decision trees are induced from a training set using Quinlan’s C4.5 system [J. R. Quinlan, C4.5: Programs for Machine Learning, Morgan Kauffman (1993)]. The system has been tested on a set of 138 nocturnal flight calls from nine species of birds [W. R. Evans, personal communication]. Some calls are faint, and interfering insect noise is present in others. Tenfold resampling was used to classify the calls unseen. Seventy-eight percent of calls were identified correctly, 4% incorrectly and 18% were placed in an ‘‘uncertain’’ category. Neural network-based classifiers are commonly used in this general domain and would likely produce similar accuracy, but use of symbolic machine learning offers two important advantages: Training time is linear in the number of examples and the resulting classifier is less opaque. Both significantly ease classifier construction.