| Bird species identification is a time-consuming, labor-intensive process. However, the resulting information is very important to the work of many environmental conservation and wildlife studies. This study presents an automated method of identifying five common song bird species. Unedited recordings of song birds were automatically pre-processed to normalize signals and remove noise. This resulted in an optimized set of coefficients from a wavelet transform. Six feature extraction methods were used to create a feature vector of each signal. Neural networks were used to identify and classify species. A backpropagation neural network resulted in an 89.33 percent recognition rate. A radial basis neural network resulted in an 88.67 percent recognition rate. |
