The habitat and reproduction surveys of raptors are conducted for environmental impact assessment in various areas. The purpose of this study was to propose a new cost-effective method of raptors survey which makes a small impact for raptors. A target species of this study was one of the raptors, Northern goshawk (Accipiter gentilis). This is categorized as “Near Threatened” in the Japanese Red List.
We used a system which can automatically classify three call patterns. This system was constructed by fine-tuning the Convolutional Neural Network model using images which are converted from sounds. Then we examined the following three factors for achieving practical use; (1) generalization performance of the system, (2) availability of the method in different places, and (3) optimal distance from the nest to place a recorder. As a result, we found that this method allows us to presume whether they inhabit or not and their approximate reproductive state.