Abstract:Birds are an important part of the ecosystem and play an indispensable role in assessing the state of the ecosystem, so the survey and monitoring of birds are crucial for protecting the ecosystem and preserving biodiversity. Traditional means of bird monitoring consume a lot of workforce and resources, and the accuracy and reliability of the results could be improved. To address this problem, a deep learning-based bird sound data analysis and recognition system is proposed, and an application demonstration is carried out in the Cuihu National Urban Wetland Park in Beijing to verify the performance and accuracy of the system. Using traditional sound signal processing methods, the system first preprocesses the audio captured by the front-end pickups. Then, it extracts audio features and classifies them using residual neural networks to achieve automatic identification of the species’ information contained in the target sounds. During the system’s operation, 200 044 effective bird sound clips were successfully monitored, and its recognition accuracy reached 93%. The system identified a total of 52 species of wild birds in 9 orders and 16 families, among which there are 6 species of national class II wildlife under key protection, namely, Cygnus cygnus, Anser cygnoides, Otus sunia, Athene noctua, Luscinia svecica, and Falco peregrinus; and a total of 22 species of Beijing’s key wildlife under key protection, namely, Ardea cinerea, Nycticorax nycticorax, Phalacrocorax carbo, and Anas platyrhynchos, etc. The system was able to automatically identify the species of wild birds in 9 orders and 16 families. The top 5 species with a higher relative plurality in the monitored audio clips are Ardea cinerea (26%), Anser cygnoides (16%), Nycticorax nycticorax (13%), Pycnonotus sinensis (11%), and Phalacrocorax carbo (8%), in that order. The experimental results show that the system achieves automatic collection and analysis of bird sounds, significantly improves the efficiency of bird monitoring, and provides strong support for the rational planning of landscape gardens, the retention of ecological nodes, and the enhancement of landscape sustainability.