Unlocking the Potential of Deep Learning for Migratory Waterbirds Monitoring Using Surveillance Video

A deep learning framework achieves high accuracy (85.59%) and stability (standard error = 0.0004) in automating migratory waterbird monitoring using surveillance video, addressing the challenges of manual analysis in coastal wetlands like Tiaozini, China. The two-step approach—video segmentation into panorama images followed by density estimation via a depth density estimation network—processes entire sites in about one minute, outperforming other deep learning methods and enabling efficient, ecologically meaningful data extraction for biodiversity monitoring.

Wu, E., Wang, H., Lu, H., Zhu, W., Jia, Y., Wen, L., Choi, C.-Y., Guo, H., Li, B., Sun, L., Lei, G., Lei, J., & Jian, H. (2022). Unlocking the Potential of Deep Learning for Migratory Waterbirds Monitoring Using Surveillance Video. Remote Sensing, 14(3), 514. https://doi.org/10.3390/rs14030514