BirDrone: a novel dataset for enhanced drone and bird detection using YOLOv9

• Main Authors: Muhamad Zamri, Fatin Najihah, Gunawan, Teddy Surya, Yusoff, Siti Hajar, Mohd. Mustafah, Yasir, Kartiwi, Mira, Md Yusoff, Nelidya
• Format: Proceeding Paper
• Language: English
• Published: IEEE 2024
• Subjects: T Technology (General); T10.5 Communication of technical information
• Online Access: http://irep.iium.edu.my/115182/7/115182_BirDrone%20a%20novel.pdf; http://irep.iium.edu.my/115182/; https://ieeexplore.ieee.org/document/10675527; https://doi.org/10.1109/ICSIMA62563.2024.10675527

Drones present substantial detection challenges due to their capacity to operate in various conditions, including low lighting, harsh weather, and similar objects like birds. Existing datasets frequently fail to address all of these challenges comprehensively. The BirDrone dataset, specifically designed to improve the accuracy of distinguishing between drones and birds, is introduced to address this issue, with a particular emphasis on small-scale objects. The dataset comprises images with intricate backgrounds and lighting conditions to enhance detection reliability. By utilizing the YOLOv9 model to assess the dataset, we achieved a high level of accuracy and significantly reduced the number of false alarms. The BirDrone dataset's development process, data augmentation methodologies, and performance outcomes of YOLOv9 are all detailed in this paper, which serves as a testament to its efficacy in practical applications.