Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{181554,
        author = {Rubi Bency .M and Jasmi Priya .J and Dharshana .VS and Ebi T Prabha},
        title = {Transformer Augmented YOLOv8 For Small Drone Detection Using Deep Learning},
        journal = {International Journal of Innovative Research in Technology},
        year = {2025},
        volume = {12},
        number = {1},
        pages = {4503-4507},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=181554},
        abstract = {The increasing misuse of drones poses significant safety and security risks, including illegal transportation of prohibited goods, interference with manned aircraft, and threats to public safety. This has raised concerns about the increased use of unmanned aerial vehicles (UAVs) due to their small size. Addressing these concerns has sparked significant research into developing effective drone detection systems. Deep learning, especially YOLO, is known as a lightweight model that offers real-time detection capabilities. Attention mechanisms have proven effective in many studies for detecting objects. This research focused on optimizing the YOLOv8n-based model by incorporating the Attention Module into the neck and improving the detection head by adding a tiny detection head, making the model work efficiently in detecting objects of tiny size. To obtain the most effective model, multiple training sets have been experimented with involving different types of attention modules, such as the Convolutional Block Attention Module (CBAM), ResBlock CBAM, Global Attention Mechanism (GAM), and Efficient Channel Attention (ECA). Therefore, based on the results, YOLOv8n + ResCBAM + high-resolution detection head, called P2-YOLOv8n-ResCBAM significantly improves the mean Average Accuracy (mAP) from 90.3% to 92.6%. Although the increased model complexity reduced frames per second (fps) from 263 to 166, the detection speed remains suitable for real-time applications. The proposed model effectively distinguishes drones from birds and recognizes them at long distances, demonstrating its potential for enhancing aerial surveillance and security measures.},
        keywords = {},
        month = {June},
        }