Enhanced Small Drone Detection Using Optimized YOLOv8 with Attention Mechanisms is an open-access journal research article by Fatin Najihah Muhamad Zamri, Teddy Surya Gunawan, Siti Hajar Yusoff, Ahmad A. Alzahrani, and Mira Kartiwi.

The growing misuse of drones presents significant safety and security risks, including the illegal transportation of prohibited goods, interference with manned aircraft, and threats to public safety. These concerns have escalated due to the increased use of unmanned aerial vehicles (UAVs), mainly because of their small size. To address these issues, substantial investigação has been conducted to develop effective drone detection systems. Deep learning, especially through the YOLO framework, is recognized for its lightweight architecture and real-time detection capabilities.

Attention mechanisms have also shown great promise in object detection across various studies. This research focuses on optimizing the YOLOv8n model by integrating an Attention Module into the neck and enhancing the detection head with an additional tiny detection head, thereby improving the model’s efficiency in detecting small objects. Multiple training sets were used to identify the most effective model configuration, incorporating different attention modules such as the Convolutional Block Attention Module (CBAM), ResBlock CBAM, Global Attention Mechanism (GAM), and Efficient Channel Attention (ECA). The results indicate that the optimized model, YOLOv8n + ResCBAM + high-resolution detection head—referred to as P2-YOLOv8n-ResCBAM—significantly improved mean Average Precision (mAP) from 90.3% to 92.6%. Although this enhancement in model complexity reduced the frames per second (fps) from 263 to 166, the detection speed remains adequate for real-time applications. The proposed model effectively distinguishes drones from birds and accurately detects them at long distances, showcasing its potential to strengthen aerial surveillance and security measures.

Data de publicação– July 2024

Enhanced Small Drone Detection Using Optimized YOLOv8 With Attention Mechanisms contains the following major sections:

  • Introdução
  • Visual Drone Detection System
  • Experiment Setup and Data Analysis
  • Experimental Results and Discussion
  • Model Validation and Deployment Result
  • Conclusion and Future Works

Post Image- The evolution of the YOLO family and a simplified design of the YOLO mechanism (Post Image Credit: Authors)