University of Tehran Press Earth Observation and Geomatics Engineering 2588-4352 8 1 2024 06 01 A Comparative Analysis of YOLO and Faster R-CNN Algorithms for Micro-UAVs Detection in Surveillance Videos 101297 10.22059/eoge.2025.389458.1167 EN Safa Khazai Civil Engineering Dept., Imam Hussein university Shahin Mirzaei Imam Hussein Comprehensive University Journal Article 2025 01 26 A micro-unmanned aerial vehicle (micro-UAV) is a type of unmanned aerial vehicle characterized by compact dimensions, typically spanning a few centimeters, and is capable of autonomous operation. Micro-UAVs are employed in a diverse range of applications, thereby presenting potential security threats. Consequently, the initial step in mitigating these threats involves the accurate and rapid detection of micro-UAVs. The advent of artificial intelligence (AI) technology has significantly enhanced the detection and efficiency of micro-UAVs. Among the most prominent AI algorithms for detecting micro-UAVs in the visible spectrum are YOLOv8 and Faster R-CNN. This study aims to compare the efficiency of YOLOv8 and Faster R-CNN, focusing on their trade-offs between detection accuracy and processing speed for micro-UAV surveillance. To this end, the performance of YOLOv8 and Faster R-CNN has been evaluated in terms of detection accuracy and processing speed. The dataset utilized comprises a comprehensive collection of 3,492 images gathered by micro-UAVs during environmental monitoring operations, categorized randomly into three distinct subsets: 70% for training, 20% for validation, and 10% for testing. Experimental results indicate that the YOLOv8 algorithm achieves a true detection rate of approximately 98.6% in detecting micro-UAVs, whereas the Faster R-CNN algorithm attains a true detection rate of approximately 99.6%. Furthermore, YOLOv8 requires an average of 0.03 seconds to process each frame, whereas Faster R-CNN necessitates 2.5 seconds. The comparative analysis reveals that the YOLOv8 algorithm is more suitable for real-time applications and surveillance systems that necessitate rapid image processing due to its significantly higher speed. Conversely, the Faster R-CNN algorithm is a preferable choice for applications where high accuracy is the primary priority, as it offers superior detection accuracy despite requiring more processing time. https://eoge.ut.ac.ir/article_101297_127a7b0e730175859b27c64605a08994.pdf