سامانه پشتیبانی خدمات اطلاعات

  اخبار و اعلانات

 آمار

تعداد نشریات163
تعداد شماره‌ها6,878
تعداد مقالات74,135
تعداد مشاهده مقاله137,878,665
تعداد دریافت فایل اصل مقاله107,237,393
khazai, safa, Asghari Beirami, Behnam, Rahbari, Hossein. (1404). Evaluation of Deep Learning Methods for Human Detection in Drone Thermal Images. , (), -. doi: 10.22059/eoge.2025.392236.1171
safa khazai; Behnam Asghari Beirami; Hossein Rahbari. "Evaluation of Deep Learning Methods for Human Detection in Drone Thermal Images". , , , 1404, -. doi: 10.22059/eoge.2025.392236.1171
khazai, safa, Asghari Beirami, Behnam, Rahbari, Hossein. (1404). 'Evaluation of Deep Learning Methods for Human Detection in Drone Thermal Images', , (), pp. -. doi: 10.22059/eoge.2025.392236.1171
khazai, safa, Asghari Beirami, Behnam, Rahbari, Hossein. Evaluation of Deep Learning Methods for Human Detection in Drone Thermal Images. , 1404; (): -. doi: 10.22059/eoge.2025.392236.1171

Evaluation of Deep Learning Methods for Human Detection in Drone Thermal Images

Earth Observation and Geomatics Engineering
مقالات آماده انتشار، پذیرفته شده، انتشار آنلاین از تاریخ 15 شهریور 1404
نوع مقاله: Original Article
شناسه دیجیتال (DOI): 10.22059/eoge.2025.392236.1171
نویسندگان
safa khazai* 1؛ Behnam Asghari Beirami2؛ Hossein Rahbari2
1Civil Engineering Dept., Imam Hussein university
2Department of surveying Eng., Faculty of Civil Engineering, Imam Hussein University
چکیده
Today, fast and accurate human detection has found many applications. Using drone imaging technology, significant progress has been made in improving the efficiency of target detection, especially human targets. Detecting humans at night is one of the basic applications of drones with thermal imaging capabilities. Deep learning algorithms offer a modern approach, surpassing classical methods in speed and accuracy.YOLOv8, Faster R-CNN, and RetinaNet algorithms are among the new AI algorithms for detecting human in the thermal spectrum. This study evaluates these algorithms for human detection in drone thermal images, using 2295 images from Roboflow and Kaggle, plus 500 custom-labeled images. We introduce a Balanced Performance Index (BPI) to balance accuracy and speed, achieving BPI values of 0.946 for YOLOv8, 0.709 for Faster R-CNN, and 0.654 for RetinaNet at 90% training split. YOLOv8 outperforms with mAP@0.5 of 0.892 and 30 FPS, making it ideal for real-time drone tasks like search and rescue.
کلیدواژه‌ها
Drone؛ thermal imaging؛ human detection؛ BPI
آمار
تعداد مشاهده مقاله: 43





سامانه مدیریت نشریات علمی. قدرت گرفته از سیناوب