نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشکده برق/ دانشگاه جامع امام حسین (ع) / تهران/ ایران
2 دانشگاه جامع امام حسین (ع)
3 دانشکده برق / دانشگاه جامع امام حسین (ع) / تهران/ ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The YOLO object detection algorithm is one of the most famous and widely used algorithms in the field of computer vision. This algorithm learns from abundant training images and can achieve remarkable performance in target detection. However, in special conditions such as military, medical, and hazardous environments where acquiring real target images is time-consuming and challenging, there are limitations that researchers seek alternative solutions to address the data scarcity issue and improve the accuracy of the algorithm. One innovative approach to solving this problem is the use of artificially generated data produced by game engines. This method reduces the reliance on real data and improves the performance and accuracy of the algorithm. By utilizing synthetic data, the limitation of the number of training data for the algorithm is overcome, and the algorithm's performance in complex and dangerous conditions is enhanced. Additionally, the use of synthetic data leads to a reduction in time, human resources, and costs required for data collection. In general, the use of synthetic data as an alternative solution in training object detection algorithms has gained attention from researchers and users due to advantages such as time reduction, reduction in human resources, and cost reduction. To achieve desirable performance in tank identification, artificially generated data produced by the Unreal Engine game engine has been used. The generated dataset has been tested on four versions of the YOLOv5 network, namely YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x, and their results have been reported.
کلیدواژهها [English]
)