With advancement in AI technology, it is very crucial to detect ship images for naval surveillance and maritime security. So, to address challenges of the accurate ship detection capability we can use the emerging technology of deep learning for object detection models, particularly You Only Look Once (YOLO). It achieves very high performance in real time scenario which results in significant popularity of the YOLO for object detection. In this we focused on ship detection using different architectures of YOLO to apply for the aerial ship images detection. We have trained the different versions of YOLO models, evaluated and compared their performances on same aerial dataset. Major performance metrics, namely class loss for detecting the object class, distribution focal loss for the correct identifying of the location of bounding box have been investigated. The results show that the version-10 of YOLO models improves both computational efficiency and the accuracy of detection.