YOLOv3 [24] is an object detector proposed by Joseph et al., and it takes the detection procedure as a regression task. This method increases the speed of detection and accepts input pictures of different sizes. YOLOv3 use Darknet-53[24] for performing feature extraction. Darknet-53[24] is much more powerful than Darknet-19[22] but still more efficient than ResNet-101[31] or ResNet-152[31]. YOLOv3 uses multi- scale prediction, which means it is detected on multiple scale feature maps. For this reason, the accuracy of target detection is improved