RD-YOLO Target Detection For Night Images

Due to the problems of blurred background, dim light and less feature details in night images, the accuracy of visual object detection is reduced and the difficulty of visual detection is increased. To solve the above problems, this paper proposes a RD-YOLO target detection algorithm, which combines the feature extraction ability of ResNet network, the image generation ability of DCGAN network and the YOLOv5 target detection ability to effectively improve the night target detection effect. Experimental verification on typical night data sets shows that the RD-YOLO network proposed in this paper can achieve 92% mAP and 270FPS detection speed, which is significantly higher than the original network, and meets the actual requirements of the night scene target detection task.