To solve the problems of low detection accuracy for remote sensing images in complex scenes with complex background and small and dense objects, an improved YOLOv3 algorithm is proposed in this paper. Based on YOLOv3, our algorithm is combined with dense connection network (DenseNet) and uses the dense connection blocks to extract deep features and enhance feature propagation. Meanwhile, Distance-IoU (DIoU) loss is introduced as the loss function of coordinate prediction, making the location of the bounding box more accurate. Besides, aiming at the situation of mutual occlusion between targets, we use DIoU instead of IoU in the improved non-maximum suppression algorithm to overcome the problem of false suppression. The proposed algorithm is tested on three classical remote sensing datasets, and the experimental results show that the detection method in this paper has higher detection accuracy.