基于U-Net网络的飞机遥感图像分割

Aircraft Remote Sensing Image Segmentation Based on U-Net Series Network

图像分割技术在城镇建设、自动驾驶、医学图像以及遥感图像等领域有着广泛应用。基于此,分割手工标注飞机遥感图像数据集的图像,先预处理图像和标签,再分别使用全卷积网络(Fully Convolutional Networks,FCN)、U-Net、U-Net++网络进行训练。经过对比分析发现,U-Net效果最佳,故选择U-Net作为最终模型。实验结果表明,该模型的mIOU为0.8432,Acc为0.9971,取得了较好的效果,能够有效处理遥感图片信息,实现飞机与周围环境的精准分割。

Image segmentation technology has a wide range of applications in urban construction,automatic driving,medical image,remote sensing image and other fields.Based on this,the images of manually annotated aircraft remote sensing image data sets are segmented.Firstly,the images and labels are preprocessed,and then the Full Convolutional Networks(FCN),U-Net and U-Net++networks are used for training.Through comparative analysis,it is found that U-Net has the best effect,so U-Net is selected as the final model.The experimental results show that mIOU reaches 0.8432 and Acc reaches 0.9971,which has achieved good results.The method proposed in this paper can effectively process remote sensing image information and achieve accurate segmentation between the aircraft and the surrounding environment.