基于HU-Net的高分辨率遥感影像道路提取

Road extraction from high-resolution remote sensing images based on HU-Net

针对高分辨率遥感影像背景信息复杂、道路提取干扰因素多的特点,提出一种基于HU-Net的高分辨率遥感影像道路提取方法。该方法首先使用在ImageNet数据集上预训练的ResNet34作为模型的特征编码器,加强模型的特征提取能力,提取更多道路特征。然后,在模型的中间部分增加混合卷积单元连接编解码器,增强模型提取空间上下文信息的能力,保留更多道路的细节特征。最后,使用转置卷积和普通卷积构建模型的解码器。结果表明:该方法在DeepGlobe道路数据集上的IoU和F1分数分别达到了0.6750和0.8060,可以较完整且准确地从高分辨率遥感影像中提取道路,并减少周围地物和树木遮挡对道路提取的影响。

In view of the complex background information of high-resolution remote sensing images and the many interference factors of road extraction,a road extraction method based on HU-Net for high-resolution remote sensing images is proposed.This method first uses ResNet34 pre-trained on the ImageNet dataset as the feature encoder of the model to strengthen the feature extraction capability of the model and extract more road features.Then,a hybrid convolution unit is added to the middle part of the model to connect the codecs,which enhances the model's ability to extract spatial context information,retain more detailed features of the road.Finally,the decoder of the model is constructed using transposed convolutional and ordinary convolutional.The experimental results show that the IoU and F1-score of this method on the DeepGlobe road dataset reached 0.6750 and 0.8060,which can extract road from high-resolution remote sensing images more completely and accurately,and reduce the influence of surrounding ground objects and tree shade on road extraction.