基于改进的UNet连云港海岸线提取研究
2020-12-23杨坤



摘 要: 精确快速地提取海岸线对于海岸带的开发规划、实时监测等具有重要意义。使用遥感图像进行海陆分割是提取海岸线的一项基础工作。本文利用深度学习技术的语义分割方法来进行海陆分割,提取海岸线。网络模型以UNet为基本结构,并使用残差块来代替其中的卷积层,最后一层再加入一个残差ASPP进一步提升分割精度。使用残差块在加深了网络的同时,提高效率提升了网络的性能。残差ASPP结构能同时捕获遥感图像的光谱信息和上下文信息。在一组来自高分一号遥感影像的实验表明,基于我们提出的网络模型结构与其他深度学习的网络模相比有着较好的结果,能够实现海岸线的提取与应用。
关键词: 遥感图像;深度学习;残差网络;空洞卷积;海陆分割;UNet
中图分类号: TP391.41 文献标识码: A DOI:10.3969/j.issn.1003-6970.2020.10.054
本文著录格式:杨坤. 基于改进的UNet连云港海岸线提取研究[J]. 软件,2020,41(10):208212
【Abstract】: Accurate and rapid extraction of coastline is of great significance for the development planning and real-time monitoring of coastal zone. Land sea segmentation using remote sensing image is a basic work of Coastline Extraction. This paper uses the semantic segmentation method of deep learning technology to segment land and sea, and extract coastline. The network model takes u-net as the basic structure, and uses the residual block to replace the convolution layer. In the last layer, a residual ASPP is added to further improve the segmentation accuracy. The residual block not only deepens the network, but also improves the efficiency and performance of the network. The residual ASPP structure can capture both spectral and contextual information of remote sensing images. Experiments on a set of remote sensing images from gaofen-1 show that the proposed network model structure has better results than other deep learning network models, and can realize Coastline Extraction and application.
【Key words】: Remote sensing image; Deep learning; Residual network; Hole convolution; Land sea segmentation; UNet
0 引言
海岸线是海洋与陆地的分界线,是重要的地物类型之一。我国是一个海洋大国,中国海岸线包括大陆海岸线与岛屿海岸线,横跨了22个纬度带。其中,中国大陆海岸线长达1.8万多千米,岛屿海岸线长达1.4万多千米,总海岸线长度达到了3.2万千米之多。海岸带是实现海洋高质量发展战略要地的关键地区。在这样的条件下,我国有着丰富的海洋资源。由于我国东部沿海发展迅速,人口主要集中在这片区域,国民生产总值也主要集中在沿海区域,所以海岸带经济在中国的经济总量中占有十分重要的地位。……
