基于高分辨率无人机影像和U-Net网络的乡村地区不透水面提取方法研究
Research on Impervious Surface Extraction Method in Rural Areas Based on High Resolution UAV Images and U-Net Network
余博洋 1秦淑洁 1沈定涛 2李思青 1王结臣3
作者信息
- 1. 南京大学地理与海洋科学学院,江苏南京,210023;自然资源部国土卫星遥感应用重点实验室,江苏南京,210023;江苏省地理信息技术重点实验室,江苏南京,210023
- 2. 华中师范大学城市与环境科学学院,湖北武汉,430079;地理过程分析与模拟湖北省重点实验室,湖北武汉,430079
- 3. 南京大学地理与海洋科学学院,江苏南京,210023;自然资源部国土卫星遥感应用重点实验室,江苏南京,210023;江苏省地理信息技术重点实验室,江苏南京,210023;江苏省地理信息资源开发与利用协同创新中心,江苏南京,210023
- 折叠
摘要
本文综合考虑乡村地区高精度不透水面提取的主要难点,选用无人机(unmanned aerial vehicle,UAV)影像作为数据源,设计了一种融合阴影去除和U-Net网络的提取方案.首先,针对影像中阴影像素较多产生干扰的问题,通过多级Otsu阈值法检测阴影像素,而后应用线性相关矫正方法进行阴影补偿.其次考虑到乡村地区不透水面分布较零散且占比较少的问题,引入U-Net网络模型提取不透水面.对比实验表明,本文方法能有效克服不透水面样本不足和阴影像素的影响,实现乡村地区小微不透水面的高精度提取.
Abstract
In response to the major challenges of high-preci-sion impervious surface extraction in rural regions,we use UAV image as the data source and design an extraction strate-gy that incorporates shadow removal and U-Net network. First,the shadow pixels are identified using the multi-stage Otsu thresholding approach to handle the problem of interfer-ence arising from the high number of shadow pixels in the im-age,and then the linear correlation correction method is ap-plied to compensate for the shadows. Second,considering the dispersed distribution of impervious surface in rural regions and their low proportion,the U-Net network model is em-ployed to extract impervious surface due to the dispersed dis-tribution of impervious surface in rural regions and their low proportion. Comparative studies demonstrate that the ap-proach presented in this paper can successfully overcome the impacts of shadow pixels and insufficient sample to produce high accuracy extraction of small micro impervious surface in rural regions.
关键词
无人机影像/不透水面/阴影去除/U-Net网络Key words
UAV image/impervious surface/shadow removal/U-Net network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据