面向地基云自动化观测的天空图像理解

项目名称： 面向地基云自动化观测的天空图像理解

项目编号： No.61271429

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 肖柏华

作者单位： 中国科学院自动化研究所

项目金额： 88万元

中文摘要： 云是大气中热力平衡和水气循环的重要环节，对气候模式和天气预报有着重要影响。目前主要通过地面人工目测和卫星遥感进行云的系统观测，而人工目测方式严重制约着我国跟进世界气象观测自动化的潮流和云资料进入数值模式的深度。本项目以地基云自动化观测的迫切需求为牵引，搜集整理我国典型气候区域的地基云图样本库，从大气代表性、应用需求、机器模式识别的特点等方面研究建立适合地基器测的天空云状分类标准；在已有基础上，以数据为驱动，结合领域先验知识，采用计算机视觉、模式识别和机器学习的手段，研究地基测云天空图像序列中的云检测与提取方法、云状和天空类型识别方法，从而准确获取云图的总云量、低云量、云状和天空类型等重要信息。本项目的研究一方面可以为地基云自动化观测的业务应用提供基础技术支撑，为卫星云图提供验证和融合的数据基础，提高云观测资料的全面性和数据质量；另一方面也为自然目标客观场景的图像理解开辟了一个新的研究领域。

中文关键词： 云状识别；云检测；图像理解；地基云图；

英文摘要： Cloud is one of the most important forces of Earth's heat balance and hydrological cycle. The net effect of clouds, however,is still unknown and it plays important role in climate models,climate predictions and meteorological services.Nowadays satellite remote sensing and ground based visual observation are commonly used, but the cloud data is still unsufficient for meteorological reseach and development. As the urgent demand for automated ground based cloud observation is coming, in this project, we will collect large scale sky image databases with ground based instruments from several typical climate zones, and investigate the specification on representations of cloud and sky types suitable for ground based cloud observation instruments. From the computer vision and pattern recognition's point of view, the analysis and understanding of cloud sky images are targetted to cloud object detection and extraction as well as cloud and sky type recognition. We will develop and evaluate novel cloud image segmentation and recognition models learning from data and field knowledge priors, to get total cloud amount, lower cloud amount, cloud types and sky type from the observed sky image sequences. If successful, the results are expected to provide technical support for automated ground based cloud observation, and data sup

英文关键词： Cloud type recognition；cloud detection；image understanding；ground-based cloud image；