摘要
全天相机云图是监测云量的重要手段,提出了一种新的云量测量量化指标——云分布密度(Cloud Distribution Density of ASI Images,ASICDD),并基于该指标建立全天相机云图自动分类系统。首先对云图进行去噪,利用最大类间方差法(Otsu)分割云区域;然后对去除背景的云区域图像使用云分布密度计算云量;最后使用4种传统的分类器(支持向量机、K最近邻、决策树和随机森林)根据计算数值进行自动分类并评估各分类器的性能。结果表明,云分布密度可作为评判全天相机云图云量的数值指标;基于云分布密度建立的云图自动分类系统实现了较高的识别准确率,其中随机森林法的分类效果最好,各类云图的识别准确率达到95%以上。
All-sky imager(ASI)images are an important means for the astronomy community to monitor cloud cover at present,but there is no definite cloud cover calculation index.Therefore,this paper proposes a new quantitative index for cloud amount measurement—Cloud Distribution Density of ASI images(ASICDD),and establishes an automatic classification system for ASI images based on this index.Firstly,all the images in the dataset from the Key Laboratory of Optical Astronomy at the National Astronomical Observatories of Chinese Academy of Sciences(CAS)are denoised and the cloud area is segmented from the sky by OTSU algorithm.Secondly,the cloud amount for ASI image with the background remove is calculated by ASICDD proposed in this paper.Finally,we use four traditional classifiers(SVM,KNN,Decision Tree and Random Forest)to classify ASI images automatically according to the calculated value and evaluate the performance of each classifier.The results show that ASICDD can be used as a numerical index for judging the cloud cover of ASI images;the automatic ASI images classification system based on ASICDD achieves a high recognition accuracy rate.Meanwhile Random Forest has the best classification effect—the recognition accuracy rate of various cloud images has reached more than 95%.
作者
张雨昕
邱波
石超君
李梦慈
相冠杰
Zhang Yuxin;Qiu Bo;Shi Chaojun;Li Mengci;Xiang Guanjie(Hebei University of Technology,Tianjin 300401,China)
出处
《天文研究与技术》
CSCD
2021年第3期413-420,共8页
Astronomical Research & Technology
基金
国家自然科学天文联合基金(U1931134)资助.
