基于纹理和SVM的地面凝结现象观测方法研究  

A Texture-based SVM Observation of Ground Condensation Phenomenon

作　　者：陈留 杨笔锋[1,2] 谢欢 马尚昌[1,2] CHEN Liu;YANG Bifeng;XIE Huan;MA Shangchang(College of Electronical Engineering,Chengdu University of Information Technology,Chengdu 610225,China;The Key Laboratory of China Meteorological Administration,Chengdu 610225,China)

机构地区：[1]成都信息工程大学电子工程学院,四川成都610225 [2]中国气象局大气探测重点开放实验室,四川成都610225

出　　处：《成都信息工程大学学报》2022年第6期622-626,共5页Journal of Chengdu University of Information Technology

摘　　要：地面凝结现象传统观测方法为人工观测,存在观测时效性差和主观性强等问题,影响观测数据质量。通过图像处理技术识别天气现象是当前研究的热点,但是霜露容易受到温度和光照的影响而消失,霜露在图像中的分布特征容易受到干扰,提取特征难度较大,因此该方法在地面凝结现象识别率方面还有较大的提升空间。对此提出一种高清CCD拍照自动识别地面凝结现象的方案,首先提取图片的感兴趣区域,然后将感兴趣区域通过Canny边缘检测提取纹理特征等一系列处理,最终利用支持向量机将结霜、结露、干燥3种现象进行识别。这种方法不受光线影响,可以同时支持在白天和夜晚观测。通过和人工观测对比,最终综合识别准确率为86.5%。The traditional observation method of ground condensation phenomenon is manual observation,which has problems such as poor observation timeliness and strong subjectivity,which affect the quality of observation data.Recognizing weather phenomena through image processing technology is a current popular trend,butdew and frost are easy to disappear due to temperature and light.The distribution of frost and dew in the image is disturbed easily.So,it is difficult to extract features.Therefore,this method still has more room for improvement in the recognition rate of ground condensation phenomena.This article introduces a high-definition CCD camerato automatically identify the ground condensation phenomenon.First,extract the interested region of the picture,then,extract the feature of the interested region by canny edge detection.Finally,identify the three phenomena of frost,dew,and dry by the Support Vector Machine.This method is not only unaffected by light,but also supports observation during the day and night.Compared with manual observation,the final comprehensive recognition accuracy rate is 86.5%.

关 键 词：边缘检测 纹理特征 支持向量机 图像识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] P412[自动化与计算机技术—计算机科学与技术]