Research On Weather Classification Of Sunny Day Or Cloudy Day Based On Single Outdoor Color Image

With the continuous development of computer vision technology,the detection and recognition of weather phenomena in outdoor environment have put forward an urgent need.The judgment of weather types for outdoor color images are widely applied in weather prediction,disaster warning,automatic driving and image restoration.Therefore,weather classification of outdoor color images has become an urgent problem in computer vision technology.Based on the comprehensive analysis of the current research situation at home and abroad,aiming at the deficiency of the existing research results of weather classification,the weather related features of outdoor color images are digged out and the relevant theories are made full use of,such as machine learning,etc.The classification of sunny or cloudy weather in outdoor color images was intensive studied.Firstly,the basic concepts of outdoor color images and the related knowledge of weather features are introduced,and the algorithms of machine learning which this paper used are briefly introduced,mainly including supervised and unsupervised learning,feature selection and its main process,random forest classifier and K-means related theoretical knowledge.Secondly,a weather classification of sunny or cloudy method for outdoor color images based on random forest is proposed.First,by deeply excavating weather correlation feature of outdoor color images,the sky frequency histogram feature,shadow energy feature and transmittance feature is proposed.Then Fisher-RF feature importance calculation method is put forward.Through screening the existing weather related features,the features with high importance and low feature dimension are selected to form the weather feature.In the last,the weather feature is input into the random forest classifier to realize the weather classification.Thirdly,by researching the characteristics of unsupervised clustering,a weather classification of sunny or cloudy method for outdoor color images based on unsupervised clustering is proposed.First,feature is reduced by using variance and covariance matrix.Then the selection of K-means clustering center point is improved accordingly.In the last,the weather classification is realized by inputting the reduced dimension feature matrix into the improved K-means algorithm.Finally,the proposed method is compared with the existing classification algorithms in detail.