Study On Hydrometeor Classification Method Of Dual-polarization Radar Based On Incremental Bayes

As a kind of high-tech,newly developed radar,dual-polarization weather radar can detect the spatial phase,size,orientation and other information of precipitation particles,so it is of great significance in weather detection and weather warning.At present,the algorithm used by dual-polarization radar to classify precipitation particles is mainly fuzzy logic.The fuzzy logic algorithm describes the degree of similarity between radar echo data and a certain type of precipitation particles through a membership function,and then uses rules to judge and classify.However,the membership functions and weight coefficients used in this method all need to rely on expert experience values,which have strong limitations.With the development of machine learning,the Bayesian classification algorithm has been continuously developed for its high efficiency,no need for expert experience,and only the advantages of extracting features from data.However,this method requires a large amount of labeled,high-quality data,and the method is difficult to obtain,and the classifier will not change once the training is completed,and the generalization is poor.Incremental learning effectively alleviates this problem.Incremental learning can add data samples containing useful information to the original data set,and the extracted useful information will update the classifier,so as to realize the process of gradually and dynamically mastering new information while expanding the old knowledge,adding to the training data set a large amount of high-quality labeled data,and the ability to adjust the classifier in time to improve the generalization and accuracy of the classifier is of great significance.Therefore,this thesis proposes a dual-polarization weather radar precipitation particle classification method based on incremental Bayes.The main contents of this paper are as follows:First,it introduces the polarization parameters obtained by using dual-polarization weather radar echo data and its calculation methods,studies the meteorological echo precipitation particle characteristics represented by rain,snow,hail,and studies non-meteorological echoes of the main cause of mid-ground clutter.Secondly,to solve the problems of insufficient number of labeled samples,difficulty in obtaining,and insufficient generalization of the classifier,a precipitation particle classification algorithm based on incremental naive Bayes is proposed.First,the data is discretized,and then a naive Bayes classifier is constructed using the training sample set,and then the classifier is incrementally learned,and finally precipitation particle classification is realized.The naive Bayes classifier after incremental learning can expand the number of samples and dynamically adjust the classifier to more accurately classify precipitation particles and enhance the generalization of the classifier.Third,aiming at the problem of ground clutter in non-meteorological echoes,a precipitation particle classification algorithm based on texture parameter incremental Bayesian Networks(BNT)is proposed.First,the discretized training sample set is used to learn the structure of attribute nodes including texture parameters,then the conditional probability table is calculated,the texture parameter incremental BNT classifier is obtained through incremental learning,and finally the precipitation particles are classified.The texture parameter incremental Bayesian network algorithm can distinguish more ground clutter,and the classification effect is more accurate.