Classification of Precipitation Particles Types Using Images from Precipitation Microphysical Characteristics Sensor

Abstract

To make full use of the images data from a newly designed precipitation micro-physical characteristics sensor (CPMS), a classification method of precipitation particles based on support vector machine (SVM) techniques is presented in this paper. Firstly, a set of descriptors including fall velocity, size, shape, and pixel parameters of precipitation particles is calculated. Secondly, the descriptors of one-minute sample are calculated by the mean values of 16 feature descriptors from all particles in one minute. Thirdly, the proposed classification model identifies the following five types of precipitation particles: small crystal snowflakes, dendric snowflakes, columnar snowflakes, aggregated snowflakes, and raindrops. More than 4,000 images of precipitation particles are divided into a training set with 94 samples and a testing set with 117 samples with 1-min resolution. The results show that the SVM classification model have good performance, the OA and K are 94% and 0.92 respectively, and the OA values of each type are more than 85%. Above results demonstrate the PMCS’s capability to classify the types of precipitation particles, which can be used as an automatic observation system for present weather, water monitoring, etc.