Application of Artificial Intelligence on the Image Identification of Icing Weather Phenomena

摘要

Based on field experiments at Nanyue Mountain Meteorological Station and Huaihua National Reference Climatological Station in Hunan Province, the camera images of icing weather phenomena, such as glaze, rime and mixing rime, are collected minutely from January to March in 2018. The convolution neural network technology is employed for modelling and training using the camera images of the icing field experiment at Nanyue station, and the results of identification are examined by the camera images. Furthermore, based on deep learning, the environmental layout requirements of ice accretion image identification are discussed. The main conclusions are as follows. When identifying icing weather phenomena at Nanyue station, the probability of correction (PC) is 99.21%, the false acceptance rate (FAR) is 0.28%, and the probability of omission (PO) is 0.51%. The probability of icing identification increases significantly in the initial stage of ice accretion, while that in the sustained stage is stably around 99.0%, and in the dissipation stage it gradually decreases. False acceptance and omission occur occasionally during the initiation and dissipation stages, the transition period between daytime and night, and the nighttime when the pictures are not clear enough. The test results show that the artificial intelligence identification model established in this paper can extract the key features of icing in different stages of an icing lifetime, and the identification result is good. In addition, the false acceptance and omission can be further eliminated by using the meteorological conditions criteria and judging the consistency of identification. This method can provide important technical support for the automatic observation of icing weather phenomena.