Based On The Monitoring And Analysis Of Weather Radar Digital Pattern Recognition Methods

After ten-year deployment of weather radar monitoring network in China, there are 131 next generation weather radars put into practice nowadays. This monitoring network provides a wide variety of observing product of high spatial and temporal resolution for monitoring, early warning and forecasting of medium and small scale weather systems, and becomes an indispensable tool of weather warning. Because of several causes, such as radar hardware quality and electromagnetic interference, there are so many abnormalities in radar images. These abnormal images will directly affect succeeding radar mosaic and weather forecasting. In view of this reason, these abnormal images are excluded in meteorological operation. At present, manual methods are inefficient for eliminating abnormal weather radar images, and unable to satisfy meteorological operation.In this thesis, an algorithm for automatic detecting abnormal weather radar images is developed. It consists of four modules:preprocessing, edge detection, feature extraction and artificial neural network based classification. In connection with characteristics of weather radar images, a novel, fast and efficient edge detection algorithm is proposed. In the feature extraction module, an improved line detection algorithm based on chain code is proposed to process radar images better, and a circle generation algorithm is used in arc feature extraction. In addition, several algorithms are constructed respectively to extract feature of angle-missing, color, noise and ultra-refraction. Experimental and operational results show that our method can effectively solve abnormal detection of weather radar images.