Identification of Abnormal Weather Radar Echo Images Based on Stacked Auto-Encoders

Abstract

Due to various factors, including climate, hardware failure etc., abnormal radar echoes were brought out, resulting in inconvenience for forecasting and warning applications. In order to improve the efficiency of the anomaly identification in radar echo image, a novel method combining the theory of classical image processing and deep learning was proposed. In the proposed method, firstly, radar echo image was converted to log-polar coordinates for better describing radar echo image rotation; secondly, based on integration projection theory, the radar echo image in the log-polar coordinates was integrated in horizontal and vertical directions respectively to extract the features of abnormal echo images; lastly, deep learning algorithm based on stacked Auto-Encoders was utilized to train and classify the abnormal echo images by the features extracted previously. The experimental results show that recognition rate of the proposed method can reach up to more than 95%, which can successful achieve the goal of screening abnormal radar echo images; also, the computation speed of it is fairly satisfactory.