Thin-Line Detection in Meteorological Radar images Using Wavelet Transforms

Abstract

The enhancement of thin-line features in meteorological radar reflectivity images is addressed using a wavelet-based analysis. Thin-line features in reflectivity correspond to surface wind convergence lines that can potentially lead to the initiation of thunderstorms. The automated detection of thin-line features is desired as input to expert systems being developed for automated thunderstorm nowcasting and as aids to human nowcasters. Any automated detection system requires enhancement of the thin line feature as a preliminary step to classifying the feature. Enhancement of the thin lines is based on characteristics of a two-dimensional wavelet transform. The reflectivity image is projected onto a directionally selective wavelet basis element for various scales and orientations and for all possible positions. The resulting wavelet transform images are reduced to a single enhanced image through a combination of fuzzy thresholding and averaging at each pixel. Each pixel in the enhanced image has an intensity proportional to the potential that the pixel lies on a thin-line feature. The projection information itself, or the resulting single enhanced image, may be passed to an expert system or neural network to complete the feature identification.