THIN-LINE DETECTION IN METEOROLOGICAL RADAR IMAGES USING WAVELET TRANSFORMS

The enhancement of thin-line features in meteorological radar reflecti vity images is addressed using a wavelet-based analysis. Thin-line fea tures in reflectivity correspond to surface wind convergence lines tha t can potentially lead to the initiation of thunderstorms. The automat ed detection of thin-line features is desired as input to expert syste ms being developed for automated thunderstorm nowcasting and as aids t o human nowcasters. Any automated detection system requires enhancemen t of the thin line feature as a preliminary step to classifying the fe ature. 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 scal es and orientations and for all possible positions. The resulting wave let transform images are reduced to a single enhanced image through a combination of fuzzy thresholding and averaging at each pixel. Each pi xel in the enhanced image has an intensity proportional to the potenti al that the pixel lies on a thin-line feature. The projection informat ion itself, or the resulting single enhanced image, may be passed to a n expert system or neural network to complete the feature identificati on.