A Technique for Filtering Ground Clutter in Radar Images

This paper deals with the detection and filtering of rain cells extracted from images taken by a weather radar. The intended impact is: on one hand, to extract the correct rain rate contained by these images and on the other hand, to increase the performance of these types of remote sensing devices. This is to automatically identify cloud masses, characterize them and predict their evolution over the time. To analyze this phenomenon, we used 512 x512 pixels radar images taken in the region of Setif in eastern Algeria. Our study consists of eliminating parasite echoes resulting from the ground in particular due to the mountain surrounding the radar, by using textural parameters. For every type of echoes we calculate the parameters of Unser into a 5 $\times $ 5 pixels window by sweeping the entire image. Then we present the histograms of these parameters and determine the discrimination thresholds for each one of them. The simulations results obtained from the local homogeneity, the variance of the differences, the entropy, the entropy of the sums, the entropy of the differences, the energy, the energy of the sums and the energy of the differences are very satisfactory with a probability of difference factor greater than 95%. We have found that the most dominant textural parameters are the Energy of differences and Variance of the sum as the major part of the ground echoes have been successfully removed without altering the precipitation echoes. The reason is that for these two parameters, the probability of difference factor was over 98%, which is better that the rest of the parameters.