Electrical Engineering-Systems Dept.

*** SEMINAR ***

Dani Cherkassky

(M.Sc. student under the supervision of Prof. Hagit Messer-Yaron)

on the subject:

Precipitation Classification Using Measurements from Commercial Microwave Links

Commercial Microwave Network (CMN) have been recently proven to be an effective tool for precipitation monitoring, mainly for accurate rainfall estimation and high resolution rainfall mapping. We address the challenge of precipitation classification, where wet/dry periods are first identified followed by the classification of wet periods into (pure) rain or sleet. Given the measurements of the Received Signal Level (RSL) obtained from spatially distributed wireless commercial microwave links, serving as a sensors network, a basic Linear Pattern Recognition system is implemented to serve as a baseline for classification performance comparison. The rough preprocessing of the RSL measurements in CMN motivated us to inquire a more sophisticated, non-Linear processor for extracting the discriminating features out of the available RSL samples. Principally, the non-Linear processor was implemented by first applying a non-Linear mapping of the input samples into a high dimensional Feature Space followed by Linear Discriminant analysis in this space. Practically, a Kernel based learning technique was used to implement the Kernel Fisher Discriminant (KFD) analysis. We propose several ways to deal with practical challenges that rise on implementing KFD on CMN RSL measurements, including the usage of high dimensional input patterns (RSL measurement from several links) while preventing over fitting of the learning algorithm, and improving the stability of the algorithm by the diagonal loading technique.

The above mentioned feature based non-Linear Pattern Recognition system has been tested with real world data on two different storms that took place in the northern part of Israel (Ortal Mountain) during Dec.2010 and Jan.2012. We can report that classification accuracy of about 93% was achieved.