Enhanced Signal Processing for Echo Detection Using Support Vector Machine, Weber’s Law Features, and Local Descriptors (LDP and LOOP)

Abstract

Removing ground echoes from weather radar images is a critical task due to their substantial influence on the accuracy of processed meteorological data. These echoes often obscure the true atmospheric signals, particularly precipitation, which is essential for weather forecasting and analysis. In this study, we aim to develop advanced methods that not only eliminate ground echoes but also preserve precipitation signals, ensuring accurate meteorological observations. To achieve this, we explore the use of Local Descriptors based on Weber’s Law Descriptor (WLD) and combine it with the Local Binary Pattern (WLBP) descriptor, as well as introducing two novel descriptors: Local Directional Pattern (LDP) and Local Optimal-Oriented Pattern (LOOP). These descriptors are employed to capture various local features and patterns within the radar images that are crucial for distinguishing between ground echoes and precipitation. To automate the classification of these echo types, we leverage Support Vector Machine (SVM) classifiers, which have proven to be effective in high-dimensional pattern recognition tasks. Our proposed methods are rigorously tested at the Setif and Bordeaux sites, allowing for comprehensive evaluation under different weather conditions. The results from these tests demonstrate the effectiveness of the proposed techniques in accurately identifying and eliminating ground echoes while preserving precipitation. Specifically, the integration of LDP and LOOP significantly enhances the ability to differentiate between echoes, improving the robustness of the classifier in challenging environments. The outcomes indicate that these methods show considerable promise for practical applications in meteorological data processing, providing a reliable solution for improving the quality of weather radar data and supporting more accurate weather predictions.