Robust Ensemble Classifier for Advanced Synthetic Aperture Radar Target Classification in Diverse Operational Conditions

Abstract

This study introduces an innovative ensemble technique to enhance synthetic aperture radar (SAR) automatic target recognition (ATR) by integrating the strengths of three diverse classification methods: support vector machine (SVM), template matching, and convolutional neural networks (CNN). The proposed framework leverages the unique advantages of each technique to significantly improve SAR target classification accuracy. CNN, particularly AlexNet, excels in standard operating conditions (SOC), while template matching proves more effective in extended operating conditions (EOCs). SVM, known for handling overfitting in parametric classification, adds robustness to the system. In this approach, binary target regions are extracted and matched to predefined template classes, while 16 target region properties are selected as features and processed through SVM for classification. The final decision is generated using a majority voting technique, ensuring the highest possible accuracy. Experimental validation was conducted using the MSTAR dataset, demonstrating the superior performance of the ensemble method. The combined approach achieved significant improvements in classification accuracy, with average accuracy rising to 90.30% in SOC and 87.22% in EOC-1, outperforming individual classifiers. These results showcase the ensemble's robustness and effectiveness, offering a novel solution for SAR target recognition in varied operational scenarios.