Machine Learning Based Aircraft Detection using SAR & Optical Images

Abstract

Aircraft detection in remote sensing imagery remains a critical challenge for surveillance and intelligence operations, with single-sensor approaches facing significant limitations that compromise operational effectiveness. This research addresses the integration of Synthetic Aperture Radar (SAR) and Optical imagery for enhanced aircraft detection through a novel computational framework. The study uses TanDEM-X SAR data (X-band, HH polarization) with up to 25 cm resolution in spotlight mode. Optical imagery is sourced from Google Earth Pro, providing high-resolution satellite data from platforms such as Landsat, Copernicus, and commercial providers. Computational models perform well even with small datasets, which is crucial when working with limited pre-paired SAR and optical data. This study proposes a three-phase computational framework that leverages complementary strengths of both sensors. The methodology encompasses aircraft template creation using edge detection and feature fusion, region detection via combined edge maps and SAR saliency, and multi-stage classification using One-Class SVM (OCSVM) and Isolation Forest algorithms. Feature-level fusion is implemented by extracting HOG, LBP, and statistical descriptors from optical imagery, and GLCM, texture, and intensity features from SAR data. The two-stage classification approach utilizes optical-based classification for geometric clarity, followed by SAR-optical fusion and Non-Maximum Suppression for improved reliability. The framework achieved high aircraft detection accuracy, effectively handling challenges like limited data and cluttered backgrounds. This research contributes to the advancement of multi-sensor remote sensing applications by providing a practical, efficient solution for operational aircraft detection scenarios.