Aerial Target Recognition Using Neural Network for Enhanced Situational Awareness

Abstract

Accurate and timely identification of aerial targets is critical for modern air surveillance and defense systems. This paper presents a novel approach for aerial target recognition using radar cross section (RCS) signatures enhanced with advanced signal processing and machine learning techniques. Traditional radar-based identification methods face significant challenges due to RCS variability caused by aspect angle changes, atmospheric conditions, and target characteristics. To address these limitations, we propose an integrated framework combining empirical mode decomposition (EMD), principal component analysis (PCA), and artificial neural networks (ANNs) for robust target classification. A synthetic dataset was generated using statistically modeled RCS signatures following Rayleigh distribution patterns for eight fighter aircraft types, with 80 trajectories per aircraft containing 1,200 sequential samples each. EMD decomposes RCS time series into intrinsic mode functions, from which 10 statistical features are extracted. PCA reduces the feature space to fewer principal components while retaining 95% of the variance. An optimized ANN classifier with a compact architecture of only 198 trainable parameters achieves over 98% training and test accuracies. The key contribution is demonstrating effective target classification using synthetic RCS data without requiring classified radar measurements. However, the study relies entirely on synthetically generated data; although the Rayleigh-based models are physically motivated by open-source RCS estimates, validation against empirical radar measurements is necessary to confirm operational generalizability. The proposed framework is computationally efficient and capable of rapid inference suitable for real-time integration; robustness evaluation under realistic noise and clutter conditions remains a key direction for future work.