Analysis of deformable and classic convolution in the classification of radio-frequency images

Target identification is one of the fundamental challenges in radar systems, involving the classification of detected objects based on their type (e.g., human, vehicle, or other living beings) and motion state (stationary or moving). Traditional approaches rely on analyzing variations in the target's motion vector parameters. However, advancements in computational power and parallel processing algorithms have enabled the integration of artificial intelligence (AI) methods to enhance identification accuracy and efficiency. While AI-based approaches require extensive and computationally intensive training phases, their inference process is typically fast and scalable. In this work, we propose a machine learning-based target identification framework that leverages convolutional neural networks (CNNs). Specifically, we investigate the application of both classical and deformable convolutional layers to process radio-frequency (RF) images. Deformable convolutions introduce adaptive receptive fields, allowing for enhanced feature extraction and improved robustness to target variations. Comparative analysis of these architectures is performed to evaluate their effectiveness in radar-based classification tasks. Experimental results demonstrate the advantages of incorporating deformable convolutional layers in improving target classification accuracy while maintaining computational efficiency.