Leveraging RF Signal Transformation and Sigmoid Calibration for Optimized Drone Class Prediction

With the increasing use of drones in various sectors, accurately identifying drone types has become crucial for security and surveillance systems. This paper presents a novel RF-based framework for accurately classifying drone types. The proposed approach integrates advanced feature extraction techniques to extract discriminative information from raw RF signals, followed by a sigmoid calibration layer to enhance the confidence and reliability of multiclass predictions. This calibration layer addresses the inherent challenges in multiclass classification by adjusting the predicted probabilities to reflect true likelihoods more accurately. By meticulously analyzing the extracted features and refining the classification models, our goal is to significantly improve the prediction accuracy and robustness of drone-type identification. Experimental results demonstrate the efficacy of our proposed method in effectively distinguishing between various drone classes, thereby contributing to enhanced drone detection and identification systems.