Deep Feature Fusion from Spectrogram and Scalogram for Enhanced RFI Classification

Abstract

Radio Frequency Interference (RFI) poses a significant challenge to the reliability of modern wireless communication systems. This research introduces a hybrid deep learning-based framework that utilizes spectrogram and scalogram transformations to detect and classify RFI with high accuracy. Raw signal data is preprocessed and converted into time-frequency images, which are then analyzed using pre-trained networks—ResNet50 for spectrograms and AlexNet for scalograms. Extracted features are fused to form a comprehensive representation of each signal sample. A Convolutional Neural Network (CNN) integrated with an attention mechanism is employed for final classification, allowing the model to focus on critical features. The proposed method achieved an impressive accuracy of 98%, outperforming traditional detection techniques. The Emperor Penguin Colony (EPC) algorithm is also used for optimal hyperparameter tuning. The approach demonstrates robustness against complex interference patterns and adaptability to diverse RFI sources. These findings indicate strong potential for application in wireless networks, satellite communications, and radio astronomy.