Imbalance-Aware MobileNet for Automated Classification of Radio Technologies in Spectrograms

This work investigates the automatic classification of radio access technologies from spectrograms using the MobileNetV3-Small architecture. To address the large variability in image size, a spectrogram partitioning strategy is adopted to preserve relevant time-frequency patterns while enabling compatibility with the classifier input. In addition, the impact of class imbalance is analyzed through the use of a weighted loss formulation. The results indicate that the proposed approach achieves effective classification performance with low computational cost, making it suitable for resource-constrained scenarios while improving the recognition of underrepresented classes.