<![CDATA[Monitoring of radio frequency spectrum is an important activity in managing frequency usage to prevent interference between users. This study proposes a deep learning-based radio frequency monitoring system using SDR# spectrum images with RTL-SDR receiver implemented in MATLAB. The system integrates GoogLeNet with transfer learning to extract visual features from spectrum images, Autoencoder to reduce feature dimensions to a 32-dimensional latent space, and K-Means algorithm to classify spectrum conditions into Cluster A (good signal), Cluster B (moderate signal), and Cluster C (weak signal). AM/FM signal modulation identification is performed using the OCR method on the SDR# image header with FM frequency range 87-108 MHz and AM 109-137 MHz. The system was tested at 20 measurement points in 4 cardinal directions with distance variations of 2 km to 10 km using 909 SDR# spectrum images. Test results show an OCR Success Rate of 16.4% as a measure of frequency text-reading success, indicating that SDR# is not optimal as an image source in this approach. Among the images successfully recognized by OCR, signal modulation classification based on frequency range achieved 100% accuracy, precision, and recall as a measure of classification correctness. The system also produced spectrum condition clustering using K-Means (k=3) with a distribution of Cluster A (56.8%), Cluster B (30.4%), and Cluster C (20.7%), and RF Amplifier recommendation based on FM signal dominance. Future research is recommended to use GQRX to improve system performance.]]>
Copyrights © 2026
