Detection and Classification of Multi-Hop Frequency Hopping Signals in Complex Electromagnetic Environments

Abstract

In complex electromagnetic environments, existing frequency hopping signal detection methods suffer from insufficient adaptability and severe performance degradation under low signal-to-noise ratio conditions. To address this challenge, this paper proposes a detection and classification method based on the connected domain structure of time-frequency images. The method first obtains the signal's time-frequency distribution via Short-Time Fourier Transform. Subsequently, K-means clustering is employed to suppress noise and fixed-frequency interference, enhancing the anti-noise capability. Following this, 8neighborhood connected domain labeling is used to extract timefrequency region features, combined with adaptive morphological filtering to refine the time-frequency image; swept-frequency interference is then filtered out using a dedicated strategy. Finally, FH signal detection and classification are achieved through DBSCAN clustering. Simulation results demonstrate that, under complex conditions with SNR above −6 dB and various interferences, the proposed method achieves a detection probability of 100% and maintains good classification performance even in signal collision scenarios.