Robust ENF-Based Inter-Grid Geo-Localization via Real-Time Online Multimedia Data

The electric network frequency (ENF) serves as a vital criterion in geographical localization because its frequency fluctuations remain consistent within the same power grid. However, the performance of existing ENF-based audio geo-localization methods is limited when dealing with real-world scenarios, such as short audio durations and noisy environments. Moreover, the size of available ENF data is still small. To address these issues, we propose a novel audio inter-grid geo-localization method utilizing real-time online multimedia data. First, we construct the China-Online-Data dataset using online data, which integrates enhancement and harmonic selection to reduce noise and improve ENF estimation accuracy. Subsequently, we propose an ENF-based Dual-Channel Geo-Localization Network (DC-GLNet), which leverages both time and time-frequency domain information to improve feature extraction and classification performance. Experimental results demonstrate that the proposed method outperforms existing methods, particularly in short audio scenarios, achieving superior accuracy for inter-grid geo-localization.