<p>Currently, conventional live-line voltage verification in power systems primarily relies on contact-based methods, which introduce safety hazards and exhibit low operational efficiency. Non-contact alternatives, such as infrared and ultrasonic techniques, are highly susceptible to environmental interference, thereby limiting accurate voltage measurement. This paper presents an airborne non-contact voltage detection algorithm based on millimeter-wave radar (MMWR) and a deep spatiotemporal network (DSTN). The proposed method enables high-precision voltage measurement through electromagnetic field inversion and dynamic signal modeling. First, using unmanned aerial vehicle (UAV) power inspection platforms, an MMWR-electromagnetic field coupling model is established. A multi-scale spatiotemporal feature fusion network integrating convolutional neural networks (CNN) and long short-term memory (LSTM) networks is designed. Subsequently, a lightweight embedded deployment scheme is developed for practical implementation. Experimental results indicate voltage measurement errors below 2% across the 10–100&#xa0;kV range. The system’s interference resistance substantially exceeds that of infrared (&gt; 15% error) and ultrasonic (&gt; 10% error) methods. Key technical advances include enhanced dynamic adaptability (stable performance under wind speeds up to 15&#xa0;m/s) and real-time processing capability (with inference latency of 50 ms per frame). This technology offers a core solution for UAV-assisted grid inspection, which helps reduce operational costs, improve grid reliability, minimize outage risks, and support the development of next-generation power systems, thereby advancing intelligent grid maintenance practices.</p>