Researchers at the National Space Science Center, Chinese Academy of Sciences, have developed a novel deep learning-based framework for the automatic identification and precise localization of space typhoons. Space typhoons are rare magnetospheric phenomena characterized by cyclonic auroral structures, plasma convection vortices, zero velocity eyes, intense precipitating electron rain, elevated electron temperatures, and upflowing ions. Traditional detection relies on manual analysis of tens of thousands of satellite images, a process that is both time-consuming and prone to subjective error. To address these limitations, the research team constructed a comprehensive dataset utilizing 300,000 far-ultraviolet auroral images captured by the DMSP satellites between 2005 and 2021. The dataset comprises 570 verified space typhoon events as positive samples, paired with an equal number of negative samples, including visually similar but non-typical auroral formations to enhance model robustness. The team evaluated six mainstream deep learning architectures, implementing transfer learning, systematic hyperparameter optimization, and dynamic learning rate scheduling to maximize predictive performance. For spatial targeting, the researchers integrated a YOLOv8-based object detection framework, achieving an accuracy of 0.92, precision of 0.99, and recall of 0.92. Leveraging these results, the team engineered an interactive detection platform capable of ingesting multi-source data, executing real-time analysis, and exporting geolocated findings. This system streamlines operational workflows and supports scalability for future space weather monitoring initiatives. The innovation significantly accelerates space typhoon surveillance, providing a reliable foundation for polar space weather modeling and hazard assessment. The findings were peer-reviewed and published in the journal Space Weather. The project received financial support from the National Natural Science Foundation of China and the Subauroral projects.