Abstract: The identification and tracking of coal and rock horizons is an important problem in intelligent coal mining. Conventional coal horizon detection methods have problems such as low accuracy and poor real-time performance. This paper proposes a coal rock horizon recognition and tracking method based on high-frequency air-coupled radar., Compared with the traditional method, the method can realize non-contact detection, and can track and identify the horizon in real time. Firstly, the echo reflection characteristics of high-frequency radar waves in the “air-coal-rock” layer under the condition of antenna suspension coupling are analyzed by forward modeling, and the positioning methods of “air-coal” and “coal-rock” layers are proposed; Secondly, the interface detection accuracy and error under different coal seam thicknesses are analyzed, the precise positioning method of “coal-rock” seam is proposed, and the positional relationship model of the seed layer is established. According to the influence of coal-rock horizon dynamic detection process, the calculation algorithm of coal seam thickness is deduced. Thirdly, according to the “coal-rock” horizon seed point, a coal-rock horizon tracking algorithm with three-level “window operator” as the core is proposed. The fast tracking of coal and rock layers is realized, and the stability of the system is improved. Finally, physical model experiments and field detection experiments are carried out. The results show that the average error in the physical model detection is ±0.12 cm, and the average error percentage is 2.18%. The average error value of detection is ±0.71 cm, and the average error percentage is 3.53%. Based on the 1.2 GHz high-frequency air-coupled radar, the information of coal and rock layers within 1 m can be dynamically obtained in real time under suspended conditions, and the dynamic detection accuracy can reach centimeter level. The results provide technical support for the intelligent mining of coal mines and the dynamic update of transparent geological models.