Abstract: The automatic picking method for ground-penetrating radar wave groups at coal-rock interfaces is crucial for achieving online intelligent interpretation of coal-rock strata. Current research often assumes that coal-rock interfaces are smooth surfaces, neglecting the electromagnetic interference effects caused by rough surfaces. This limitation makes such methods unsuitable for high-frequency radar antennas or automatic tracking of coal-rock strata under complex geological conditions. To address this, a new method for automatic picking of ground-penetrating radar wave groups at rough coal-rock interfaces is proposed. This method employs a local extremum approach and an adaptive adjustment method based on local amplitude statistics to automatically extract and filter anchor points of potential coal-rock interface curves. Considering the impact of electromagnetic wave interference on the distribution characteristics of anchor points, a method for nearby growth within anchor point neighborhoods is introduced, using the amplitude coefficient of variation of anchor points as the criterion for terminating curve growth. The method also employs a RANSAC iterative fitting algorithm and waveform feature matching to classify and identify interfering hyperbolas and coal-rock interface curves. The spatial relationship of the vertices of the fitted curves is used to filter out interfering hyperbolas near rough interfaces. For the problem where the local trend of the coal-rock interface manifests as hyperbolas under specific geological conditions, amplitude attenuation models for rough coal-rock interface curves and interfering hyperbolas are constructed. By compensating for amplitude changes of co-curved anchor points while considering electromagnetic wave geometric diffusion effects, false detection of interfering hyperbolas is controlled. Forward modeling and experimental results show that this method can accurately detect the position of rough coal-rock interfaces. Its variable parameters are derived solely from known detection conditions and electrical properties of the rock layers, enabling automatic picking of ground-penetrating radar wave groups at coal-rock interfaces.