Abstract: To address the challenges of low extraction accuracy and high dependence on manual intervention in obtaining the pumping stroke frequency of reciprocating drilling pumps under strong noise and complex modulation environments, an adaptive extraction strategy integrating the Hilbert envelope demodulation, sequence autocorrelation analysis, and automatic multiscale-based peak detection (AMPD) is proposed in this paper. The method follows a progressive processing framework of “demodulation characterization–feature enhancement–adaptive extraction”. Firstly, the down-sampling threshold is set to 5000 Hz following the principle of effective carrier band coverage to balance computational efficiency and feature fidelity. Hilbert envelope demodulation is adopted to preliminarily remove the modulation characteristics of the signal. Subsequently, autocorrelation analysis is introduced to suppress non-periodic noise and enhance weak fundamental frequency components. Finally, the AMPD algorithm is utilized to achieve automated and precise identification of the pumping frequency. Multi-scenario field experiment results show that the proposed method has excellent robustness and generalization ability. In field tests involving BW-250, CS-10-800, and F-1600HL pump models under various working conditions, the method achieves high extraction success rates of 93.33%, 100%, and 91.67%, respectively. The results confirm that compared with spectrum analysis, cepstrum analysis and traditional autocorrelation methods, the proposed algorithm exhibits superior accuracy under low signal-to-noise ratio and variable working conditions, providing a reliable frequency benchmark for the precision diagnosis of the drilling pump hydraulic end.