Abstract: Noise detection and suppression of oil well pumping units pose challenges in data processing for mature exploration areas.The conventional method in the industry is to identify pumping unit noise through manual interactions and then suppress it as high-amplitude interference.However,manual identification wastes manpower and yields low detection accuracy,often resulting in missed detections.Hence,based on the noise characteristics of pumping units,this study conducted noise detection on seismic data containing pumping unit noise using deep learning methods.It then estimated the bandwidth of the detected noise using mathematical morphology techniques to determine the final position and distribution pattern of the noise.This allows for adaptive parameter support for the anomalous amplitude attenuation(AAA) method to achieve automatic detection and efficient suppression of pumping unit noise.The processing results of actual seismic data reveal that the methodology used in this study enables intelligent detection of pumping unit noise,significantly reducing the manual effort required for noise identification,improving the detection accuracy,and enhancing the fidelity and robustness of the AAA method.