Abstract: In response to the shortcomings of the two-step difference method for extracting seismic microwave anomalies proposed in 2011, an improved two-step difference method which incorporates hierarchical clustering and wavelet analysis is proposed to overcome the problems resulted from ignoring the brightness temperature differences caused by ground cover types and the lack of robustness of the brightness temperature background field in the original method, thus making the anomaly results more reasonable in terms of spatial pattern and anomaly amplitude. The method is applied to the microwave anomaly extraction of the M_S7.4 earthquake in Maduo County, Qinghai Province on May 22, 2021. The results show that a NE-trending brightness temperature increase strip with a length of about 900 km and an amplitude of 12 K appeared three months before the Maduo earthquake, which has high similarity in spatial distribution and evolution characteristics with the anomaly characteristics of the M_S7.1 Yushu earthquake on April 14, 2010. Furthermore, the appearing time and magnitude of the anomalies are related to the magnitude of the earthquakes. The anomalies of M_S7.4 Maduo earthquake appeared earlier than M_S7.1 Yushu earthquake with larger scale, and the temperatureincrease of microwave brightness due to both earthquakes reached more than 8 K. Preliminary mechanism analysis suggests that the pre-earthquake microwave anomaly stripes are consistent with the spatial distribution of NEward tensional fractures and extensional fracture zones in the region, and the appearance of the anomalies may be related to the geothermal activity of the fracture zones and the emissivity changes of surface rocks before the earthquakes.