Accurate arrival time picking of microseismic events is crucial for understanding subsurface processes and assessing associated risks. However, the low signal-to-noise ratio (SNR) inherent in field microseismic data poses significant challenges. To address the above issues, this article proposes a novel methodology based on sequence self-similarity and multicenter fuzzy clustering (MFC). Initially, the similarity among subsequences at different time points is leveraged to extract the self-similarity feature from the raw data, attenuating the impact of noise and enhancing the data quality. Subsequently, the fuzzy category is introduced between the microseismic waveform and the noise categories to avoid misclassification caused by the traditional binary picking methods. Finally, temporal information and neighboring information are incorporated to reclassify the fuzzy category, thus improving waveform integrity and the accuracy of first arrival picking. We conducted tests using synthetic and field microseismic data to validate the reliability of the proposed method. The results indicate that the proposed method both outperforms traditional statistical methods in picking accuracy and provides automatic high-quality labels for deep learning model training. This work offers a promising tool for real-time monitoring and early warning systems in various subsurface applications.