Automatic Seismic Phase Picking Method APP++ based on Deep Learning and Array Strategy and Its Application in Sichuan-Yunnan Region

On the basis of APP method, a further improved algorithm, APP++, is proposed by using U-Net ++ model, which is used to automatically pick up seismic phase on time under the array. The test results show that APP++ model has the same picking ability as APP, but also can avoid misidentification better, and has the ability of cross-region generalization. The method was applied to the fixed seismic network and mobile seismic array in Sichuan-Yunnan region, and the continuous data from 2014 to 2019 were scanned. 73 291 earthquakes were detected, which fit the G-R relationship well, and the complete magnitude was in the range of 1/5～2. The spatial distribution of earthquakes is in good agreement with the faults in this region. The calculated B-value distribution map also shows good agreement with the local fault morphology and stress state. The result show the practicability of APP++ method very well.