YuYu1015-PhaseNet-50Hz-268k-v1

50 Hz causal PhaseNet phase picker (P / S / Noise), 268,443 parameters. Distributed in SeisBench format (.json.v1 config + .pt.v1 weights).

🌐 English · 繁體中文

Event #1 zoom (710 – 750 s)	Event #2 zoom (890 – 925 s)

---

English

Model spec

Architecture	PhaseNet U-Net
Parameters	268,443
Sampling rate	50 Hz
Input window	3001 samples (60 s)
Phases	P, S, Noise
Edge blinding	100 samples (= 2 s) each side of the output window

Quick start

from huggingface_hub import snapshot_download
import seisbench.models as sbm
from obspy import read

# Download checkpoint (cached on first run)
local = snapshot_download(
    repo_id="YuYu1015/YuYu1015-PhaseNet-50Hz-268k-v1")
model = sbm.PhaseNet.load(
    f"{local}/phasenet_50hz_ethz", version_str="1")
model.eval()

# Load + resample mseed to 50 Hz
st = read("your_waveform.mseed")
st.resample(50)

# Pick
out = model.classify(st, P_threshold=0.3, S_threshold=0.3)
for pk in out.picks:
    print(pk.phase, pk.peak_time, f"prob={pk.peak_value:.3f}")

Files

File	Purpose
phasenet_50hz_ethz.pt.v1	PyTorch state_dict — model weights
phasenet_50hz_ethz.json.v1	SeisBench config — required for loading

Citation

@misc{yuyu1015_phasenet_50hz_268k_v1,
  title  = {YuYu1015-PhaseNet-50Hz-268k-v1},
  author = {YuYu1015},
  year   = {2026},
  url    = {https://huggingface.co/YuYu1015/YuYu1015-PhaseNet-50Hz-268k-v1}
}

Underlying architecture and toolbox:

Zhu, W., & Beroza, G. C. (2019). PhaseNet: a deep-neural-network-based seismic arrival-time picking method. Geophysical Journal International, 216(1), 261–273.

Woollam, J., et al. (2022). SeisBench - A Toolbox for Machine Learning in Seismology. Seismological Research Letters, 93(3), 1695–1709.

License

Apache License 2.0 — permissive open-source license. Free to use, modify, and distribute (including commercially) provided the copyright and license notice are preserved. Includes a patent grant from contributors; software is provided "as is" with no warranty.

See https://www.apache.org/licenses/LICENSE-2.0 for the full text.

⬆ 回到頂部 / Back to top

---

繁體中文

模型規格

架構	PhaseNet U-Net
參數量	268,443
採樣率	50 Hz
輸入視窗	3001 samples (60 秒)
拾取相位	P、S、Noise
邊緣遮蔽	輸出視窗兩端各遮 100 samples (= 2 秒)

快速開始

from huggingface_hub import snapshot_download
import seisbench.models as sbm
from obspy import read

# 下載模型 (首次執行才下載,之後讀 cache)
local = snapshot_download(
    repo_id="YuYu1015/YuYu1015-PhaseNet-50Hz-268k-v1")
model = sbm.PhaseNet.load(
    f"{local}/phasenet_50hz_ethz", version_str="1")
model.eval()

# 讀 mseed 並重新採樣到 50 Hz
st = read("your_waveform.mseed")
st.resample(50)

# 拾取
out = model.classify(st, P_threshold=0.3, S_threshold=0.3)
for pk in out.picks:
    print(pk.phase, pk.peak_time, f"prob={pk.peak_value:.3f}")

檔案

檔案	用途
phasenet_50hz_ethz.pt.v1	PyTorch state_dict — 模型權重
phasenet_50hz_ethz.json.v1	SeisBench 設定檔 — 載入時必須

引用

@misc{yuyu1015_phasenet_50hz_268k_v1,
  title  = {YuYu1015-PhaseNet-50Hz-268k-v1},
  author = {YuYu1015},
  year   = {2026},
  url    = {https://huggingface.co/YuYu1015/YuYu1015-PhaseNet-50Hz-268k-v1}
}

底層架構與工具包:

Zhu, W., & Beroza, G. C. (2019). PhaseNet: a deep-neural-network-based seismic arrival-time picking method. Geophysical Journal International, 216(1), 261–273.

Woollam, J., et al. (2022). SeisBench - A Toolbox for Machine Learning in Seismology. Seismological Research Letters, 93(3), 1695–1709.

授權

Apache License 2.0 — 寬鬆開源授權。可自由使用、修改、散布(包含商 業用途),條件是保留版權與授權聲明。授權含貢獻者的專利授予,軟體 「as is」提供,無擔保責任。

完整條文見 https://www.apache.org/licenses/LICENSE-2.0。

⬆ 回到頂部 / Back to top