Volcanic Deformation Detection using InSAR data

---

To run detection

Usage for testing all interferograms (converted to grayscale and save in png format) in data folder:

DATA_DIR="data/"
OUT_DIR="output/"
MODEL_NAME="models/Model1.pd"

python getProbmap_fn.py --out_dir="$OUT_DIR" --data_dir="$DATA_DIR" --model_name="$MODEL_NAME"

Download models:

• [Paper1] Model1
• [Paper2] Model2

---

To retrain

We trained our models with Matlab. You can download the pretrained model here: model2.mat. Dataset is saved in two folders named 'deform' and stratified' under the main data folder 'data/deform' and 'data/stratified'. For more details, please see comments in runTrain.m.

The new model in mat file can be converted to newmodel.pd to use with Python Tensorflow as follows.

% In Matlab
mode_name = 'new_model'
modeldir = 'results/'
modeldir = 'models/'
load([modeldir, modelname, '.mat']);
exportONNXNetwork(netFineTune, [modeldir, modelname, '.onnx']);

# In Python, read onnx model and convert to pd graph
import onnx
from onnx_tf.backend import prepare

modeldir = 'results/'
mode_name = 'new_model'
onnx_model = onnx.load(modeldir + model_name + ".onnx") 
tf_rep = prepare(onnx_model)  
tf_rep.export_graph(modeldir + model_name + ".pd") 

---

References:

[Paper1] Application of Machine Learning to Classification of Volcanic Deformation in Routinely Generated InSAR Data, N Anantrasirichai, J Biggs, F Albino, P Hill, D Bull Journal of Geophysical Research: Solid Earth, 2018. [https://doi.org/10.1029/2018JB015911]

[Paper1] A deep learning approach to detecting volcano deformation from satellite imagery using synthetic datasets, N Anantrasirichai, J Biggs, F Albino, D Bull Remote Sensing of Environment 230, 2019. [https://doi.org/10.1016/j.rse.2019.04.032]