The program MTInSAR_convlstm_pixel_selection can be used for selecting PS pixels in multi-temporal InSAR processing.
Input: A stack of interferograms in .mat format (exported from WabInSAR software developed by Manoochehr Shirzaei). WabInSAR software is open access and can be downloaded from the following link:¶ https://sites.google.com/vt.edu/eadar-lab/software
Output: A map with labels 0 and 1, 0 denoting non-PS pixels, and 1 denoting PS pixels
The interferograms are divided into image patches of 100 by 100 pixels and then fed to the network for training. Further, the time series is sampled to a range of 20-30 time steps for better learning.
-
Use requirements.txt to install necessary libraries for the program. Follow these steps:
In your shell (or command prompt)
(i) Go to the directory where requirements.txt is located (ii) activate your virtualenv (if you create a separate virtual envcironment for this program) (iii) run pip install -r requirements.txt
-
The input data for this program can be downloaded from the following link: https://doi.org/10.7294/23478236
-
The example dataset has two files: (i)interferogram stack and (ii) labels (measurement pixel or not). The file ph_im.mat contains time series of interforgrams generated from WabInSAR software v5.3. The dimension is whn, where w=width, h=height and n=number of SAR interferograms. The file elpx.mat contains pixel locations in image (elpx_imloc) and row (elpx_loc) forms for the selected measurement points after the pixel selection step.
Tiwari, Ashutosh; Shirzaei, Manoochehr (2023). Deep learning for efficient selection of measurement pixels in multi-temporal InSAR processing. University Libraries, Virginia Tech. Dataset. https://doi.org/10.7294/23478236
Manoochehr Shirzaei: Conceptualization, input data preparation, validation and funding. Avadh Bihari Narayan: Data preparation, testing and validation for earlier versions of convolutional LSTM model. Onkar Dikshit: Funding support for earlier version