SAR SLC groundtruthing program

A program used to process SLC images from netCDF4 format and generate geotiff, as well as groundtruth and generation of chips from SNAP. This repo also contains functions that may not be utilized in the actual ground truthing program.

How to use

The main python script to run when using this program is processnetCDF.py under processimage folder.

config.yaml

The config.yaml file is used to specify the netCDF file locations, shpfile location, output folder location and chip parameters in the following format.

Note: We assume that the program takes the image as range for row and azimuth for column, if the image is not in this orientation set transpose to True

netcdfPathlist:
  - 'data/SAR_CPLX_20190823071330_9.6G_HH_12_pres_2_fdc_246.sar.rgo.sig.nc'
  - 'data/SAR_CPLX_20190823071330_9.6G_VH_11_pres_2_fdc_246.sar.rgo.sig.nc'

shpPath: 'chiptest.shp'
transpose: True
outputFolder: for_training
range_res: 0.7
az_res: 0.7
range_px: 0.1
az_px: 0.1
size: 20
winout: 'hanning'
sat: 2e-2

The first time the program is run, there will be no shpfile detected as the geotiff images need to be generated first. A new folder with 4 subfolders will be created at outputFolder. The 4 subfolders:

• cropped-image: Cropped chips in geotiff and npy formats. Chips numbered in generation order.
• cropped-label: Row and Column coordinates of the mask used to crop the chip relative to the rows and columns of the corresponding geotiff images in large-label. Saved as csv.
• large-image: Full geotiff images from the processing of netCDF data. Naming convention follows naming convention of original netCDF files.
• large-label: Information of each chip from the corresponding geotiff:
  • chipnumber
  • latitude and longitude of chip centre
  • size of chip
  • aspect_n
  • aspect_r
  • grazingangle
  • window
  • rangeres and azimuthres
  • rangepixelsize and azimuthpixelsize
  • 4 corner chip coordinates

Note: A sample chiptest.shp file has been provided in order to test the program. Replace this file with the actual shpfile of the cropped chips when using the program.

Snap

Once geotiffs have been generated, using snap, create a new vector container under Vector -> New Vector Data Container. Create polygons to groundtruth targets, saving them to the vector container created. Within the vector container with masks, fill out the individual chip details.

Note: currently the program only uses the geometry and aspect_n column, the other parameters are either calculated by the program or set in config.yaml. However, the program can be modified to use these inputs by modifying the dictionary in processnetCDF.py.

Once the chips are all masked, export the vector container under Vector -> Export -> Geometry as Shape file. Save the shape file and change shpPath to the location of the shp file. Running processnetCDF.py with the new shp file will process the images and create a new yaml file chips.yaml within outputFolder. This file will contain the parameters of each chip which will be used to crop out the corresponding chips.

Note: Pixel size and resolution are currently fixed for all chips, that means that netcdfPathlist needs to be adjusted to crop chips of different sizes and resolutions. E.g. cropping for X-band chips first, followed by cropping of KU-band chips with seperate resolution and pixel size.