To improve Model, making data collection faster is imperative. The bottleneck in training data generation pipeline is the rasterization of files. The current approach follows the following steps:
raster full disk image using QGIS (15 - 20 mins)
Load the shapefiles from HMS database onto QGIS
locate where the plumes are coming from
Draw a shapefile or Modify the existing Shapefile
export the shapefile.
In this method, using qgis software is not optimal as it cannot process a fulldisk raw file in-memory. So, every scroll through the data in the software may potentially make the software reload the data from the disk. this may slow down the labelling process.
To overcome this:
We can use the info from NOAA shapefiles to narrow down approximate extent of a smoke plume.
We use those co-ordinates to crop and warp (using gdal calls) a smaller section of raw file to create a GeoTiff Raster and store it on disk. (BAND 1 and BAND 3 will be used for this)
Re draw Shapefiles and build dataset using the generated GeoTiff in QGIS. this should be much faster than loading a raw fulldisk.
To improve Model, making data collection faster is imperative. The bottleneck in training data generation pipeline is the rasterization of files. The current approach follows the following steps:
In this method, using qgis software is not optimal as it cannot process a fulldisk raw file in-memory. So, every scroll through the data in the software may potentially make the software reload the data from the disk. this may slow down the labelling process.
To overcome this: