This repository is a fork of https://github.com/AndreaCodegoni/Tiny_model_4_CD This code was used by "dataeaze systems" for training model for changes in construction in a city (Pune), from images satellite imagert from NRSC. These experiments and models are stritcly for non-commercial and research purposes only.

========================================

1. Tiny_model_4_CD contains traning.py and test_ondata.py for tranining and testing. For more detail read readme inside the Tiny_model_4_CD.
2. Models take images of 256 x 256, whereas we have two tif files of pune you can acces them as p1ntif and p2msetif (https://drive.google.com/drive/folders/1NmEFHYcYTcXpGoOTX5ufm-Z-7xkZaeJ9?usp=drive_link).
3. Keep only intersection area of 2 images (Before and After) using qgis.
4. Use pngconv.py to convert into png.
5. If required do histogram matching using histogramMatch.py.
6. Do image regestration using image_reg_msecalc.py (Avoid step 3, 5and 6 if you are using gdrive p1ntif and p2msetif as they are already preprocessed).
7. Use split.py to split them into 256 x 256.
8. Train Model and get results.
9. Merge result using resmerger.py.
10. Convert result into tif by png2georef.py.
11. For reverse geocoding reversegeocode.py can be used.
12. Desktop application can created using qttut2.py also high level web page can made using qgis2web plugin.