Hi Nikam. My only experience is using eCognition and ArcGIS Pro.

I’ve created this eCognition guide to OBIA – https://gisgeography.com/nearest-neighbor-classification-guide-ecognition/

I haven’t created a guide yet for ArcGIS Pro. But the process includes using the Mean Segment Shift algorithm. From there, you use spectral characteristics to classify objects.

The main disadvantage of Esri is the limited number of statistical properties compared to eCognition. I find eCognition much more superior than ArcGIS Pro.

Let me know if you have any questions about any of these two… Unfortunately, I can’t be much assistance for Jupyter Notebooks

This is definitely a common issue in any type of image classification

I have a problem when its come to classifying informal settlements, since some of the material they are using are exactly the same as those that are used for buildings in CBDs.

See (a few quick random links that are hopefully useful):
1) https://grasswiki.osgeo.org/wiki/Image_classification#Further_reading_on_OBIA_with_GRASS_GIS
2) https://sagatutorials.wordpress.com/segmentationobia-videos/
3) http://wiki.awf.forst.uni-goettingen.de/wiki/index.php/Object-based_classification_(Tutorial)

A Google search will generate many more examples of OBIA through QGIS.

I don’t think that it’s possible to work in QGIS using OBIA. Esri has a new OBIA technique called the “Segment Mean Shift”. But this isn’t open source.

Other than this, there are Python libraries with segmentation modules. For example, the Remote Sensing and GIS Software Library (RSGISLib) is available free of charge – provided freely under a GPL3 license.