I started this course with thought of resolving problem from work. The idea was not so simple: recognition and outlining using vector lines aeronautical objects at the airports on satelite maps. This task was hard, because airports are not the same in every region – let’s compare for example Frankfurt and Dubai airport:
As it is obviously visible – objects are photographed under different angles, colors of runways, airways, control tower and even the level of sunlight is different.
At the beggining this problem was unsolvable for me, despite the fact that my speciality during studies was Multimedia and image processing should be easy.
The idea was to use neural networks for this task. My first trials landed into area of geospatial analysis: https://whatis.techtarget.com/definition/geospatial-analysis#:~:text=Geospatial%20analysis%20is%20the%20gathering,are%20applied%20to%20geographic%20models.
and this free github project: https://github.com/mapbox/robosat
I found company ESRI: https://www.esri.com/en-us/maps-we-love/gallery/satellite-map and thought that maybe there would be a solution of this problem. Unfortunately after analysis they refused to take this project.
Then I found this lecture: https://dbc.wroc.pl/Content/2676/PDF/kubik_klas_obr_rastrowych_optima.pdf
And I thought about my thesis superior. After contacting him I found out about AI Bay society, which is lead at my old uni. Every summer they lead „Summer School of AI” for free. It is mainly dedicated for managers and administration, but in fact everybody can attend. I strongly recommend- more info: https://aibay.ai/
Thanks to series of briliant lectures and some more „hobby” research, now I finally understand how neural networks work, what are types of them, what are the most popular Machine Learning models and what could be their appliance. I understand also how to build datasets and how to validate them.
I have some ideas now, how to solve my work’s problem. However the first trials will be of course tested on my blog using different pet projects.