After reading this article, it was obvious for me to use YOLO CCN to detect objects on pic and use MobileNet model to classify them. https://medium.com/nanonets/how-to-easily-detect-objects-with-deep-learning-on-raspberrypi-225f29635c74
I used follwoing Github repo. I have built docker image and tried to run it.
After taking pictures of several types of food in my home I used also this program – Images labeling and annotations:
Unfortunately due to some problems in docker image, incompatible version of TensorFlow-presumably 1.16 and some other missing python libraries, I gave up after 2 days of fight with this approach. After 2 days of little break and watching Korean series (mostly recommended if you got stuck with IT!;), I found another approach to use YOLO CNN and MobileNet model on RaspberryPI, which will be tried soon.
However I found excelent description of YOLO algorithm: