问题
I'm trying to make a network that identifies missing atoms from an image, and is then able to count them.
So far, I have created a CNN that is able to output an image like below that solely highlights such atoms 1, and I have found an OpenCV library tat I think would be able to count all the individual white spots (https://www.geeksforgeeks.org/white-and-black-dot-detection-using-opencv-python/).
However I would like to be able to count the missing atoms for a given structre: eg I've circled in red the clusters of dots that I would like my network to identify and count as a structure with 8 atoms for example. I have thought the easiest method might be to:
- Apply bounding boxes based on segmentation, crop the image and count the atoms per box,which would provide a labelled image, however a big issue would be the inclusion of additional, non-connected atoms in the box
If this approach was taken, how would I be able to create more specific-shaped boxes? I could try coloring each circle based on the area, and ignoring the other color when cropped to the bounding box size?
Personally I think the question I'm struggling with is how do I segment this image and process each cluster, or is there a smoother technique that I could use or integrate into my ML model?
来源：https://stackoverflow.com/questions/59769220/count-the-number-of-atoms-in-a-cluster