How to use ratio of pixel to recognite object

I have a binary image as below. I want to take out the green object and eliminate the rest. I have already tried to eliminate some object using "regionprops":

- size of object (200<x<6600 pixel)

- MajorbyMinorAxis (1.25< x < 1.75)

Therefore that properties will not work.

I have an idea that using a bounding box for each object and count the number of pixel of each object in its own bounding box (number of pixel of object/ number of pixel of bounding box).

For example: the object on the top left corner will get around 15% of its bounding box and the green object can make > 75%. Even the size of bounding box for each object is different (compare the top left corner and top right corner object) but the ratio (number of pixel of object/ number of pixel of bounding box) will be similar from 15-30%

But it is just an idea. I don't know how to make the code to make it work and whether it will work correctly.

Can anyone help. but I still accept even if cannot work perfectly but giving an improvement