Hello everyone,
I am trying to make a machine learing model that can be used to tell if a seedling is of good or bad quality based on an image or a movie (so with the help of one or two cameras). A good seedling has one or multiple brown terminal bud at the top of the stem. This terminal bud should not have oped up and continued to grow further (reflux); this can be recognized by a light green color at the top of the seedling. Also the growing shape is important. If the main stem splits into two main stemps, the seedling is bad. Also, the length of the stem and the diameter is important. Finally, the root of the seedling is important. There should be white root tips present, otherwise the seedling is unhealthy. The rootplug should be firm and not missing more than 1/3th of the top of the plug. I was planning on using a decision tree algorithm to classify the seedlings as good seedlings or bad seedlings at the end.
How could I best subract the foreground or remove the background? Do you think I could best use a movie or an image in the input? The seedlings are processed on a conveyor belt and are transported in rows of several seedlings as in the picture below. At the moment I am a little overwhelmed by all the segmentation techniques and do not know which I could use. It is important the computer vision grading is performed at a high speed so the algorithm should not have a long runtime. Also if it is impossible or really time consuming to get all of this information (terminal bud, reflux, basal fork, length, diameter, and root shape), could someone tell me what is easiest to combine?
Thank you for your help!
Reflux:
Split stem (basal fork):
Plug problem (not dense):
0 Comments
Sign in to comment.