I’m having some difficulty properly splitting up the leaf colors from the non leaf colors.
As an example, here is an attempt at converting an rgb image to lab and attempting to threshold using otsu’s method.
I’ll need to keep testing to see if there is a way to improve seperating plants from non-plants in an image.
Looking at the different types of plants being analyzed, it seems like different plants have different types of light that they reflect, resulting in different types of filtering being more effective than others. For now, what I can do is run a pipeline that does all of the options in parallel, since I don’t have a good method for determining whether or not an image is filtered to detect the plants. This would be a lot more manual than I’d like it to be ideally. So far, I’ve had the best luck with peas, which seems to work well with the workflow that I’ve already established.
I made a graph of leaf area over time for the peas images. The graph seems to suffer considerably from noise, so I’m going to investigate ways to reduce the amount of noise in the image.
After working on some numerical methods to remove the noise, these are the graphs that I’ve generated
Here is the path to the script I used to run this