Ideal workflow in RC for this method

I’d work with masks: open the enhanced JPGs in Photoshop, and using “actions” build a custom alpha mask for each image, then save as PNG (PNG can have alpha channel, which RC will use as masks).

Your actions could include: big “rectangular” mask, masking everything except the center of the image (since your cashew nut is always right in the center of the image) and adding to this mask by masking everything in the orange colour range (the putty on which your cashew rests) - I think in Photoshop this is “selection” -> colour range. By making a custom action for this particular dataset, you can build masks in this manner for all your images automatically.