I was able to test the algorithm on a large set of images from 2 different cage setups. The results are mixed but generally positive. Most of the waste pellets are detected by the segmentation algorithm however dark spots, shadows, and the exposed cage floor still prove to be challenges. I am shifting the focus of development to creating methods to help filter out these false positives.

In parallel, we are developing a validation setup that includes a separate higher definition camera to help visually verify the presence of detected waste pellets. This setup will only be used to evaluate the module's accuracy using the different perspective and higher definition camera to more easily view the cage floor.