This research develops a methodology to automate the process of counting the number of granules that remains in a toilet after being flushed (ASME A112.19.2-2018/CSA B45.1-18). This work integrates a U-Net convolutional network with a variation of the connected component algorithm. The training set consisted of 3678 images. Results show an accuracy above 98% between 0 and 180 granules. The methodology has been implemented in the production line.

This research develops a methodology to automate the process of counting the number of granules that remains in a toilet after being flushed (ASME A112.19.2-2018/CSA B45.1-18). This work integrates a U-Net convolutional network with a variation of the connected component algorithm. The training set consisted of 3678 images. Results show an accuracy above 98% between 0 and 180 granules. The methodology has been implemented in the production line.