We have developed a 2-stage computer vision model compromising of (1) region proposal using YOLOv5 to identify the liquid formulation in the image (also referred to as “crop detection”) and (2) classification of whether the formulation is stable or not.