The Good, The Bad, and The Ugly: Machine Vision Based Robotic System for Sorting & Saving Imperfect Produce

Close to 30 percent of fresh produce is wasted because they do not meet strict cosmetic standards, even though they are perfectly edible, and retain full nutritional value. Hence, eliminating wastage of imperfect produce in the food supply chain is a sustainable means for increasing food supply for a growing global population, without increasing the climate stressors from agriculture. This can be achieved through automated sorting of good, bad and imperfect produce using advances in machine vision and robotics. In this project I design, develop, and prototype such a robotic system for using machine vision to sort fruits and vegetables into good, bad and imperfect categories. The Intel RealSense camera has been used for capturing produce images, and the YOLO-3 deep learning object detection model has been used for produce classification. A crowdsourcing app has been developed for clicking and grading of produce images and an open image data set has been published on the Harvard dataverse for use by researchers and practitioners. A DORNA robot was used for its fast speed and a vacuum gripper was selected for gentle handling of produce items.This system has wide applicability across a variety of fruits and vegetables and can play a critical role in saving imperfect produce, as a way to improve food security for a growing global population