Development of a machine vision system to estimate the physical attributes of potato tubers on-the-go at the post-harvest stage

Abstract

This study presents a deep learning and image processing-based machine vision system for sampling and sizing full-size potato tubers on post-harvest conveyors. First, we present a method for sampling fully visible potato tubers running on post-harvest conveyors in the Laboratory and the field, overcoming challenges such as occlusion and varying lighting conditions. This method utilizes Mask R-CNN and image feature-based machine learning models, achieving high sampling accuracy and segmentation quality that averaged over 90% even in field conditions. Subsequently, a machine vision system designed to estimate the size of potato tubers sampled on the post-harvest conveyor is proposed. To validate the efficacy of this proposed system, two distinct methods were employed: static and dynamic conveyor experiments. The outcomes of these experiments revealed a minimum coefficient of determination of 0.77 for the estimation of the minor diameter of the potato tubers when they were in free-rolling motion on the conveyors, regardless of their orientation and spatial arrangement within clusters. Furthermore, the dimension errors observed across all scenarios remained consistently below 10%, affirming the system's accuracy and robustness in size estimation.