Use of Technical Vision for Automatic Separation of Defective Potato Tubers

Abstract

The paper describes a mock-up sample of an automatic sorting machine designed to recognize external defects of potato tubers and their automatic inspection by compressed air jet. The basic requirements for vision and automatic inspection systems for the separation of substandard potato tubers are formulated. The recognition process consisted of three main modules: segmentation, tracking of potatoes moving in the frame on the conveyor belt and classification using a trained artificial neural network. For real-time segmentation of potato tubers against the background of a conveyor belt, a method based on color threshold calculation was used. Centroid tracking algorithm was used for tracking moving potato tubers. To train the artificial neural network, a custom dataset consisting of images of marketable and defective potato tubers was created.