In-field machine vision system for identifying corn kernel losses

• Published in: Computers and electronics in agriculture Vol. 174; p. 105496
• Main Authors: Monhollen, Nolan S., Shinners, Kevin J., Friede, Joshua C., Rocha, Eduardo M.C., Luck, Brian D.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.07.2020; Elsevier BV
• Subjects: Accuracy; Clearing; Corn losses; Decks; Field study; Field tests; Headers; Image analysis; Imaging; Kernels; Machine vision; Residues; Vision systems; Machine vision; Imaging; Corn losses
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2020.105496

•A system for removing residue to expose lost kernels was developed.•An optical system and machine learning program was used to quantify lost kernels.•The image analysis program achieved an average kernel detection precision of 0.90.•The combined residue clearing and image systems achieved a system accuracy of 82%.•The system identified significant differences in losses created by deck plate spacing. Losses from the combine corn header result in decreased yield and profit. The development of improved corn headers to reduce losses is hampered by lack of sufficient tools for kernel loss assessment. A loss assessment system was developed that consisted of a residue clearing process to expose lost corn kernels on the ground, and a machine vision image system to quantify the exposed kernels. A mower deck was used to size-reduce and remove residue with minimal kernel displacement. The vision system consisted of an optical system for imaging the ground area and an image analysis program to identify lost kernels. The image analysis corn kernel detection system achieved an average precision of 0.90. A further assessment of system accuracy using random images from additional field tests resulted in an accuracy of 0.91. The combined residue clearing and machine vision systems achieved an overall system accuracy of 0.82 in field tests evaluating staged losses using known quantities of kernels. The loss analysis system was able to distinguish statistically significant (P < 0.05) differences in losses created by different corn header deck plate spacing, while requiring less time and labor than conventional assessment methods.