ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

Machine Learning Approaches for Corn Ear Phenotyping with Unwrapped Images

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600855.(doi:10.13031/aim.202600855)
Authors:   Mike Tran, Haibo Yao, Chengkai Li, Yanbo Huang, Ebrahiem Babiker, John Brooks, Dong Chen
Keywords:   corn ear phenotyping, hyperspectral imaging, machine learning, kernel numbers

Abstract. Corn is the number one crop in the United States. The US is also the largest corn producer in the world, representing one-third of world production. Corn breeding is an important process to develop and identify corn hybrids with important traits such as high yield, drought tolerance, resistance to diseases and insects, etc. One essential task in corn breeding is ear assessment after hand harvest of corn ears. This task involves evaluation of traits such as ear shape and size, kernel row number, kernel number, surface distribution of kernels on the ear, to potentially link these traits to genetic markers. However, traditional ways for measuring and recording corn ear traits are manually based, which is subjective, inconsistent, time consuming, and expensive. In addition, it is difficult for a manual approach to record certain information such as kernel surface distribution on the ear. Thus, it is imperative to develop methods to automate the above manual approach for whole corn ear assessment and phenotyping. This research used a previously developed novel hyperspectral imaging system for whole corn ear imaging. The hyperspectral imaging system implemented pushbroom line scanning and was able to scan the entire corn ear surface using a rotational stage. The resulting hyperspectral image of a whole corn ear is a flattened surface image of the ear, which provides a two-dimensional (2D) digital representation of a whole corn ear surface from a three-dimensional (3D) whole corn ear body. Secondly, an automated procedure was developed to visually identify key phenotypic traits, such as the number of kernels, number of rows, and distribution, using machine learning and 2D images. The modified Faster R-CNN model achieved 0.97 accuracy and 0.97 recall in kernel detection. It is expected that spectral information in the hyperspectral images could also be used to identify other pertinent traits such as nutritional and chemical constituents of kernels, insect damage, fungal infection status, and genotypic expression under fluorescence excitation.

(Download PDF)    (Export to EndNotes)