Rapid assessment of corn seed viability using short wave infrared line-scan hyperspectral imaging and chemometrics

Authors: WAKHOLI, COLLINS - Chungnam National University; KANDPAL, LALIT - Chungnam National University; LEE, HOONSOO - Us Forest Service (FS); BAE, HYUNGJIN - Chungnam National University; SEO, YOUNG-WOOK - Chungnam National University; Kim, Moon; MO, CHANGYEUN - Korean Rural Development Administration; LEE, WANG-HEE - Chungnam National University; CHO, BYOUNG-KWAN - Chungnam National University

Submitted to: Sensors and Actuators B: Chemical
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/3/2017
Publication Date: 10/30/2017
Citation: Wakholi, C., Kandpal, L., Lee, H., Bae, H., Seo, Y., Kim, M.S., Mo, C., Lee, W., Cho, B. 2018. Rapid assessment of corn seed viability using short wave infrared line-scan hyperspectral imaging and chemometrics. Sensors and Actuators B: Chemical. 255:498-507.

Interpretive Summary: This research presents a hyperspectral imaging and analysis technique that can be used to nondestructively determine seed viability, which is a critical factor for seed value and for production yields. Of 600 viable corn seeds, half were subjected to microwave heat treatment to render them non-viable while the other half remained untreated, and then hyperspectral shortwave infrared images of all the seeds were acquired. Classification models using three different analysis methods were tested for differentiating viable and non-viable seeds. The three models tested performed with 83% to 100% classification accuracies for the seeds in this study. These results show that hyperspectral short-wave infrared imaging can be the basis for development of a fast and effective method for non-destructive large-scale sorting of corn seeds for viability, which will help producers and supplier ensure seed quality and help farmers maximize product yields.

Technical Abstract: Knowledge of the viability status of seeds before sowing is important to farmers and seed suppliers. However, a myriad of factors can reduce viability of seeds or completely render seeds non-viable during pre- and post-harvest operations. Spectral imaging has shown potential for determining seed viability, because of its ability to non-destructively and rapidly assess the internal conditions of seeds, making it ideal especially for industrial sorting applications. This study was conducted to determine suitable classification model(s) for classifying corn seeds based on their viability using hyperspectral imaging. For this study, half of the corn samples were treated using microwave heat treatment and the other half were not microwave heated. Hyperspectral imaging data from all the samples were then collected using a shortwave infrared hyperspectral imaging system with a spectral range of 1000–2500 nm. Three classification models, linear discriminant analysis (LDA), partial least squares discriminant analysis (PLS-DA), and support vector machines (SVM), coupled with some pre-processing methods, were tested to determine the most suitable among them. The SVM model resulted in the highest spectral classification of up to 100% accuracy for the samples under this investigation. This model applied to hyperspectral seed images produced accurate binary classification images of viable seeds and demonstrated that hyperspectral imaging could be used to accurately classify viability of corn. These results serve as a key step towards development of a fast and non-destructive large-scale hyperspectral-based sorting system for corn viability determination.