Rapid measurement of soybean seed viability using kernel-based multispectral imaging analysis

Authors: BAEK, INSUCK - University Of Maryland; KUSUMANINGRUM, DEWI - Chungnam National University; KANDPAL, LALIT - Chungnam National University; LOHUMI, SANTOSH - Chungnam National University; MO, CHANGYEUN - Korean Rural Development Administration; Kim, Moon; CHO, BYOUNG-KWAN - Chungnam National University

Submitted to: Sensors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/8/2019
Publication Date: 1/11/2019
Citation: Baek, I., Kusumaningrum, D., Kandpal, L., Lohumi, S., Mo, C., Kim, M.S., Cho, B. 2019. Rapid measurement of soybean seed viability using kernel-based multispectral imaging analysis. Sensors. 19(2):271. https://doi.org/10.3390/s19020271.
DOI: https://doi.org/10.3390/s19020271

Interpretive Summary: Viability is an important seed quality factor that affects seed germination and crop yield. Many existing testing methods for seed viability still rely on time-consuming and sample-destructive measurements. This study developed a viability testing method for soybean seeds based on shortwave infrared (SWIR) hyperspectral imaging (HSI) technique. The method was developed from data extracted from HSI images of two hundred viable seeds and two hundred seeds that were artificially aged to render them non-viable. While the original images contained 256 spectral wavebands, the final classification method was determined to need only seven optimally-selected wavebands for imaging and analysis. Testing of the classification method showed over 95% accuracy in identifying viable and non-viable soybean seeds, suggesting great potential for use of this non-destructive test method by seed producers and distributors to effectively and rapidly assess soybean seed viability.

Technical Abstract: Viability is an important quality factor influencing seed germination ability and crop yield. Many seed viability testing methods still rely on time-consuming and sample-destructive measurements. Therefore, the aim of this study was to detect viable and non-viable soybean seeds by using a shortwave infrared hyperspectral imaging (HSI) technique. HSI measurements were acquired for two hundred samples each of viable and non-viable soybean seeds. Data extracted from the HSI measurements were further analyzed by partial least squares discrimination analysis (PLS-DA) for classification of viable and non-viable soybean seeds. In addition, variable importance in projection (VIP) was used as a waveband selection method to develop a multispectral imaging model. First, the spectral profile of each pixel was subjected to PLS-DA analysis which yielded a reasonable classification accuracy; however, a large number of pixels were misclassified. Therefore, the image processing based method was then adapted with an optimum detection rate strategy to classify the seed viability. Using PLS-DA-based binary images, we successfully distinguished between viable and non-viable seeds; the image-based classification of seeds showed an accuracy >95% even when using only 7 bands (selected by VIP method). Our results showed that using only 7 wavebands for image-based classification and optimum detection rate strategy can be used to classify viable and non-viable soybean seeds with 96% classification accuracy for each. This proposed multispectral image-processing based method may be used as an effective and accurate technique for discrimination of soybean seed viability in a rapid and non-destructive manner.