A portable method for predicting the germination percentage of sorghum × sudangrass seed using multispectral images

Abstract

The high quality of sorghum × sudangrass [Sorghum bicolor (L.) Moench. × S. sudanense (Piper) Stapf.] seed is an important prerequisite for its application in animal husbandry, and germination percentage is one of the most routine indicators used to test seed quality. This study proposes a method for the rapid and nondestructive measurement of sorghum × sudangrass seed germination percentage based on multispectral image technology. We constructed target region in sorghum × sudangrass seed samples, and after white board calibration and ratio conversion, the spectral reflectance of each group of seeds was obtained at five wavebands. A seed germination test was performed in an incubator, and germination percentage was obtained from 100 sorghum × sudangrass seed samples. Using the neural network and the Levenberg–Marquardt method, spectral reflectance and germination percentage data from the 100 seed samples were used to establish a predictive model of seed germination percentage. The input neurons were reflectance in five wavelength bands, and the output neuron was seeds germination percentage. Experimental data from 80 samples were randomly selected for training, and data from the remaining 20 nontraining samples were imported into the predictive model for simulation verification. The fitting correlation coefficient of the model was .73202, representing the relevant closing degree, and the correlation coefficient between the predicted value and the simulation value from 20 nontraining samples was .7533, which referred to the relationship between variables. The model was able to predict the seed germination percentage with acceptable accuracy. Therefore, the nondestructive method described here may be suitable for rapid detection of sorghum × sudangrass seed germination percentage in the context of seed production.

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.