Single-kernel FT-NIR Spectroscopy For Simultaneously Detecting Cucumber (Cucumis Sativus L.) Seeds At Multiple Maturation Stages

The maturity of the seeds at harvest determines their inherent quality characteristics,such as seed longevity and seed vigor.In the seed industry,these characteristics play a decisive role in seed germination,seed storage and corresponding crop yields.In terms of cucumber seed industry,the output value of cucumber seed production in China in 2014 was1.771 billion yuan,and the output value of cucumber seed industry in greenhouse was 861 million yuan.However,in the actual production of seeds,seed maturity is usually visually estimated by the experience of the grower.This practice has a strong subjectivity,it is difficult to ensure the consistency of seed quality batches,and does not meet the development trend of the seed industry.Near-infrared spectroscopy as an optical fast non-destructive testing technology is in line with the requirements of seed quality testing in the new era.In order to identify and classify the maturity of single cucumber seeds,this paper sets five mature states.Firstly,the nearinfrared spectroscopy acquisition equipment suitable for single seeds was determined.Then,a classification method capable of simultaneously detecting the maturity of multiple cucumber seeds was proposed and its prediction performance was verified.Compared to diode array near-infrared spectrometers,Fourier transform near-infrared spectrometers have higher signal-to-noise ratio and higher spectral resolution.The Fourier transform near-infrared spectrometer could be used to measure the near-infrared absorption spectrum of a single seed by using a suitable single-particle measurement accessory.The average spectrum and the second derivative of the original spectra were obtained to obtain the average spectrum and the second derivative spectrum.The corresponding instrumental analysis results showed that the content of the main components of cucumber seeds fluctuated with the increase of seed maturity.The theory of molecular vibration correlated spectral data with chemical composition analysis results and obtains one of the results: a single kernel nearinfrared spectrometer could be used to identify the maturity of a single cucumber seed.Based on the former conclusion,an exploratory analysis of the sample was conducted to determine the specific classification strategy.By using appropriate spectral pre-processing methods,it was possible to visually see the correlation between different categories of samples in the principal component analysis scores plot.First,in the principal component analysis(PCA)model,the sample residual threshold was used to distinguish the one type of sample with the greatest difference in maturity,and then the three partial least squares discriminant analysis(PLS-DA)binary classifiers were constructed as a hierarchical classification model for predicting samples of the remaining categories.The proposed hierarchical classification model showed satisfactory performance in the prediction of independent external test set samples,and the overall classification accuracy rate reached 98.2%.Finally,in order to better evaluate the effectiveness of this classification strategy,Cohen’s kappa,a multivariate statistical indicator,was calculated with a result of 97.7%.This arrived at the second conclusion: the proposed hierarchical classification strategy was very effective for simultaneously predicting cucumber seed samples of five maturity stages.Seeds of different maturation stages showed significant individual differences in the nearinfrared absorption spectrum.However,bulk sample analysis weakened this difference between individuals and could not reflect the near-infrared absorption characteristics of single kernel seed samples.In this paper,a near-infrared spectrum acquisition method suitable for single cucumber seeds was used.Combined with the results of chemical composition analysis,the effectiveness of this collection method was verified by molecular vibration theory.Furthermore,in order to conduct simultaneous prediction of samples of five different maturation stages,a hierarchical classification strategy was proposed.Based on the prediction results of the independent validation set,calculating a statistical indicator for multi-class evaluation,the results showed that this hierarchical classification model had achieved a very good prediction performance.