The tests commonly used to determine seed vigor are often laborious and time-consuming; thus, rapid methods are highly required for identifying high-vigor seeds among different batches. In this paper, we describe a novel approach able to distinguishing among batches of soybean seeds of different physiological quality based on their nutrient content measured by laser-induced breakdown spectroscopy (LIBS) assisted by multivariate analysis and machine learning algorithms. These include principal component analysis (PCA), support vector machine learning (SVM), linear and quadratic discriminant analyses (LDA and QDA), and nearest neighbor methods (KNN). A total of 92 measurements, 46 collected from batches marketed as low-vigor seeds and 46 as high-vigor seeds, were analyzed. The SVM method performed the best in discriminating among the batches. In particular, the quadratic SVM function could classify correctly 100% of the high-vigor samples and 97.8% of the low-vigor samples, whereas the cubic function yielded the opposite result; i.e., 97.8% of the high-vigor samples and 100% of the low-vigor samples were classified correctly. The best LIBS spectral region for the analysis was in the range of 350–450 nm, with calcium being the main distinguishing element. Thus, the LIBS technique combined with machine learning classification methods showed a promising potential for classifying soybean seed batches according to their physiological quality.

通常用于确定种子活力的测试通常费力且费时。因此，迫切需要快速的方法来识别不同批次中的高活力种子。在本文中，我们描述了一种新方法，该方法能够通过多变量分析和机器学习算法辅助的激光诱导击穿光谱法（LIBS）测量其营养成分，从而区分不同批次生理品质不同的大豆种子。其中包括主成分分析（PCA），支持向量机学习（SVM），线性和二次判别分析（LDA和QDA）以及最近邻方法（KNN）。总共分析了92种测量值，从以低活力种子销售的批次中收集了46种，以高活力种子销售的批次中收集了46种。SVM方法在区分批次方面表现最佳。特别地，二次SVM函数可以正确分类100％的高活力样本和97.8％的低活力样本，而三次函数得出相反的结果。也就是说，正确分类了97.8％的高活力样本和100％的低活力样本。用于分析的最佳LIBS光谱区域在350–450 nm范围内，其中钙是主要区别元素。因此，LIBS技术与机器学习分类方法的结合显示了根据大豆种子批次的生理质量对大豆种子批次进行分类的潜力。正确分类了8％的高活力样本和100％的低活力样本。用于分析的最佳LIBS光谱区域在350–450 nm范围内，其中钙是主要区别元素。因此，LIBS技术与机器学习分类方法的结合显示了根据大豆种子批次的生理质量对大豆种子批次进行分类的潜力。正确分类了8％的高活力样本和100％的低活力样本。用于分析的最佳LIBS光谱区域在350–450 nm范围内，其中钙是主要区别元素。因此，LIBS技术与机器学习分类方法的结合显示了根据大豆种子批次的生理质量对其进行分类的潜在潜力。