The maturity of seeds at harvest determines their intrinsic quality characteristics such as longevity and vigor, and these characteristics are dominant factors for seed quality evaluation in the seed industry. However, little information is available on how to identify and further classify the maturation stage of seeds in a way that is nondestructive, precise, rapid, and inexpensive, while also exactly meeting the need for the uniform control of seed performance in the seed industry to improve crop yield. This study demonstrated a nondestructive method for detecting seed maturity by using the single-kernel near-infrared spectroscopy (SK-NIRS) technique. The results showed that five classes of cucumber seeds with different maturation levels can be distinguished successfully. A tree-structured hierarchical classification strategy consisting of one soft independent modeling of class analogy (SIMCA) model and three partial least squares discriminant analysis (PLS-DA) models were proposed ending up with 99.69% of the overall classification accuracy and 0.9961 of Cohen’s kappa in the test set, and its predictive performance was superior to both SIMCA and PLS-DA for direct multiclass classification. SK-NIRS in combination with a multiclass hierarchical classification strategy was proved to be both intuitive and efficient in classifying cucumber seeds according to maturation levels.
Single-Kernel FT-NIR Spectroscopy for Detecting Maturity of Cucumber Seeds Using a Multiclass Hierarchical Classification Strategy
