Hyperspectral and imagery integrated analysis for vegetable seed vigor detection

Seed vigor is a primary determinant of seed quality. Since the traditional methods detecting the seed vigor are time-consuming and costly, hyperspectral imaging technology in this study was used to develop a method that was rapid and nondestructive to detect the vigor of vegetable seeds. Hyperspectral images of two kinds of vegetable seeds, broad beans and hyacinth beans, were collected by measuring spectral reflectance from 400 to 1000 nm. The raw spectral data were first preprocessed with Savitzky-Golay (SG) smoothing and multiplicative scatter correction (MSC). And then, principal component analysis (PCA) and uninformative variable elimination (UVE) were carried out to select optimal wavelengths, and image features were also extracted simultaneously from RGB image that combined the central wavelengths at 660, 560 and 480 nm. Finally, support vector ma-chines (SVM) and random forest (RF) model were employed to build the classification models based on spectral data, image data and the integrated data from both spectral and image data. The results showed that the SVM model had better classification performance and its accuracy achieved 83 to 91 %, when the spectral data were selected through UVE and combined with the image data. An integration of both spectral data and image data can improve the accuracy of the model compared to those used alone.