Estimation of Soybean Internal Quality Based on Improved Support Vector Regression Based on the Sparrow Search Algorithm Applying Hyperspectral Reflectance and Chemometric Calibrations

Abstract

The nutritional components of soybean, such as fat and protein, directly decide soybean quality. The fast and accurate detection of these components is significant to soybean industries and soybean crop breeding. This study developed an improved SSA-SVM (support vector regression based on the sparrow search algorithm) for the rapid and accurate detection of the fat and protein in soybean seeds using hyperspectral reflectance data. In this work, 85 soybean samples were selected. After their fat and protein contents were analyzed using chemical methods, a total of 85 groups of hyperspectral image data were collected using the hyperspectral imaging system. An effective data preprocessing method was applied to reduce the noise for enhancing the prediction models. Some popular models, including partial least-square regression (PLSR), random forest regression (RFR), and support vector regression based on the genetic algorithm (GA-SVR), were also established in this study. The experimental results showed that the improved SSA-SVM model could predict the nutrient contents of the soybean samples with accuracies of 0.9403 and 0.9215 and RMSEs of 0.2234 and 0.325 for the fat and protein, respectively. The convergence speed was improved significantly. Therefore, hyperspectral data combined with the SSA-SVM algorithm presented in this study were effective for evaluating the soybean quality.