Embedded System Design and Control of a Portable Vis/NIR Spectroscopy with Hybrid XGBoost-ANFIS for Soybean Seed Quality Prediction

This study proposes a portable intelligent system for non-destructive evaluation of soybean seed quality using Vis/NIR spectroscopy integrated with a hybrid machine learning approach. Traditional quality inspection relies on destructive laboratory tests and expert involvement, which are time-consuming and inefficient, creating a demand for rapid, field-deployable alternatives. The methodology consists of two sequential stages: (1) prediction of seed quality parameters, moisture content (MC), germination rate (GR), and electrical conductivity (EC) using Extreme Gradient Boosting (XGBoost) regression; and (2) classification via Adaptive Neuro-Fuzzy Inference System (ANFIS) directly using the regression outputs as inputs. The research contribution is a novel embedded portable system that integrates automated spectral preprocessing, hybrid machine learning, and interpretable fuzzy logic for seed quality assessment. Spectral optimization was automated using the Nippy module with combination 9 operator. Experimental results from 800 soybean seed samples show that XGBoost achieved R² values of 0.961 for MC, 0.970 for GR, and 0.964 for EC, outperforming traditional chemometric methods (PLS) by 1401–1653%. The ANFIS classifier achieved 100% accuracy on the test set and 98% on external validation, with R² = 0.9996 ± 0.0002, demonstrating robust generalization through independent validation and mitigating overfitting concerns. The proposed system provides a rapid, non-destructive, and interpretable alternative to conventional seed testing, filling a critical gap in portable agricultural sensing with strong potential for real-time quality assessment in precision agriculture.