Non-Destructive Detection of Elasmopalpus lignosellus Infestation in Fresh Asparagus Using VIS–NIR Hyperspectral Imaging and Machine Learning

The early detection of internal damage caused by Elasmopalpus lignosellus in fresh asparagus constitutes a challenge for the agro-export industry due to the limited sensitivity of traditional visual inspection. This study evaluated the potential of VIS–NIR hyperspectral imaging (390–1036 nm) combined with machine-learning models to discriminate between infested (PB) and sound (SB) asparagus spears. A balanced dataset of 900 samples was acquired, and preprocessing was performed using Savitzky–Golay and SNV. Four classifiers (SVM, MLP, Elastic Net, and XGBoost) were compared. The optimized SVM model achieved the best results (CV Accuracy = 0.9889; AUC = 0.9997). The spectrum was reduced to 60 bands while LOBO and RFE were used to maintain high performance. In external validation (n = 3000), the model achieved an accuracy of 97.9% and an AUC of 0.9976. The results demonstrate the viability of implementing non-destructive systems based on VIS–NIR to improve the quality control of asparagus destined for export.