A feasibility study on improving the non-destructive detection accuracy of Huping jujube (Ziziphus jujuba Mill. cv. Huping) damage degree using near infrared spectroscopy

Near infrared (NIR) spectroscopy is promising for fruit quality assessment but faces robustness challenges in damage detection, as surface reflectance alone cannot fully characterize internal and external damage features. To overcome this limitation, we propose combining NIR spectroscopy with multi-position light scattering information to improve the accuracy of non-destructive jujube damage grading. The Huping jujube was impacted and the damaged jujube was taken as the sample. The NIR spectra of three kinds of samples with different damage grades are collected. With the damage degree as the reference index, five machine learning algorithms of Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbor (KNN), Radial Basis Function network(RBF), and Long Short-Term Memory (LSTM) are combined to construct the damage degree identification model of single-position spectral and multi-position detection data fusion. The test set accuracy of the optimal multi-position spectral modeling (MPSM) method is 100.00 %. Compared with the single-position spectral modeling (SPSM) method, the stability of the MPSM fusion method is significantly improved, and the accuracy rate is increased by more than 13.89 %. This study established a reliable non-destructive detection method for subtle fruit damage, demonstrating the effectiveness of multi-position spectral fusion in capturing sub-surface damage and providing a transferable framework applicable to other bruise-prone delicate fruits.