Peanut Frostbite Detection Method Based on Near Infrared Hyperspectral Imaging Technology

Peanuts were susceptible to frost damage during harvesting, transportation, storage, and processing due to temperature and humidity changes, which could affect the quality of peanuts and their products. In order to explore the mechanism of peanut frost damage and improve the detection efficiency of frost-damaged peanuts, this study used near-infrared hyperspectral technology to study the feasibility of non-destructive detection of peanut frost damage, optimization methods based on feature variable screening discriminant models, and the mechanism of peanut frost damage. The effects of five preprocessing methods, including standard normalized variate (SNV), multiplicative scatter correction (MSC), Savizkg-Golag (SG) smoothing, SG smoothing-SNV, and SG smoothing-MSC, on the original data were experimentally studied. Then, eight variable selection methods, including competitive adaptive reweighted sampling (CARS), random frog (RF), variable importance in projection (VIP), successive projections algorithm (SPA), Monte Carlo uninformative variable elimination (MC-UVE), iteration retention information variable (IRIV), variable combination population analysis-iteration retention information variable (VCPA-IRIV), and variable combination population analysis-genetic algorithm (VCPA-GA), were used to screen the feature wavelengths related to peanut frost damage. Support vector machine (SVM) was used to select the feature wavelengths that reached the discrimination accuracy threshold of 90% as the feature wavelengths of peanut frost damage. The results showed that the detection of peanut frost damage based on near-infrared hyperspectral imaging technology was generally feasible and had high accuracy. All variable selection methods can effectively screen the feature wavelengths related to frost damage. Among them, the VCPA-GA algorithm selected the least 7 feature wavelengths, accounting for only 3.125% of all wavelengths in the dataset. The accuracy of the training set and the test set were 91.60% and 90.12%, respectively. The selected frostbite characteristic wavelength reflects information about oleic acid and protein, verifying that excessively low temperatures can lead to a decrease in oleic acid content and an increase in protein content in peanuts. This study provides a theoretical basis and technical support for the rapid non-destructive detection of peanut frost damage.