光源光斑直径对苹果霉心病近红外检测的影响

为实现中早期霉心病苹果的有效剔除以提高苹果的整体品质，该研究利用近红外光谱技术对苹果霉心病进行快速无损检测，从光谱和分类模型两方面探究光源光斑直径对苹果霉心病检测的影响。在30、50 及70 mm光源光斑直径条件下采集了苹果样本的透射光谱，分析不同光源光斑直径下健康苹果和霉心病苹果的光谱差异，然后应用支持向量机（support vector machines，SVM）和粒子群算法优化-最小二乘支持向量机（particle swarm optimization-least squares support vector machine，PSO-LSSVM）方法建立苹果霉心病的分类模型，并对不同光源光斑直径下的分类模型性能进行对比。在此基础上，采用竞争自适应重加权采样（competitive adaptive reweighted sampling, CARS）方法筛选特征波长变量并建立分类模型。研究结果表明，30 mm光源光斑直径对苹果霉心病的检测效果最好，建立的SVM和PSO-LSSVM分类模型性能均最优。30 mm光源光斑直径下，最优PSO-LSSVM模型的预测集的灵敏度、特异度和正确率分别为89.5%、95.5%和92.7%。CARS-PSO-LSSVM分类模型性能比全波段的分类模型性能略有下降，预测集的灵敏度、特异度和正确率分别为89.5%、90.9%和90.2%，但建模变量数仅占原波长变量数的4.2%，有效地简化了分类模型。该研究为苹果霉心病的快速无损高精度检测提供技术支撑。

Effects of light source spot diameter on the near-infrared detection ofmold heart diseases in apples

Moldy heart disease is one of the most serious diseases in apples. However, there are hidden symptoms of fruit heart disease in the early and middle stages. The mixing of healthy apples has posed a great risk to the overall quality of apples, resulting in the healthy apples rotting together. It is very urgent for rapid and nondestructive testing on the moldy heart disease in apples. In this study, near-infrared spectroscopy was used to rapidly and non-destructively detect the mold heart disease with higher precision in apples. The influence of the light source spot diameter on the detection of mold heart disease in apples was explored, in terms of both spectra and classification models. The 127 red Fuji apple samples were selected in the experiment, which was purchased in March 2023 in Luochuan County, Yan''an City, Shanxi Province. The transmission spectra of apple samples were collected using a near-infrared spectrometer (NIRS) with a wavelength range of 200-1100 nm at light source spot diameters of 30, 50, and 70 mm. A systematic analysis was first made to determine spectral differences between healthy and mold heart apples under different light source spot diameters. Then, the classification models were established using support vector machines (SVM) and particle swarm optimization-least squares support vector machine (PSO-LSSVM) combined with different pre-processing methods. The performance of classification models was also compared under different light source spot diameters. Competitive adaptive reweighted sampling (CARS) was used to screen the characteristic wavelength variables, in order to optimize the classification model. The results show that the spot diameter of a 30 mm light source had the best effect on the detection of mold heart disease in apples, indicating the best performance of SVM and PSO-LSSVM classification models. Specifically, the performance of classification models decreased, as the spot diameter of the light source increased. Meanwhile, the path of the light source mainly carried the information around the apple core at the spot diameter of 30 mm. The spectra collected from this region also shared more information about the diseased tissue of apples. By contrast, the path of the light source covered most or all of the apple at the spot diameters of 50 or 70 mm. More information was then collected about the healthy apple tissues, leading to interference with the detection of mold heart disease in apples. The Sensitivity, Specificity, and Accuracy of the best PSO-LSSVM were 89.5%, 95.5%, and 92.7%, respectively, in the prediction set at 30 mm light source spot diameter. The CARS effectively screened the feature wavelengths. The performance of the CARS-PSO-LSSVM classification model was slightly inferior to that of the full-wavelength model, where the Sensitivity, Specificity, and Accuracy in the prediction set were 89.5%, 90.9%, and 90.2%, respectively. Therefore, the classification model was effectively simplified with high recognition accuracy and strong robustness, although the number of modelled variables was only 4.2% of the original. This finding can provide the technical support for the rapid, non-destructive, and high-precision detection of apple mold heart disease.