可移动式苹果内部品质果园产地分级系统

为满足苹果内部品质产地检测分级需求，本研究研发出检测模块和分级模块，构成可移动式苹果内部品质果园产地分级系统。在此系统的基础上，以苹果糖度和霉心病为代表品质指标，提出一种基于乘法效应消除（Multiplicative Effect Elimination，MEE）的光谱校正方法，用于消除苹果物理属性差异导致的有效光程变化对光谱的影响。利用该系统获取苹果600~900 nm漫透射光谱数据，分别采用多元散射校正（Multiple Scattering Correction，MSC）、标准正态变量变换（Standard Normal Variate Transform，SNV）和MEE算法对苹果光谱预处理后，建立糖度偏最小二乘回归（Partial Least Squares Regression， PLSR）预测模型和霉心病偏最小二乘判别（Partial Least Squares - Discriminant Analysis，PLS-DA）模型。结果表明，MEE算法相比于MSC和SNV算法建模结果更好，糖度预测模型的校正集相关系数（Rc）、校正集均方根误差（Root Mean Square Error of Calibration，RMSEC）、预测集相关系数（Rp）和预测集均方根误差（Root Mean Square Error of Prediction，RMSEP）分别为0.959、0.430%、0.929和0.592%；霉心病判别模型的校正集敏感性、校正集特异性、校正集准确率、预测集敏感性、预测集特异性和预测集准确率分别为98.33%、96.67%、97.50%、100.00%、90.00%和95.00%。将建立的最佳预测模型导入分级系统进行试验，结果表明该系统的分级准确率为90.00%，分级速度约3个/s。该系统具有成本低、结构简单、移动方便等优点，可以满足苹果内部品质果园产地检测分级需求。

Development of Mobile Orchard Local Grading System of Apple Internal Quality

The detecting and grading of the internal quality of apples is an effective means to increase the added value of apples, protect the health of residents, meet consumer demand and improve market competitiveness. Therefore, an apple internal quality detecting module and a grading module were developed in this research to constitute a movable apple internal quality orchard origin grading system, which could realize the detection of apple sugar content and apple moldy core in orchard origin and grading according to the set grading standard. Based on this system, a multiplicative effect elimination (MEE) based spectral correction method was proposed to eliminate the multiplicative effect caused by the differences in physical properties of apples and improve the internal quality detection accuracy. The method assumed that the multiplication coefficient in the spectrum was closely related to the spectral data at a certain wavelength, and divided the original spectrum by the data at this wavelength point to achieve the elimination of the multiplicative scattering effect of the spectrum. It also combined the idea of least-squares loss function to set the loss function to solve for the optimal multiplication coefficient point. To verify the validity of the method, after pre-processing the apple spectra with multiple scattering correction (MSC), standard normal variate transform (SNV), and MEE algorithms, the partial least squares regression (PLSR) prediction models for apple sugar content and partial least squares-discriminant analysis (PLS-DA) models for apple moldy core were developed, respectively. The results showed that the MEE algorithm had the best results compared to the MSC and SNV algorithms. The correlation coefficient of correction set (Rc), root mean square error of correction set (RMSEC), the correlation coefficient of prediction set (Rp), and root mean square error of prediction set (RMSEP) for sugar content were 0.959, 0.430%, 0.929, and 0.592%, respectively; the sensitivity, specificity, and accuracy of correction set and prediction set for moldy core were 98.33%, 96.67%, 97.50%, 100.00%, 90.00%, and 95.00%, respectively. The best prediction model established was imported into the system for grading tests, and the results showed that the grading correct rate of the system was 90.00% and the grading speed was 3 pcs/s. In summary, the proposed spectral correction method is more suitable for apple transmission spectral correction. The mobile orchard local grading system of apple internal quality combined with the proposed spectral correction method can accurately detect apple sugar content and apple moldy core. The system meets the demand for internal quality detecting and grading of apples in orchard production areas.