针对市场中存在广陈皮年份造假、 以次充好等问题, 提出一种基于黑寡妇优化算法(BWO)和支持向量机模型(SVM)的广陈皮陈化年份高光谱鉴别方法。 以四类陈化年份(5~20年)的广陈皮作为实验对象, 采集样本的高光谱图像数据(385~1 014 nm波长), 通过镜头校准和反射率校准后提取样本感兴趣区域的平均光谱数据。 首先, 采用多项式平滑算法(SG), 结合多元散射校正算法(MSC)和去趋势算法(detrend)对数据进行降噪处理; 然后, 分别采用连续投影算法(SPA)、 竞争性自适应重加权与逐步回归混合算法(CARS_SR)筛选出特征波段; 最后, 提出以均方根误差(RMSE)为适应度函数, 分别使用偏最小二乘判别分析模型(PLS)、 粒子群算法优化SVM模型(PSO-SVM)和蝗虫算法优化SVM模型(GOA-SVM)对广陈皮的陈化年份进行鉴别, 并通过采用BWO算法优化SVM模型(BWO-SVM)来得到鉴别模型的最优参数。 结果发现: SG_detrend算法对广陈皮高光谱数据具有较好的降噪能力, CARS_SR算法具有较好的特征信息提取能力; 与PLS、 PSO-SVM和GOA-SVM相比, BWO-SVM算法可以得到更好的鉴别模型控制参数; SG_detrend-CARS_SR-BWO-SVM模型对广陈皮陈化年份的鉴别准确率达到97.59%, RMSE为0.060 2, R2为0.952 9。 该工作为实现广陈皮陈化年份的快速无损鉴别提供了新方法, 也为便携式鉴别仪器或在线生产设备研发提供了理论依据。

In response to the problems, including Cantonese tangerine peel appearing on the market with shoddy and year of aging falsification, a hyperspectral identification method based on the Black Widow Optimization (BWO) algorithm that supports vector machine (SVM) was proposed to address these problems. In the current study. The Samples hyper-spectral image data (385~1 014 nm) were collected with the Cantonese tangerine peel with four aging years (5~20 years) as the experimental object. The average spectral data of the sample region of interest were extracted by lens and reflectance calibration. Firstly, Savitzky-Golay Smoothing (SG), the Multiple Scattering Correction (MSC), and the Detrended Fluctuation Analysis (detrend) algorithm were utilized to perform spectral noise reduction for the data. Furthermore, the successive projections algorithm (SPA) and the competitive adaptive reweighting sampling mixed stepwise regression (CARS_SR) algorithm were used to extract the feature wavelengths. Finally, the root mean square error (RMSE) was proposed as the fitness function. The partial least-regression (PLS), the particle swarm optimization (PSO)-SVM, and the grasshopper optimization algorithm (GOA)-SVM were used to identify the aging year of Cantonese tangerine peel. Additionally, the identification models optimal parameters were obtained using the BWO algorithm optimized SVM model (BWO-SVM). It was found that the SG_detrend algorithm has a relatively excellent noise reduction ability for the hyperspectral data of Cantonese tangerine peel. The feature wavelengths could be extracted via the CARS_SR algorithm. Compared with PLS, PSO-SVM, and GOA-SVM, more optimal control parameters for the identification model could be gained using BWO-SVM. The accuracy of 97.59%, RMSE of 0.060 2, and R2 of 0.952 9 for the identification of aged vintage Cantonese tangerine peel were achieved with the SVM model. This research provides a novel method to achieve rapid and nondestructive identification of aged vintage Cantonese tangerine peel and also provides a theoretical basis for the development of portable identification instruments and online production equipment.