基于高光谱成像技术的蓝莓糖度无损检测模型优化研究

Optimization research on non-destructive detection model of Vaccinium spp. sugar content based on hyperspectral imaging technology

中文关键词:  蓝莓糖度  无损检测  高光谱成像技术  机器学习

英文关键词:Vaccinium spp. sugar content  non-destructive detection  hyperspectral imaging technology  machine learning

中文摘要:

      目的 基于高光谱成像技术优化蓝莓糖度的无损检测模型。方法 以产自丹东的L25品种蓝莓为实验对象, 采用高光谱成像技术获取900~1700 nm波长范围内的蓝莓高光谱图像, 并计算感兴趣区域的平均光谱作为原始数据。通过多元散射矫正(multiple scatter correction, MSC)、标准正态变量变换(standard normal variable, SNV)和Savitzky-Golay (SG) 3种预处理方法改善光谱数据质量。基于预处理后的全波长数据, 分别建立偏最小二乘回归法(partial least squares, PLS)、反向传播神经网络(back propagation neural network, BPNN)、支持向量机回归法(support vector regression, SVR)等糖度预测模型。结果 在MSC和SNV预处理下, PLSR模型在糖度预测过程中表现出较好的预测性能, 均方根误差(root mean squared error of prediction, RMSEP)分别为0.3586、0.3599。结论 本研究优化了基于高光谱成像技术的蓝莓糖度无损检测模型, 为蓝莓糖度的快速、准确预测提供了有效的技术支持, 具有较强的实际应用潜力。

英文摘要:

      Objective To optimize a non-destructive detection model for predicting Vaccinium spp. sugar content using hyperspectral imaging technology. Methods The L25 variety of blueberries from Dandong was selected as the subject, and hyperspectral imaging technology was acquired in the wavelength range of 900–1700 nm. The average spectrum of the region of interest was calculated as the raw data. The 3 kinds of preprocessing methods, including multiple scatter correction (MSC), standard normal variate (SNV) and Savitzky-Golay (SG), were applied to improve the spectral data quality. Non-destructive sugar content prediction models were established using partial least squares regression (PLSR), back propagation neural network (BPNN), and support vector regression (SVR) based on the full-wavelength data after preprocessing. Results The experimental results demonstrated that the PLSR model, with MSC and SNV preprocessing, exhibited the best performance, achieving root mean square error of prediction (RMSEP) values of 0.3586 and 0.3599, respectively. Conclusion This study provides an optimized non-destructive detection model for Vaccinium spp. sugar content, offering effective technical support for rapid and accurate sugar content prediction with significant practical potential.