水果内部复杂非均相体系综合质量全透射高光谱成像无损检测方法研究

Full-transmittance hyperspectral imaging technique combining with spectroscopic analysis approach of complex heterogeneous systems and model population analysis strategy provides a new way for accurately and robustly assessing the chemical concentrations and tissue properties in fruits. In view of the challenge of accurately assessing the quality of complex heterogeneous systems in fruits, using typical fruits as research objects, this project focused on simultaneous detection of nutritional quality and safety index of fruits. Based on full-transmittance hyperspectral imaging technique, the novel model population analysis strategy and binary matrix sampling combined with chemometric algorithms were used for hyperspectral data mining. The multiplicative effects models from full-transmittance hyperspectral light scattering were built. Optical path length estimation and correction method was used to account the multiplicative parameters for multiplicative effects in the spectral measurements of the calibration samples. And then, the dual-calibration fusion models based on fruit quality parameters and multiplicative parameters were developed. Based on the proposed method, multiplicative effects of physical properties of fruit samples were effectively separated from hyperspectral variations related to the chemical compositions and tissue properties. Finally, a set of the novel nondestructive detection method system for the assessment of comprehensive quality of complex heterogeneous systems in fruits based on full-transmittance hyperspectral imaging was formed to improve the nondestructive detection accuracy and robustness of internal qualities of fruits. The research results of this project can be extended to other agricultural and animal products quality detection and food engineering fields.

全透射高光谱成像技术结合复杂非均相体系光谱分析和模型集群分析策略提供了一条准确、稳定检测水果内部化学组分含量和组织特性的新途径。项目针对水果内部复杂非均相体系质量难以准确评估的行业难题，以我国典型水果为研究对象，聚焦食用营养品质和安全指标的同步评估，以全透射高光谱成像技术为检测手段，采用新型模型集群分析策略和二进制矩阵采样方法，结合多种化学计量学算法进行数据挖掘，构建全透射高光谱散射乘子效应模型，利用改进光程估计和校正方法估测样本的乘子效应参数矢量，构建快速的乘子效应参数和水果质量参数的双校正融合模型，实现水果组分含量和组织特性所引起的高光谱贡献与组织物理性质差异所引起的光散射乘子效应之间的有效分离，最终形成一套新颖的基于全透射高光谱成像技术的水果内部复杂非均相体系综合质量评估的方法体系，提高水果质量无损检测精度和稳定性。本项目的研究方法可以推广到其它农、畜产品质量检测和食品工程等领域。

水果内部复杂非均相体系质量准确、稳定评估是水果品质快速检测分级的关键，本研究采用高光谱成像技术，以典型厚皮柑橘类水果橙子为研究对象，系统性地开展了相关研究：（1）基于全透射高光谱成像技术的橙子内部糖度定量预测。研究结合可见-近红外全透射高光谱成像技术，采用蒙特卡洛异常样本检测法识别并剔除异常样本，采用自适应重加权采样和连续投影算法的组合变量选择法对有效波长进行选择，基于不同的输入变量构建了线性和非线性校正模型，同时在模型构建中，引入了高光谱散射乘子效应模型，利用改进光程估计和校正方法估计了样本光谱的乘子效应参数，最终确定了可用于橙子内部糖度准确预测的多参数补偿和乘子效应校正的非线性最小二乘-支持向量机模型，该模型对橙子内部糖度预测其决定系数为0.94，预测均方跟误差为0.33；（2）基于可见-近红外全透射高光谱成像技术对橙子早期腐烂果进行检测研究。采用主成分分析和伪彩色图像变换获得了可用于早期腐烂果有效检测的第三主成分图像和用于果梗识别的第二主成分图像，通过改进分水岭分割算法和果梗识别算法实现了柑橘早期腐果的快速检测，腐烂果和正常果成功识别率分别为93%和96%；（3）基于漫反射高光谱成像技术对橙子早期腐烂果快速检测研究。通过主成分分析和权重系数分析获得了可用于腐果检测的7幅有效波长图像，二次主成分分析获得的第六主成分图像作为目标图像，采用二维经验模态分解和重构的方法对图像进行去噪和感兴趣区域增强，通过研发的改进分水岭分割算法实现了腐烂病斑的有效分割，研究表明融合多光谱主成分图像、二维经验模态分解和图像重建以及改进分水岭分割算法可实现脐橙早期腐果的检测识别，腐烂果和正常果识别正确识别率分别为97.3%和100%。相关研究成果为柑橘类水果品质无损检测的快速实施提供了帮助和有意义的参考，同时促进了柑橘类水果综合品质高端分级装备的研发。