指状青霉和意大利青霉感染的早期腐烂柑橘高光谱成像快速检测方法研究

With the development of citrus industry, the postharvest decay ratio of citrus fruit is very high which leads to a weak competitiveness in the citrus market. The rapid spread of early decay caused by fungi infection of Penicillium digitatum and Penicillium italicum was the most major cause for citrus fruit decay. Therefore, it is crucial to explore a fast nondestructive method for detecting the early fungal infected citrus. Gannan navel orange and Nanfeng sweet orange were selected as the research objects. The fast nondestructive detection methods of early decay citrus were studied by using the Vis-NIR and NIR line-scanning hyperspectral imaging technologies. The relationship between hyperspectral image acquisition parameters and detection result of decay citrus fruit was analyzed. The distribution of spectral reflection and absorption intensity from different peel areas of the same sample due to citrus surface curvature change was studied and the corresponding spectra energy compensation algorithm was developed. The target image set and target variable set used for detecting the early decay citrus were determined based on the hyperspectral image information and the proposed novel image-spectra fusion method. Finally, detection method and classification model for identifying the early decay citrus were obtained by integrating the image-spectra information of hyperspectral images. This study will provide a solid theoretical foundation for developing the multispectral rapid detection equipment to identify the early decay citrus caused by fungal infection with Penicillium digitatum and Penicillium italicum, and will be quite helpful to reduce the citrus postharvest economic loss, enhance market competitiveness, and promote the healthy development of citrus industry in China.

随着我国柑橘产业的发展，柑橘采后腐烂损失率居高不下，市场竞争力严重不足。其中，指状青霉和意大利青霉真菌感染引起的早期腐烂的快速传播是最主要的原因。因此，探索该真菌感染早期腐烂柑橘的快速无损检测方法至关重要。本项目以经济价值较高的赣南脐橙和南丰蜜桔为研究对象，率先使用可见-近红外和近红外两种线扫描高光谱成像技术对早期腐烂果的快速检测方法进行研究。分析高光谱图像采集参数对感染柑橘检测的影响，探明由于柑橘表面曲率变化引起不同区域对光谱反射和吸收强度差异的分布规律，开发光谱能量补偿算法。基于高光谱-图像信息通过新颖的图谱融合方法确定可以用于早期腐烂柑橘识别的目标图像集和目标变量集。最终建立针对高光谱图谱信息融合的一体化检测方法和分类模型。项目研究成果将为该类真菌感染早期腐烂果的多光谱快速检测设备开发提供理论依据和方法基础，对减少柑橘采后减损、增强市场竞争力和促进我国柑橘产业健康发展具有重要意义。

真菌感染引起的腐烂是柑橘类水果采后最严重的缺陷，目前缺乏快速、有效的检测技术对此类缺陷进行准确检测。基于此，本课题开展了基于高光谱成像技术的柑橘早期腐果快速检测方法的研究。通过主成分分析获得了可以用于早期腐果识别的4个特征波长575、698、810和969 nm，并构建了多光谱组合图像，利用均值归一化对组合图像中水果表面高光谱能量分布不均进行了快速校正，结合两步伪彩色图像增强方法，成功实现了不可见真菌感染区域的快速可视化检测，对所有的样本包括正常果、指状青霉感染果和意大利青霉感染果进行识别，获得了98.3%的整体识别精度。同时，基于可见-近红外和长波近红外高光谱数据，构建了支持向量机SVM、极限学习机ELM、K最近邻KNN、线性判别分析LDA、簇类独立软模式分类法SIMCA等模型对南丰蜜桔早期腐烂组织进行分类识别，研究发现基于波长范围为1000~2500 nm的高光谱成像技术结合SVM模型能够对指状青霉感染组织、意大利青霉感染组织和正常果组织进行有效地分类识别，精度达98.33%。进一步，本研究还提出基于紫外诱导荧光成像技术的柑橘腐果快速检测系统及方法，重点解决了紫色光光污染和低荧光下腐烂病斑难分割的问题。这些研究成果对研发高端柑橘类水果质量分级装备、减少柑橘采后损失、增加农民收入具有重要意义。此外，依托本项目，还开展了基于长波近红外高光谱成像技术的梨果内部可溶性固形物无损检测方法研究，所构建的最优模型MC-UVE-SPA-PLS对梨果内部可溶性固形物预测相关系数R为0.88，预测均方根误差RMSEP为0.35；开展了基于高光谱成像技术的大桃表面缺陷快速检测方法研究，提出了图像比的大桃表面常见缺陷检测算法，获得了整体96.6%的检测精度；开展了基于高光谱技术的大桃内部可溶性固形物预测稳健模型构建方法的研究，构建了最优的多区域组合特征波长模型，该模型可以成功用于大桃内部全局可溶性固形物含量预测，预测相关系数R为0.86，均方根误差RMSEP为0.67。相关研究成果为水果品质无损检测的快速实施提供了新的参考方法和解决途径。