基于高光谱图像的蓝莓糖度和硬度无损测量

Nondestructive measurement of firmness and sugar content of blueberries based on hyperspectral imaging

为了对蓝莓硬度和糖度进行无损检测,采用近红外光谱仪(900~1700 nm)分别对490个"蓝丰"蓝莓的果柄侧和花萼侧进行高光谱成像,并测量整个果实的硬度和糖度。应用偏最小二乘回归法分别对果柄侧、花萼侧和整个果实的平均光谱建立硬度和糖度预测模型。试验结果表明,蓝莓硬度呈双峰分布,表明实际生产中有望分为2类;蓝莓糖度呈正态分布;硬度和糖度的相关性仅为?0.15,说明不能通过二者之中的任何一个来估计和评价另一个。采用整个果实的平均光谱数据建模效果最好,硬度的校正集相关系数RC和验证集相关系数RV达到0.911和0.871,糖度的为0.891和0.774,但主成分数都有所增加。结果表明,采用高光谱技术对蓝莓硬度和糖度进行快速、无损检测是可行的。

Blueberry is an important fruit in the world, and its consumption has increased significantly in recent years because of its flavor and antioxidant capacity for anti-aging. In China, blueberry has increased greatly with rapid development of internet shopping and express. However, fruits with low firmness are easily to be damaged with juice leaking out, which is unacceptable by the consumer. Sugar content is also important to consumer. Accurate determination of blueberry quality is challenging since individual fruit are small and dark in color, and vary greatly in external and internal quality characteristics. Traditionally, blueberry quality was inspected by human in situ at the sorting line, which was inefficient and unreliable. Moreover, it's difficult to sort fruit by human based on sugar content and firmness, two quality attributes that are not only important to the consumer, but also directly impact the shelf life of blueberries. Therefore, hyperspectral imaging technique for predicting the firmness and sugar content of blueberries was researched. A pushbroom hyperspectral imaging system was used to acquire hyperspectral reflectance images from 490 ‘bluecrop' blueberries in two fruit orientations(i.e., stem and calyx ends) for the spectral region of 900-1 700 nm. Each fruit was then segregated by building a binary mask to recognize the fruit from the background using threshold segmentation in the hypercube. This was accomplished on the spectral image at 1 542 nm, which gave the maximum contrast between the fruit and the background. After that, each berry was identified by combining tilting and labeling operations on the masked image. From the regions of interest of each segmented blueberry image, mean reflectance was computed by averaging over all pixels. Finally, prediction models were developed based on partial least squares method using cross validation and were externally tested with 25% of the samples. Effect of fruit orientations on hyperspectral imaging prediction was evaluated by designated the spectral data in three treatment(stem end, calyx end, and whole fruit which averaged two mean spectra for the stem and calyx end regions to obtain on spectrum). Results showed that better firmness predictions(RC=0.911, RV=0.871) were obtained, compared to sugar content predictions(RC=0.891, RV=0.774). Fruit orientation had no or insignificant effect on the firmness and sugar content predictions. Further analysis showed that blueberries could be sorted into two classes of firmness. The result showed that hyperspectral imaging is promising for online sorting and grading of blueberries for firmness and sugar content as well.