【作者】 周水琴；

【导师】 应义斌；

【作者基本信息】 浙江大学 ， 农业电气化与自动化， 2013， 博士

【摘要】 水果含有人体所需的多种矿物质和维生素等,对维持人体正常的生理功能有着重要的作用,它是人类饮食结构的基本组成部分,在人们的日常生活中必不可少。我国是水果生产大国,水果种植面积和产量在全世界名列前茅。水果采后商品化处理水平低是影响国内水果在国际市场竞争力的主要因素之一,因此实现水果外观与内部品质的无损检测及分级已成为国内水果产业化的必要前提。目前水果无损检测与分级主要有基于可见光技术的外部品质检测与光谱技术的内部品质检测两大类,外部品质检测技术较为成熟,但对于外部轻微损伤缺陷及内部缺陷的检测还存在不足之处,如检测效果受损伤时间影响,内部缺陷受检测位置及水果大小影响等。核磁共振成像技术可反映水果内部含水量的变化,应用其进行检测具有无损、可视化、安全无辐射、不受样品大小影响等优点。梨是国内水果产量居第三位的水果。本研究利用核磁共振成像技术和图像处理技术,对梨果挤压损伤、跌落损伤、内部褐变三种不同缺陷进行无损鉴别与分级,并建立了梨果坚实度与磁共振质地系数间的相关模型。利用医用核磁共振设备,采集了梨果中的鸭梨、香梨和黄花梨的冠状面核磁共振T2加权图像,经过图像转换、图像预处理、特征提取等处理实现对鸭梨挤压损伤及跌落损伤的识别、香梨内部褐变的识别,并对鸭梨跌落损伤阶段及香梨褐变严重程度进行分级；通过统计和分析黄花梨坚实度和核磁共振图像质地系数间的皮尔森相关关系,建立了多元回归模型。本文的研究目的在于验证核磁共振成像技术检测水果外部的机械损伤及内部缺陷的可行性,排除损伤时间对检测结果的影响,并确定内部褐变的严重程度,为研发具有自主知识产权的水果品质在线检测生产线提供方法依据。本论文的主要研究内容、结果和结论如下：1)确定了用于梨果缺陷检测的核磁共振图像采集方式。分析了核磁共振设备所采集的图像,结果表明：T2加权成像的清晰度可以完成本研究所需检测的缺陷和内部品质；且梨果冠状面图像比矢状面图像采集速度快,图像处理简单方便；对于不同品种的梨果,由于大小不同,可采取不同的切片厚度和切片间距,以适合水果缺陷与品质检测。2)提出了用于鸭梨表面轻微损伤检测的角点特征法,分析了挤压损伤与正常鸭梨组织核磁共振图像灰度的差异。对于采用万能试验机模拟的轻微压伤,通过Otsu阈值分割、膨胀操作并提取边界的图像处理方法,最后对水果边界进行角点检测。试验对207幅有效鸭梨样本图像进行轻微损伤检测,轻微损伤鸭梨样本的检测正确率为92.1%；正常鸭梨截面图像的检测正确率为100%,畸形鸭梨截面图像检测正确率100%。试验还对真实轻微损伤鸭梨进行损伤识别,结果发现对32个鸭梨真实损伤样本识别时,识别率可达96.8%。结果表明,对鸭梨冠状面切片边界提取角点特征的方法可判断鸭梨是否存在轻微挤压损伤。3)提出了用于鸭梨新旧跌落损伤检测的图像处理方法,分析了正常鸭梨与不同跌落损伤阶段鸭梨组织核磁共振图像灰度的差异,发现跌落损伤新伤组织图像灰度比正常鸭梨组织图像灰度高,跌落损伤旧伤组织图像灰度比正常鸭梨组织图像低。对于从离地面40mm架子上通过自由落体运动形成的跌落损伤,通过Otsu阈值分割、去除果核、旧伤特征提取等图像处理方法可以识别出旧伤水果；然后对判断不为旧伤的水果作进一步图像处理,通过固定阈值分割、去除果核、新伤特征提取可以识别出新伤水果,余下的均为完好水果。试验对100个切片进行跌落损伤检测,60个旧伤切片中有59个被检测为旧伤,1个被检测为新伤,识别正确率为98.3%；20个新伤切片均被检测为新伤,识别正确率为100%；20个完好切片均被检测为完好水果.识别正确率为100%。结果表明,利用核磁共振成像不仅可以实现鸭梨跌落损伤的识别,还可以同时实现对新伤和旧伤的识别。比较了损伤不同阶段损伤组织灰度的变化情况,采用典型判别分析方法,利用代表鸭梨切片图像的图像直方图参数对鸭梨损伤阶段判别的可行性进行了研究。研究所选样本的分类准确率为81.3%；能够较好地区分损伤第1阶段和第4阶段的鸭梨图像,分类准确率达到100%；但对于损伤第2阶段和第3阶段的图像容易形成误判,有较大的交叉区域,第2阶段损伤鸭梨有6个被归为第3阶段损伤鸭梨,1个被归为第1阶段鸭梨(该鸭梨损伤区域非常小),第3阶段损伤鸭梨有8个被归为第2阶段损伤鸭梨。对于同一鸭梨切片,第2阶段与第3阶段图像用肉眼观察灰度差别也不是特别大,因此使用直方图参数特征无法较好区分损伤第2阶段与第3阶段这种微弱的变化。将鸭梨分为损伤初期、中期和晚期三个阶段重新进行判别分析,该方法能较好区分鸭梨损伤阶段,总的分类正确率为98.75%。结果表明,利用核磁共振成像结合直方图特征参数可实现鸭梨损伤阶段的判别,并且分为三个阶段(损伤初期、中期和晚期)效果较好。4)针对库尔勒香梨在贮藏过程中出现的内部褐变缺陷,提了基于核磁共振图像的自动图像处理方法。对室温贮藏六个月的新疆库尔勒香梨,通过定期采集图像,观察其贮藏过程中的褐变变化情况,通过Otsu阈值分割、果核/水果区域像素比、形态学操作、去除果核、提取褐变特征等图像处理步骤,可以判断香梨是否存在内部褐变。试验对贮藏过程中42个香梨的128个有效切片进行内部褐变检测,该图像处理方法对褐变切片的识别正确率达到100%,对于正常切片的识别正确率为84%,总的识别正确率为98%。同时,通过分析我们还发现,该算法对贮藏后期的香梨总体褐变识别率比较高,这可能与贮藏前期香梨内部复杂的物理化学变化有关。结果表明,利用形态学的图像处理方法可实现褐变香梨的无损识别。分析了褐变香梨区域直方图,将香梨分为完好香梨、轻度褐变、中度褐变及重度褐变四类,结合褐变识别图像处理方法,各类香梨对应的识别准确率分别为84%、95%、94.4%和100%,轻度和中度褐变香梨各有1个切片被误判。结果表明,利用核磁共振成像结合褐变香梨区域直方图技术,可实现对香梨褐变程度的定性判断。5)分析了黄花梨在成熟和贮藏过程中坚实度与磁共振质地系数间的相关关系,选择对坚实度相关性较高的质地系数对成熟过程和贮藏过程分别建立了多元回归模型,并使用该模型进行了预测评估,该模型的稳定性和重复性有待进行更多的试验研究来补充和完善。更多还原

【Abstract】 Fruit contains many kinds of minerals and vitamins which can maintain the body’s normal physiological function. It is one of the basic components of human diet and indispensable in People’s Daily life. China is a big fruit production country. Fruit planting area and yield are among the best in the world. Low commercialization postharvest process is one of the main factors that affect market competition in the international market. Therefore, nondestructive detection and classification of fruit external and internal quality becomes the necessary for fruit industrialization of our country. Fruit nondestructive detection and grading methods mainly include two categories as external quality detection based on visible light and internal quality based on spectral technology. External quality detection technology is relatively popular.It has shortcomings for the external detection in minor damage defects and internal defects, such as detection effect influenced by damaged time and internal defects affected by detecting position and the size of the fruit. Nuclear magnetic resonance (NMR) imaging technology can reflect the change of the internal water content of fruit. It has been adopted due to its advantages: nondestructive testing, visualization, no radiation, safety and not affected by sample size and so on.Pear is selected as the research object whose production ranked third in the domestic. Nondestructive inspection of fruit quality were summarized and contrasted. The existing problems were pointed out. Nuclear magnetic resonance T2weighted images of Chinese pear, Korla Fragrant pear and Huanghua pear were scanned through the medical nuclear magnetic resonance (NMR) equipment. An image processing method was proposed after the image format transformation, which includes image processing, feature extraction and so on to realize pears extrusion injury and dropping damage identification and recognition of pear internal browning defects. In addition, dropping damage stage of the pears and browning severity level were also discussed. Pearson correlation between firmness of pear and nuclear magnetic resonance image texture coefficient was analyzed through statistics. A multiple regression model was established. The purpose of this study was to verify the feasibility detecting mechanical damage of external and internal defects of fruit with nuclear magnetic resonance (NMR) imaging technology. The influence of injury time on test results can be excluded and the severity of the internal Browning can be determined with MRI. It will Provide basis to research and develop fruit quality on-line detection line.The main contents and conclusions were listed as follows:1) The type of nuclear magnetic resonance imaging to detect fruit defect was analysed and determined. The results indicated that:Through the test of nuclear magnetic resonance image acquisition and image processing, the type of T2weighted imaging resolution could satisfy the requirement to detect the defect and internal quality of the study. The image acquisition speed of coronal image was faster than sagittal image and the image processing was easier and more convenient. For different type fruits different slice thickness and sliced spacing were employed to suit the fruit defect and quality detection.2) A new nondestructive inspection method based on corner feature was proposed which can detect the compressing damage of Chinese pears. The difference of nuclear magnetic resonance images for compressing bruised and sound Chinese pears was analyzed. The slight pressure injury simulated by Instron texture instrument could be detected through the Otsu threshold segmentation, expansion operation and boundary extraction and the corner detection of the fruit image. The207effective pear images were selected as the samples to detect the slight damage defect. The research results showed that the detection accuracy of slight bruise pears image was92.1%, while the detection accuracy was100%respectively for normal and misshapen pear images. Tests on real slight bruised pears and Fuji apples indicated that the detection accuracy was96.8%for32real bruised pear images and4real bruised apple images were all identified as bruised ones. The experimental results showed that detecting subtle compressing bruises on fruits with NMR technique was feasible through corner detection for coronal slice image.3) A new image processing method was suggested which can detect the new and old falling damage of Chinese pears based on nuclear magnetic resonance imaging. The difference of normal tissue and bruised tissue was analyzed. It was found that the gray level of new bruised tissue is higher than normal tissue while the gray level of old bruised tissue was lower than the normal tissue in the MR image. For falling injury caused by free fall movement from the shelf40mm above ground, the old bruised fruits were recognized through the Otsu threshold segmentation, to remove the core of the fruit and old bruise feature extraction; Then the new bruised fruits were discriminated through the fixed threshold segmentation, to remove the core of the fruit and new injury feature extraction from the rest and the remaining are all good fruits. Drop damage detection test of100slice images showed that59old bruised slices are recognized of60old bruised ones and the accuracy rate was98.3%;20new injury slices are distinguished as the new injury ones and the accuracy rate was100%;20sound slices were testing for good ones and the accuracy rate was100%. The result showed that it was feasible to not only detect the falling damage of Chinese pears but also to distinguish the new and old damage with MRI.Fruit damage stage judgment was studied through typical discriminant analysis of histogram parameters of fruit slice image. The classification accuracy rate is81.3%; Good results are achieved for the1st and the4st damage stage of fruit images and the classification accuracy comes to100%; but for the2nd and the3rd damage stage of the fruit image is easy to misclassify each other. Six fruit slices of the2nd damage phase were misclassified as the3rd phase damage fruit and one was misclassified as the1st phase one (the fruit damage area is very small). The8fruit slices of the3rd damage stage of the fruit image were misclassified as the2nd damage stage fruit. For the same fruit slice, there was no obvious difference of the2nd stage and the3rd stage image with the naked eye. The slight change could not be distinguished using histogram parameter features. Bruised pears were divided into three stages as early, middle and late stage to classify with discriminant analysis again. The method could distinguish fruits damage stage better and the overall classification accuracy was98.75%. Results indicated that the use of nuclear magnetic resonance imaging (MRI) combined with the histogram characteristic parameters which could realize fruit damage phase discrimination and it was better to divide into three stages (early, middle and late stage).4) A new nondestructive inspection method based on nuclear magnetic resonance imaging was suggested which can detect the internal browning of Korla fragrant pear caused during storage. During six months storage at room temperature for xinjiang korla fragrant pear, nuclear magnetic resonance images were scanned regularly in order to observe the browning process. Browning was recognized through the Otsu threshold segmentation, ratio of core and fruit area pixels, morphological operation, to remove core and browning characteristic extraction. The128valid slices were selected as the test objects from42pears for internal browning inspection. The analysis results showed that the image processing method was suitable for browning identification. The accuracy reached100%and84%for browning and sound slices respectively. The total recognition accuracy rate is98%. We also found that the algorithm was more appropriate for pears stored longer which would get higher accuracy. It may relate to the internal complicated physical and chemical change of fragrant pear during storage at the beginning. The results indicated that it was feasible to detect the internal browning of fragrant pears based on morphology.The area histogram of browning fragrant pear was analyzed. The fragrant pears were divided into four categories as fine, mild Browning of fragrant pear, mild Browning and severe Browning. Combining with Browning recognition and image processing method, the corresponding identification accuracies were84%,95%,94.4%and84%for each kind of fragrant pear respectively. There was one slice image for mild and moderate Browning pears that was misclassified. Results indicated that the qualitative judgment of fragrant pear browning degree can be realized with nuclear magnetic resonance imaging (MRI) based on the browning fragrant pear area histogram technique.5) The relation of magnetic resonance texture coefficients and firmness of Huanghua pear was analyzed during the ripe and storage process. The multiple regression model was built based on the texture coefficients with higher correlation for the ripe and storage stage respectively. Prediction and assessment was made using the model. The stability and repeatability of the model needs to be more experimental research to supplement and complete.更多还原