【作者】 屈年巍；

【导师】 马本学；

【作者基本信息】 石河子大学 ， 农业机械化工程， 2011， 硕士

【摘要】 库尔勒香梨是新疆在海内外最负盛名的果品之一,也是新疆出口创汇的重要产业之一。目前香梨分级主要是依靠人工分级,由于分级人员长期大量依靠目测分级,容易产生视觉疲劳和情绪波动,而且每个人对颜色的标准等感官不一,难免产生误差,造成同一个香梨有不同检测结果,这不仅影响了香梨的果品价值,也不利于保证水果质量和提高消费满意度。在很多同类水果中,着色好且均匀一致的水果商品价值较高。颜色也在一定程度上间接反映了水果的糖度、酸度及口感等。同时,香梨表面的轻微碰伤由于颜色和正常区域相近,容易被忽视。因此,香梨按颜色分级和轻微碰伤的检测也是香梨自动化检测中的两项重要内容。本文提出了基于机器视觉技术对库尔勒香梨的颜色和轻微碰伤进行无损检测,主要方法及内容如下：通过对香梨进行自动化分级的必要性和国内外在水果智能检测领域的研究现状进行分析,提出了利用机器视觉技术对香梨进行颜色分级和碰伤检测的方法,并针对香梨颜色和轻微碰伤检测的研究内容确定了研究方法。根据研究内容设计相关实验系统,并对系统的构成和各部件选取进行了分析和优化,概述了实验所用的实验仪器和软件系统。利用机器视觉技术对香梨颜色进行分级的方法进行了研究,找到了对香梨背景分割的方法：利用HSI模式下的H分量作为香梨分割的模板,再进行闽值分割和二值化,得到香梨和背景的分割图,利用该分割图与原香梨图像进行融合,得到去除背景的香梨图像,并对其表面像素和平均值做统计,将统计结果作为人工神经网络的输入,颜色信息作为输出,分别比较了基于BP人工神经网络和RBF人工神经网络的香梨颜色分级,两种方法的分级准确率均为95%。同时,针对香梨碰伤特点,设计了香梨的碰伤实验装置,对香梨进行人工碰伤。利用机器视觉技术和高光谱图像技术对香梨碰伤进行研究,使用了主成分分析法和独立分量分析法,分别利用主成分分析法和独立分量分析法对获取的高光谱图像数据处理分析,发现香梨第四主成分包含碰伤的主要信息,直接对该主成分进行图像处理和分析,碰伤提取的准确率为81.0%,通过提取特征向量发现其的主要贡献波段分别为528nm,662nm和740nm,利用光谱图像融合技术对主要贡献波段进行融合并处理分析,检测准确率为70.9%。利用独立分量分析可以检测到在第四分量下的香梨碰伤区域,对该分量进行图像处理,提取碰伤区域,检测准确率为82.5%。最后,归纳了本文的主要结论和有待于进一步研究的内容。更多还原

【Abstract】 Kurle fragrant pear is one of the most popular fruits in abroad producted by Xinjiang. It’s a very important local Economic fruit. Right now, main grading method is by manpower. For a long time working, the works would too busy to keep precision, and everyone would have different result for the same pear. Different standards not only influence the value of fragrant pear, but also influence the quality of fragrant pears and customs’ contentment. Some same kind of fruits will have a good price when they have the same color. Color reflects the mature degree of fruits. The bruising of fragrant offen has the same color with normal area, they were easy to neglect. So the color grading and bruising inspection are two important targets of fragrant pears automatic grading. The paper provides a new algorithm of the Kurle fragrant pears’ color and bruising non-destructive based on machine vision. The contents of each chapter are described as follows:A method for color grading and druises detection was introduced. The research status of automated classification and the need for domestic and foreign intelligence testing was analyzed.According to the study the experimental systems was desined and optimized. The equipment and software system was introduced. The method of pear grading based on machine vision was researched:H component of HSI was used to be the template of image segmentation. The image of pear apart from background was gain after thresholding and binaryzation. The statistics of average and mean square deviation of surface pixels was as the import of ANN, the color information was as the output.The rate of accuracy were both 95% by two methods.According to the characteristics of pear bruising, the bruising equipment was designed and used to simulate bruise. The machine vision technology and hyperspectral imaging technology were used to detect the slight bruisings of pear. PCA and ICA was used to analysis the dates. PCA4 was found with mainly information of bruising. The rate of accuracy was 81.0% by directly processing PCA 4. The rate of accuracy was 70.9% by processing the image which fused by 528 nm,662 nm and 740 nm. The rate of accuracy was 82.5% by processing ICA 4.At least part discusses the main achievements and conclusions and some problems for further study.更多还原