【作者】 李庆中；

【导师】 汪懋华；

【作者基本信息】 中国农业大学 ， 农业电气化与自动化， 2000， 博士

【摘要】 高品质、绿色农产品的生产是21世纪农业发展的重要方向之一。在国际市场上，高品质意味着高收益。目前在我国，因缺乏农产品品质检测标准和手段，使各种农产品混杂上市，在国际市场上缺乏竞争力，初级农产品得不到增值，给国家和农民造成了巨大的经济损失，中国加入WTO后，这个问题会显得更加突出。因此，现在迫切需要进行农产品品质、农业化学品残留等污染物的监测新技术的开发和研究。水果的外部品质主要涉及大小、形状、颜色和表面缺陷。目前，国外在利用机器视觉技术进行水果大小、形状和颜色自动分级方面，已达到实用化水平；但因缺陷检测的复杂性，外部缺陷的快速自动检测还一直处在继续研究阶段。国内从90年代开始进行这方面的研究，但主要集中在水果静态图象处理和识别方法的研究上，因算法处理速度的限制，至今尚未开发出自动分级的商品化产品。本文以提高基于计算机视觉的图象信息处理速度和探索水果自动分级机械的产业化开发为目标，应用计算机视觉信息技术与机电一体化技术，对苹果在线状态下，计算机视觉信息快速获取和处理技术进行了研究，完成的主要研究成果如下： （1）图象低层处理算法是利用计算机视觉进行水果在线快速分级的基础。本文首先对图象低层处理中的平滑、增强算法进行了有效性和实时性研究，得到了适合苹果在线快速分选的图象增强算法和平滑算法。在图象边缘检测方面，提出了基于分形特征的图象边缘检测新方法，该方法具有较强的抗噪声性能。对基于类别方差和信息熵的两种图象分割算法进行了研究，提出了适合水果在线状态下，果实与背景快速分割的算法。 （2）水果表面缺陷的识别复杂而费时，一直是实现水果自动化分级的障碍。为提高缺陷分割的速度，本文提出一种基于参考图象的缺陷快速分割算法，可有效、快速地分割出在线状态下水果表面的缺陷。真正缺陷区与梗萼凹陷区的快速识别是国内外研究的另一热点，因为这是造成分级误差的主要原因。为此，本文提出基于有向分形维数和人工神经网络的快速识别新方法，并提出双金字塔数据形式的盒维数快速计算方法。以上两个难题的解决，为实现水果表面缺陷的自动化检测奠定了基础。 （3）国内研究者提出的水果形状、大小识别算法，都是围绕水果中心轴线来提取特征参量，而水果在分选生产线上的状态是随机的，故这些特征参量的提取方法不再适合于在线状态。为此，本文提出基于参考形状的苹果果型相关评判和果径计算新方法，并利用改进的遗传算法提高处理速度，试验表明这是一种快速而有效的方法。 （4）水果颜色的在线快速分级，要求系统在快速获取表面颜色信息的同时，对 博士 论文 一 水果的着色率进行实时评价。本文对颜色信息快速获取的方法和有效颜色特征的 提取方法进行了研究，并利用遗传算涪实现了颜色分级神经网络的结构设计和权 值学习，克服以往神经网络设计时，结构人为确定带来的弊病，可有效缩短神经 网络的设计和训练时间。 ＊）为了尽快开发利用计算机视觉进行水果自动化分级的实用化产品，本文 通过多视觉、多光谱、光学成象以及机电技术的有机集成，提出了一种快速获取 水果全视角图象的新方法，并完成了该系统的研制。该系统由苹果输送机构、机 器视觉成象、处理系统和照明室三部分组成，可完成苹果在线状态下图象信息的 4”””“”—’—”——””一—”“”『“““’”””“————”’””——’—””“—’一’”‘”“”””‘“”””－”‘“””“ 快速采集和处理，为开发商品化产品提供了可行的途径。 …）根据苹果外部品质各项指标在线检测实现的方法和实时分级的速度要求， 确定了合理的在线分级方案。用 Visual＋＋5．0编程，综合设计了苹果形状、大 小、颜色和表面缺陷自动分级的软件系统。最后利用开发的硬件和软件系统对3 种苹果进行了检测、分级综合试验。试验结果表明：系统分级的正确率在 85％以 上，利用联想Pill—500微机系统检测分级1个苹果表面缺陷所用的时间为320ms， 检测分级 1个苹果的形状、大小和颜色，所用的时间平均小于 2 10 ms。分级速度 已基本达到在线分级的要求。 本研究对于我国开展基于视觉技术的水果自动分选系统高新技术产品的开 发，具有重要参考价值，对高品质水果的增值和推进创汇农业科技活动，提高提 高农业国际竟争能力有重要的经济意义。更多还原

【Abstract】 The production of high qualitative and unpolluted farm produce is one of the important developing trends of agriculture in the new century. High quality means high profit in international market. However, the farm produce with different varieties and different quality have cause tremendous losses in economy due to lacking the post- harvest inspecting standards and measures in China, and this problem will become more grave with China entering the WTO. So it is urgent to carry out the researches on agricultural produce quality measurement and chemical deposits detection. The inspection items for fruit external quality involve size. shape. color and surface defects. Although some quality inspection operations such as color, size, and shape are performed by automated system in western countries, the automation of the defect sorting process is still a challenge subject due to the complexity of the problem. In China, the researches for fruit automatic grading based on computer vision were started in 1990. However, these researches were only concentrated on methodology development for static fruit images. Up to now no automatic grading system is put into practical use in fruit packinghouse in China due to the limitation of processing speed. The main objective of this paper is to design and construct a on-line fruit sorting experimental hardware system based on computer vision, and develop the corresponding image processing and recognition algorithms suitable for real-time apple grading operation. Based on this goal, the following research achievements have been fulfilled: (1) Some fast image preprocessing methods suitable for on-line apple automatic grading including image smoothing .. enhancement were determined through investigations and experiments. Moreover, a new edge detection method with higher resistance to noise was developed and a fast segmentation method for on-line apple grading was also proposed. (2) Real-time fruit surface defect inspection and recognition is still a challenging subject due to its complexity. Based on a reference image of apple, a novel defect segmentation algorithm was developed. The method can compensate the reflectance intensity gradient on curved objects and provide flexibility to cope with the variations in brightness and size of fruit The test results demonstrated the effectiveness of the proposed algorithm for defect segmentation on apples. This method can be applied to other fruits with curved shape. Moreover, an effective approach for box-dimension estimation based on a dual-pyramid date structure was developed. Utilizing traditional fractal dimension and 4 oriented fractal dimensions as input values, a BP neural network is designed for identifying fruit defect area and stem, calyx concave area. Experimental results show that the approach is effective for real-time defect extraction and identification. (3) Based on correlative theory a fast and effective shape judgement method and a size calculating method have been developed. Genetic algorithm was used to reduce the processing time for this method. The test results demonstrate the effectiveness of the proposed algorithm for on-line fruit shape grading. (4) A method using HIS color feature and neural network techniques for fruit color inspection was developed. A GA-based training algorithm was introduced to find optimal structure and connection weights of artificial neural network. This approach can overcome the drawbacks of the traditional designing method of artificial neural network in which the structure was predetermined. The更多还原