【作者】 黄秀玲；

【导师】 郑加强； 赵茂程；

【作者基本信息】 南京林业大学 ， 机械设计及理论， 2013， 博士

【摘要】 鉴于目前我国苹果存在品质差、价格低、附加值低、进出口贸易逆差大等问题，并且针对苹果果梗、萼区的处理是分级难点的现实，本文采用集机器视觉、神经网络、自动旋转定向输送等技术为一体的苹果品质实时检测方法，研究尝试基于HALCON和VC++平台的神经网络综合分级技术以及探索使用单摄像头双平面镜同时拍摄物体3幅不同侧面图像方法，开发并研制了一套苹果分级机械装置及基于神经网络的自动分级软件系统。研究成果对促进苹果果业健康发展，具有非常现实的社会和经济意义，同时可对基于神经网络的苹果分级生产线的研制奠定理论基础，为我国苹果参与国际市场竞争提供技术保障。主要创新性成果有：（1）研究并设计了基于自动旋转定向的苹果分级机械装置，使苹果能自动进出料、自动转正。苹果自动转正后，其果梗和花萼轴线位于竖直位置，为后续图像处理提高实时性奠定基础。（2）在苹果分级机械装置的基础上，创新提出了集成设计苹果保鲜剂自动喷涂装置的思路。苹果在进行分级的同时，完成保鲜剂喷涂，将可极大提高生产率，降低生产成本，全面提高苹果品质。（3）首次研究并设计了单摄像头拍摄同时获取苹果3幅图像的系统，并用补偿镜补光照明方法获取优质图像。利用两个对称布置的平面镜，在一次拍摄过程中，得到苹果的3个不同侧面的图像，与其它分级系统相比，节省了至少2个摄像头，节约了成本。利用补偿镜补光照明方法，使获得的苹果图像清晰，有利于后续图像处理并提高分级实时性。（4）对图像拍摄误差进行分析，首次建立了单摄像头、双平面镜条件下采集苹果图像所造成的误差数学模型，结果表明，图像盲区拍摄误差为5％，且这部分盲区位于方案设计中不进行检测的果梗和花萼区。对图像拍摄景深进行分析和计算，为一次拍摄清晰的3幅苹果图像提供了理论依据。（5）研究了苹果颜色分级的新方法。研究用苹果H分量在0～60o以及210o～255o进行分割，并将分割得到的苹果红色区域面积与苹果总表面积之比作为苹果颜色分级指标，该方法具有计算速度快的优点。首次研究了苹果图像在L*a*b*空间的特征，发现每一种等级的苹果其a*b*值范围有很大区别，可以得出用a*b*作为颜色分级指标以及介于0.9~1之间为特级、0.8~0.9之间为一级、小于0.8为二级或等外的结论。（6）提出了基于Kirsch算子提取苹果果面缺陷的新方法。对于获取的苹果图像，分别用Kirsch算子、LOG算子及Sobel算子对苹果果面缺陷进行提取，Kirsch算子提取出的果面缺陷效果最好，Sobel算子次之，LOG算子最差。（7）提出将多线程技术用于苹果分级系统。将多线程技术用于苹果分级系统实现并行处理，使苹果分级图像处理速度得到了较大提高，1s可以处理20帧以上苹果图像，完全满足实时生产的需要。（8）对苹果综合分级方法进行研究，首次研究并设计了基于HALCON和VC++平台的神经网络苹果综合分级系统。用本文苹果分级系统对30个苹果进行分级，并用人工分级进行对比，结果表明，分级系统分级正确率达到96.7%，满足苹果分级要求。更多还原

【Abstract】 Due to poor apple quality and low price in China, and identifying stem/calyx of apple was difficult inapple grading, an apple grading machine and software based on techniques such as machine vision, neuralnetwork, and automatic orientation etc. was developed in this paper to improve the quality of apple in China.Integrated grading technique based on HALCON and C++and neural network was discussed. Method ofcapturing three images at the same time using one camera with two mirrors was researched. The results ofstudy could give basis of manufacturing apple grading production line.The innovative results and conclusions were list as follows:(1) An apple grading machine was designed through autorotation orientation, where apple could beautomatically feed, discharged, and orientated. The stem and calyx of apple was orientated invertical position which could save time of image processing.(2) Automatic preservatives spraying device was innovatively brought forward to be integrated withthe apple grading machine to improve productivity and apple quality, cut the production cost.(3) An image capturing system with one camera and two mirrors was designed to capture threeimages of one apple at the same time. The other four fill-in mirrors were used to reflect light,where clear image would be captured. Compared with other kind of machine vision systems, thesystem in this paper saved at least two cameras, and real-time performance was enhanced also.(4) Error of image capturing was researched. Mathematical model of error of image capturing systemby one camera and two mirrors was established. It was shown that the error of blind area of imagecapturing was only5%which was in the stem and calyx of apple and where didn’t need to detectby designing. Formula of depth of focus was established, so it could get three images of one appleclearly by one capturing.(5) New method of apple color grading was discussed. Using hue component in0～60o and210o～255o to segment apple image. The ratio of the segmentation area by using hue component in0～60o and210o～255o to the whole area of image was introduced as color grading parameter. Thismethod had advantage of fast calculation speed. L*a*b*color mode was discussed to grade applecolor, where value a*b*at0.9~1was top grade, value a*b*at0.8~0.9was first grade, valueless than0.8was second grade.(6) New method to extract apple surface defects based on the Kirsch operator was discussed. Kirsh,LOG, Sobel was used to segment surface detect of apple respectively, where Kirsch operaterextracting surface detects of apple is the best, LOG is the worst.(7) Multi-threading technology was used in apple grading system. The speed of grading was morequickly by using multi-threading technology, more than20apple images could be processedwithin1s, which could meet the needs of real-time production.(8) Integrated grading technique was researched. Integrated grading system based on HALCON andC++and artificial neural network was designed.30apples were graded using this system, gradingaccuracy was96.7%compared with manual grading.更多还原