脐橙表面缺陷的机器视觉快速检测研究及嵌入式系统应用

发布时间：2018-05-03 17:27

本文选题：机器视觉 + 脐橙　；参考：《浙江大学》2017年博士论文

【摘要】：缺陷检测是水果分级的重要环节之一。由于水果表面缺陷呈多样性和复杂性等特点,水果表面缺陷的快速检测一直是学术界和工业界的研究热点。近年来,计算机视觉技术逐渐被应用于农产品的外部质量检测。本文以脐橙为对象,利用机器视觉技术详细地研究探讨脐橙表面8种类型常见缺陷(蓟马果、溃疡果、裂伤果、日灼果、药伤果、风伤果、虫伤果、介壳虫果)检测方法并提出相应的算法,同时开发了脐橙缺陷的机器视觉自动化检测试验系统,所提出的检测方法对研发快速脐橙缺陷分级装备奠定了基础。论文的主要研究内容及成果如下:1)搭建适合水果表面缺陷检测的视觉试验系统平台,包括基于传统PC计算机视觉为基础的可见光RGB成像系统、基于嵌入式系统的嵌入式机器视觉系统。2)针对类球形水果表面亮度分布不均匀干扰检测现象,提出了一种新颖的脐橙表面缺陷快速多阈值边缘分割算法。该方法能成功检测出表面缺陷区域。此分割算法对蓟马果、溃疡果、裂伤果、日灼果、药伤果、风伤果、虫伤果、介壳虫果等8类型缺陷脐橙的5008个感兴趣区域进行分割,获得了92%分割精度。3)提出了一种新颖的脐橙表面灰度局部阈值快速分割算法。该方法能克服类球形水果表面亮度分布不均匀问题,该算法将积分图理论和局部阈值计算相结合并成功分割脐橙表面缺陷区域,此分割算法对蓟马果、溃疡果、裂伤果、日灼果、药伤果、风伤果、虫伤果、介壳虫果等类型缺陷脐橙进行检测,获得了95.2%检测正确率,每一幅离线图像处理时间是38.5ms。4)提出了一种新颖的脐橙表面亮度分布不均的快速自适应亮度矫正及单阈值快速水果缺陷分割算法。该方法能使正常水果组织表面区域被矫正为高灰度区域,而水果表面的缺陷区域仍然保持低灰度区域,矫正克服了类球形水果表面亮度分布不均导致缺陷检测误检的问题,这也为单阈值脐橙表面缺陷快速检测提供了可能性。依据不同光照成像环境下试验表明,该算法可以直接对脐橙表面整体亮度进行自适应矫正,并且该自适应亮度变换算法比现有文献水果亮度变化算法相比速度快10倍以上。5)提出一种基于低成本小型多核嵌入式机器视觉的在线脐橙缺陷检测自动化系统,并研究了其软件和硬件系统的实现,包括嵌入式在线图像采集实现,光照成像环境实现,Linux嵌入式系统千兆网工业相机等相关硬件底层驱动设计,在线机器视觉的图像算法设计,Linux嵌入式系统在线检测软件架构与设计等。试验结果表明该系统可以在单通道每秒7个脐橙的速度下,检测正确率达到95.8%。
[Abstract]:Defect detection is one of the most important steps in fruit grading. Because of the diversity and complexity of fruit surface defects, rapid detection of fruit surface defects has been a hot spot in academia and industry. In recent years, computer vision technology has been gradually applied to the external quality detection of agricultural products. In this paper, eight common defects on navel orange surface (thrips, ulcer fruit, laceration fruit, sunburn fruit, medicinal fruit, wind fruit, insect fruit) were studied in detail by using machine vision technology. A machine vision automated testing system for navel orange defects was developed, which laid a foundation for the development of fast navel orange defect grading equipment. The main contents and results of this paper are as follows: (1) A visual test system platform for fruit surface defect detection is built, including a visible light RGB imaging system based on traditional PC computer vision. An embedded machine vision system based on embedded system. 2) A novel fast multi-threshold edge segmentation algorithm for surface defects of navel orange is proposed to detect the uneven luminance distribution on the surface of spherical fruit. This method can detect the surface defect area successfully. The segmentation algorithm was used to segment 5008 regions of interest in eight types of defective navel oranges, such as thrips, ulcerated fruit, lacerated fruit, sunburn fruit, medicinal fruit, wind fruit, insect fruit, mesoderma fruit, etc. The accuracy of 92% segmentation is obtained. 3) A novel fast segmentation algorithm for navel orange surface grayscale local threshold is proposed. This method can overcome the problem of uneven luminance distribution on the surface of globular fruit. The algorithm combines the integral graph theory with the local threshold calculation and successfully divides the surface defect area of navel orange. The algorithm is used to segment thrips, ulcers and cracked fruit. The accuracy of detecting navel orange with sunburn fruit, drug wound fruit, wind wound fruit, insect fruit, mesoderm fruit, etc., was obtained by 95.2%. The processing time of each offline image is 38.5 ms.4) A novel fast adaptive luminance correction algorithm with uneven luminance distribution on the surface of navel orange and a fast fruit defect segmentation algorithm with single threshold are proposed. This method can make the normal fruit tissue surface area be corrected as a high gray area, while the defect area of the fruit surface remains a low gray level area, which overcomes the problem that the uneven luminance distribution on the surface of globular fruit results in the false detection of defects. This also provides the possibility for rapid detection of single threshold navel orange surface defects. Experiments under different illumination imaging conditions show that the proposed algorithm can be used for adaptive correction of the overall brightness of navel orange surface directly. In addition, the adaptive luminance transform algorithm is more than 10 times faster than the existing fruit luminance change algorithm. (5) an online navel orange defect detection automation system based on low-cost multi-core embedded machine vision is proposed. The realization of software and hardware system is studied, including the realization of embedded online image acquisition, the realization of Linux embedded system gigabit network industrial camera, and the realization of illumination imaging environment. Image algorithm Design of online Machine Vision and Linux embedded system On-line Detection Software Architecture and Design. The experimental results show that the system can detect correctly at the speed of 7 navel oranges per second in a single channel.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TS255.7;TP391.41

【参考文献】