• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >Machine vision algorithms for sensing orientation, shape and surface defects on bell peppers.
【24h】

Machine vision algorithms for sensing orientation, shape and surface defects on bell peppers.

机译：机器视觉算法，用于感测甜椒的方向，形状和表面缺陷。

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Manual inspection and sorting of bell peppers is costly and unreliable due to its subjective nature. Traditional contacting techniques seem to be unsuitable for this extremely irregular-shaped fruit. Computer vision is proposed as a promising alternative. A vision-based fresh bell pepper sorting system was developed which analyzes gray level images of six orthogonal views of the pepper for features indicative of pepper quality.; The initial phase involved the development and evaluation of image processing algorithms for computing orientation and shape classification of bell peppers. A procedure was developed for computing the orientation using the Hough transform on six orthogonal views to locate stem and blossom end centers. Novel image analysis algorithms were devised for sensing abnormal concavities (using paired gradients and their histogram adjusted to orientation) and asymmetry (using a modified medial axis technique lacking its predecessor's limitations). A two-way shape classification with a success rate of 88.6 percent was achieved.; The second phase involved detection of widely varying types of pepper defects. Unlike most industrial applications, no model pepper exists which can provide a prototype for matching purposes. The complexity mounts in the presence of non-defects such as lobe lines, blossom end pattern, stem, stem calyx and cavities. Hence, this second phase entailed the development of algorithms that effectively locate the surface defects without misinterpreting the non-defects as defects. Segmentation was achieved via a series of procedures that includes background removal, edge detection, shaded edge suppression, weak edge elimination by automatic thresholding, peripheral edge removal, edge thinning, and edge linking using a best-first search algorithm, A{dollar}sp*{dollar}.; An intelligent cost function was developed to be used in conjunction with A{dollar}sp*{dollar} algorithm. Two new cost measures namely NGSC (Neighborhood Gray-level Similarity Cost) and TEC (Texture in terms of Edgeness Contrast) were developed and a strategy to combine these two measures (NGEC) was proposed. An admissible, first ever for a closed boundary, heuristic function was developed. A search efficiency improvement in the order of seven times was demonstrated without any significant reduction in the segmentation success rate.; The final phase involved the development of algorithms for feature extraction, feature ranking, feature grouping and classification. Many textural, contextual and shape features were computed, ranked, and grouped. A successful blob classification scheme with an error rate of 8.81 percent was achieved. A two-way pepper classification with a success rate of 75.86 percent was demonstrated. A much higher success rate of 90.34 percent was achieved when a tolerance of {dollar}pm{dollar}1 class was allowed in a four-way pepper classification scheme.
机译：由于其主观性，手工检查和分类甜椒是昂贵且不可靠的。传统的接触技术似乎不适用于这种极其不规则形状的水果。提出了计算机视觉作为有希望的替代方案。开发了一种基于视觉的新鲜灯笼椒分选系统，该系统可以分析六个正交视图的灰度图像，以显示指示辣椒品质的特征。初始阶段涉及开发和评估图像处理算法，以计算甜椒的方向和形状分类。开发了一种程序，该程序使用Hough变换在六个正交视图上计算方向以定位茎和花的末端中心。设计了新颖的图像分析算法，用于检测异常凹面（使用成对的梯度及其直方图调整为方向）和不对称性（使用缺乏前身局限性的改良的中间轴技术）。实现了双向形状分类，成功率为88.6％。第二阶段涉及检测多种多样的辣椒缺陷。与大多数工业应用不同，没有模型胡椒可以提供用于匹配目的的原型。复杂性在于存在无缺陷的情况，例如叶线，开花末端图案，茎，茎萼和腔。因此，第二阶段需要开发一种算法，该算法可以有效地定位表面缺陷，而不会将非缺陷解释为缺陷。分割是通过一系列程序实现的，包括背景去除，边缘检测，阴影边缘抑制，通过自动阈值消除弱边缘，外围边缘去除，边缘细化以及使用最佳优先搜索算法A {dollar} sp的边缘链接*{美元}。;开发了智能成本函数，以与A {dollar} sp * {dollar}算法结合使用。制定了两种新的成本措施，即NGSC（邻域灰度相似性成本）和TEC（边缘对比度方面的纹理），并提出了将这两种措施相结合的策略（NGEC）。开发了一种可允许的启发式功能，这是有史以来第一个封闭边界的功能。证明了搜索效率提高了7倍左右，而细分成功率没有任何显着降低。最后阶段涉及开发用于特征提取，特征排名，特征分组和分类的算法。计算，排序和分组了许多纹理，上下文和形状特征。成功的Blob分类方案实现了8.81％的错误率。展示了一种双向胡椒分类，成功率为75.86％。在四向胡椒分类方案中，允许{dollar} pm {dollar} 1类的耐受性时，成功率高达90.34％。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号