胡发焕+董增文+

摘要：为实现利用机器视觉代替人工视觉对脐橙进行品质分级检测，采用数字形态学方法把脐橙从背景中分离出来，并提取脐橙的体积、果面缺陷、颜色和纹理等几个主要特征；以这些特征量为支持向量机的输入特征向量进行ＳＶＭ分类器训练；最后用该分类器进行脐橙分级检测。结果表明，该分类器正确识别率可达９０．５％，单个脐橙处理时间为１６５ ｍｓ，具有识别率高、实时性好的特点，适合于实时环境下的脐橙分级检测。

关键词：机器视觉；支持向量机；品质分级；脐橙

中图分类号：ＴＰ３９１文献标识码：Ａ文章编号：0439－８114（２０14）09－2160-05

Online Grade Detection of Navel Orange Quality Based on Machine Vision

HU Fa-huan1,2,DONG Zeng-wen2

(1. School of Mechanical & Electrical Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, Jiangxi, China; 2.School of Mechanical & Electrical Engineering, Nanchang University, Nanchang 330031, China)

Ａｂｓｔｒａｃｔ： To replace artificial vision with machine vision and realize grade detection of navel orange quality, mathematical morphological was used to separate navel orange from background. The bulk features, surface defect features, colour features and texture features were extracted and used as the input feature vectors of the support vector machine, support vector machine was used in training and classification of those feature. The trained classifier was used to detect the navel orange. Results showed that the classifier had high rate of correct identification and real-time , which could be used in real-time detection of navel orange.

Ｋｅｙ ｗｏｒｄｓ： machine vision; support vector machine; quality grade; navel orange

脐橙的品质分级检测是脐橙产后处理和市场销售的一个重要环节，我国水果传统分级是采用人工识别的方式。随着计算机图像处理技术和机器视觉技术的发展，国内外专家学者对利用机器视觉进行农产品的分级检测进行了大量的研究，但主要侧重于利用颜色、形状、大小、纹理等某一个或几个特征进行分级［１，２］，因特征信息较少，分类效果不是很理想。为此，采用提取脐橙的大小、颜色、表皮纹理、果面缺陷等综合信息特征，建立了基于支持向量机的脐橙综合评价模型，实现对脐橙品质分类检测。江西赣州是我国主要的脐橙生产地，采用赣州出产的纽荷尔脐橙为对象，利用机器视觉技术，实现对脐橙的自动分级检测，较人工方法提高了效率，降低了加工成本，为脐橙的品质分级检测提供了新的方法。

１试验设备和仪器

设备采用三星ＳＣＣ－１３１Ａ摄像头，腾龙１／３ ＣＣＴＶ ＣＳ ＴＡＭＲＯＮ镜头，ＭｙＶｉｓｉｏｎ９９图像采集卡，图像采集处理系统如图１。……