【作者】 杨国彬；

【导师】 赵杰文；

【作者基本信息】 江苏大学 ， 农产品加工及贮藏工程， 2003， 硕士

【摘要】 本文的研究目的是实现应用计算机对自然场景下的成熟西红柿进行自动识别，为将来能够实现用利用机器人对水果进行自动化采摘打下基础。本文的主要研究内容如下： 1．提出了基于颜色特征对自然场景下的成熟西红柿进行识别的方法及对其质心位置的确定。 根据对在自然场景下拍摄的西红柿图像中的成熟西红柿、未成熟西红柿、叶子(枝干)等的颜色特征进行了仔细地分析和研究，分别提出了基于两种颜色模型对西红柿进行识别的方法，即：HIS系统识别方法和RGB系统识别方法。利用颜色识别方法对西红柿图像进行处理后，为了准确地确定成熟西红柿质心位置，提出了黑体检出算法对经过识别处理后的西红柿图像中出现的一些误判的小面积区域进行消除处理。提出了划定西红柿区域范围的分割算法，使整个西红柿图像缩小到一个包含成熟西红柿区域在内的较小区域，排除了大量的干扰信息，为进一步识别西红柿提供了非常有利的条件。通过空洞填充算法，对在特殊条件下识别出来的西红柿区域中出现的空洞进行填充处理。然后采用了矩方法计算质心坐标，确定出所识别出来的西红柿区域在二维图像中的准确位置。对识别过程中出现的一些误判现象也作了细致的分析。 2．为了使将来的机器人采摘范围更加广泛，本文对基于形状特征对自然场景下西红柿外形轮廓检出也进行了探索。 为了提取图像中各对象的形状特征，必须对图像进行平滑、去噪、边缘检测、细化、消去短枝及补缺等一系列的前期处理过程。本文采用中值滤波对图像进行了平滑、去噪处理。通过改进的Sobel算子对图像进行了边缘检测处理。提出一种双侧双链表分割算法对图像进行分割。研究了西红柿、叶子、枝干的形状特征，通过傅立叶描绘子，提取基于傅立叶系数导出的如圆形度、细长度、凹度、密集度等形状特征。再根据统计的特征值建立适当的分类器模型对西红柿外形轮廓进行检出。由于本文是对自然场景下的彩色西红柿图像进行处理，其情况复杂，处理难度很大，因此本文没有提出利基于形状特征对西红柿轮廓进行检出的具体方法，只是提出一种轮廓检出的设计方案，为最终能够实现基于形状特征对西红柿外形轮廓进行检出做了一些基础性的研究工作。更多还原

【Abstract】 The objective of this study was that mature tomatoes in natural outdoor scenes could be automatically discriminated from images by computer, and grounded for automatically picking fruits by robots .The main contents are:1. Proposing discriminating methods that discriminated the red color of the tomato based on color features in natural outdoor scenes and finding the tomato’s centroid.Putting forward two classifying method under two color model , namely HIS discriminating method and RGB discriminating method ,based on studying and analyzing the color features of mature tomatoes , immature tomatoes and leaves in natural outdoor scenes . After processing tomatoes images , advancing a picking black body arithmetic which expunged a lot of small district which were wrongly discriminated for calculating the coordinate of the tomato’s centroid by rule and line .Advancing a filling in inanition arithmetic for filling in inanition of district of tomatoes which was discriminated in especial circumstances . Making sure the exact address of the district of tomatoes which were discriminated . Putting forward a fixing on bound of the tomatoes’ districts arithmetic for reducing bound of images , after processing , it offered a very advantaged condition for discriminating tomatoes more . Analyzing carefully some wrongly discriminating phenomena too.2. More discussing a discriminating project that discriminated the tomato’s profile based on shape features.In order to extract shape features , images were processed by smoothing , wiping off noise , extracting confine , thinning , expunging small branches and filling a vacancy processes and so on . In this paper, median filter was used to smooth images. Advancing improved sobel arithmetic for extracting confine. Putting forward a Double linked List with two sides arithmetic for segmenting images . On the basis of analyzing shape features of tomatoes , leaves and branches ,we extract features of roundness , slightness , concavity degree and denseness depending on Fourier description . So we will found an appropriate classifying model to discriminate tomato’s profile from images. Because color images in natural outdoor scenes was very difficult to be processed , this paper didn’t put forward a concrete discriminating tomato’s profile way based on shape features and only brought forward a discriminating project and did some basic studies for realizing to discriminate tomatoes based on shape features at last.更多还原