【作者】 李军良；

【导师】 王从庆；

【作者基本信息】 南京航空航天大学 ， 模式识别与智能系统， 2011， 硕士

【摘要】 本文以苹果为对象,研究了基于机器视觉和近红外光谱技术的水果品质分级。建立了苹果静态漫透射近红外光谱糖度预测模型;研究了结合DSP机器视觉系统的苹果近红外光谱动态检测方法;研究了苹果外部特征与糖度的分级模型,以及基于信息融合技术的分级模型。首先对静态漫透射近红外光谱进行分析,研究了不同的预处理方法对预测模型的影响,选定合适的预处理方法;采用反向间隔偏最小二乘法对光谱变量区间进行初步定位,减少建模变量空间;采用量子进化算法对经过初步筛选的变量进一步进行优选,并与遗传算法优选效果进行了比较,最终建立了静态漫透射近红外光谱与苹果糖度的预测模型。其次,在现有的水果分级设备上,搭建了动态漫透射近红外光谱检测平台,并与DSP机器视觉系统结合,采集苹果动态近红外光谱,研究动态光谱糖度预测模型;将近红外检测功能应用在DSP水果分级系统中,编写分级软件,使系统能够同时检测水果的外部品质和内部品质,从而对水果综合品质分级。然后,采用图像处理算法,从苹果样品图像中提取出水果的大小、颜色、果形、纹理等特征,并对这些特征进行选择,建立外部特征与苹果糖度的神经网络分级模型和支持向量机分级模型;实验结果表明,采用支持向量机建立的分级模型分级准确率高于神经网络分级模型。最后,从苹果样品的近红外光谱信息中提取出光谱数据的主成分特征,做为近红外信息特征,然后采用支持向量机融合水果的外部特征和近红外主成分特征,建立了基于机器视觉和近红外信息融合的糖度分级模型,实验结果表明,采用信息融合技术建立的分级模型准确率,高于采用单一DSP图像处理系统或单一近红外检测系统所建立的糖度分级模型正确率。更多还原

【Abstract】 The thesis choses the apple as research object, studies the apple’s quality assessment based on machine vision and near infrared spectrophotometer (NIRS) technologys. This research establishes the predication model of apples’suguar based on the static diffuse transimission NIR spectrophotometer, studies the application of dynamic NIR on fruit sorting system combined machine vision, buildes the sorting model of apples’suguar with outward features and the sorting model based information fusion of outward and internal features.Firstly, the static diffuse transimission NIR spectrophotometer of apples was analysed.It studied different preprocessing methods and chosed appropriate ones. Spectroscopy’s regions were preselected by using the backward interval partial least squares (BiPLS), and quantum-Inspired evolutionary algorithms(QEA) was applied in the selection of NIR spectroscopy variables for the prediction of apple sugar degree.At the sametime, QEA model was compared to genetic algorithms(GA) model.Secondly, the detecting platform of dynamic NIR was constructed on the fruit sorting equipment. Dynamic NIR spectrophotometer was acquired with the equipment combined DSP machine vision system, and dynamic prediction model of apples’suguar was studied.The NIR detecting function was added to the sorting equipment, so it can detect outward quality and internal quality.Then,the outward features of apples including size,colour,shape and texture were extracted.Suituable outward features were chosed to build sorting models of apples’suguar with artificial neural netowrk( ANN) and support vector machines(SVM).Tht SVM model is better than the ANN model.At last, principal component features were extracted out from the NIR spectrophotometer as apples’interal features.Then outward features and internal features were fused with SVM to constructed sorting model of apples’suguar.The result shows that the sorting model with fusion technology is better than the model with single sensor.更多还原