基于小波描述子的水果果形分类  

Fruit shape classification based on wavelet descriptor

作　　者：刘仰龙[1] 王从庆[1] 高珏 许荣华 

机构地区：[1]南京航空航天大学自动化学院,江苏南京210016 [2]扬州福尔喜果蔬汁机械有限公司,江苏扬州225105

出　　处：《浙江大学学报（农业与生命科学版）》2010年第3期322-328,共7页Journal of Zhejiang University:Agriculture and Life Sciences

基　　金：江苏省科技支撑计划资助项目(BE2008353)

摘　　要：水果果形是水果分级的重要指标之一,该文提出了一种基于小波描述子的水果果形分类方法.通过提取水果轮廓计算出半径序列,并进行归一化处理,对归一化后的半径序列进行小波变换提取小波描述子;分别截取小波描述子12、20、36和67个系数点对水果边界进行重建.结果表明:用36个系数点就可较好地重建果形,匹配率为98.64%,用67个系数点可达相当高的精度,为99.96%;选取36个系数点作为果形特征,并运用核主成分分析(KPCA)提取分类所需的7个主要特征输入径向基函数(radial basis function,RBF)神经网络进行分类,发现该方法分级准确率可达90%,效果优于傅里叶描述子,是一种有效地描述水果果形的方法.Shape is one of the most important features in fruit grading.A method for fruit shape classification based on the wavelet descriptor was proposed.Wavelet descriptor was extracted to describe the fruit shape from wavelet transform of the normalized radius sequence which was calculated from the fruit contour.Twelve,twenty,thirty-six and sixty-seven wavelet descriptor coefficients were used on the reconstruction of the fruit border,respectively.It was found that thirty-six coefficients with matching rate of 98.64% were sufficient for representing the shape features,and more accuracy was obtained with sixty-seven coefficients with matching rate of 99.96%.The seven main features were selected as the radial basis function （RBF） neural network input from the thirty-six coefficients using kernel principal component analysis （KPCA）.The result indicated that the identification accuracy reached 90% which was better than the Fourier descriptor,showing that this method based on the wavelet descriptor was effective to describe the fruit shape.

关 键 词：水果分级 果形分析 小波变换 小波描述子 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] S661.1[自动化与计算机技术—计算机科学与技术]