摘要
本文分析了库尔勒香梨果形分类和质量分级技术现状及目前存在的问题。目前,由于香梨分级手段落后,效率低,造成香梨果形参差不齐,果实大小不一,严重影响其在国际水果市场的竞争力。因此,对香梨进行严格的果形分析和外观质量分级,对于采取不同的贮藏、销售和加工措施是非常有必要的。本文通过采用数字图像处理技术对香梨的果形特征参数、形状系数和几何特性参数进行了定量分析,利用基于聚类分析的综合人工神经网络算法和统计分析的方法,实现了香梨的畸形果判别分类、标准果形分类,并确定了预测香梨质量的数学模型,为香梨质量自动分级的设计和发展提供研究基础。具体研究内容和结论如下:
1.基于聚类分析的综合人工神经网络算法的香梨畸形果判别方法研究。采用均分直线法得到20个香梨的果形特征参数,选用的BP网络有一个隐含层,分别为网络隐含层和输出层选取S型对数激活函数和线性激活函数,输入为香梨的果形特征参数,输出节点为3个。利用设计的BP网络对香梨进行训练和测试。实验结果表明,平均正确率80%,分级效果较好。
2.香梨标准果形的分类方法研究。根据纵截面的形状香梨的标准果形分为近圆形、卵圆形和纺锤形三类,以纵横比Ar作为量化指标,近圆形香梨纵横比在0.98-1.12的范围内变化;而卵圆形和纺锤形香梨的纵横比的变化范围分别为1.12-1.29和1.28-1.39。根据横截面的形状香梨的标准果形分为圆状和椭圆状,以圆度比Rr作为量化指标,圆状香梨的圆度比变化范围为1.00-1.10;椭圆状香梨的圆度比变化范围为1.10-1.15;同时当香梨横截面的圆度比Rr<1.10且纵截面的纵横比Ar<1.29时,香梨果形为标准果形,而此区域之外被认为非标准果形。
3.香梨质量预测方法研究。香梨的轴径参数、投影面积参数和体积均与其质量有较好的相关性。在基于轴径的质量预测模型中,以香梨长轴α和中轴b或短轴c为自变量的线性质量预测模型更为适宜;在基于投影面积的质量预测模型中,以投影面积Pab或Pac为自变量的乘幂形式质量预测模型更为适宜;在基于体积的质量预测模型中,则推荐使用以香梨的体积值Vosp为自变量的乘幂形式模型,可进行香梨的质量准确预测。从高相关性和测量的简便性两方面考虑,最终确定以投影面积Pab为自变量的乘幂形式,即M=0.0079Pab1.197,R2=0.9899作为最佳的质量预测模型。
4.库尔勒香梨果形分析与外观质量自动分级软件开发。编写了基于Matlab的库尔勒香梨果形分析及外观质量自动分级软件系统,采用面向对象的程序设计方法,各模块分别实现香梨的畸形果判别、标准果形的分类和香梨的质量预测,实现了库尔勒香梨果形分析和外观质量的自动分级。
Since current methods for classification and analysis of fruit shape and mass for Korla fragrant pear are behindhand, this study aimed to develop feasible methods to determine the classification of Korla fragrant pear shape and models for the mass predicting. The shape characters and geometrical parameters are determined using image processing methods. The discriminate on normal and misshapen fruit shape, the classification of normal fruit shape, models for mass prediction of Korla fragrant pears based on these geometrical parameters are identified. All above can supply basis for the research of mass classification using machine vision method. Main contents and conclusions of this study were listed as follows:
The study of discriminate of normal and misshapen fruit shape based on the integrated neural network ensemble (InNNE) based on clustering technology. The 20 shape characters are determined by the straight-line divided method from the edge of Korla fragrant pear, which are the input layer of the BP neural network. The activation function of output layer are the logarithm and linear on, which include 3 nodes.. It can be concluded that the average accuracy is about 80%.
The study on classifying for normal fruit shape of Korla fragrant pear:three normal fruit shapes in Korla fragrant pear based on the A.R are sub orbicular, ovoid and spindle. The A.R of sub orbicular pear range from 0.98 to 1.14, and for ovoid and spindle pear is 1.15 to 1.27 and 1.28 to 1.39.
The study on predicting models of the mass for Korla pear:Different linear and nonlinear regression models for mass prediction of Korla fragrant pears based on these geometrical parameters were identified. The results indicated that among the mass prediction models of Korla fragrant pear based on diameters, the linear model with the independent variables as major and intermediate diameter (or minor diameter) had higher correlation. Also, among the mass prediction models based on the projected area, the polynomial model with the independent variable as the projected area which includes the major diameter can be optimum. The results showed that it is feasible to predict the mass of Korla fragrant pear use the model M=0.0079Pab1.197, R2= 0.9899, which is more feasible.
The development of automatic classification software for Korla fragrant pear:the Object oriented programming method is used to design the automatic classification software for Korla fragrant pear based on MATLAB, All the modules realized the discriminate of normal and misshapen fruit shape, classifying of the normal fruit shape and mass predicting.
引文
[1]库尔勒香梨产业化发展研究课题组.库尔勒香梨产业化发展的思路和建议[EB/OL]. http://kjj.xjb z.gov.cn/html/2006/2008-4/9/2008_04_09_668.html,2008-04-09.
[2]高启明,侯江涛,李疆.库尔勒香梨生产现状与与研究进展[J].中国农学通报,2005,21(2):233-236.
[3]张强,祁春节,彭抒昂.中国梨果国际竞争力的研究[J].世界农业,2006,12(8):34-37.
[4]Currie A J, Ganeshanandam S, Notion D A, et al. Quantitative evaluation of apple (Malus× d omestica Borkh.) fruit shape by principal component analysis of Fourier descriptors [J]. Euphyt ica,2000,111:219-227.
[5]Rashidi M, Seyvan K. Modeling of Kiwifruit mass based on outer dimensions and projected area[J]. American-Eurasian J.Agric.& Environ Sci,2008,3(1):26-29.
[6]Keramat Jahromi M, Jafari A, Rafiee S, et al. Some physical properties of date fruit (cv. Lasht)[J]. Agricultural Engineering International,2007,3(5):1-7.
[7]Sadrnia H, Rajabipour A, Jafary A, et al. Classification and analysis of fruit shapes in long type watermelon using image processing [J]. International Journal of Agriculture & Biology,2007,9(1): 68-70.
[8]Beyer M, Hahn R, Peschel S, et al. Analysing fruit shape in sweet cherry [J]. Sceintia Horticulturae, 2002,96:139-50.
[91 United States Department of Agriculture. United States Standards for Grades of Watermelons[S].
[10]Platt R G. Avocado varieties recently registered with the California avocado society [Z]. California Avocado Society,1975,59:41-51.
[11]梨NY/T440-2001.中华人民共和国农业行业标准[S].
[12]库尔勒香梨NY/T585-2002.中华人民共和国农业行业标准.[S].
[13]高敦禄.库尔勒香梨也有“公”和“母”[J].新疆农垦科技,1988,10(2):12.
[14]亚合甫·木沙,热洋古丽·木沙.PBO对库尔勒香梨果形与品质的影响试验[J].山西果树,2007,(1): 9-10.
[15]张大海,徐庆岫,李利民等.库尔勒香梨果形变化规律研究[J].新疆农业科学,1999,(6):261-263.
[16]周祖锷.农业物科学[M].北京:农业出版社,1994:55-62.
[17]林开颜,吴军辉,徐力鸿.基于计算机视觉技术的水果形状分级方法[J].农业机械化学报,2005,5(6):72-74.
[18]Rashidi M, Seyvan K, Gholami M. Determination of Kiwifruit volume using image processing [J]. ARPN Journal of Agricultural and Biological Science,2007,2(6):17-22.
[19]高华,王雅琴,基于计算机视觉的农产品分级系统研究[J].计算机工程与与应用2004,6(14):228-229.
[20]Wright M E, Tappan J H, Sistler F E. The size and shape of typical sweet potatoes [J]. Tra nsactions of the ASAE,1986,29(3):678-682.
[21]Pitts M J, Hyde G M, Cavalieri R P. Modeling potato tuber mass with tuber dimensions [J]. Transactions of the ASAE,1987,30(4):1154-1159.
[22]Tabatabaeefar A, Rajabipour A. Modeling the mass of apples by its geometrical attributies [J]. Scientia Horticulture,2005,105:373-382.
[23]Lorestani A N, Tabatabaeefar A. Modelling the mass of kiwi fruit by geometrical attributes [J]. Agrophysics,2006,20:135-139.
[24]赵燕超,徐丽明.基于多元聚类分析的草莓形状分类算法[J].中国农业大学学报,2008,13(1):77-80.
[25]宇传华SPSS与统计分析[M].北京:电子工业出版社,2007:459460.
[26]苏金明,张莲花,刘波等MATLAB工具箱应用[M].北京:电子工业出版社,2004,102-14
[27]张润楚.多元统计分析[M].北京:科学出版社,2006,241-244.
[28]马锐.人工神经网络原理[M].北京:机械工业出版社,2010,86-92.
[29]高隽.人工神经网络原理及仿真实例[M].北京:机械工业出版社,2007,43-50.
[30]丛爽.面向Matlab工具箱的神经网络理论与应用[M].北京:中国科学技术大学出版社,2009,204-232.
[31]袁曾任,卢振中.由神经网络提取规则的一种方法及其应用[J].信息与控制,1997,26(1):61-66.
[32]窦柳明,廖宁放等.涂料颜色配方预测的人工神经网络模型实验研究[J].光学技术,2008,34(1):116-120.
[33]李耀勇,郑南宁.前馈神经网络的隐结点个数与网络推广能力的关系[J].西安交通大学学报,1996,30(5):22-30.
[34]Li Yaoyong, Zheng Nanning. ImProved dynamical model for neuralnetwork and the stabilizatio n of associative memory [J]. PROGRESS IN NATURAL SCIENCE,1998,8(5):610-619.
[35]包晓安,张瑞林,钟乐海.基于人工神经网络与图像处理的苹果识别方法研究[J].农业工程学报,2004,20(3):109-113.
[36]潘磊庆,屠康等.基于计算机视觉和神经网络检测鸡蛋裂纹的研究[J].农业工程学报,2007,23(5): 154-159.
[37]李小昱,汪小芳等.基于机械特性BP神经网络的苹果贮藏品质预测[J].农业工程学报,2007,23(5): 150-154.
[38]陈旭生,李渊博.基于BP网络的车牌数字识别[J].光盘技术,2009,23(8):37-40.
[39]李凯,黄厚宽.一种基于聚类技术的选择性神经网络集成算法[J].计算机研究与发展,2005,42(4):594-598.
[40]李国正,杨杰,孔安生,等.基于聚类算法的选抒性神经网络集成[J].复旦大学学报,2004,43(5):689-692.
[41]李锦卫,廖桂平.基于数字图像处理的油菜种子形状特征参数提取及模糊聚类分析[J].中国农业科学,2006,38(9):186-190.
[42]齐新战,刘丙杰,冀海燕.基于聚类分析的综合神经网络集成算法[J].计算机仿真,2010,27(1):166-170.
[43]应义斌.水果性状的傅立叶描述子研究[J].生物数学学报,2001,6(2):234-240.