Study on the Model of Excessive Staminate Catkin Thinning of Proterandrous Walnut Based on Quadratic Polynomial Regression Equation and BP Artificial Neural Network

Xianping WANG,Guishou CAO,Xiaohua YANG,Qianru ZHANG,Kai LI,Hongyan LI,Zemin DUAN

Pomology Institute,Shanxi Academy of Agricultural Sciences/Shanxi Provincial Key Laboratory of Fruit Germplasm Innovation and Utilization,Taiyuan 030031,China

Study on the Model of Excessive Staminate Catkin Thinning of Proterandrous Walnut Based on Quadratic Polynomial Regression Equation and BP Artificial Neural Network

Xianping WANG,Guishou CAO,Xiaohua YANG,Qianru ZHANG,Kai LI,Hongyan LI,Zemin DUAN*

Pomology Institute,Shanxi Academy of Agricultural Sciences/Shanxi Provincial Key Laboratory of Fruit Germplasm Innovation and Utilization,Taiyuan 030031,China

The excessive staminate catkin thinning(emasculation)of proterandrous walnut is an important management measure for improving yield.To improve the excessive staminate catkin thinning efficiency,the model of quadratic polynomial regression equation and BP artificial neural network was developed.The effects of ethephon,gibberellin and mepiquat on shedding rate of staminate catkin of proterandrous walnut were investigated by modeling field test.Based on the modeling test results,the excessive staminate catkin thinning model of quadratic polynomial regression equation and BP artificial neural network was established,and it was validated by field test next year.The test data were divided into training set,validation set and test set.The total 20 sets of data obtained from the modeling field test were randomly divided into training set(17)and validation set(3)by central composite design(quadric rotational regression test design),and the data obtained from the next-year field test were divided into the test set.The topological structure of BP artificial neural network was 3-5-1.The results showed that the prediction errors of BP neural network for samples from the validation set were 1.355 0%, 0.429 1%and 0.353 8%,respectively;the difference between the predicted value by the BP neural network and validated value by field test was 2.04%,and the difference between the predicted value by the regression equation and validated value by field test was 3.12%;the prediction accuracy of BP neural network was over 1.0%higher than that of regression equation.The effective combination of quadratic polynomial stepwise regression and BP artificial neural network will not only help to determine the effect of independent parameter but also improve the prediction accuracy.

Walnut;Staminate catkin of walnut(SCW);Thinning;BP artificial neural network;Regression;Prediction

W alnut(Juglans regia)originated in China.Scientific investigation and geological excavations have proven that more than 2 500 years ago or in earlier period,there had been six walnut species. China is one the world’s three major centers of origin of walnut.In the 1st-2ndcentury,the economic cultivation of walnut appeared.A total of 380 walnut germplasms have been identified[1], and they are widely distributed in the northwestern,northern and eastern provinces.The acreage and yield of walnut in China all rank first in the world.In 2010,the harvested acreage of walnut was about 300 000 hm2,and the yield was about 3 541.2 kg/hm2. Compared with the general yield(6-7 t/hm2)of walnut in America,the gap is very obvious[2].

Walnut is a monoecious and cross-pollinated plant,including proterandrous and protogynous types.

Materials and Methods

Field test

The test was carried out in the Pomology Institute of Shanxi Academy of Agricultural Sciences in 2004.The tested walnut species was 28-year-old Liaohe No.1.According to the design requirements by BP neural network, the field tests were divided into BP modeling test and BP model validating test.Based on the competition of BP modeling test,the validating test was carried out next year.The substances used for emasculation included 95% ethephon(Eth),99%gibberellin(GA) and 95%mepiquat(Pix).

Design and conducting of mathematical modeling testThe test adopted the quadratic general rotary utilized design of central composite design(CCD).A total of 3 factors,5 levels and 20 test combinations were designed.Based on X1(Eth),X2(GA), X3(Pix)and effective concentration ranges of ethephon,gibberellin and mepiquat,the encoding was performed using linear transformation according to following equations(Table 1):

Wherein,Z2j,Z1jand Z0jrepresent the upper,lower and zero levels of each factor;△j and R represent the variation interval and asterisk arm value of each tested factor.

The test was carried out in Jinzhong at the expanding-elongation phase of staminate catkin of walnut. The walnut branches with uniform growth were selected.The test adopted the randomized block design. There were 3 replicates for each test combination,and there were 80-120 male flowers in each replicate.After a 24-h treatment,the shedding rate of staminate catkin in each test combination was investigated day by day in the first 5-7 d.The effect of each test treatment was expressed as the average shedding rate of staminate catkin.

Design and conducting of mathematical model validation test

Based on the modeling test,the model validating test was carried out next year.The tested material,test location and investigation method were all as described above.According to the need to retain an appropriate proportion of staminate catkin and the results of modeling test,the shedding rate of staminate catkin of walnut was controlled in the range of 85%to 95%.The water treatment was treated as CK. The encoded values of X1,X2,X3were 0.087 9,-0.504 5 and 0.025 6,respectively.

Design of BP artificial neural network

Constitution of training set,validation set and test set of BP neural networkAccording to the requirements by BP neural network design, all the test data obtained from the 2-year field tests were divided into training set,validation set and test set. The 20 set of data obtained by BP modeling test were divided into training set and validation set.The training set was composed of 17 sets of data,and the validation set was composed of 3 sets of data.All the 3 sets of data obtained by BP model validating test in the next year were divided into the test set.

Topological constitution of BP neural networkThe BP neural network was composed of input layer,output layer and hidden layer.The number of nodes in the input layer was the number of tested factors(n=3);and the number of nodes in the output layer was the number of response index (m=1);the number of nodes in the hidden layer was determined by comparing the effects of different network parameters on fitting residual.

Analysis and calculation of test results

The quadratic polynomial regression analysis of test results,the analysis of test results by BP neural network model and the generation of figures and tables were all performed using the DPS 14 software.The prediction error of different model was calculated according to the following formula:

Results and Analysis

Test results

The effects of various tested factors(X1,X2,X3)and treatment combinations on shedding rate of staminate catkin of walnut(Y,%)were shown in Table 2.For the central composite design,the change range of average staminate catkin shedding rate of walnut reached 24.29%.So it was indicated that the results of field test were affected by a variety of factors.

Regression analysis of test results

In accordance with Table 1,the tested factors of X1,X2,X3were treated as independent variables,and the average staminate catkin shedding rate of walnut was treated as dependent variable.Then the quadratic polynomial stepwise regression analysis was performed. The obtained mathematical regression formula of the objective function was as follows:

The values of multiple correlation coefficient(R),determination coefficient(R2),residual standard deviation (SSE),adjusted correlation coefficient(Ra)and adjusted determination coefficient(Ra2)were 0.909 869, 0.827 862,11.826 6,0.865 109 and 0.748 414,respectively.

As shown in the mathematical regression model of the objection function,after the stepwise regression analysis and calculation,the X2was deleted in the linear term and quadratic term,indicating that the X2only played a meaningful role in the interaction.

In the mathematical regression model,the action coefficients of various tested factors were analyzed.The results showed that the regression coefficient b1of linear term X1was＞0; the regression coefficient b3 of linear term X3was＜0;the regression coefficient of quadratic term X32was＜0; among the interaction terms,the b13＞b12＞b23.It suggested that among the linear terms and interaction terms,the X1and X3play major roles.Their main effects and interaction effect were shown in Fig.1 and Fig.2.

In a given range(R),the step size of X1,X2and X3was all assigned as 1, and the objective function(y)was assigned between 85%and 95%.According to the mathematical model,a total of 125 combination programs were obtained.Among them,a total of 23 combinations were in line with the given intervals of the objective function.The statistics results of frequency analysis can provide a reference for production practice.

Analysis of test results using BP artificial neural network model

Determination of parameters of BP artificial neural network model

The BP neural network was composed of three layers,including input layer, output layer and hidden layer.After comparing the effects of different network structures and parameters on the fitting residuals of training samples,the topological structure of 3-5-1 was selected for the BP neural network.The raw data was normalized and then iteratively trained 1 000 times,and the fitting residual was 0.002 205 450 385 154 9.The fitting results by the BP neural network were analyzed(Table 3),and the results showed that the fitting residual of the BP neural network met the requirements by this test.

Comparison of application effect between quadratic regression model and BP neural network modelThe investigation results of staminate catkin shedding rate of walnut in validating field test(Table 4)showed that large amounts of male flowers fell off on the 4thd after the test treatment. The average shedding rate reached 84.08%,which was increased by over 70%compared with that of the CK, meeting the requirements by control target(85%-95%).

The predicted values by the regression equation and BP neural network and the results of validating field test were compared(Table 5).It showed that the difference between the predicted value by BP neural network and the actual value in field testwas 2.04%,and the difference between the predicted value by regression equation and the actual value in field test was 3.12%.The prediction accuracy of BP neural network was over 1.0%higher than that of regression equation.

Table 1Types,concentrations and encoding schemes of substances used for emasculation of walnut mg/kg

Table 2Test programs and test results

Table 3Simulation validation of BP neural network structure%

Table 4Shedding rates of staminate catkin of walnut in validating field test%

Table 5Comparison of shedding rate of staminate catkin among regression prediction,BP prediction and validating field test

Discussion

During the rapid growth and numerous blooming of staminate catkin of proterandrous walnut,the female flower development is at a critical stage.The consumption of large quantities of nutrients and moisture by staminate catkin affects the development and fruit setting of female flowers.Moreover,walnut has large amounts of male flowers with large amounts of pollens,and the male flowers are all long-distance transmitted wind-pollinated flowers.However, 90%of the male flowers of walnut are invalid.Zhao et al.[21]found that removing 90%of the male flower buds at the germination phase or removing 60%and 90%of the male flower buds at the elongation period,along with fertilization at the flowering stage, could significantly improve the fruit setting rate,thereby improving yield. Zhang et al.[22]conducted a test in Yangbi County,Yunnan Province.The results showed that after the excessive staminate catkin in walnut was thinned,the female flowers obtained more nutrients,so their development and fruit setting were improved.Compared with that of the control,the fruit setting rate of the treatment group was increased by 12%-17%.But so far, manual operation is still the main method of walnut emasculation.According to the survey,the 18-year-old walnut tree has around 2 000 male flowers,and the 70-80-year-old walnut tree has about 3 150 male flowers, sometime even up to 12 741.Even worse,the duration suitable for walnut emasculation usually lasts for only 7-10 d in spring.Therefore,in the production and management of walnut, there are rare orchards which carry out excessive staminate catkin thinning.In 1996,some domestic scholars used alcohol for excessive staminate catkin thinning in walnut,and up to 51.1%of the male flowers had been thinned. Wang et al.[3]applied the ethephon in the walnut emasculation,and they carried out validating field test the following year and validated the feasibility of the technique.They pointed out that under the premise of saving the cost of production,in accordance with the appropriate mathematical model, the balanced combination of the 2 kinds of chemicals with growth-inhibiting effect and shedding effect is entirely feasible for walnut emasculation.

The classical mathematical theory points out that when the regression equation is significant,the difference between the predicted value and actual value is not only related to the adopted statistical significance level and adopted sample size for statistical analysis but also related to the value of observation point.In general,only when the value of observation point is near the average value of observation points,the prediction makes sense. Moreover,the value of observation point must be in the sampling range for fitting regression equation,and cannot be extrapolated.The studies and practices have all shown that the prediction,application and analysis of mathematical model widely used by regression analysis can not go beyond the restriction of original data and the background conditions generated by original data,such as varieties,culti-vation and management technical measures and ecological environment. On one hand,the statistical model is established based on a large amount of data or test model.In the case of that the test data is less than that required by modeling,the modeling cannot be completed or the established model is out of work.Even under the condition of same basic data,the regression analysis results are usually different,or even differ significantly due to different mathematical models adopted for regression analysis.On the other hand,some undesirable accidents often occur in actual agricultural production,and traditional mathematical methods are difficult to describe the complex system of agricultural production.Therefore,only the appropriate selection and utilization of mathematical method can relatively accurately reflect the practical features of large agricultural production system.

After decades of research,BP neural network has around widespread attention due to it being able to solve complex nonlinear function approximation problem.It has been demonstrated that the three-layer forward network(including a hidden layer)can approximate any multivariate function.During the application, the network layers,each neuron number,fitting error,learning rate and sample data all should be determined according to the specific circumstances[11].Yi et al.[9]pointed out that although the overall prediction effect of regression analysis is relatively ideal, the prediction effect of BP neural network is very satisfactory.Yao et al.[23]fount that the BP neural network model had a strong learning ability.When the human activities or environment factors were greatly changed,it does not require special tests and identification parameters;the new information is only needed to be input and retrained,thus the changes in the system all can be tracked.However,BP neural network also has defects of slow convergence of learning process, poor global search ability,easy falling into local minimum,poor robustness and poor network performance[13]. Moreover,the personal experience and subjective judgements of data processor play an important role.This effect is produced not only on the design of network topology but also on the selection of network training sample data,selection of training parameters and comparison of error.In addition,the selection of samples for the training process of BP neural network has great effect on model determination and predictive application.So the intrinsic characteristics and laws of overall samples must be taken into account[24].The network training sample data includes the results obtained by complete design[17],orthogonal design[19],composite design[18]and surface design[16],as well as the accumulated observation(survey)data.Li et al.[25]pointed out that under the premise of large sample size,the accuracy of training results of BP neural network is higher than those of other mathematical models.The small sample size in researches and relatively insufficient training samples in the training all have certain effect on the prediction accuracy.In this study,the data was composed of training sample(17), model input(3)and model output(1), and the appropriate training parameters were selected.The overfitting and overtraining of the model were avoided,meeting the requirements by predictive application.

Conclusions

The quadratic polynomial stepwise regression analysis is adopted for the field test results.Thus the minor factors can be removed,but the important factors affecting the objective function can be retained.In addition, through analyzing the main effects, quadratic effects and interaction effects of important factors,the practical utilization value of the mathematical model is cleared.

In the premise of no requirements for establishing complex mathematical models and analyzing effects of various factors,the BP neural network model can get relatively accurate predictions by determining the reasonable network structure and training parameters.

The effective combination of quadratic polynomial stepwise regression analysis and BP artificial neural network not only can determine the effects of various factors but also can obtain relatively accurate predictions.

[1]XI RT(郗荣庭),ZHANG YP(张毅萍). Chinese Fruit Trees:Walnut(中国果树志核桃卷)[M].Beijing:China Forestry Publishing House(北京:中国林业出版社),1996.

[2]PAN YH(潘月红),ZHOU AL(周爱莲). Analysis of the development status, prospects and countermeasures of Chinese walnut industry(中国核桃产业发展现状前景及对策分析)[J].Food and Nutrition in China(中国食物与营养), 2012,18(5):22-25.

[3]DUAN ZM(段泽敏),WANG XP(王贤萍), CAO GS(曹贵寿).Chemical male flower thinning technique in walnut(核桃化学去雄技术研究)[J].Journal of Shanxi Agricultural sciences(山西农业科学), 2005,33(1):39-42.

[4]REN GW(任广伟),WANG FL(王凤龙), GAO HJ(高汉杰),et al.Applying BP neural network to predict tobacco virus diseases transmitted by aphids(BP神经网络在烟草蚜传病毒病预测中的应用) [J].Acta Tabacaria Sinica(中国烟草学报),2004,10(4):23-26.

[5]HU XP(胡小平),YANG ZW(杨之为),LI ZQ(李振岐),et al.Prediction of wheat strip rust in Hanzhong area by BP neural network(汉中地区小麦条锈病的BP神经网络预测)[J].Acta Agriculturae Boreali-Occidentalis Sinica(西北农业学报),2000,9(3):28-31.

[6]ZHOU M(周曼),ZHOU MQ(周明全).Automatic rice pest insects regognition based on BP neural network(基于BP神经网络的水稻害虫自动识别)[J]. Journal of Beijing Normal University (Natural Sciences)(北京师范大学学报(自然科学版)),2008,44(2):165-167.

[7]HU XP(胡小平),LIANG CH(梁承华), YANG ZW(杨之为),et al.Development and application of the BP neural network rediction system on plant diseases and pests(植物病虫害BP神经网络预测系统的研制与应用)[J].Journal of Northwest Agriculture&Forestry University(Natural Sciences)(西北农林科技大学学报(自然科学版)),2001,29(2): 73-76.

[8]HAN L(韩磊),LI R(李锐),ZHU HL(朱会利).Comprehensive evaluation model of soil nutrition based on BP neural network(基于BP神经网络的土壤养分综合评价模型)[J].Transactions of the Chinese Society of Agricultural Machinery (农业机械学报),2011,42(7):109-115.

[9]YI XS(易湘生),LI GS(李国胜),YIN YY (尹衍雨).Establishment and comparison of pedotransfer functions of soil moisture constant in the Three-River Headwaers Region of Qinghai Province (青海三江源地区土壤水分常数转换函数的建立与比较)[J].?Chinese Journal of Eco-Agriculture(中国生态农业学报),2012,20(8):1096-1104.

[10]LI S(李珊),MA LL(马丽丽),GE CX(贺超兴),et al.Simulation study between water evaporation of cultivation substrate and environmental factor of greenhouse(温室栽培基质耗水量与环境因子相关性的研究)[J].Chinese A-gricultural Science Bulletin(中国农学通报),2011,27(8):144-149.

[11]ZHANG B(张兵),YUAN SQ(袁寿其), CHENG L(成立),et al.Model for predicting crop water requirements by using L-M optimization algorithm BP neural network(基于L-M优化算法的BP神经网络的作物需水量预测模型) [J].Transactions of the CSAE(农业工程学报),2004,20(6):73-76.

[12]NIU ZX(牛之贤),LI WP(李武鹏), ZHANG WJ(张文杰).Prediction of grain yield using AIGA-BP neural network(基于AIGA-BP神经网络的粮食产量预测研究)[J].Computer Engineering and Applications(计算机工程与应用),2012,48(2):235-237.

[13]SU B(苏博),LIU L(刘鲁),YANG FT(杨方廷).Comparison and research of grain production forecasting with methods of GM(1,N)gray system and BPNN(GM(1,N)灰色系统与BP神经网络方法的粮食产量预测比较研究)[J]. Journal of China Agricultural University (中国农业大学学报),2006,11(4):99-104.

[14]MOSTAFA KHAJEH,MANSOUR GHAFFARI MOGHADDAM,MOHAMMAD SHAKERI.Application of artificial neural network in predicting the extraction yield of essential oils of Diplotaenia cachrydifolia by supercritical fluid extraction[J].Journal of Supercritical Fluids,2012,69:91-96.

[15]JUN X,XUE YJ,XU YX,et al.Artificial neural network modeling and optimization of ultrahigh pressure extraction of green tea polyphenols[J].Food Chemistry,2013,141(1):320-326.

[16]KA SINHAA,PAPITA DAS SAHAA, SIDDHARTHA DATTAB.Response surface optimization and artificial neural network modeling of microwave assisted natural dye extraction from pomegranate rind[J].Industrial Crops and Products,2012,37:408-414.

[17]KOSTIC MILAN D,JOKOVIC NATASA M,STAMENKOVIC OLIVERA S, et al.Optimization of hempseed oil extraction by n-hexane[J].Industrial Crops and Products,2013,48:133-143.

[18]KEKA SINHA,SHAMIK CHOWDHURY,PAPITA DAS SAHA,et al. Modeling of microwave-assisted extraction of natural dye from seeds of Bixa orellana(Annatto)using response surface methodology(RSM)and artificial neural network(ANN)[J].Industrial Crops and Products,2013,41:165-171.

[19]ZENG XY(曾祥燕),ZHAO LZ(赵良忠), SHENG Y(盛岩).Optimization of extraction technology of rutin from Sophora japonica based on BP neural network(BP神经网络优化槐花中芦丁的提取工艺)[J].Natural Product Research and Development(天然产物研究与开发),2013,25:312-316.

[20]HAN W(韩伟),XUN XH(孙晓海),LUO WF(罗文峰).Optimization and BP neural network model of extraction of polysaccharide from Albizia julibrissin Durazz(合欢皮多糖提取工艺优化及BP神经网络模型)[J].Journal of Nanjing University of Technology(Natural Sciences)(南京工业大学学报(自然科学版)),2013,35(5):57-62.

[21]ZHAO ZQ(赵增强).Measures to improving the fruit-setting of walnut grown in Qin-ba Mountains(秦巴山区提高核桃坐果率的研究)[J].Shaanxi Forest Science and Technology(陕西林业科技),2009,4:38-39.

[22]ZHANG HY(张怀玉),ZHANG J(章健). Reason analysis of low yielding of Shangluo walnut and measures to improve yield and efficiency(商洛核桃低产原因分析及增产增效措施)[J].Modern Horticulture(现代园艺),2012,7: 57-58.

[23]YAO RJ(姚荣江),YANG JS(杨劲松), ZOU P(邹平),et al.BP neural network model for spatial distribution of regional soil water and salinity(区域土壤水盐空间分布信息的BP神经网络模型研究)[J].Acta Pedologica Sinica(土壤学报),2009,46(5):788-794.

[24]LI X(李向),GUAN T(管涛),XU Q(徐清).The evaluation of soil heavy metal pollution based on the BP neural network:Taking soil environmental quality assessment in Baotou as an example (基于BP神经网络的土壤重金属污染评价方法-以包头土壤环境质量评价为例)[J].Chinese Agricultural Science Bulletin(中国农学通报),2012,28(2): 250-256.

[25]LI WF(李文峰).Application of BP artificial neural network on prediction of soil water content(BP神经网络在许昌土壤墒情预测模型的应用)[J].Chinese A-gricultural Science Bulletin(中国农学通报),2013,29(32):238-241.

[26]ZAI SM(宰松梅),GUO DD(郭冬冬), HAN QB(韩启彪),et al.Soil moisture prediction based on artificial neural network model(基于人工神经网络理论的土壤水分预测研究)[J].Chinese Agricultural Science Bulletin(中国农学通报), 2011,27(8):280-283.

Responsible editor:Tingting XU

Responsible proofreader:Xiaoyan WU

雄先型核桃雄花疏除的二次回归与BP神经网络模型研究

王贤萍，曹贵寿，杨晓华，张倩茹，李凯，李鸿雁，段泽敏*
（山西省农业科学院果树研究所/果树种质创制与利用山西省重点实验室，山西太原030031）

雄先型核桃雄花疏除（去雄）是提高产量的重要管理措施，为提高核桃去雄的效率，建立二次回归与BP神经网络模型。分别以乙烯利、赤霉素和甲哌鎓为自变量和核桃雄花脱落率为响应指标，进行田间建模试验，建立了二次多项式回归方程和BP神经网络模型，并于翌年进行BP模型田间确认试验。试验数据分为训练集、确认集和试验集，中心组合（二次旋转回归试验设计）田间建模试验得到的20组数据随机划为训练集（17）和确认集（3）数据，试验集为翌年田间确认试验得到的数据，BP神经网络的拓扑结构为3-5-1。①BP神经网络对确认集样本的预测值误差分别为1.3550%、0.4291%、0.3538%；②BP神经网络的预测值与田间确认试验结果相差为2.04%，回归预测值与田间确认试验结果相差为3.12%；③BP神经网络预测比回归预测提高预测精度1.0%以上。将二次多项式逐步回归分析和BP神经网络方法有效的结合使用，既可明确各因子的作用效应亦可得到相对准确的预测结果。

核桃；雄花序；疏除；BP神经网络；回归；预测Most of the cultivated walnut species are proterandrous.The excessive staminate catkin thinning(emasculation)is a traditional technique to improve the fruit setting rate and yield.In general,at the germination period, 90%of staminate catkin is removed. Thus the fruit setting rate will be significantly improved,and the yield of walnut will be also increased by over 10%.However,the emasculation of walnut is currently carried out by hand.To improve the excessive staminate catkin thinning efficiency, alcohols were first adopted in 1996 by some Chinese scholars.Wang et al. ever applied the ethephon in the excessive staminate catkin thinning of walnut and established the corresponding mathematical model[3].In recent years,with the rapid development of artificial neural network theory,the BP(back propagation algorithem)neural network has been widely used in the prediction of crop pests and diseases[4-7],soil nutrients and moisture content[8-11]and grain yield[12-13]and optimization of extraction process of plant functional ingredients[14-20].However,the application of BP artificial neural network in the emasculation and standardized cultivation of walnut has not been reported.This study aimed to investigate the BP neural network model of excessive staminate catkin thinning of walnut based on the field test results of walnut emasculation so as to provide technical basis for improving the yield and efficiency of walnut.

山西省科技厅科技攻关项目“核桃化学去雄技术”（002023）。

王贤萍（1961-），女，山西祁县人，研究员，从事农产品安全与果品加工研究，E-mail：Wangxpzls@163.com。*通讯作者，研究员，从事果树栽培生理与果品加工研究，E-mail：duanzmzls@163.com。

2015-02-10

修回日期 2015-05-25

Supported by Key Science and Technology Program of Shanxi Province,China (002023).

*Corresponding author.E-mail:duanzmzls@163.com

Received:February 10,2015 Accepted:May 25,2015