Classification of Winter Rapeseed Cultivars and their Yield Characters with the Common Vector Approach

Transcription

1 Classifiation of Winter Rapeseed Cultivars and their Yield Charaters with the Common Vetor Approah Nurdilek Gülmezoğlu 1, Zehra Aytaç 2 and M. Bilginer Gülmezoğlu 3 1 Department of Soil Siene and Plant Nutrition, Faulty of Agriulture, Eskişehir Osmangazi University, 26480, Eskişehir, Turkey, dgulmez@ogu.edu.tr 2 Department of Field Crops, Faulty of Agriulture, Eskişehir Osmangazi University, 26480, Eskşehir, Turkey, zehrak@ogu.edu.tr 3 Department of Eletrial and Eletronis Engineering, Faulty of Engineering and Arhiteture, Eskişehir Osmangazi University, 26480, Eskişehir, Turkey, bgulmez@ogu.edu.tr Abstrat In this study, five ultivars (Ceres, Zorro, Falon, Express, and Samourai) of winter rapeseed were lassified by using the ommon vetor approah (CVA). For this purpose, seven yield haraters (plant height, number of branhes per plant, number of pods per plant, number of pods on main stem, number of seeds per pod, pod length and thousand seed weight) of eah ultivar were used. The seven and six yield haraters taken from five ultivars were lassified by using CVA. 100% lassifiation rate is guaranteed for the training set of both studies. For the test set, the lassifiation of five ultivars has low performane, but the lassifiation of seven and six yield haraters gave satisfatory results. It is onluded that the CVA method was suessful in the lassifiation of different varieties belonging to any plant and/or of different haraters belonging to any variety. Keywords: Charater lassifiation, ommon vetor approah, rapeseed lassifiation. 1 Introdution Rapeseed is an important oilseed rop in the agriultural systems of many arid and semiarid areas. Agronomi and quality advantages of new varieties have enlarged their prodution areas worldwide (Gül et al. 2007). Rapeseed in Turkey is mostly ultivated as a winter annual for oil prodution and rarely livestok feed. If planted in spring, they an be grown as summer rop but the seed yield would be dereased due to short growing season and lak of enough water at the end of growing season, thus, winter ropping is preferred. The anola ultivars are slow growing espeially in winter and most of them will omplete their life yle in 210 to 270 days (Sharghi et al. 2011). There are wide variations among the ultivated anola ultivars with respet to seed and oil yields per unit area at different planting dates as well as irrigation regimes. The seed yield and maturity of anola is greatly influened by fertility management, seeding rate and seeding date (Grant and Bailey, 1993). 713

2 Computer-based algorithms have been extensively used in agriulture in order to lassify various plants and their haraters or samples. Classifiation of plant varieties with omputer algorithms has been beome popular in reent years. The ommon vetors representing the invariant features of the plants an be extrated by eliminating the differenes in eah lass of plants (Gülmezoglu 1999). Then these ommon vetors are used for the lassifiation of varieties and haraters of plants. Different methods were used in order to derive features or parameters from plant varieties (Wang et al., 1999; Zayas et al., 1996; Utku and Köksel, 1998; Delwihe & Massie, 1996; Neuman and Bushuk, 1987; Shuaib et al., 2010). Some lassifiations were analyzed for haraters of rapeseed. Ali et al. (2012) analyzed for near infrared spetrosopy and prinipal omponent grain of rapeseed. Jankulovska et al. (2014) presented the use of different multivariate approahes to lassify rapeseed genotypes based on quantitative traits. Some of these parameters were plant height, number of primary branhes per plant, number of pods per plant, pod length, number of seeds per pod, seed weight per pod, 1000 seed weight, seed weight per plant and oil ontent. This model has been applied in agriultural sienes to identify the effet of yield harater differenes (Gülmezoğlu and Gülmezoğlu 2015). However, there is no information on the use in plant breeding programs. In this study, we onsidered five ultivars (Ceres, Zorro, Falon, Express, and Samourai) of winter rapeseed. Initially, these ultivars were lassified by using the ommon vetor approah (CVA). Seondly, seven yield haraters (plant height, number of branhes per plant, number of pods per plant, number of pods on main stem, number of seeds per pod, pod length and thousand seed weight) and six yield haraters exepting number of branhes per plant were lassified by using CVA. 2 Material and Method This researh was arried out over three years during 2005 at the Faulty of Agriulture of Eskisehir Osmangazi University, Eskisehir (39 o 48 N; 30 o 31 E; 789 m in elevation). The field experiments inluded five winter rapeseed ultivars (Ceres, Zorro, Falon, Express, Synergy and Samourai). The experiment was planned in a Randomized Complete Blok Design with four repliations. The individual plots were 3 m long and onsisted of five rows. The ultivars were sown on the first week of September, using a seed rate of 10 kg ha -1 in 40 m spaed lines on a well prepared seed bed. The experiment was fertilized respetively with 150 kg N ha -1 as ammonium nitrate: and 50 kg P 2 O 5 ha -1 as di-ammonium phosphate: The plants were irrigated one during emergene and thinned at the rosette stage. The weeds were ontrolled by hand weeding. Eah ultivar was represented with seven yield haraters whih are plant height, number of branhes per plant, number of pods per plant, number of pods on main stem, number of seeds per pod, pod length and thousand seed weight. Eah harater inludes 20 plant samples whih were taken from field study onduted during growing year. As in all lassifiation methods, CVA has training and testing stages. In the training stage, a ommon vetor whih represents ommon or invariant properties of 714

3 eah lass is alulated and an in differene subspae for eah lass is onstruted. Let the vetors a,a,...,a 1 2 m Є R n be the feature vetors for a variety-lass C in the training set where m n. Then eah of these feature vetors whih are assumed to be linearly independent an be written as a i = ai, dif + a om +εi for i=1,2,, m (1) where the vetor aidif, indiates the differenes resulting from limati effets and alien-pollination, and the vetor a om is the ommon vetor of the variety or harater lass C, and ε i represents the error vetor (Gülmezoğlu et al. 2001). The ommon vetor an be obtained from the subspae method. Let us define the ovariane matrix of the feature vetors belonging to a variety or harater lass as m T Φ= ( ai aave)( ai a ave) (2) i= 1 where a ave is the average feature vetor of C th lass whose ovariane matrix is to be alulated and T indiates the transpose of a matrix. The eigenvalues of the ovariane matrix Ф are non-negative and they an be written in dereasing order: λ1 λ2 K λ n. Let u,u,...,u 1 2 n be the orthonormal eigenvetors orresponding to these eigenvalues. The first (m-1) eigenvetors of the ovariane matrix orresponding to the nonzero eigenvalues form an orthonormal basis for the differene subspae B (Gülmezoğlu et al. 2001). The orthogonal omplement, B, is spanned by all the eigenvetors orresponding to the zero eigenvalues. This subspae is alled the indifferene subspae and has a dimension of (n-m+1). The diret sum of two subspaes B and B is the whole spae, and the intersetion of them is the null spae. The ommon vetor an be shown as the linear ombination of the eigenvetors orresponding to the zero eigenvalues of Ф (Gülmezoğlu et al. 2001), that is, a,, om = ai um um + L + ai un un i=1,2,,m (3) From here, the ommon vetor aom is the projetion of any feature vetor onto the indifferene subspae B. The ommon vetor represents the ommon properties or invariant features of the variety or harater lass C. The ommon vetor is independent from index i. Therefore, the ommon vetor is unique for eah lass and all the error vetors εi would be zero. During the lassifiation stage, the following deision riterion is used: 2 n T distane = argmin ( x i ) j j 1 C S j=m a -a u u (4) 715

4 where a x is an unknown or test vetor and S indiates the total number of lasses. If the distane is minimum for any lass C, the feature vetor a x is assigned to lass C. Classifiation algorithm given above an be summarized as follows: Step 1: Construt feature vetors by using samples taken for eah harater belonging to any ultivar. Be sure that number of samples in eah feature vetor (or dimension of eah feature vetor) is greater than number of feature vetors (or haraters) for eah ultivar. Step 2: Find the ovariane matrix (Eq. (2)) for eah ultivar by using feature vetors belonging to that ultivar. Step 3: Find the eigenvalues λi and orresponding eigenvetors u i for eah ovariane matrix. Step 4: Find the ommon vetor (Eq. (3)) for eah ultivar by using the (n-m+1) eigenvetors orresponding to zero eigenvalues. Step 5: When an unknown feature vetor ax is given, lassify this vetor by using Eq. (4). 3 Results In the first study, eah of five ultivars forms one lass in the CVA method. Seven haraters eah of whih inludes 20 samples for eah ultivar or lass form seven feature vetors of that lass. Therefore, there are five lasses and eah lass has seven feature vetors. When the feature vetors (haraters) used in the training stage were tested, all lasses (ultivars) were orretly lassified, i.e., 100% orret reognition rate was obtained. When the leave-one-out strategy was used in the testing stage, that is, when six feature haraters were used in the training stage and remaining one harater was tested, 25.7% orret reognition rate was obtained as average of leave-one-out steps. The results obtained from this study are given in Table 1. The average sore obtained in the test set is very low beause samples inluded in the haraters representing different ultivars are very lose to eah other. In the seond study, the haraters were lassified by using CVA. First of all, seven haraters were onsidered and eah of seven haraters forms one lass in the CVA method. Twenty samples taken from eah ultivar for any harater form one feature vetor of that harater. Therefore, there are seven lasses and eah lass has five feature vetors. When the feature vetors, eah of them inludes 20 samples, used in the training stage were tested, all lasses (haraters) were orretly lassified, i.e., 100% orret reognition rate was obtained. When the leave-oneout strategy was used in the testing stage, that is, when six feature haraters were used in the training stage and remaining one harater was tested, 77.1% orret reognition rate was obtained as average of leave-one-out steps. The results obtained for this study are given in Table

5 Table 1. Corret reognition rates of five ultivars as perentage. Varieties Training set Test set Samourai Zorro Falon Ceres Express Average Table 2. Corret reognition rates of seven yield haraters as perentage. Yield Charaters Training Set Test Set Plant Height Number of Branhes per Plant Number of Pods per Plant Number of Pods on Main Stem Number of Seeds per Pod Pod Length Thousand Seed Weight Average Seondly, six haraters exepting number of branhes per plant were lassified. All haraters were orretly lassified (100% orret reognition rate was obtained) in the training set and 90% reognition rate was obtained for the test set. These sores are remarkable beause samples taken from ultivars for eah harater are lose to eah other and well represent that harater. These results are given in Table 3. Table 3. Corret reognition rates of six haraters as perentage Yield Charaters Training Set Test Set Plant Height Number of Pods per Plant Number of Pods on Main Stem Number of Seeds per Pod Pod Length Thousand Seed Weight Average

6 4 Disussion It is known that varieties of different plants have been suessfully lassified by using various omputer-based algorithms. Espeially, lassifiation of wheat varieties with omputer algorithms has been beome popular in reent years (Zayas et al. 1996, Utku and Köksel 1998, Neuman and Bushuk 1987, Gülmezoğlu and Gülmezoğlu 2015). Therefore, in this study, first of all, five rapeseed varieties were lassified by using CVA method. In spite of 100% orret reognition rate in the training set, very low reognition rate (25.7%) was obtained in the test set. The reason is that samples inluded in the haraters representing different varieties are very lose to eah other. Thus, ommon properties or invariant features of eah variety annot be orretly extrated and indifferene subspae annot be onstruted effiiently. Additionally, haraters were lassified by using CVA method. Initially, seven haraters are applied to the lassifiation proess and 100% and 77.1% reognition rates were obtained for the training and test sets respetively. The reason of low sore for the test set is that the samples belonging to number of branhes per plant harater are similar to samples belonging to other haraters. When the number of branhes per plant harater is disarded, that is, when the remaining six haraters are lassified, 90% reognition rate is obtained for the test set. 5 Conlusion It is onluded that the CVA method was very suessful in the lassifiation of different varieties belonging to any plant and/or of different haraters belonging to any variety. Suh lassifiations an be very helpful in assignment of unknown varieties or unknown haraters to identify plant. When more speifi haraters are extrated for eah variety of plants, good performane an be ahieved from the lassifiation proess. As a future work, number of varieties for any plant and the number of haraters will be inreased. Satisfatory results are also expeted from this work. Referenes 1. Ali, I., Shah, S.A., Ahmed, H.M., Rehman K.U., Ahmad, M. (2012) Studies On Geneti Diversity For Seed Quality In Rapeseed (Brassia Napus L.) Germplasm of Pakistan Through Near Infrared Spetrosopy And Prinipal Component Analysis. Pak. J. Bot., 44: , Speial Issue Marh Delwihe, S.R., Massie, D.R. (1996) Classifiation of wheat by visible and nearinfrared refletane from single kernels. Analytial Tehniques and Instrumentation 73(3)

7 3. Grant, C.A., Bailey, L.D. (1993) Fertility management in anola prodution. Canadian Journal of Plant Siene 73: Gül, M.K. Egesel, C.Ö., Kahriman, F., Tayyar, Ş. (2007) Investigation of Some Seed Quality Components in Winter Rapeseed Grown in Çanakkale Provine. Akdeniz University Journal of the Faulty of Agriulture., 2007, 20(1), Gülmezoğlu M.B., (1999), A Novel Approah to Isolated Word Reognition, IEEE Trans. Speeh and Audio Proessing, vol. 7, No.6, pp Gülmezoğlu, M.B., Dzhafarov, V., Barkana, A. (2001) The Common Vetor approah and its Relation to Prinipal Component Analysis. IEEE Trans. Speeh and Audio Proessing 9(6), Gülmezoğlu, M.B., Gülmezoğlu, N. (2015) Classifiation of bread wheat varieties and their yield haraters with the ommon vetor approah. International Conferene on Chemial, Environmental and Biologial Sienes (CEBS-2015), Marh, 18-19, 2015, Dubai, BAE. 8. Jankulovska, M., Ivanovska, S., Marjanovi-Jeromela, A., Bolari, S., Jankuloski, L., Dimov, Z, Bosev, D., Kuzmanovska, B. (2014) Multivariate Analysis Of Uantitative Traits Can Effetively Classify Rapeseed Germplasm. Genetika, Vol. 46, No.2, Neuman M. and Bushuk, W. (1987) Disrimination of wheat lass and variety by digital image analysis of whole grain samples. Journal of Cereal Siene 6(2) Sharghi, S., Shirani, A.M.R., Noormohammadi, G. Zahedi, H. (2011) Yield and yield omponents of six anola (Brassia napus L.) ultivars affeted by planting date and water defiit stress. Afrian Journal of Biotehnology. 10(46): Shuaib, M, Jamal, M., Akbar, H., Khan, I., Khalid, R. (2010) Evaluation of wheat by polyarylamide gel eletrophoresis. Afrian Journal of Biotehnology 9(2) Utku, H., Köksel, H. (1998) Use of statistial filters in the lassifiation of wheats by image analysis. Journal of Food Engineering 36(4) Wang, D., Dowell F. E., Laey, R. E. (1999) Single wheat kernel olor lassifiation using neural networks. Transations of the ASABE 42(1) Zayas, I. Y. Martin, C. R. Steele J. L., Katsevih, A. (1996) Wheat lassifiation using image analysis and rush-fore parameters. Transations of the ASABE 39(6) eligible 719