Germplasm image analysis of Astragalus maritimus and A . verrucosus of Sardinia ( subgen . Trimeniaeus , Fabaceae )

Mattana, E., Grillo, O., Venora, G. & Bacchetta, G. 2008. Germplasm image analysis of Astragalus maritimus and A. verrucosus (subgen. Trimeniaeus, Fabaceae). Anales Jard. Bot. Madrid 65(1): 149-155. The relationships between A. verrucosus Moris and A. maritimus Moris, exclusively endemics of Sardinia, are studied with a germplasm image analysis system. Morphometric and colorimetric features of the seeds and fruits of the two taxa have been studied and statistically elaborated to verify and confirm the validity of these species and to improve the performance of the correct classification of the image analysis system, previously elaborated to be a tool for taxonomic studies.

Morphometric and colorimetric features of the seeds and fruits of the two taxa have been measured and statistically analyzed, to confirm the validity of the two species and to carry out a first approach to investigate their taxonomic position inside the subgenus Trimeniaeus Bunge.

Materials and Methods
The fruits of the two species, were collected in the field (Tab. 1) or obtained from ex situ cultivation in the Botanical Garden of Cagliari, and the seeds were stored at the Sardinian Germplasm Bank, BG-SAR (Mattana & al., 2005).
The image of the samples was acquired by a flatbed scanner (Epson GT-15000), with a resolution of 200 dpi, with a scanning area not exceeding 1024 × 1024 pixel, before their entrance into the dehydration room (15 °C at the 15% of R.H.), in order to avoid any possible variation in dimension, shape and colour.The acquisition method was performed according to Bacchetta & al. (2008).For the analysis of every accession, a sample constituted by 100 seeds and 20 fruits (Tab. 1) was randomly prepared and sent by email at the Stazione Sperimentale di Granicoltura per la Sicilia fot the image processing.The image analysis system was KS-400 V. 3.0 (Carl Zeiss, Vision, Oberkochen, Germany).The KS-400 can be customized for specific applications by editing appropriate image analysis algorithms in 'macros', that allow to make the analysis automatically.In order to achieve the relative dimensions, shapes and colour (RGB -Red Green Blue and HLS -Hue Light Saturation channels) measures of the individual objects (seeds and fruits) a macro, formerly developed for identify diaspores of wild plants species (Bacchetta & al., 2008), was specially modified, adding 20 new seed features.The 34 measured features for each accession are shown in Table 2, 33 were recorded on seeds and 1 specific on fruits.
Statistical analysis of the seeds images variables was performed by Stepwise Linear Discriminant analysis method, it was used to find the best features to classify the two species, an algorithm included in the SPSS software package, release 14 (SPSS Inc. 1989-2006).
The selection for enter a new variable in the model is based on the higher value of F and least values of Wilks' Lambda.For each variable in the model, the F to remove and Wilks' Lambda statistics are useful for describing what happens if the variables is removed from current model (given that others variable remain).Moreover the procedure of cross validation was applied to verify the performance of the classifier developed.
For SEM pictures, seeds were dried with silica-gel and mounted on metal stubs using double-stick tape and taken by a FEI ESEM QUANTA 200.

Results
Seeds of the two taxa (Fig. 2) show a relative shape uniformity.They are flat, reniform shaped and they appear smooth, without sculptures and ornaments.In Table 3 the comparison among all morphometric and colorimetric parameters measured is presented.The Shape factor (0.87 ± 0.04 and 0.89 ± 0.04) and the Roundness factor ( 0.69 ± 0.06 and 0.72 ± 0.06) values confirm how the shape of these seeds is similar, but  4,293,328.23 ± 772,713.71 144,454,893.95 ± 47,109,687.29 2,44,.817.74 ± 504.274.51 67,627,344.73 ± 19,112,492   mm with an area of 4.54 ± 0.82 mm 2 and 8.84 ± 1.25 mm 2 respectively.The Stepwise Linear Discriminant analysis carried out for all the features measured on this work, increase the performance of the correct classification between the seeds of the two taxa at the 100% (Fig. 3), on respect of the 98.5% obtained in Bacchetta & al. (2008).
The best features selected by Stepwise method and its power in contributing to the discriminating process are reported in Table 4.The discrimination was carried out principally by colorimetric features (RGB and HLS channels) and the only one morphometric is the Area at the tenth position.
The SEM images (Fig. 4) of A. maritimus seeds highlight how they are angular and have more bor-  ders irregularities and widespread sags and swellings while A. verrucosus seeds are flatter, smoother and rounder.
Fruits of the two taxa (Fig. 5), conforming to the literature, look very differents.A. maritimus pods are brown-coloured, C-shaped and smooth; A. verrucosus pods are cream-coloured, oblongs with tubercles and with two hooks at an extremity (Corrias, 1978;Bacchetta, 2001).The image analysis carried out on the fruits (Table 3), confirm this high dissimilarity on size (Area 155.80 ± 22.18 mm 2 for A. verrucosus and 76.53 ± 9.67 mm 2 for A. maritimus), shape (Shape factor 0.51 ± 0.05 and 0.27 ± 0.02 respectively) and colour (see values for RGB and HLS channels).

Discussion
The germplasm image analysis system carried out to identify diaspores of wild plant species (see Bacchetta & al., 2008) and here modified with the adding of new 20 features and also applied on fruits, confirm to be a helpful tool on taxonomic studies with high performances of correct classification and the comparison with SEM images show also how this system is able to discriminate very similar seeds.
In fact, how in this case, the shape and dimensions are not significantly different, the integration of morfometric with colorimetric features in the analysis, assure in any case high percentages of correct classification.
E. Mattana & al.The morphometric and colorimetric characterization carried out by image analysis of the seeds confirm the taxonomic distance between these taxa and the validity of the two species.The results of this work represents also a first approach to investigate the taxonomic relationship of these two species inside the subgenus Trimeniaeus.
Studies on the life form and other plant traits of these two species are in progress to verify their position on the sections Platyglottys and Drepanodes.

Fig. 3 .
Fig. 3. Plot of the discriminating analysis.The Linear Discriminant Analysis verified by cross validation shows a performance of 100%.

Table 2 .
List Bacchetta & al. (2008)orphometric and colorimetric features; in bold the new features not considered inBacchetta & al. (2008).Feature Description and units of measurementArea Area of the object (mm 2 ) Perimeter Perimeter of the object (mm)

Convex Perimeter Convex Perimeter of the object (mm) Crofton Perimeter Crofton Perimeter of the object (mm) Ratio Co. P./Cr.P Ratio between Convex and Crofton Perimeters Max
. diameter Maximum diameter of the object (mm) Min.diameter Minimum diameter of the object (mm)

Ratio between ends distances b Ratio between ends distances Diameter ratio Ratio between diameters Shape
factor = (4 × × area)/perimeter

2 Diameter Eq. Ellipse Ellipse diameter of Equivalent area Ellipse Eq. Max. and Min. Ellipse of equivalent area, Maximun and Minimum axis Roundness
factor = (4 × area)/( × max.diameter 2 ) Red colour a Channel Red model RGB Std.

dev. Red Standard deviation Red channel Green colour
a Channel Green model RGB Std.

dev. Green Standard deviation Green channel Blue colour
a Channel Blue model RGB Std.

dev. Blue Standard deviation Blue channel Hue
a Channel Hue model HLS Std.

dev. Hue Standard deviation Hue channel Light
a Channel Light model HLS Std.

dev. Light Standard deviation Light channel Saturation
a Channel Saturation model HLS Std.

dev. Saturation Standard deviation Saturation channel Mean Density Mean grey level Density Std. dev. Density Standard deviation Density Skew Grey level asymmetry Kurtosis Grey level dispersion Energy Power of increasing intensity Entropy Power of dispersion Sum density Sum of density grey level Sum square density Sum square density grey level
a Grey levels (Range 0 -255).b For fruits A. maritimus seeds are smaller (1.37) 2.22 × 2.89 (3.90) mm than A. verrucosus (2.26) 3.07 × 3.96 (4.74)

Table 3 .
Morphometric and colorimetric measures of Astragalus verrucosus and A. maritimus.

Table 4 .
Ranking of selected features after fourteen cycles of stepwise selection (SPSS Linear Discriminant Analysis -Stepwise method).