The productivity of citrus can be affected by several factors, such as pest and disease management, not irrigated orchards, narrow genetic base and lack of technological support. The chemical control is quite difficult to be made due to the systemic characteristics of several diseases, their etiology and their easy dispersion. Some diseases, such as citrus variegated chlorosis (CVC), citrus canker, citrus leprosis, citrus tristeza, black spot (CBS), and also pests like the leprosis acari, are the most common biotic factors limiting the citrus culture. This paper focuses on the black spot disease, a fungal disease caused by Guignardia citricarpa Kiely (anamorph Phyllosticta citricarpa), which is characterized for presenting sunken lesions in the rind of fruits causing precocious maturation, accented fall and depreciation for in natura fruit market (Bonants et al. 2003). One of the main features of CBS is the delay for the symptom appearance, making the fungal presence identification necessary as soon as possible.

The method most frequently used for spore identification is by means of collecting at orchards suspended particles blown on discs, and analyze them under the microscope. This task takes a long time and human errors may affect the results. There are other scientific works dealing with the identification shapes problem by imaging analysis (Vanhoutte et al. 1995; Wilkinson, 2000; Araya-Kroff, 2004; Takemura, 2004), however, this work is a first initiative to identify the fungal spore. The identification process based on shape analysis face some difficulties such as the shape, colour (by special dye) and texture similarity among other species. The use of a computer aided vision system to assist the spores identification is one of the strategies to speed up this process. The first step to obtain an efficient computer aided vision system is to evaluate shape analysis methods.  Considering that a specific hardware will be developed and put to work in loco (in orchard), the techniques for identifying ascospores must be implemented on it. So, some shape analysis methods have been studied and applied to this problem and a comparative study among them using spectrum analysis has been conducted.

Two types of spore can spread the fungus: the asexual (conidia), which develops in fruits and leaves fixed at the plant, and the sexual (ascospores), which develops in leaves in decomposition. Usually it infects fruits either on the same plant or on neighbouring plants. The ascospores are spread not only on short, but also on long distances: the wind can spread it, infecting orchards at kilometres of distance (Goes, 2002). Although conidia may cause infection of citrus, the ascospores are seen to be the primary source of infection where citrus is cultivated. The ascospores are unicellular, hyaline, aseptate, broader in the middle, cylindrical, 8-17.5 x 3.3-8 μm, ends obtuse with colourless terminal mucoid appendage (EPPO/CABI, 1997; Baldassari et al. 2001).

Several techniques for shape description and characterization have been developed, and to choose and apply they depend on the features desired to detect. The methods used in this work are based on the shape signature, projection and curvature, including Fourier spectrum analysis.

Experiments

To collect the samples a device that blows particles on a disc was strategically positioned to cover the entire orchard. Each device takes 24 hrs to complete a disc with samples. A special dye was applied on disc to colour the hyalines structures. The images of discs were acquired using a digital camera connected to a conventional microscopy with lens 40 x 5, with resolution around 640 x 512 pixels.

Experiments were performed applying three different methods simultaneously with the power spectrum analysis. Before applying shape analysis techniques, the segmentation and contour extraction were applied as follows. The segmentation process was performed applying the threshold method in the acquired image, automatically determined using the iterative selection method (Parker, 1997). However, noises appeared and, to circumvent this, a morphologic filter was applied for both to noise removal and contour smoothing. This filter is a combination of opening and closing morphologic operations (Gonzalez and Woods, 2002). Another fact observed during this segmentation stage was the connections among objects contained in the disc. In this case, the watershed segmentation method was applied for segregation of connected particles, using the immersion approach (Vincent and Soille, 1991). Finally, contour following algorithm (Costa and Cesar, 2001) was applied in each object. The algorithm consists in tracking the object contour given an initial point, and storing respective coordinate in a linked list.

The applied methods were shape signature, horizontal projection and curvature. After applying these methods on contour particles, the power spectrum was estimated and the features were selected to compound the feature vectors. For each one, the n first and the n last coefficients were considered in the feature vector composition. The number of descriptors (n) varied for each experiment due to the signal width generated by each method, and it was defined by visual analyzing of the Fourier spectrum obtained to choose the coefficients with significant information. Alternatively, the accumulated energy function (Costa and Cesar, 2000) was computed to validate the values chosen: the values were discarded as accumulated energy stabilizes (high frequencies coefficients). The selection of the descriptors was always performed on the low-frequency coefficients, located in the spectrum extremities, using the Discrete Fourier Transform result. Considering that the Fourier Transform is applied on descriptors (i.e. signatures), the use of low frequencies aims to discard variations and details avoiding difficult to compare them. In this case, the descriptors keep their features, circumventing the signal shift problem. The ascospore identification process for these experiments was based on the distance estimate among the average vector (M), defined previously, and feature vectors that represent the particles found on discs. The Euclidian norm between P_s and M vectors was performed for estimating the distance between two vectors. A threshold was defined for separating the particles, considering as ascospores the particles with distance less than or equal to this threshold. The threshold was defined considering the results obtained during the experiments, towards minimizing errors (non-identification) and minimizing the false positive.

Results and Discussion

The results were obtained according to the estimated distance among feature vectors and the average vector and the identification decision was given by threshold. The set used for these experiments consisted of 60 samples, where 20 were Guignardia citricarpa ascospore samples and 40 were other particles also found on the collection discs. The results and their threshold are presented as follows. Three experiments were performed: shape signature based, horizontal projection based and shape curvature based. Fourier descriptors (power spectrum analysis) were used in the experiments as feature. The experiment based on the shape curvature method identified 55 out of 60 cases correctly (about 92%). This result is considerable more than satisfatory compared to other two experiments that identified correctly 73% and 86%, respectively. This result refers to a small difference among descriptors obtained by applying the power spectrum and curvature methods. The estimated distance makes better ascospore identification possible. However, there are still some samples with small differences in relation to the average vector identified as ascospores (false-positive).

In these experiments, false-negative cases occur due to deformations in the ascospore shape caused in the collection process or in the image acquisition process. The ascospores present cylindrical shape with central dilatation and the collection disc has a transparent and adherent product on the surface. These factors lead to different perspectives of spores depending on the angle that it adheres on the disc. Other factors, such as microscope focus variation, may also cause deformations, which appear after the segmentation process. Consequently, this kind of mistake is also possible to occur when a specialist analyzes the disc. The great number of particles contained in a single image is another difficulty faced by specialists.

Therefore, it is important to emphasize that although there are factors, which can harm the identification process, the third experiment had a considerable success rate using only one shape description method - curvature. This method extracted features capable of identifying most of the samples correctly, even with shape perspective variations and deformations. This result was not obtained in former experiments through shape signature and horizontal projection applied separately, but perhaps better results can be obtained through synergy among different methods.

References

ARAYA-KROFF, P.; AMARAL, A.L.; NEVES, L.; FERREIRA, E.C.; PONS, M.N.; MOTA, M.; ALVES, M.M. Development of image analysis techniques as a tool to detect and quantify morphological changes in anaerobic sludge: I. Application to a granulation process. Biotechnology and Bioengineering, 2004, vol. 87, no. 2, p. 184-193.

BALDASSARI, Ricardo B.; GOES, Antonio; SANTOS, Jaime M. and TIMOSSI, Adriano J. Microscopia eletrônica de varredura de isolados de Guignardia citricarpa obtidos de plantas cítricas. Summa Phytopathologica, 2001, vol.27, no. 1, p. 88-92.

BONANTS, Peter J.M.; CARROLL, George C.; WEERDT, Marjanne; BROUWERSHAVEN, Ilse R. and BAAYEN, Robert P. Development and validation of a fast PCR-based detection method for pathogenic isolates of the citrus black spot fungus, Guignardia citricarpa. European Journal of Plant Pathology, 2003, vol. 109, no. 5, p. 503-513.

COSTA, Luciano F. and CESAR JR, Roberto M. Shape analysis and classification: theory and practice. Boca Raton, CRC Press, 2001, 659 p. ISBN 0849334934.

EPPO/CABI. Guignardia citricarpa. In: Quarantine Pests for Europe. 2nd ed. CAB International, Wallingford (GB), 1997, p. 773-781.

GOES, Antonio. Efeito da combinação de fungicidas sistêmicos e protetores no controle da mancha preta dos frutos cítricos causada por Guignardia citricarpa. Summa Phytopathologica, 2002,vol.28, no. 1, p. 09-13. (in Portuguese)

GONZALEZ, Rafael C. and WOODS, Richard E. Digital image processing. Upper Saddle River, Prentice Hall, 2002, 793 p. ISBN 0201180758.

PARKER, J. R. Algorithms for image processing and computer vision. New York, Wiley Computer Pub, 1997, 432 p. ISBN 0471140562.

SEBER, G.A.F. Multivariate Observations. Wiley, 1984. 686 p. ISBN 0471691216.

TAKEMURA, C.M.; CESAR- JR, R.M.; ARANTES, R.A.T.; COSTA, L.F.; HINGST-ZAHER, E.; BONATO, V.; REIS, S.F. Morphometrical data analysis using wavelets. Real-Time Imaging, 2004, vol.10, no. 4, p. 239-250.

VANHOUTTE, B.; PONS, M.N.; THOMAS, C.R.; LOUVEL, L. and VIVIER, H. Characterization of Penicillium chrysogenum physiology in submerged cultures by color and monochrome image analysis. Biotechnology and Bioengineering, 1995, vol. 48, no. 1, p. 1-11.

VINCENT, L. and SOILLE, P. Watersheds in digital spaces: An efficient algorithm based on immersion simulations. IEEE Trans. Pattern Analysis and Machine Intelligence, 1991, vol. 13, no. 3, p. 583-598.

WILKINSON, M.H.F.; ROERDINK, J.B.T.M.; DROOP, S. and BAYER, M. Diatom contour analysis using morphological curvature scale Spaces. In: Proceedings International Conference on Pattern Recognition. (15º, 3^rd-8^th September, 2000, Barcelona, Spain). 2000, vol. 3, p. 652-655.

Note: Electronic Journal of Biotechnology is not responsible if on-line references cited on manuscripts are not available any more after the date of publication.