News & Politics

Development of a semi-automatic system for pollen recognition

Development of a semi-automatic system for pollen recognition
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
   Aerobiologia  18:  195–201, 2002.© 2002  Kluwer Academic Publishers. Printed in the Netherlands.  195 Development of a semi-automatic system for pollen recognition Alain Boucher 1 , ∗ , Pablo J. Hidalgo 2 , Monique Thonnat 1 , Jordina Belmonte 3 , Carmen Galan 2 ,Pierre Bonton 4 & R´egis Tomczak  4 1  INRIA, Sophia-Antipolis, 2004 route des Lucioles, B.P. 93, F-06902 Sophia-Antipolis Cedex, France; 2  Department of Plant Biology, University of C´ ordoba. Campus Universitario de Rabanales, 14071-C´ ordoba,Spain;  3 Unit of Botany, Autonomous University of Barcelona, 08193 Bellaterra (Cerdanyola del Vall`es), Spain; 4  LASMEA, UMR 6602 du CNRS, Blaise Pascal University, F-63117 Aubi`ere Cedex, France ( ∗ author for correspondence: Phone: +33 492387657;Fax: +33 492387939;E-mail: Received 2 February, 2001; accepted 17 June 2002 Key words:  aerobiology,automaticpollen counting, automatic pollen recognition,colourimage analysis, 3D study,light microscopy, pollen morphology Abstract A semi-automatic system for pollen recognition is studied for the european project ASTHMA. The goal of sucha system is to provide accurate pollen concentration measurements. This information can be used as well by thepalynologists, the clinicians or a forecast system to predictpollen dispersion. At first, our emphasis has been put onCupressaceae,  Olea , Poaceae and Urticaceae pollen types. The system is composed of two modules: pollen grainextraction and pollen grain recognition. In the first module, the pollen grains are observed in light microscopy andare extracted automatically from a pollen slide coloured with fuchsin and digitized in 3D. In the second module,the pollen grain is analyzed for recognition. To accomplish the recognition, it is necessary to work on 3D imagesand to use detailed palynological knowledge. This knowledge describes the pollen types according to their mainvisible characteristerics and to those which are importantfor recognition.Some pollen structures are identified likethe pore with annulus in Poaceae, the reticulum in  Olea  and similar pollen types or the cytoplasm in Cupressaceae.The preliminary results show the recognition of some pollen types, like Urticaceae or Poaceae or some groups of pollen types, like reticulate group. 1. Introduction This study is developed in the frame of an EuropeanProject named ASTHMA (Advanced System of Tele-detection for Healthcare Management of Asthma).The ultimate objective of this project is to providenear real time accurate information on aeroallergensand air quality to the sensitive users on an individualbasis and at specific locations to help them in optimis-ing their medication and improve their quality of life.The main goal of this study is to build a prototypesystem for pollen recognition to be used as a part of the pre-operational integrated forecast system of theASTHMA project.Classically, airborne pollen analysis is done bytechnicians who search, locate, recognize and countthe pollen grains encountered. This pollen analysisis a time-consuming process in which the technicianspends many hours at the light microscope collectingthe pollen data on a paper sheet. Later, these datahave to be translated into a digital database for furtherprocessing and analysing them. Recently, the applica-tion of computer science to pollen count facilitatesthe reading process by means of automatic counters(Bennett, 1990), instead of classical mechanical coun-ters. With these systems the resulting data of thehuman count are directly typed to a computer (Eisnerand Sprague, 1987). Recently, Massot et al. (2000)used a voice recognition system to avoid the routinehandling of the data. Nevertheless, the contributionof a technician during the pollen searching, detec-tion and recognition can not be avoided until now.  196Other studies try to differentiate aerobiological sporesby image analysis (Benyon et al., 1999) or identifypollen texture by neural networks (Li and Flenley,1999). However, an automated system able to detectand recognize pollen grains from a slide has not beendescribed previously to our knowledge.In this paper, the current work for the developmentof a semi-automatic system for pollen recognition ispresented. This system can be seen as two comple-mentary modules: slide analysis and pollen recogni-tion. The noveltyof this system is the use of 3D imageprocessingand palynologicalinformationforthe iden-tification of pollen grains. The idea of a 3D imageprocessing is not new (Erhardt et al., 1985) and itseems to be a good candidate instead of the classicstatistical analysis in 2D. 2. Studied pollen types Four pollen types have been selected for this studycorresponding to the frequent and highly allergenicpollen types in the Mediterranean area, i.e. Cupres-saceae,  Olea ,  Parietaria  and Poaceae pollen types.Cupressaceae pollen, inaperturate, with gemmaeirregularly distributed in the exine and with a thick intine, belongs to cultivated and wild trees andshrubs widely distributed in the Mediterraneanregion. Parietaria  and Poaceae are fundamentally herbs,abundant in most Mediterranean urban and wildenvironments. Parietaria pollen is distinctly small,psilate andtriporate,(heptaporateinonespecies), withvisible vestibulum under each pore. Poaceae pollenis medium to large in size, with exine ornamenta-tion from psilate to verrucate and monoporate, witha well defined circular annulus around the pore.  Olea pollen belongs to olive trees, widely cultivated in theMediterranean countries.  Olea  pollen type is mediumsize, presents a coarse reticulum and is tricolpate(tricolporoidate). Besides these four pollen types, atotal of 10 similar and also frequent pollen types wereincluded in the image reference database.  Populus pollen type was included to avoid possible confu-sions with the other inaperturate and winter pollinatedtype, i.e. the Cupressaceae type. Other reticulatetypes were considered in order to avoid confusionswith  Olea  type, i.e.  Ligustrum, Fraxinus, Phillyrea ,Brassicaceae and  Salix   types.  Celtis  and  Coriaria types (both porate) were included in order to avoidconfusions with the monoporate pollen of Poaceaetype. And finally,  Morus  and  Broussonetia  pollentypes to avoid confusions with the small pollen of  Parietaria  type. The system works with usual aerobio-logical slides colouredwith fuchsin. A standardizationin the fuchsin concentration was done in order toadjust the best colouring for the detection and recog-nition of the pollen grains. The best concentrationwasestablished in 4  µ g of fuchsin /100 ml of colouringmedium. 3. First module: Image acquisition The first module of the system performs the globalslide analysis that has to isolate the pollen grains onthe slide and do their digitization in three dimensions(Tomczak et al., 1998). A workstation was designedfor both automatic and manual handling and readingof the slides (Figure 1). The hardware of the system iscomposed of an optical light transmitted microscopeequipped with a x60 lens (ZEISS Axiolab), a CCDcolour camera (SONY XC711) with a framegrabbercard (MATROX Meteor RGB) for image acquisition,and a micro-positioning device (PHYSIK INSTRU-MENTE) to shift the slide under the microscope.These components are driven by a PC computer. Agraphic interface was developed to allow the techni-cian to easily operate on the system.Two problems arise when one wants to extractinformation about pollen grains from image data in anautomated way. First, autonomous image acquisitionin microscopyrequires to adjust sharpness in real timebefore acquiring image data. To achieve this, an auto-mated focusing algorithm was conceived. It is basedon a sharpness criterion computed from image dataand a maximum criterion searching strategy. It allowsthe computation of the best focusing position for agiven sample, from a few measuring positions in realtime. Once the image has been focused, the secondproblem is the detection of pollen grains in the scene.The slides are stained with fuchin to colour the pollengrains in pink. This is necessary to differentiate pollengrains from other particles on the slide. However,variations of coloration intensity among the pollentypes are important and some other airborne particlesare also sensitive to the colorant. Simple segmenta-tion techniques (for instance, techniques only basedon chrominance analysis) are not efficient enoughto localise and isolate pollen grains. To solve thisproblem, a localisation algorithm based on a splitand merge scheme with markovian relaxation wasconceived. It includes three steps: colour coding  197 Figure 1.  Slide analysis workstation. following the best colour axes as computed by a prin-cipal componentanalysis on the RGB image (Noriega,1996); split-and-merge segmentation with markovianrelaxation; detection and extraction of pollen grainsby chrominance and luminance analysis (Tomczak etal., 2000). Figure 2 shows an example of detectionand extraction of the pollen grains. The localisationrate is estimated to be over 90% of the encounteredpollen grains on the slides, which is higher than othermethodslike the one developedbyFrance et al. (1997)which obtains a localisation rate of 80%. Once thepollen grain is found, the last step is to digitize it asa sequence of 100 colour images showing the grain atdifferentfocus (with a step of 0.5 microns – see Figure3). This set of images allows the identification using3D characteristics. 4. Knowledge acquisition General and specific palynological knowledge isnecessary for the development of the recognitionmodule. General knowledge of palynology like view,ornamentation, apertures, etc. is considered in a firststep of the data supply. Table 1 shows the charac-teristics considered in the general knowledge acquisi-tion for the four main pollen types. Not only isdetailed palynological information taken into accountbut also other characteristics such as the floweringtime.Thisgeneralknowledgeisneededbutnotsufficientfor the recognition of pollen grains based on images.For example, the notion of a pore can lead to variousinterpretations for an image processing system. It canbe seen differently depending on the pollen types, thegrain orientation, the grain quality, etc. So, one needsto analyze different examples for each pollen typeto extract some more specific knowledge that can beadapted for an automatic system.To bridge the gap between general knowledge andspecific needs for an automatic system, a softwaretool was designed to interchange information betweenpalynologists and computer scientists (see Figure 4).Using this tool, the experts in palynologycan annotatethe relevant specific characteristics at each level of the3D sequence. The annotations refer to more practicalinformation used for recognition. 5. Second module: Pollen recognition Once a pollen grain has been digitized, the next stepis to recognize its type. To achieve this, differentimage processing techniques can be applied, whichinclude various levels of knowledge from the applica-tion (Crevier and Lepage, 1997). The identification of the pollen grain type is based on palynological knowl-edge (see previous section). It can be achieved byfollowing two steps: compute global measures on thecentral image of the grain (2D) and search for specificpollen characteristics on the complete sequence (3D).Themaindifferencesbetweenthesetwostepslieinthelevel of knowledge needed and in the processing time.Globalmeasurescanbecomputedquicklywithoutanyknowledgeaboutthe pollengrain. All globalmeasuresare computed to guide the selection of a few possiblespecific characteristics that will be tested. The searchof these characteristics demand more time to computeand are specific to one or some pollen types.5.1  Global measures of the grain The strategy for recognition first isolates the grain onthe central image using colour histogram thresholdingtechniques. Then some global measures are computedon the grain (see Table 2). These measures are similarto what have been used in other applications to recog-nize different kind of objects, like fungal spores(Benyon et al., 1999) or planktic foraminifera (Yu atal., 1996).These measures are analyzed to give first estima-tions about the possible types of the grain. This can  198 Figure 2.  Detection and extraction of pollen grains from a slide. (a) Original image and computed areas of interest. (b) Splitting result. (c)Merging result. (d) Interpretation result. (e) Extracted images from areas of interest. (f) Post-processed images of extracted pollen grains. Figure 3.  Image digitization in three dimensions. (a) For each pollen grain, a sequence of 100 colour images is taken, showing the grainat different focus (with a step of 0.5 microns). (b–d) Images at different focus of an  Olea  grain, showing different details needed for itsidentification. Table 1.  Abstract of the general characteristics of the main pollen types considered in this study. This information is a part of the generalknowledge on palynology.Pollen type Cupressaceae Olea Poaceae ParietariaApertures Inaperturate Tricolporoidate Monoporate TriporateSize (m: medium, s: small, b: big) m s-m s-m-b sPolarity Apolar or heteropolar Isopolar Heteropolar Isopolar or apolarSymmetry Radial Radial Radial RadialShape e: circular e: circular-elliptic e: circular-elliptic e: circular(p: polar view; e: equatorial view) p: circular p: circular p: circular-elliptic p: circularExine ornamentation Microgemmate ReticulateScabrate to verrugate PsilateExine thickness 1  µ m 2–3.5  µ m 1–1.5  µ m  < 1  µ mFlowering period All the year April–July All the year All the year(more frequent) (September–June) (spring–summer) (February–July)  199 Figure 4.  Example of the tool developed for the specific palyno-logical knowledge acquisition in order to exchange informationon pollen characteristics. The expert in palynology can look intothe sequence like through a microscope and annotate the relevantspecific characteristics for the recognition of the grain at each level(Poaceae here). Table 2.  Description of the computed feature measurements.These measures are done on the global pollen grain to initializea first type estimation and also on the different characteristicsextracted from the images to validate them. Measure FormulaMean colour (M)  For red, green and blue colours Area (A)  Number of pixels Perimeter (P)  Perimeter length Compactness (C)  C  =  P  24 π 4 Equivalent circular diameter (D)  D  =   4 Aπ Moments of inertia (m)  m ij   =  (x  − ¯ x) i (y  − ¯ y) j  Eccentricity (E)  E  = m 20 + m 02 +   (m 20 − m 02 ) 2 + 4 m 211 m 20 + m 02 −   (m 20 − m 02 ) 2 + 4 m 211 Convex hull area (CHA)  Area of convex hull Convex hull perimeter (CHP)  Perimeter length of convex hull Concavity (CC)  CC  =  CHA − A Convexity (CV)  CV   =  CHP P  Solidity (S)  S   =  ACHA help not only to guide the system in its choices, butalso to identify quickly some special cases (brokenCupressaceae grains for example can be identifiedeasily by their non-circular shape). Moreover, theinformationof the samplingdate andlocation, if avail-able, can beusedto computethese first estimations, byreducing the list of possible pollen types.One next step that can be done without any a prioriknowledge is the selection of the images of interestto process. The system does not need to analyze allthe 100 images of the digitized sequence. Only 5 to10 images of interest can be enough to identify thepollen type, but which images depend on the pollengrain. The selection of the images of interest is doneusing SML operator (Sum Modified Laplacian) whichprovides local measures of the quality of image focus(Nayar and Nakagawa, 1994).One value is plotted per image to indicate its clear-ness. The local maximum peaks of the curve for thecomplete sequence indicate the images of interest,representing the clearest images with details on focusand high contrast images with colour variations.5.2  Type-specific characteristics Using the first estimations, the system now looks andtests for more specific pollen characteristics. Differentalgorithmsaredevelopedtoidentifyasinglecharacter-istic with differentappearances.For example,a poreisseendifferentlyfromthepolarviewthantheequatorialview. Two different algorithms are needed to detectbothcases. Also a regionof interest, where the charac-teristic can be found, can be defined for an algorithm(exine, inside the grain, near the thickest intine,  ... ).An algorithm which detects a specific character-istic must include the following elements: •  Which images will be processed?  The algorithmmust select one or several images where thecharacteristic can be found. For a pore, all theimages of interest will be tested, because thisfeature may appear at different position on thegrain. For a reticulum, the images located nearthe border of the grain when analysing the opticalsection and in the center when analysing the uppersurface (see Figure 5) will be tested, because clearviews of the reticulum can be found there. •  Which image processing techniques will be used? Depending on the characteristic, various tech-niques can be used (thresholding, Laplacian of Gaussian, region segmentation, colour analysis,etc.) (Pal and Pal, 1993). Using these techniques,
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks