Quality grading of cork planks with classification models based on defect characterisation J. R. Gonzalez-Adrados, F. Lopes, H. Pereira A quality grading system for cork planks is presented using a classification model based on data obtained from the image analysis of transverse and tangential sections of cork planks and taking into account different types of defects. Three types of defects may be distinguished in cork by their geometrical parameters: pores (lenticular channels), physiological anomalies (nail, clay and ‘‘bofe’’) and pathogenic anomalies (insect galleries of Coroebus undatus (F.) and Crematogaster scutellaris Ol.). A discriminant analysis was used to construct the classification model: the variable with the highest discriminant power was the total porosity in the transverse section, and the other variables were transformations of the specific width in the tangential section and the aspect ratio in the transverse section. The model classified two thirds of the samples in accordance with the manual classification currently used in the industry and all cases of disagreement referred to changes between contiguous quality classes. Klassifizierung von Korkplatten mittels automatisierter Fehlererkennung Ein Klassifizierungsmodell fu ¨r Korkplatten wird vorge- stellt, das auf Ergebnissen einer Bildanalyse beruht, wobei verschiedene Fehlertypen in Quer- und Tangentialschnit- ten beru ¨cksichtigt werden. Drei Fehlertypen ko ¨nnen unterschieden werden aufgrund ihrer geometrischen Parameter: Poren (Kana ¨le zwischen den Lentizellen), physiologische Anomalien und pathogene Anomalien (z.B. Insektenfraß). Eine Diskriminanten-Analyse wurde einge- setzt, um ein Klassifizierungsmodell zu entwickeln. Die Variable mit dem sta ¨rksten Einfluß auf die Klassifizierung war die gesamte Porosita ¨t, gemessen am Querschnitt. Die andern Variablen betreffen Transformationen der spezifi- schen Breite der Poren im Tangentialschnitt und ihr Aspekt-Verha ¨ltnis im Querschnitt. Das Modell klassifiziert zwei Drittel der Proben in U ¨ bereinstimmung mit der in der Industrie u ¨blichen visuellen Klassifikation. In allen anderen Fa ¨llen beziehen sich die Abweichung auf jeweils benachbarte Klassen. 1 Introduction The evaluation of cork quality is at the present stage one of the most important steps of the processing cycle of cork for the production of stoppers. This evaluation is made at two different stages: (1) after water boiling of the crude cork planks, when they are classified into quality grades before the mechanical boring of stoppers; (2) as a final quality grading of the finished stoppers. The classification of stoppers is nowadays mostly done automatically, at least in a first step, with commercially available sorting machines that use an optical reading of the stoppers surface. The automatic sorting is usually followed by manual classification, at least for the better quality classes or whenever an ‘‘error-free’’ grading is aimed at. However, the classification criteria are subjective and the sorting relies on the use of stoppers as standards and on an operator-adjusted calibration. It has been shown that the degree of coincidence for the classification of stoppers by different operators or by manual vs. auto- matic classification in one single industrial unit is under 50% (Melo and Pinto 1989). For quality evaluation of the cork planks, only visual appreciation is carried out by an experienced operator. Here, the definition of criteria for the quality classes is more complex, since the raw-material itself is more het- erogeneous than the final stopper (Barros and Pereira 1987). The quality of the cork planks effects their price and exerts an influence on the yield of the produced stoppers as well as on their quality profile (Pereira et al. 1994). This last aspect is the most relevant from an eco- nomic point of view, since the price difference between good and medium quality stoppers is even larger than for cork planks. Recent work has shown the possibility of using image analysis techniques to characterise the porosity of cork planks and to classify it by means of a porosity coefficient (Pereira et al. 1996). Moreover, it is possible to distinguish between several types of defects by applying a discrimi- nant analysis based on the results of image analysis (Gonzalez-Adrados and Pereira 1996). When other defects are present in addition to the lentic- ular channels that constitute the ‘‘normal’’ porosity of cork, Originalarbeiten Æ Originals Holz als Roh- und Werkstoff 58 (2000) 39–45 Ó Springer-Verlag 2000 J. R. Gonzalez-Adrados INIA-CIFOR, Aptdo 8 111, 28080 Madrid, Spain H. Pereira (&), F. Lopes Departamento de Engenharia Florestal Universidade Te ´cnica de Lisboa 1399 Lisboa Codex, Portugal This work was financed by the concerted action ‘‘Procork’’, contract AIR2-CT93–1591 within the AIR programme of EC- DGXII and by a scholarship given to the second author by JNICT through the PRAXIS XXI programme. We acknowledge the help of CORGOM – Correia & Gomes, Lda (Portugal) with the manual classification of the cork planks. 39