Evaluation of a standardised procedure to assess the shape of pellets using image analysis

The influence of threshold definition, number of pellets counted, image mag nification and lightning technique on the assessment. of pellet shape has b een investigated using three batches of pellets and an image analysis syste m. The pellet parameters measured were 'aspect ratio', 'circularity', 'proj ection sphericity','e(R),' and 'Feret diameter.' The methodical error, repr oducibility and repeatability of the results were chosen as statistical tes t parameters. The position of the light source is crucial in providing an a ccurate particle size value. Top light was identified as the illumination t echnique that gave a mean pellet size similar to the true pellet size. The use of a light table produced significantly larger pellet size values. A mi nimum pixel resolution appears necessary for an accurate shape parameter de finition. One pixel should not cover more than 30 mu m for pellets of an av erage particle size of 1.2 mm. Shape descriptors, which are based on a mult iple combination of area and perimeter data such as the circularity, are gr eatly dependent on the number of pellets counted. Shape factors, which do n ot (aspect ratio) or only as a single value do involve an area or perimeter measurement (e(R), projection sphericity) are, however, nearly independent of the number of pellets counted, as long as the magnification is sufficie ntly large and the pellets are randomly drawn from the batch. For nearly sp herical particles, the methodical error is below 1%, but for elongated part icles this error can reach 5%. The repeatability is also very good for near ly spherical particles (< 2%), but increases to very large values if the pa rticles are clearly elongated. The limiting values for the various shape fa ctors should be reconsidered. An upper value for the aspect ratio of 1.1 an d a lower value of 0.6 for e(R) are recommended. The circularity should not be used as the shape factor to characterise spheres, because errors in ima ge recognition can affect strongly the applicability of this shape factor. The projection sphericity has only a limited sensitivity to variations in p article shape. (C) 1999 Elsevier Science B.V. All rights reserved.