Determining wood chip size: image analysis and clustering methods

• Main Authors: Paolo Febbi, Corrado Costa, Paolo Menesatti, Luigi Pari
• Format: Article
• Language: English
• Published: PAGEPress Publications 2013-09-01
• Series: Journal of Agricultural Engineering
• Subjects: image analysis, wood chip, size distribution, clustering, sieve result.
• Online Access: http://www.agroengineering.org/index.php/jae/article/view/344
• ISSN: 1974-7071; 2239-6268

One of the standard methods for the determination of the size distribution of wood chips is the oscillating screen method (EN 15149- 1:2010). Recent literature demonstrated how image analysis could return highly accurate measure of the dimensions defined for each individual particle, and could promote a new method depending on the geometrical shape to determine the chip size in a more accurate way. A sample of wood chips (8 litres) was sieved through horizontally oscillating sieves, using five different screen hole diameters (3.15, 8, 16, 45, 63 mm); the wood chips were sorted in decreasing size classes and the mass of all fractions was used to determine the size distribution of the particles. Since the chip shape and size influence the sieving results, Wang’s theory, which concerns the geometric forms, was considered. A cluster analysis on the shape descriptors (Fourier descriptors) and size descriptors (area, perimeter, Feret diameters, eccentricity) was applied to observe the chips distribution. The UPGMA algorithm was applied on Euclidean distance. The obtained dendrogram shows a group separation according with the original three sieving fractions. A comparison has been made between the traditional sieve and clustering results. This preliminary result shows how the image analysis-based method has a high potential for the characterization of wood chip size distribution and could be further investigated. Moreover, this method could be implemented in an online detection machine for chips size characterization. An improvement of the results is expected by using supervised multivariate methods that utilize known class memberships. The main objective of the future activities will be to shift the analysis from a 2-dimensional method to a 3- dimensional acquisition process.