A novel image processing and machine learning approach for wood pellet size classification using shadow analysis
Year: 2025
Type: article
Abstract: The increasing use of wood pellets as a renewable energy source has driven the need for improved quality assessment methods. While current standards, such as ISO 17829, define dimensional analysis procedures, they rely on manual measurements that may not fully capture the variability of a sample. This study presents a novel patent-pending image-processing-based approach for classifying pellet dimensions using shadow analysis. A prototypal system, composed of a camera, light sources, and a computing unit, was developed to measure shadows cast by stacked pellets under controlled lighting conditions.Pellet samples were analyzed using both traditional caliper measurements and the proposed system. The extracted shadow-based features were assessed through statistical analysis, Pearson correlation, and Principal Component Analysis to identify relationships between shadow characteristics and actual pellet dimensions. Machine learning models, including K-Nearest Neighbors (KNN), Support Vector Machines, and Logistic Regression, were trained and validated.Results indicate that the light source position significantly influences shadow characteristics and classification accuracy. The best model, KNN, achieved an accuracy of 71% in classifying pellet length classes. These findings suggest that shadow-based image analysis could provide a more efficient and representative alternative to manual dimensional analysis, with potential applications in pellet storage and combustion systems.
Source: Renewable Energy
Authors: Thomas Gasperini, Elena Leoni, Nicolò Bartolini, Giacomo Ciccone, Giuseppe Toscano, Carmine De Francesco
Institution: Marche Polytechnic University
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,460
OpenAlex ID: W4416919276