Semi-automatic detection and segmentation of wooden pellet size exploiting a deep learning approach

Roberto Pierdicca, Mattia Balestra, Giulia Micheletti, Andrea Felicetti and Giuseppe Toscano

Renewable Energy, 2022, vol. 197, issue C, 406-416

Abstract: The production of wood pellets was born as a response to the need to manage the residual sawdust of wood processing. Nowadays, the standard UNE EN ISO 17225-2:2021 determine the general requirements of the fuel specifications and their classes. Among these, the length of the pellet plays an important role in defining its behaviour in different contexts, starting from the way the spaces are occupied both in static (storage) and dynamic (feeding) conditions. The geometric-dimensional aspects of pellets are of particular importance for the density, the energy density, and the effective thermal capacity of thermal plants, affecting also the flowability. Despite the extreme importance of such parameters, the pellet measurement is carried out using a precision caliper on a group of individual pellets taken from laboratory samples. This method is time-consuming and returns dimensional values in very small quantities, raising the issue of sample representativeness. Considering the impact of the quality parameters, it is important to examine alternative solutions. In this light, this work has the task of testing and verifying the efficiency of a system that uses a deep neural network, to determine the geometric-dimensional parameters of wood pellets. Thus, the implemented system detects, segments, and determines the dimensions of wood pellets in a bunch. This problem is not trivial, due to the irregular lighting conditions that affect the quality of the images and the overlapping of the wood pellets. To evaluate the performance of the deep neural network approach, several experiments have been carried out, in different lighting conditions and for validation purposes, we considered also PVC pellets, which have a known and fixed dimension. The comparison between real environment data and the validation set, despite a slight tendency towards underestimation in length, shows great performances in terms of RMSE.

Keywords: Pellet size; Biomass; Deep learning; Object detection; Climate change (search for similar items in EconPapers)
Date: 2022
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (1)

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0960148122011089
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:renene:v:197:y:2022:i:c:p:406-416

DOI: 10.1016/j.renene.2022.07.109

Access Statistics for this article

Renewable Energy is currently edited by Soteris A. Kalogirou and Paul Christodoulides

More articles in Renewable Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().