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.

Vision-based size classification of iron ore pellets using ensembled convolutional neural network

Detection and segmentation of iron ore green pellets in images using lightweight U-net deep learning network

Facilitating Ferrous Oxide Prediction: Enabling Sintering Forecasting with Orthogonal Basis-Based Implicit Subspace Identification

A new weakly supervised learning approach for real-time iron ore feed load estimation

Ore image classification based on small deep learning model: Evaluation and optimization of model depth, model structure and data size

Stacked Deep Ensemble for Automated Iron Ore Image Characterization

Machine learning for particle size prediction in iron ore grinding process

Deep Fusion Feature Extraction and Classification of Pellet Phase

Pellet Roasting Management System Based on Deep Learning and Internet of Things

Analysis of Particle Size Distribution of Coke on Blast Furnace Belt Using Object Detection

Blast furnace raw material granularity recognition model based on deep learning and multimodal fusion of 3D point cloud

Improved Online Prediction of Silicon Content in Ironmaking Process Using Support Vector Regression with Novel Outlier Detection

Counting Steel Rods Online Using Lqv Neural Network In Real-Time Images

OreFormer: Ore Sorting Transformer Based on ConvNet and Visual Attention

Research on application of on-line melting in-situ visual inspection of iron ore powder based on Faster R-CNN

Deep learning based data augmentation for large-scale mineral image recognition and classification

Image-Based Size Analysis of Agglomerated and Partially Sintered Particles via Convolutional Neural Networks

Implementation of near-infrared spectroscopy and convolutional neural networks for predicting particle size distribution in fluidized bed granulation

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

Deep learning-based size prediction for optical trapped nanoparticles and extracellular vesicles from limited bandwidth camera detection

Research on quality prediction of iron ore green pellets based on optimized neural network