Developing Computer Vision Models for Classifying Grain Shapes of Crushed Stone

In the construction industry, along with traditional approaches for the visual and instrumental assessment of building materials, methods based on intelligent algorithms are increasingly appearing; in particular, machine learning and neural network technologies. The utilization of modern technologie...

Full description

Saved in:
Bibliographic Details
Main Authors: Alexey N. Beskopylny, Evgenii M. Shcherban’, Sergey A. Stel’makh, Alexandr A. Shilov, Irina Razveeva, Diana Elshaeva, Andrei Chernil’nik, Gleb Onore
Format: Article
Language:English
Published: MDPI AG 2025-03-01
Series:Sensors
Subjects:
Online Access:https://www.mdpi.com/1424-8220/25/6/1914
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the construction industry, along with traditional approaches for the visual and instrumental assessment of building materials, methods based on intelligent algorithms are increasingly appearing; in particular, machine learning and neural network technologies. The utilization of modern technologies enables us to enhance building processes to a new quality level, decreasing the construction pace without precision losses compared to traditional methods. This research introduces a novel method for characterizing crushed stone grain morphology using the application of specially designed three-dimensional computer vision neural networks to point data clouds. Flakiness affects the strength, adhesion, and location of crushed stone grains. So, calculating this indicator by determining the planar dimensions of each particle in the crushed stone is necessary for the assessment of its suitability for various types of construction work. Architectures based on PointNet and PointCloudTransformer are chosen as the basis for the classification algorithms. The input data were 3D images of crushed stone grains, the shapes of which were divided into needle-shaped, plate-shaped, and cubic classes. The accuracy quality metric achieved during the training of both models was 0.86. Using intelligent algorithms, along with grain analysis methods via manual selection, sieve analysis, or using special equipment, will reduce manual labor and can also serve as an additional source for verifying the quality of building materials at various stages of construction.
ISSN:1424-8220