Identification and analysis of seashells in sea sand using computer vision and machine learning

Abstract

Due to the shortage and high price of river sand, the use of sea sand as a fine aggregate for concrete is gradually being considered. Seashells are fragile and have an undesirable effect on the compressive strength of concrete. However, the exact effect of seashells is still unclear and quality control of concrete is not possible since there are no effective methods for seashell characterization. In this study, we investigated the feasibility of segmenting photos of sea sand and analyzing seashells by using three typical machine learning methods, i.e., PointRend, DeepLab v3 + , and Weka. A new imaging method was proposed to avoid overlapping sea sand particles and preserve the smallest particles with sufficient resolution. A total of 960 photos were captured, and 2199 seashells were labeled, of which 80% and 20% were used for model training and validation, respectively. As a result, PointRend could efficiently recognize seashells with different shapes, sizes, and surface textures. It also had the highest Intersection over Union (IOU) and pixel accuracy (PA) scores due to the well-defined boundaries of the seashells, followed by DeepLab v3 + and Weka. From the segmentation results, the size of the seashells showed a left-skewed distribution with a mean diameter of 0.747 mm, which was smaller than the size of the sea sand. There was also considerable variation in the irregularity and roundness of the seashells. As the size of the seashells increased, their shapes became more irregular. The automated analysis of the seashells can provide further insights into the effect of shells on the properties of concrete.

More information

Identifiers

ORCID for Ye Li: orcid.org/0000-0002-2379-5497

Catalogue record

Export record

Altmetrics