An industrial carbon block instance segmentation algorithm based on improved YOLOv8

Automatic cleaning of carbon blocks based on machine vision is currently an important aspect of industrial intelligent applications. The recognition of carbon block types and center point localization are the core contents of this task, but existing instance segmentation algorithms perform poorly in...

Full description

Saved in:
Bibliographic Details
Published in:Scientific reports Vol. 15; no. 1; pp. 8147 - 16
Main Authors: Shi, Runjie, Li, Zhengbao, Wu, Zewei, Zhang, Wenxin, Xu, Yihang, Luo, Gan, Ma, Pingchuan, Zhang, Zheng
Format: Journal Article
Language:English
Published: London Nature Publishing Group UK 09.03.2025
Nature Publishing Group
Nature Portfolio
Subjects:
639/705/117
639/705/258
Algorithms
Carbon
Carbon block instance segmentation
Humanities and Social Sciences
Industrial automation
Localization
Machine vision
multidisciplinary
Reinforce feature fusion
Science
Science (multidisciplinary)
YOLO
YOLO
Carbon block instance segmentation
Machine vision
Reinforce feature fusion
Industrial automation
ISSN:2045-2322, 2045-2322
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Automatic cleaning of carbon blocks based on machine vision is currently an important aspect of industrial intelligent applications. The recognition of carbon block types and center point localization are the core contents of this task, but existing instance segmentation algorithms perform poorly in this task. This paper proposes an industrial carbon block instance segmentation algorithm based on improved YOLOv8 (YOLOv8-HDSA), which achieves highly accurate recognition of carbon block types and edge segmentation. YOLOv8-HDSA designs a Selective Reinforcement Feature Fusion Module (SRFF) that utilizes Hadamard product and dilated convolution to enhance the feature representation of carbon block regions and suppress background noise, fully utilizing the complementary advantages of semantic and detail information to enhance feature fusion capabilities. YOLOv8-HDSA adds a convolutional self-attention mechanism with residual structure to the head, preserving important local information of carbon blocks and improving the ability to extract fine-grained edge details and global features of carbon blocks. YOLOv8-HDSA introduces Focaler-IoU as a loss function to dynamically adjust sample weights to optimize regression performance. The experimental results showed that YOLOv8-HDSA improved the average recognition accuracy of carbon blocks by 7.2% and the segmentation accuracy by 3.8% on real industrial datasets.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
content type line 23
ISSN:2045-2322
2045-2322
DOI:10.1038/s41598-025-91495-x