Multimodal Collaborative Intelligence for Scrap Steel Grading in Industrial IoT for Metallurgy
Year: 2026
Type: article
Abstract: As a key raw material for electric arc furnace (EAF) steelmaking, the automated and precise grading of scrap steel is crucial for achieving low-carbon, intelligent production in the era of the Industrial Internet of Things (IIoT). However, existing methods suffer from issues such as unimodal perception, limited generalization ability, and limited processing strategies. To address these challenges, we propose an intelligent scrap steel grading solution based on a multimodal image-point cloud framework. We introduce a novel joint “pose-contour” state representation to classify scrap steel into five categories. Based on this, we develop the pose-contour joint state recognition network (PCJS-Net), whose dual-trunk architecture collaborates with the intramodal heterogeneous feature fusion module (IHFFM) to integrate local image details with global contextual information effectively. To enhance robustness in industrial settings, we employ a Bayesian optimization-based data augmentation strategy to mitigate sample imbalance. Concurrently, we design a multistate point cloud thickness estimation module (MS-PCTEM) that utilizes PCJS-Net outputs as prior information, integrating spatial data of scrap steel for adaptive thickness estimation. On our self-built dataset, this method achieves 93.6% 0.45% accuracy and 92.0% 0.50% recall while reducing processing time per frame by 3.55 s compared to traditional point cloud methods, significantly outperforming existing approaches. This work provides an efficient and reliable solution for IoT-driven intelligent scrap steel grading, creating substantial value for the green and low-carbon transformation of the steel industry.
Source: IEEE Internet of Things Journal
Authors: Xin Li, Yao Xiao, Haicheng Zhao, Liya Wang, Jie Li
Institution: North China University of Science and Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7201848144
