LC-DETR-Based Target Detection Algorithm for Separating Point of Ore Belt in Spiral Concentrator

The key to utilising the intelligence of a spiral concentrator, a type of gravity beneficiation apparatus, is identifying the mineral zones in its sorting fluid. A form of LCDETR based on the enhanced RT-DETR algorithm is proposed for detecting the separation point of the spiral concentrator mineral zone in order to address the issue of complex backgrounds and fuzzy targets that are challenging to detect when detecting the target of the separation point of the spiral concentrator mineral zone. First, the HDRAB module and CSP architecture are combined with the HDCSPNet backbone network to enhance the model's detection accuracy in a high-noise environment. The second is a novel network architecture called BIMAFPN, which combines the MAFPN and BIFPN networks to enhance the model's anti-interference capabilities. To decrease the number of parameters in the approach and increase computational efficiency, the VoVGSCSP module and GSConv lightweight convolution are introduced. The model's capacity to handle the target specifics of the ore belt separation point in the spiral concentrator is essentially improved when the EUCB module finally realises the dimensionality and resolution adjustment of the stage feature map. The accuracy of the LC-DETR model is 90.5 %, according to experiments, which is 6 % to 10 % better than that of the conventional model. In comparison to the YOLO series algorithms, Faster R-CNN, and RetinaNet models, it meets industrial real-time requirements and has optimal detection accuracy and stability. Furthermore, the accuracy of LC-DETR remains consistent at roughly 90 % in a variety of complex scenarios, including those involving noise, fuzzy mineral zones, and varying lighting. This indicates a strong capacity for generalisation, which can reliably locate and identify the target, lessen the occurrence of leakage and misdetection, and meet the need for automatic identification of mineral zoning features in factory spiral concentrators.

Authors

Institutions

Publication Details

Published
2025-04-11
DOI
https://doi.org/10.1109/ainit65432.2025.11035792
Citations
1
Primary Topic
Image and Object Detection Techniques
Type
article
Field-Weighted Citation Impact
1.28

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

LC-DETR-Based Target Detection Algorithm for Separating Point of Ore Belt in Spiral Concentrator

Ruoyun Zhang, Yunchen Zhang, Yang Liao, Huizhong Liu et al.
1 citations
Image and Object Detection Techniques
1.28
article

LC-DETR-Based Target Detection Algorithm for Separating Point of Ore Belt in Spiral Concentrator

Ruoyun Zhang, Yunchen Zhang, Yang Liao, Huizhong Liu, Jun Hu, Fei Liu
article en
1 citations

Abstract

The key to utilising the intelligence of a spiral concentrator, a type of gravity beneficiation apparatus, is identifying the mineral zones in its sorting fluid. A form of LCDETR based on the enhanced RT-DETR algorithm is proposed for detecting the separation point of the spiral concentrator mineral zone in order to address the issue of complex backgrounds and fuzzy targets that are challenging to detect when detecting the target of the separation point of the spiral concentrator mineral zone. First, the HDRAB module and CSP architecture are combined with the HDCSPNet backbone network to enhance the model's detection accuracy in a high-noise environment. The second is a novel network architecture called BIMAFPN, which combines the MAFPN and BIFPN networks to enhance the model's anti-interference capabilities. To decrease the number of parameters in the approach and increase computational efficiency, the VoVGSCSP module and GSConv lightweight convolution are introduced. The model's capacity to handle the target specifics of the ore belt separation point in the spiral concentrator is essentially improved when the EUCB module finally realises the dimensionality and resolution adjustment of the stage feature map. The accuracy of the LC-DETR model is 90.5 %, according to experiments, which is 6 % to 10 % better than that of the conventional model. In comparison to the YOLO series algorithms, Faster R-CNN, and RetinaNet models, it meets industrial real-time requirements and has optimal detection accuracy and stability. Furthermore, the accuracy of LC-DETR remains consistent at roughly 90 % in a variety of complex scenarios, including those involving noise, fuzzy mineral zones, and varying lighting. This indicates a strong capacity for generalisation, which can reliably locate and identify the target, lessen the occurrence of leakage and misdetection, and meet the need for automatic identification of mineral zoning features in factory spiral concentrators.

Jiangxi University of Science and Technology (CN)
National Natural Science Foundation of China
Openalex Percentile: Top 19%
Image and Object Detection Techniques
1.28
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.