SPD-YOLO: a reliable model for imported supermarket products detection

Accurate detection of imported supermarket products is essential for automated retail management and intelligent inventory inspection. However, in practical supermarket scenarios, detection performance is often affected by dense product arrangements, highly regular vertical layouts, scale variation, partial occlusion, and complex background interference. To address these challenges, we propose SPD-YOLO, a lightweight detector for accurate and efficient imported product detection. First, the backbone incorporates a C3K2-CSSC feature enhancement module, which integrates column attention, spatial attention, and a self-calibration branch to strengthen column-wise consistency modeling and spatial feature representation for vertically arranged products. Second, we introduce a C2DA feature processing module between the backbone and the feature fusion layer. By embedding deformable attention into the C2PSA structure, C2DA enables adaptive feature sampling and enhances the representation of fine-grained and partially occluded products. Third, we adopt Slim-Neck, a lightweight feature fusion architecture that combines GSConv and VOVGSCSP to promote efficient multi-scale feature interaction while reducing computational redundancy. Finally, we replace the original detection head with EfficientHead, a lightweight decoupled head that uses grouped convolutions for efficient feature processing and Distribution Focal Loss for more accurate bounding box regression. Experiments on the self-constructed imported supermarket product dataset show that SPD-YOLO achieves 91.4% mAP50 and 75.8% mAP50-95 with only 2.7 M parameters and 5.9 GFLOPs, outperforming the YOLO11n baseline and other compared detectors. Compared with YOLO11n, SPD-YOLO reduces GFLOPs from 7.0 to 5.9 and improves inference speed from 330.1 FPS to 358.6 FPS while maintaining the same parameter size, demonstrating clear computational and speed advantages. Furthermore, on the public Locount Dataset, SPD-YOLO achieves 69.2% mAP50 and 52.8% mAP50-95, demonstrating its competitive generalization performance. These results indicate that SPD-YOLO achieves a favorable balance among detection accuracy, model complexity, and inference efficiency.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-18
DOI
https://doi.org/10.1038/s41598-026-80342-2
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00

Funders

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

SPD-YOLO: a reliable model for imported supermarket products detection

Li Liu, Lihua Tang, Xian Zhang, Chunmei Du
Scientific Reports
Advanced Neural Network Applications
article

SPD-YOLO: a reliable model for imported supermarket products detection

Li Liu, Lihua Tang, Xian Zhang, Chunmei Du
article en

Abstract

Accurate detection of imported supermarket products is essential for automated retail management and intelligent inventory inspection. However, in practical supermarket scenarios, detection performance is often affected by dense product arrangements, highly regular vertical layouts, scale variation, partial occlusion, and complex background interference. To address these challenges, we propose SPD-YOLO, a lightweight detector for accurate and efficient imported product detection. First, the backbone incorporates a C3K2-CSSC feature enhancement module, which integrates column attention, spatial attention, and a self-calibration branch to strengthen column-wise consistency modeling and spatial feature representation for vertically arranged products. Second, we introduce a C2DA feature processing module between the backbone and the feature fusion layer. By embedding deformable attention into the C2PSA structure, C2DA enables adaptive feature sampling and enhances the representation of fine-grained and partially occluded products. Third, we adopt Slim-Neck, a lightweight feature fusion architecture that combines GSConv and VOVGSCSP to promote efficient multi-scale feature interaction while reducing computational redundancy. Finally, we replace the original detection head with EfficientHead, a lightweight decoupled head that uses grouped convolutions for efficient feature processing and Distribution Focal Loss for more accurate bounding box regression. Experiments on the self-constructed imported supermarket product dataset show that SPD-YOLO achieves 91.4% mAP50 and 75.8% mAP50-95 with only 2.7 M parameters and 5.9 GFLOPs, outperforming the YOLO11n baseline and other compared detectors. Compared with YOLO11n, SPD-YOLO reduces GFLOPs from 7.0 to 5.9 and improves inference speed from 330.1 FPS to 358.6 FPS while maintaining the same parameter size, demonstrating clear computational and speed advantages. Furthermore, on the public Locount Dataset, SPD-YOLO achieves 69.2% mAP50 and 52.8% mAP50-95, demonstrating its competitive generalization performance. These results indicate that SPD-YOLO achieves a favorable balance among detection accuracy, model complexity, and inference efficiency.

Scientific Reports
Jiangsu Vocational Institute of Architectural Technology (CN)
Jiangsu University, Government of Jiangsu Province
Advanced Neural Network Applications
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.

Laika access coming soon