Opto-Electronic Engineering, Volume. 52, Issue 2, 240280-1(2025)
Improving the lightweight FCM-YOLOv8n for steel surface defect detection
Figures & Tables(13)
Fig. 7. Comparison of detection performance on the Severstal and NEU-DET datasets
Fig. 8. Comparison of detection performance of different models on the Severstal dataset after data augmentation
Fig. 9. Comparison of detection performance on the GC10-DET dataset
Table 1. Comparison experiment between Cc-C2f and C2f
View tableView in ArticleTable 1. Comparison experiment between Cc-C2f and C2f
Module mAP@0.5/% Par/M FLOPs/G FPS C2f 75.2 3.0 8.1 181 Cc-C2f 75.9 2.6 6.9 161
Table 2. Ablation experimental data
View tableView in ArticleTable 2. Ablation experimental data
Dataset FreqFusion Cc-C2f MA mAP@0.5/% Par/M FLOPs/G FPS Severstal 72.9 3.0 8.1 153 √ 74.5 2.8 7.7 156 √ 73.0 2.6 6.9 141 √ 74.1 3.0 8.1 145 √ √ 74.0 2.5 6.6 145 √ √ √ 75.1 2.5 6.6 143 NEU-DET 75.2 3.0 8.1 181 √ 75.7 2.8 7.7 188 √ 75.9 2.6 6.9 161 √ 75.9 3.0 8.1 182 √ √ 76.0 2.5 6.6 154 √ √ √ 76.7 2.5 6.6 154
Table 3. Comparison of detection data from different algorithms
View tableView in ArticleTable 3. Comparison of detection data from different algorithms
Dataset Model mAP@0.5/% Par/M FLOPs/G FPS Severstal YOLOv3-tiny 60.2 12.1 18.9 151 YOLOv4-tiny 55.5 5.9 16.1 97 YOLOv5n 72.6 2.5 7.1 175 YOLOv5s 72.1 9.1 23.8 120 YOLOv6n 74.5 4.2 11.8 153 YOLOv7-tiny 61.5 6.0 13.1 76 YOLOv8n 72.9 3.0 8.1 153 YOLOv8s 73.7 11.1 28.4 106 Ours 75.1 2.5 6.6 143 NEU-DET YOLOv3-tiny 64.4 12.1 18.9 156 YOLOv4-tiny 64.0 5.9 16.1 120 YOLOv5n 73.2 2.5 7.1 185 YOLOv5s 75.8 9.1 23.8 106 YOLOv6n 75.9 4.2 11.8 185 YOLOv7-tiny 68.6 6.0 13.1 89 Reference [ 3 ]74.4 5.4 8.9 87 Reference [ 10 ]75.1 2.3 9.0 - Reference [ 11 ]75.7 14.4 - 109 Reference [ 12 ]75.7 7.5 16.8 94 Reference [ 13 ]76.0 3.0 - - YOLOv8n 75.2 3.0 8.1 181 YOLOv8s 75.2 11.1 28.4 108 Ours 76.7 2.5 6.6 154
Table 4. Comparison of GC10-DET dataset detection results
View tableView in ArticleTable 4. Comparison of GC10-DET dataset detection results
Model AP/% mAP@0.5/% Par/M FLOPs/G FPS Pu WI Cg Ws Os Ss In Rp Cr Wf YOLOv8n 97.9 89.2 96.0 77.9 68.2 63.0 37.5 28.1 44.7 85.6 68.8 3.0 8.1 303 FCM-YOLOv8n 98.6 89.3 96.5 78.4 69.7 67.3 29.1 36.6 60.8 90.4 71.7 2.5 6.6 270
Tools
Get Citation
Copy Citation Text
Liming Liang, Kangquan Chen, Linjun Chen, Pengwei Long. Improving the lightweight FCM-YOLOv8n for steel surface defect detection[J]. Opto-Electronic Engineering, 2025, 52(2): 240280-1
Paper Information
Category: Article
Received: Nov. 30, 2024
Accepted: Jan. 6, 2025
Published Online: Apr. 27, 2025
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20