Accesso libero

YOLOv8-A: Enhanced Lightweight Object Detection with Nonlinear Feature Fusion and Mathematical Optimization for Precision Small Target Detection in Industrial Silicon Melting Processes

TongLI,

TongLI,

  • School of Information Engineering, Beijing Institute of Graphic CommunicationBeijing, China
  • School of Computer and Communication Engineering, University of Science and TechnologyBeijing, China
Cerca questo autore su
Sciendo|Google Scholar
TongLI,
 e 
Zhen-Cheng Li

Zhen-Cheng Li

School of Information Engineering, Beijing Institute of Graphic CommunicationBeijing, China
Cerca questo autore su
Sciendo|Google Scholar
Li, Zhen-Cheng
  
17 mar 2025
YOLOv8-A: Enhanced Lightweight Object Detection with Nonlinear Feature Fusion and Mathematical Optimization for Precision Small Target Detection in Industrial Silicon Melting Processes's Cover Image
INFORMAZIONI SU QUESTO ARTICOLO
Articolo precedente
Articolo Successivo
Cita
Scarica la copertina
Pubblicato online: 17 mar 2025
Ricevuto: 27 ott 2024
Accettato: 11 feb 2025
DOI: https://doi.org/10.2478/amns-2025-0830
Parole chiave
© 2025 TongLI et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Aperture protrusions diagram during the temperature-raising process.
Aperture protrusions diagram during the temperature-raising process.

Figure 2.

YOLOv8m Structure.
YOLOv8m Structure.

Figure 3.

Improved YOLOv8-A Structural Design
Improved YOLOv8-A Structural Design

Figure 4.

Structural of GSConv
Structural of GSConv

Figure 5.

Visualization of Feature Maps
Visualization of Feature Maps

Figure 6.

Dynamic Up-sampling Based on Sampling Points
Dynamic Up-sampling Based on Sampling Points

Figure 7.

Sampling Point Generator in DySample
Sampling Point Generator in DySample

Figure 8.

Neck Feature Network Design
Neck Feature Network Design

Figure 9.

Automated Defect Detection Platform for Crystal Processing and Melting Stages.
Automated Defect Detection Platform for Crystal Processing and Melting Stages.

Figure 10.

Precision-Recall curves comparing YOLOv8-A with Faster R-CNN, YOLOv3,YOLOv5, and YOLOv8m,demonstrating YOLOv8-A’s superior precision and recall in small target detection
Precision-Recall curves comparing YOLOv8-A with Faster R-CNN, YOLOv3,YOLOv5, and YOLOv8m,demonstrating YOLOv8-A’s superior precision and recall in small target detection

Figure 11.

Comparison of mAP@0.5 and mAP@0.5:0.95 across different models
Comparison of mAP@0.5 and mAP@0.5:0.95 across different models

Figure 12.

mAP@0.5 Progression Across Training Epochs
mAP@0.5 Progression Across Training Epochs

Figure 13.

Inference Time vs. mAP@0.5
Inference Time vs. mAP@0.5

Figure 14.

Training loss curves showing faster convergence for YOLOv8-A compared to baseline models.
Training loss curves showing faster convergence for YOLOv8-A compared to baseline models.

Figure 15.

Radar chart comparing YOLOv8-A with YOLOv5, YOLOv7, and DETR across multiple performance metrics.
Radar chart comparing YOLOv8-A with YOLOv5, YOLOv7, and DETR across multiple performance metrics.

Figure 16.

Comparison of detection results for YOLOv8-A and YOLOv8m on aperture protrusion images .
Comparison of detection results for YOLOv8-A and YOLOv8m on aperture protrusion images .

Training Parameters

Parameter Value
Batch 16
Epochs 300
Initial Learning Rate 0.001
Optimizer AdamW
Input Image size 640×640
Early Stopping 50 epochs without improvement in validation accuracy
Learning rate 0.001
Learning Rate Decay Cosine Annealing
NMS IoU 0.7
Weight-Decay 0.0005

Comparison for different models, demonstrating YOLOv8-A’s superior performance and efficiency_

Model mAP@0.5 mAP@0.5:0.95 Precision Recall Params (M) Inference Time (ms) FPS
Faster R-CNN 0.768 0.722 0.815 0.856 139.24 16.8 59.5
YOLOv5m 0.933 0.775 0.819 0.989 25.06 12.9 77.5
YOLOv8m 0.947 0.774 0.862 0.988 25.86 3.8 263.2
YOLOv8-A 0.982 0.806 0.920 0.996 23.67 3.7 270.3

Ablation Study Result

Model GSConv DySample BiFPN mAP@0.5 mAP@0.5:0.95
YOLOv8m ✗ ✗ ✗ 0.941 0.779
YOLOv8-A + GSConv ✓ ✗ ✗ 0.951 0.780
YOLOv8-A + DySample ✓ ✓ ✗ 0.960 0.792
YOLOv8-A (Full) ✓ ✓ ✓ 0.982 0.806

Presents a comprehensive comparison with other models

Model mAP@0.5 mAP@0.5:0.95 Precision Recall Params (M) Inference Time (ms)
Faster R-NN 0.768 0.722 0.815 0.856 139.24 16.8
YOLOv5m 0.933 0.775 0.819 0.989 25.06 12.9
YOLOv8m 0.947 0.774 0.862 0.988 25.86 3.8
YOLOv8-A 0.982 0.806 0.920 0.996 23.67 3.7
EfficientDet 0.929 0.794 0.854 0.978 5.98 8.2
DETR 0.944 0.780 0.860 0.991 50.00 32.3

Benchmarking results comparison with other models

Device Inference Time (ms) Power Consumption (W) FPS
NVIDIA Jetson Nano 28 12 35
NVIDIA Xavier 23 15 45
Lingua:
Inglese
Frequenza di pubblicazione:
1 volte all'anno
Argomenti della rivista:
Scienze biologiche, Scienze della vita, altro, Matematica, Matematica applicata, Matematica generale, Fisica, Fisica, altro
Feed RSS della rivista