Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CFIS-YOLO: A Lightweight Multi-Scale Fusion Network for Edge-Deployable Wood Defect Detection

Jincheng Kang, Yi Cen, Yigang Cen, Ke Wang, Yuhan Liu

TL;DR

CFIS-YOLO tackles edge-deployable wood defect detection by integrating three innovations into a YOLOv10-based framework: CARAFE dynamic upsampling for better multi-scale feature fusion, FasterNet-inspired C2f_FNB blocks using Partial Convolution to reduce FLOPs, and an Inner-SIoU loss with auxiliary boxes and angular constraints to improve small-object localization. The combined design achieves $mAP@0.5 = 77.5\%$ on a public wood defect dataset while maintaining a compact footprint of about $7.17$ million parameters, and it demonstrates practical edge performance with $135$ FPS on the SOPHON BM1684X (power ~17\% of the original). Ablation and comparative experiments confirm that each module contributes meaningfully and that their synergy yields the best performance, surpassing several mainstream detectors at a similar or lower parameter budget. The edge-deployment results, including quantization to FP16 and a Gradio-based interface, validate CFIS-YOLO as a viable, real-time solution for industrial wood-processing environments, with future work focusing on dataset expansion and broader edge-device testing.

Abstract

Wood defect detection is critical for ensuring quality control in the wood processing industry. However, current industrial applications face two major challenges: traditional methods are costly, subjective, and labor-intensive, while mainstream deep learning models often struggle to balance detection accuracy and computational efficiency for edge deployment. To address these issues, this study proposes CFIS-YOLO, a lightweight object detection model optimized for edge devices. The model introduces an enhanced C2f structure, a dynamic feature recombination module, and a novel loss function that incorporates auxiliary bounding boxes and angular constraints. These innovations improve multi-scale feature fusion and small object localization while significantly reducing computational overhead. Evaluated on a public wood defect dataset, CFIS-YOLO achieves a mean Average Precision (mAP@0.5) of 77.5\%, outperforming the baseline YOLOv10s by 4 percentage points. On SOPHON BM1684X edge devices, CFIS-YOLO delivers 135 FPS, reduces power consumption to 17.3\% of the original implementation, and incurs only a 0.5 percentage point drop in mAP. These results demonstrate that CFIS-YOLO is a practical and effective solution for real-world wood defect detection in resource-constrained environments.

CFIS-YOLO: A Lightweight Multi-Scale Fusion Network for Edge-Deployable Wood Defect Detection

TL;DR

CFIS-YOLO tackles edge-deployable wood defect detection by integrating three innovations into a YOLOv10-based framework: CARAFE dynamic upsampling for better multi-scale feature fusion, FasterNet-inspired C2f_FNB blocks using Partial Convolution to reduce FLOPs, and an Inner-SIoU loss with auxiliary boxes and angular constraints to improve small-object localization. The combined design achieves on a public wood defect dataset while maintaining a compact footprint of about million parameters, and it demonstrates practical edge performance with FPS on the SOPHON BM1684X (power ~17\% of the original). Ablation and comparative experiments confirm that each module contributes meaningfully and that their synergy yields the best performance, surpassing several mainstream detectors at a similar or lower parameter budget. The edge-deployment results, including quantization to FP16 and a Gradio-based interface, validate CFIS-YOLO as a viable, real-time solution for industrial wood-processing environments, with future work focusing on dataset expansion and broader edge-device testing.

Abstract

Wood defect detection is critical for ensuring quality control in the wood processing industry. However, current industrial applications face two major challenges: traditional methods are costly, subjective, and labor-intensive, while mainstream deep learning models often struggle to balance detection accuracy and computational efficiency for edge deployment. To address these issues, this study proposes CFIS-YOLO, a lightweight object detection model optimized for edge devices. The model introduces an enhanced C2f structure, a dynamic feature recombination module, and a novel loss function that incorporates auxiliary bounding boxes and angular constraints. These innovations improve multi-scale feature fusion and small object localization while significantly reducing computational overhead. Evaluated on a public wood defect dataset, CFIS-YOLO achieves a mean Average Precision (mAP@0.5) of 77.5\%, outperforming the baseline YOLOv10s by 4 percentage points. On SOPHON BM1684X edge devices, CFIS-YOLO delivers 135 FPS, reduces power consumption to 17.3\% of the original implementation, and incurs only a 0.5 percentage point drop in mAP. These results demonstrate that CFIS-YOLO is a practical and effective solution for real-world wood defect detection in resource-constrained environments.

Paper Structure

This paper contains 24 sections, 10 equations, 11 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overall architecture
  4. Lightweight Upsampling Operator CARAFE
  5. The Kernel Prediction Module
  6. The Content-aware Reassembly Module
  7. FasterNet Network and C2f_FNB Module
  8. Inner-SIoU Bounding Box Loss Function
  9. Experimental Setup and Result Analysis
  10. Experimental Environment and Datasets
  11. Experimental Environment
  12. Datasets
  13. Performance Metrics
  14. Ablation Experiment
  15. Independent Effects of Modules
  16. ...and 9 more sections

Figures (11)

  • Figure 1: Structure of the proposed CFIS-YOLO.
  • Figure 2: Upsampling process of the CARAFE operator.
  • Figure 3: Partial Convolution.
  • Figure 4: The image displays typical wood defect samples within the dataset, including: (A) Live Knot, (B) Marrow, (C) Resin, (D) Dead Knot, (E) Knot with Crack, (F) Knot Missing, (G) Crack.
  • Figure 5: Number of various annotations in the dataset.
  • ...and 6 more figures