Related Experiment Videos
FLF-RCNN: A Fine-Tuned Lightweight Faster RCNN for Precise and Efficient Industrial Quality Inspection
Ningli An1, Zhichao Yang1, Liangliang Wan1
1The Faculty of Printing, Packaging Engineering and Digital Media Technology, Xi'an University of Technology, Xi'an 710048, China.
Sensors (Basel, Switzerland)
|March 28, 2026
Summary
A new Fine-tuned Lightweight Faster RCNN (FLF-RCNN) framework improves industrial quality inspection (IQI) efficiency and accuracy. This deep learning model enhances defect detection in manufacturing by optimizing feature extraction and adapting to varied defect scales.
Area of Science:
- Manufacturing Technology
- Computer Vision
- Deep Learning
Background:
- Industrial Quality Inspection (IQI) is crucial for intelligent manufacturing.
- Deep learning excels at feature extraction but faces efficiency and transferability challenges in IQI.
- Existing methods struggle with the accuracy-efficiency trade-off and adapting to diverse defect scales.
Purpose of the Study:
- To propose a Fine-tuned Lightweight Faster RCNN (FLF-RCNN) framework for enhanced IQI.
- To address the limitations of insufficient efficiency and poor transferability in current deep learning-based IQI methods.
- To improve the adaptability of anchor box ratios for detecting defects of various sizes.
Main Methods:
- Introduced LSNet, a lightweight backbone network with optimized architecture for enhanced receptive fields.
- Employed a collaborative mechanism using large and small kernel convolutions for efficient and precise defect representation.
- Integrated transfer learning with pretrained weights for improved generalization in data-scarce environments.
- Developed an Adaptive Anchor Box-Adjustment Module (AAB-AM) using K-means clustering to handle varied defect scales.
Main Results:
- FLF-RCNN achieved a mAP50 of 43.6% on the Tianchi dataset, outperforming MobileNet and EfficientNet backbones.
- The method surpassed the baseline Faster R-CNN by 7.9% in mAP50.
- Computational complexity was reduced by approximately 40% (98.65 GFLOPs).
- Parameter count decreased by around 30% (28.2M).
Conclusions:
- FLF-RCNN presents a feasible and practical solution for IQI in intelligent manufacturing.
- The framework achieves a superior balance between accuracy and computational efficiency.
- FLF-RCNN demonstrates enhanced generalization and adaptability for diverse industrial inspection tasks.