Related Experiment Video
Updated: Jan 4, 2026
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
978
FRD-CNN: Object detection based on small-scale convolutional neural networks and feature reuse
1School of Computer Science and Technology, Xidian University, Xi'an, 710071, China.
Scientific Reports
|November 10, 2019
Summary
This study introduces novel neural networks, fire-FRD-CNN and mobile-FRD-CNN, to address the increased model size in object detection caused by feature reuse. These networks efficiently manage parameters while maintaining performance on benchmark datasets.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Convolutional Neural Networks (CNNs) dominate recent object detection.
- Feature reuse enhances CNN performance but inflates model size.
- DenseNet mitigates parameter increase with thin layers, inspiring further research.
Purpose of the Study:
- To develop novel methods for managing model size increase from feature reuse in CNNs.
- To construct efficient neural network architectures for object detection.
Main Methods:
- Employed diverse feature reuse strategies on fire and mobile units.
- Developed two new neural network models: fire-FRD-CNN and mobile-FRD-CNN.
- Validated models using experiments on KITTI and PASCAL VOC datasets.
Main Results:
- Successfully constructed fire-FRD-CNN and mobile-FRD-CNN architectures.
- Demonstrated the effectiveness of the proposed models on standard object detection benchmarks.
- Achieved efficient parameter management in feature-rich CNNs.
Conclusions:
- The proposed fire-FRD-CNN and mobile-FRD-CNN offer effective solutions for parameter-efficient object detection.
- Feature reuse methods can be optimized to prevent excessive model size increase.
- Novel architectures can balance performance and efficiency in deep learning models.
