Related Experiment Video
Updated: May 12, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.0K
Retina-Inspired Lightweight Spiking Convolutional Neural Network for Single-Image Dehazing
IEEE Transactions on Neural Networks and Learning Systems
|September 26, 2024
Summary
This study introduces a lightweight, retina-inspired spiking convolutional neural network (RI-SCNN) for efficient image dehazing. The novel approach significantly improves image quality and reduces computational costs, making it ideal for visual sensor hardware.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Neuroscience
Background:
- Hazy conditions degrade image quality by scattering light.
- Existing single-image dehazing methods are computationally expensive.
- Need for efficient and high-performance dehazing techniques.
Purpose of the Study:
- To propose a lightweight spiking convolutional neural network (CNN) for image dehazing.
- To develop a computationally efficient and energy-saving dehazing solution.
- To enhance image reconstruction quality from hazy inputs.
Main Methods:
- Developed a retina-inspired spiking CNN (RI-SCNN) with ON and OFF pathways.
- Implemented a linear reconstruction mechanism for feature integration.
- Utilized discrete binary spike trains and surrogate gradient learning for training.
Main Results:
- RI-SCNN achieves superior quantitative and qualitative dehazing performance.
- Demonstrated significant improvements in energy efficiency and run speed.
- The network effectively reconstructs clear images from hazy inputs.
Conclusions:
- The proposed RI-SCNN offers a lightweight and efficient solution for image dehazing.
- Its low computational and energy demands make it suitable for visual sensor deployment.
- Retina-inspired design and spiking mechanisms enhance dehazing capabilities.
