| 摘要: |
| 为了构建适用于小型设备的轻量级图像去噪模型,提出了一种基于特征知识蒸馏的新方法. 该方法通过学习教师模型的特征图,捕捉深层知识,从而构建轻量级去噪模型,其参数量仅为教师模型的五分之一. 实验结果验证了蒸馏算法的有效性,在不同噪声水平及数据集下,都显著提升了学生模型的去噪性能,为轻量级图像去噪模型的构建提供了一种新的方向. |
| 关键词: 知识蒸馏 特征学习 卷积神经网络 图像去噪 |
| DOI：10.20192/j.cnki.JSHNU(NS).2024.04.005 |
| 分类号:TP183 |
| 基金项目: |
|
| A lightweight image denoising model based on feature knowledge distillation |
|
SHEN Yu, ZHANG Dingni
|
|
College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
|
| Abstract: |
| To construct a lightweight image denoising model suitable for small-scale devices, a novel approach based on feature knowledge distillation was proposed in this paper. Deep-seated knowledge within a teacher model was captured through learning from its feature maps by this method, resulting in the creation of a lightweight denoising model with parameters only one-fifth the size of the teacher model. Experimental results validated the effectiveness of the distillation algorithm, demonstrating significant improvements in denoising performance for the student model across varying noise levels and datasets, which introduced a promising avenue for constructing lightweight image denoising models. |
| Key words: knowledge distillation feature learning convolution neural network image denoising |
