代码链接：https://github.com/zhao62/Deep-Residual-Shrinkage-Networks

Deep-Residual-Shrinkage-Networks

The deep residual shrinkage network is a variant（变种） of deep residual networks (ResNets), and aims to improve the feature learning ability from highly noise signals or complex backgrounds. Although the method is originally（最初） developed for vibration-based（振动） fault diagnosis（诊断）, it can be applied to image recognition and speech recognition（语音识别） as well. The major innovation is the integration of soft thresholding as nonlinear transformation layers into ResNets. Moreover, the thresholds（阈值） are automatically determined by a specially designed sub-network, so that no professional expertise on threshold determination is required.

The method is implemented using TensorFlow 1.0.1, TFLearn 0.3.2, and Keras 2.2.1, and applied for image classification. A small network with 3 residual shrinkage blocks is constructed in the code. More blocks and more training iterations can be used for a higher performance.

Abstract:

This paper develops new deep learning methods, namely（换句话说）, deep residual shrinkage networks, to improve the feature learning ability from highly noised vibration signals and achieve a high fault diagnosing accuracy.

Soft thresholding is inserted as nonlinear transformation layers into the deep architectures to eliminate unimportant features.

Moreover,（而且） considering that it is generally challenging to set proper values for the thresholds, the developed deep residual shrinkage networks integrate（集成） a few specialized（专门的） neural networks as trainable modules to automatically determine the thresholds, so that professional expertise on signal processing is not required. The efficacy of the developed methods is validated（验证） through experiments with various types of noise.

Reference:

Minghang Zhao, Shisheng Zhong, Xuyun Fu, Baoping Tang, Michael Pecht, Deep residual shrinkage networks for fault diagnosis, IEEE Transactions on Industrial Informatics, 2020, 16(7): 4681-4690.

Deep Residual Shrinkage Networks for Fault Diagnosis | IEEE Journals & Magazine | IEEE Xplore

Results：

(1) The performance on the manually（手动） noised Cifar10 dataset

In the DRSN_TFLearn.py, we manually add noise to the Cifar10 dataset, and get the results of a deep residual shrinkage network. Then, we delete lines 79-88 in DRSN_TFLearn.py, make it be a deep residual network, and get its results.

Methods	Deep residual shrinkage network	Deep residual network
Training accuracy	88.96%	87.78%
Validation accuracy	84.33%	83.99%

(2) The performance on the un-noised Cifar10 dataset

If we delete lines 25-27 in DRSN_TFLearn.py, the code will conduct a deep residual shrinkage network on the Cifar10 dataset without manually added noise. Then, if we delete lines 79-88 in DRSN_TFLearn.py, the code will conduct a deep residual network on the un-noised Cifar10 dataset.

Methods	Deep residual shrinkage network	Deep residual network
Training accuracy	90.28%	89.26%
Validation accuracy	85.87%	85.57%

Additional notes

There might be some problems in the Keras code. The TFLearn code is recommended for usage（使用）.

深度残差收缩网络：（六）代码实现相关推荐

深度残差收缩网络再次理解（论文地址+代码地址+代码理解）
首先论文地址:Deep Residual Shrinkage Networks for Fault Diagnosis | IEEE Journals & Magazine | IEEE Xp ...
【深度残差收缩网络】Deep-Residual-Shrinkage-Networks模型+代码
https://github.com/zhao62/Deep-Residual-Shrinkage-Networks 代码链接 1.代码 1.1DRSN_keras.py python版本为3.6 安 ...
《Deep residual shrinkage networks for fault diagnosis》基于深度残差收缩网络的故障诊断（翻译与python代码）
基于深度残差收缩网络的故障诊断(翻译) 赵明航,钟诗胜,付旭云,汤宝平,Michael Pecht 论文连接:https://ieeexplore.ieee.org/document/8850096 ...
注意力机制+软阈值函数=深度残差收缩网络（附代码）
深度残差收缩网络是一种面向强噪声数据的深度神经网络,是由"深度残差网络"和"收缩"组成的.一方面,"深度残差网络"已经成为了深度学习领域的基 ...
注意力机制、深度残差网络和软阈值化的集成：深度残差收缩网络（附代码）
深度残差收缩网络是一种针对强噪声数据的深度学习方法,是由"深度残差网络"和"收缩"组成的.一方面,"深度残差网络"已经成为了深度学习领域的基 ...
深度残差收缩网络：（一）背景知识
总共六篇文章: 深度残差收缩网络:(一)背景知识深度残差收缩网络:(一)背景知识_马鹏森的博客-CSDN博客深度残差收缩网络:(二)整体思路深度残差收缩网络:(二)整体思路_马鹏森的博客-CSD ...
深度残差收缩网络：借助注意力机制实现特征的软阈值化
作者 | 哈尔滨工业大学(威海)讲师赵明航本文解读了一种新的深度注意力算法,即深度残差收缩网络(Deep Residual Shrinkage Network). 从功能上讲,深度残差收缩网络是一 ...
深度残差收缩网络（Deep Residual Shrinkage Networks for Fault Diagnosis ）
摘要-本文开发了新的深度学习方法,即深度残余收缩网络,提高来自高噪声振动信号的特征学习能力,并实现较高的故障诊断准确性.软阈值作为非线性转换层插入到深层体系结构中,以消除不重要的特征.此外,考虑到通常 ...
论文翻译-基于深度残差收缩网络的故障诊断 Deep Residual Shrinkage Networks for Fault Diagnosis
深度残差收缩网络是深度残差网络的一种改进,针对的是数据中含有噪声或冗余信息的情况,将软阈值化引入深度残差网络的内部,通过消除冗余特征,增强高层特征的判别性.以下对部分论文原文进行了翻译,仅以学习为目的 ...
（全文翻译）基于深度残差收缩网络的故障诊断Deep Residual Shrinkage Networks for Fault Diagnosis
M. Zhao, S. Zhong, X. Fu, B. Tang, M. Pecht, Deep residual shrinkage networks for fault diagnosis, I ...

深度残差收缩网络：（六）代码实现