The deep residual shrinkage network is a variant 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 developed for vibration-based fault diagnosis, it can be applied to image recognition and speech recognition as well. The major innovation is the integration of soft thresholding as nonlinear transformation layers into ResNets. Moreover, the thresholds 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.
This paper develops new deep learning methods, namely, 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, considering that it is generally challenging to set proper values for the thresholds, the developed deep residual shrinkage networks integrate a few specialized 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 through experiments with various types of noise.
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.
https://ieeexplore.ieee.org/document/8850096
http://homepage.hit.edu.cn/zhaominghang
(1) The performance on the manually 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% |
There might be some problems in the Keras code. The TFLearn code is recommended for usage.