lu-jf / cae-transfer

In this project, a convolutinal auto-encoder based unsupervised learning and its transfer learning are built

CAE-Transfer

In this project, a convolutinal auto-encoder based unsupervised learning and its transfer learning are built

Citation

The code is a private recurrence rather than an official code, and the reproduced paper is:
M. Xia, H. Shao, Z. Huang, Z. Zhao, F. Jiang and Y. Hu,
"Intelligent Process Monitoring of Laser-Induced Graphene Production With Deep Transfer Learning",
IEEE Transactions on Instrumentation and Measurement, vol. 71, pp. 1-9, 2022, Art no. 3516409,
doi: 10.1109/TIM.2022.3186688.

Environment

The case ideally requires:
Python>=3.8
keras>=2.6.0
numpy>=1.19.5
Scikit-learn>=0.24.1
matplotlib>=3.3.4

Nets.py

The code is used for buliding the network to be trained, including Convolutional Auto-Encoder (CAE),
Enhanced Convolutional Neural Network (ECNN) which is same with the paper. The every models will be
built and compile, to test with default parameters.

Transfer.py

The code is used for unsupervised learning and transfer learning according to the paper. The dataset
is not provided because of some reasons, but the dataset form adapted this code is given:

1. the dataset should be composed of picture;
2. the directory of dataset should submit to the following structure, or correcting the code:
   .root
   |└---source
   |----└---unlabeled
   |-----------└Train
   |-----------└Valid
   |-----------└Test
   |----└---labeled
   |-----------└Train
   |└---target
   |----└---Train
   |----└---Valid
   |----└---Test
   Besides, the training and loss curve is samply plotting in the code, in order to observe the training
   effect expediently.

Test_and_Plot.py

The code is used for testing the models and plotting the confusion matrix. The model file must be existing
before you running the code.