Efficient and Refined Deep Convolutional Features Network for the Crack Segmentation of Solar Cell Electroluminescence Images
A keras implementation.
If you are using the code/model/data provided here in a publication, please consider citing:
@article{wang2022efficient,
title={Efficient and refined deep convolutional features network for the crack segmentation of solar cell electroluminescence images},
author={Wang, Chuhan and Chen, Haiyong and Zhao, Shenshen and Rahman, Muhammad Rameez Ur},
journal={IEEE Transactions on Semiconductor Manufacturing},
volume={35},
number={4},
pages={610--619},
year={2022},
publisher={IEEE}
}
Training: python = 3.5/3.6, keras = 2.2.4, tensorflow-gpu = 1.9.0, cuda = 9.0, cudnn = 7.6.5, numpy = 1.18.5, opencv-python = 4.4.0.42
Download the public crack detection dataset is available here
DeepCrack: @article{liu2019deepcrack, title={DeepCrack: A deep hierarchical feature learning architecture for crack segmentation}, author={Liu, Yahui and Yao, Jian and Lu, Xiaohu and Xie, Renping and Li, Li}, journal={Neurocomputing}, volume={338}, pages={139--153}, year={2019}, publisher={Elsevier} }.
We have create the train.txt and test.txt.
To create crack dataset, please follow:
- extract DeepCrack.zip to ./dataset/DeepCrack,
Run train.py
Run predict_img.py
You need change the path, for expamle: model = load_model("./save_model/ERDCF/ERDCF_ep140.h5" , custom_objects={'dice_loss': dice_loss, 'F_score': F_score})
Run eval.py
We provid a pretrained model on the public crack detection dataset. ./pre/ERDCF_crack.h5
We have uploaded the prediction in ./pre/ERDCF.zip.
If you have any questions, please contact me