By Guizhong Fu, Peize Sun, Wenbin Zhu, Jiangxin Yang, Yanlong Cao, Michael Ying Yang and Yanpeng Cao.

The paper is available at |[PDF Download]

In this paper, we present a compact yet effective convolutional neural network (CNN) model, which emphasizes the training of low-level features and incorporates multiple receptive fields, to achieve fast and accurate steel surface defect classification.Our proposed method adopts the pre-trained SqueezeNet as the backbone architecture.We also construct a diversity-enhanced testing dataset of steel surface defects to evaluate the robustness of classification models. The dataset contains severe camera noise, non-uniform illumination, and motion blur.

The enhanced dataset is build on NEU dataset: A noise robust method based on completed local binary patterns for hot-rolled steel strip surface defects neural networks | |[Download]

The enhanced dataset can be download at:

Google Drive: https://drive.google.com/file/d/16HjqGvnr_OUfTF0HSN1XyNg9XPr4Edfw/view?usp=sharing

Baidu Cloud: https://pan.baidu.com/s/18OowNbcJmBIi92fTWKI2wg Password：fcmh

This code is based on Caffe. Thanks to the contributors of Caffe. Caffe: https://github.com/BVLC/caffe

Our Model also uses the pretrain model SqueezeNet-1.0 SqueezeNet: https://github.com/DeepScale/SqueezeNet

[ETE] An end-to-end steel strip surface defects recognition system based on convolutional neural networks | |[pdf]

[DECAF+MLR] A generic deep-learning-based approach for automated surface inspection | |[pdf]

If you have any questions, feel free to contact:

• Guizhong Fu ([email protected])
• Yanpeng Cao ([email protected])

Copyright(c) Guizhong Fu and Yanpeng Cao All rights reserved.