缺陷检测相关论文继续更新

转自自己的github ： https://github.com/Eatzhy/surface-defect-detection

论文可以从github上下载，记得加星收藏，谢谢哦！

---

surface-defect-detection

分享一些表面缺陷检测的文章，主要检测对象是：金属表面、LCD屏、建筑、输电线等缺陷或异常检物。方法以分类方法、检测方法、重构方法、生成方法为主。电子版论文放在了paper文件的对应日期文件下。

2019.01

[1]CNN做分类

论文题目：A fast and robust convolutional neural network-based defect detection model in product quality control

摘要：The fast and robust automated quality visual inspection has received increasing attention in the product quality control for production efficiency. To effectively detect defects in products, many methods focus on the handcrafted optical features. However, these methods tend to only work well under specified conditions and have many requirements for the input. So the work in this paper targets on building a deep model to solve this problem. The elaborately designed deep convolutional neural networks (CNN) proposed by us can automatically extract powerful features with less prior knowledge about the images for defect detection, while at the same time is robust to noise. We experimentally evaluate this CNN model on a benchmark dataset and achieve a fast detection result with a high accuracy, surpassing the state-of-the-art methods.

个人总结：2017年7月的一篇杂志文章。作者使用一个多层的CNN网络对DAGM2007数据集中的六类缺陷样本进行分类，分类结束之后，对于每一类样本进行缺陷检测。具体做法是：1.使用sliding-window方法在512×512的原图上进行采样，采样大小为128×128；2.对上部分每一类图像采样后的小图像块进行二分类（有缺陷和无缺陷）。最终实验和以前传统方法做的对比，比如SIFT+SVM，效果不错。下图为文章两次分类使用的CNN网络，两次分类的区别在于：1.全连接层的输入分别为6和2；2.输入的图像尺寸不同。

[2]图像金字塔层次结构思想和卷积去噪自编码器网络对纹理缺陷做检测

论文题目：An Unsupervised-Learning-Based Approach for Automated Defect Inspection on Textured Surfaces

摘要：Automated defect inspection has long been a challenging task especially in industrial applications, where collecting and labeling large amounts of defective samples are usually harsh and impracticable. In this paper, we propose an approach to detect and localize defects with only defect-free samples for model

training. This approach is carried out by reconstructing image patches with convolutional denoising autoencoder networks at different Gaussian pyramid levels, and synthesizing detection results from these different resolution channels. Reconstruction residuals of the training patches are used as the indicator for direct pixelwise defect prediction, and the reconstruction residual map generated in each channel is combined to generate the final inspection result. This novel method has two prominent characteristics, which benefit the implementation of automatic defect inspection in practice. First, it is absolutely unsupervised that no human intervention is needed throughout the inspection process. Second, multimodal strategy is utilized in this method to synthesize results from multiple pyramid levels. This strategy is capable of improving the robustness and accuracy of the method. To evaluate this approach, experiments on convergence, noise immunity, and defect inspection accuracy are conducted. Furthermore, comparative tests with some excellent algorithms on actual and simulated data sets are performed. Experimental results demonstrated the effectiveness and superiority of the proposed method on homogeneous and nonregular textured surfaces.

个人总结：2018年6月的一篇杂志文章。文章和4月在另一篇杂志上的《Automatic Fabric Defect Detection with a Multi-Scale Convolutional Denoising Autoencoder Network Model》一文作者是同一人。主要框架：结合图像金字塔层次结构思想和卷积去噪自编码器网络（CDAE）对纹理图像缺陷进行检测。具体实施：利用不同高斯金字塔层次的卷积去噪自编码器网络重构image patchs，利用训练patch的重构残差作为直接像素方向缺陷预测的指标，将每个通道生成的重构残差图结合起来，生成最终的检测结果。 论文是无监督的方法做缺陷检测，在布匹丝织物这种重复性背景纹理很强的图集上效果很好，在金属表面、加工部件表面数据集效果一般，甚至很差。在最后的实验部分，作者也是用了DAGM2007数据集做了测试，效果一般，远远达不到工业应用要求，但相对其他方法，部分种类效果有提升。

[3]级联自编码器(CASAE)结构用于金属表面异常的分割和定位

论文题目：Automatic Metallic Surface Defect Detection and Recognition with Convolutional Neural Networks

摘要：Automatic metallic surface defect inspection has received increased attention in relation to the quality control of industrial products. Metallic defect detection is usually performed against complex industrial scenarios, presenting an interesting but challenging problem. Traditional methods are based on image processing or shallow machine learning techniques, but these can only detect defects under specific detection conditions, such as obvious defect contours with strong contrast and low noise, at certain scales, or under specific illumination conditions. This paper discusses the automatic detection of metallic defects with a twofold procedure that accurately localizes and classifies defects appearing in input images captured from real industrial environments. A novel cascaded autoencoder (CASAE) architecture is designed for segmenting and localizing defects. The cascading network transforms the input defect image into a pixel-wise prediction mask based on semantic segmentation. The defect regions of segmented results are classified into their specific classes via a compact convolutional neural network (CNN). Metallic defects under various conditions can be successfully detected using an industrial dataset. The experimental results demonstrate that this method meets the robustness and accuracy requirements for metallic defect detection. Meanwhile, it can also be extended to other detection applications.

个人总结：2018年九月一篇杂志文章。作者提出来一种用于金属表面缺陷的检测方法，借助自编码器在图像重建上的性能，设计一种级联自编码器(CASAE)体系结构，用于金属表面异常的分割和定位。再利用CNN将分割后的缺陷区域做细分类。具体的pipeline如下，整体来说文章的思路就是语义粗略分割+卷积细分类。详细的分析可以看：https://blog.youkuaiyun.com/qq_27871973/article/details/83817694

[4]Faster R-CNN用于土木行业缺陷检测

论文题目：Autonomous Structural Visual Inspection Using Region-Based Deep Learning for Detecting Multiple Damage Types

摘要： Computer vision-based techniques were developed to overcome the limitations of visual inspection by trained human resources and to detect structural damage in images remotely, but most methods detect only specific types of damage, such as concrete or steel cracks. To provide quasi real-time simultaneous detection of multiple types of damages, a Faster Region-based Convolutional Neural Network (Faster R-CNN)-based structural visual inspection method is proposed. To realize this, a database including 2,366 images (with 500 × 375 pixels) labeled for five types of damages—concrete crack, steel corrosion with two levels (medium and high), bolt corrosion, and steel delamination—is developed. Then, the architecture of the Faster R-CNN is modified, trained, validated, and tested using this database. Results show 90.6%, 83.4%, 82.1%, 98.1%, and 84.7% average precision (AP) ratings for the five damage types, respectively, with a mean AP of 87.8%. The robustness of the trained Faster R-CNN is evaluated and demonstrated using 11 new 6,000 × 4,000-pixel images taken of different structures. Its performance is also compared to that of the traditional CNN-based method. Considering that the proposed method provid