FPGA-accelerated textured surface defect segmentation based on complete period Fourier reconstruction

Pan, Yinfei; Lu, Rongsheng; Zhang, Tengda

doi:10.1007/s11554-019-00927-1

Cited by 15 publications

(5 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…3) Shape feature-based methods. The shape feature belongs to the middle layer feature of the image, which mainly describes the contour and the regional feature [18].…”

Section: A Traditional Defect Detection Methodsmentioning

confidence: 99%

MFNet: A Novel Multi-level Feature Fusion Network with Multi-branch Structure for Surface Defect Detection

Zhu¹,

He²,

Zhou³

2023

Preprint

View full text Add to dashboard Cite

<p>Surface defect detection is an essential topic in the industrial inspection field. Many methods based on machine vision (MV) have been applied. However, it is still a challenging task due to the complexity of defects, including low-contrast, small objects, and irregular geometric boundaries. To deal with these problems, this paper proposes a novel multi-level feature fusion network (MFNet) with the multi-branch structure for surface defect detection. Firstly, we extract low- and high-level features via the encoder based on ResNet34. Secondly, an improved atrous spatial pyramid pooling module is adapted to expand the receptive field of low-level features. Then, the decoder adopts a multi-branch structure to fuse multi-level features for details, and a global attention module is introduced to strengthen the effectiveness of feature fusion and detection accuracy. Finally, the optimal result from multiple outputs can be obtained by multi-branch. Extensive experiments are carried out on three representative defect datasets: CrackForest, Kolektor, and RSDDs. The quantitatively contrastive experiments prove that our method enjoys a better semantic segmentation performance in industrial defect detection, outperforming four excellent semantic segmentation networks (CrackForest: Accuracy-98.54%, F1-Score-74.71%, IoU-60.22%; Kolektor: Accuracy-99.82%, F1-Score-83.09%, IoU-71.38%; Type-Ⅰ RSDDs: Accuracy-99.79\%, F1-Score-85.09%, IoU-74.84%).</p>

show abstract

“…3) Shape feature-based methods. The shape feature belongs to the middle layer feature of the image, which mainly describes the contour and the regional feature [18].…”

Section: A Traditional Defect Detection Methodsmentioning

confidence: 99%

MFNet: A Novel Multi-level Feature Fusion Network with Multi-branch Structure for Surface Defect Detection

Zhu¹,

He²,

Zhou³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…At present, there are many ways to speed up models, such as model weight quantization and model pruning. Pan et al [13] applied the FPGA accelerated Fourier reconstruction operator to texture surface defect segmentation, and the parallel acceleration structure of FPGA was three times that of the CPU of the same level. Although existing deep learning models use the GPU as a general-purpose computing unit, FPGA will become an attractive alternative as the technology evolves.…”

Section: Real-time Problemmentioning

confidence: 99%

A Survey of Surface Defect Detection Based on Deep Learning

Yang¹

2022

Proceedings of the 2022 7th International Conference on Modern Management and Education Technology (MMET 2022)

View full text Add to dashboard Cite

In recent years, with the rapid development of technologies such as computers and artificial intelligence, various research fields based on deep learning have been broadly used, among which industrial detection is the most important. In this paper, the definition of defects and defect detection is firstly defined. Then, several mainstream methods of surface defect detection based on convolutional neural network are introduced in recent years, and the typical application scenarios of each method are summarized. Finally, two key problems in surface defect detection are discussed.

show abstract

“…Hwang et al proposed a fast and stable Zernike moment algorithm based on the iterative relation of Zernike moment polynomials, and similarly, Pan et al proposed a fast and stable algorithm for Fourier Merlin moments. By using such an iterative relation for the calculation, the high power exponential operation in the direct calculation of 2 Advances in Mathematical Physics the higher-order moment kernel function by equation is avoided; thus, the overflow error and the computational accuracy error are well suppressed, and the computational speed and computational accuracy of the algorithm are improved at the same time [16]. In summary, although such iterative algorithms improve the overflow and computational accuracy errors, they still do not solve the problem of numerical integration errors.…”

Section: Related Workmentioning

confidence: 99%

Wavelet Analysis-Based Texture Analysis of Ceramic Surface Images

Liu

Zhou

Yang

2021

Advances in Mathematical Physics

View full text Add to dashboard Cite

This paper is conducted to explore a new characterization method as a supplement to the traditional roughness characterization. The main research includes the extraction and evaluation of damage features of ceramic surface morphology by applying wavelet methods, the extraction of damage features in surface contours by using wavelet analysis, and the quantitative evaluation of damage degree by using damage rate and damage mean spacing. By comparing various fractal dimension calculation methods, a fractal dimension method suitable for calculating the ceramic surface was selected, and the fractal method was used to describe the ceramic surface topography as a whole. By comparing different methods of calculating the fractal dimension and further verifying them with the measured three-dimensional morphology, it is found that the vibrational method is more suitable for calculating the fractal dimension of ceramic surface, and its calculation accuracy is investigated, and the results show that the method is a reliable one. Based on the fractal theory, a mathematical model of surface wear and surface sealing was established. Further study of the model shows that the surface with a large fractal dimension has a good sealing effect; the surface corresponding to the best fractal dimension is the most resistant to wear. The fractal method can characterize the complexity of the surface profile as a whole. The wavelet method can describe the ceramic surface profile from a local perspective, and the combination of the two methods can characterize the ceramic surface well. Finally, the experimental device of the ceramic surface defect detection system is constructed, and the joint debugging of hardware and software is completed. Under different light source intensities, ceramic image samples are collected, and the accuracy detection experiments of sample defective edges are conducted, and the results show that the light source has a small impact on the accuracy of ceramic defective edge detection. The results show that the light source has more influence on the accuracy of scratch detection. The results show that the system constructed in this thesis has good applicability for different ceramic sample detection.

show abstract

FPGA-accelerated textured surface defect segmentation based on complete period Fourier reconstruction

Cited by 15 publications

References 17 publications

MFNet: A Novel Multi-level Feature Fusion Network with Multi-branch Structure for Surface Defect Detection

MFNet: A Novel Multi-level Feature Fusion Network with Multi-branch Structure for Surface Defect Detection

A Survey of Surface Defect Detection Based on Deep Learning

Wavelet Analysis-Based Texture Analysis of Ceramic Surface Images

Contact Info

Product

Resources

About