Multiscale Adversarial and Weighted Gradient Domain Adaptive Network for Data Scarcity Surface Defect Detection

Song, Yiguo; Liu, Zhenyu; Wang, Jiahui; Tang, Ruining; Duan, Guifang; Tan, Jianrong

doi:10.1109/tim.2021.3096284

Cited by 17 publications

(7 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, it is worth thinking about choosing which F i to be concatenated when considering structural regularization. To analyze channel dependencies, M C _A 1,4 , M C _A 2,3 , M C _A 3,2 , M C _A 4,1 and M C _A 5,0 are generated according to the different topology structure of an attention path connection (as shown in figure 5). Besides, for a fair comparison, all experiments were performed on the MFE module modified with VGG16.…”

Section: Ablation Studymentioning

confidence: 99%

“…detection and defect classification. The former detection process locates the defective regions from a given image, when the defect relative location information and detailed area boundaries can be obtained [2][3][4]. The latter process correctly identifies and labels the types of defects detected, either at the image level [5], region level [6] or pixel level [7,8], and different levels of labeling methods will have a certain defective regions information differences.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A multiple feature-maps interaction pyramid network for defect detection of steel surface

Zhao¹,

Zhao²,

He³

2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

Diverse categories, variety of shapes and uncertainty of contrast lead to challenges in accurately detecting the fine details of defects in steel surface images. Deep learning methods have provided accurate, real-time detection algorithms in the field of defect detection of steel surface in recent year. Most deep learning-based networks fuse information from each intermediate layer simply and directly, while the intrinsic relationship of feature maps with different resolutions is lacking. Therefore, a novel approach to exploit the attention mechanism, Multiple Feature-maps Interaction Pyramid Network (MFIPNet), is proposed. MFIPNet is designed to consider both structural regularization and structural information in an integrated fashion by using the attention mechanism as a selector for multiple feature-maps. In MFIPNet, multiple feature-maps are integrated together with different contributions, which assists the network to better adapt to the complexity of the defects. The proposed method is evaluated on the public datasets both qualitatively and quantitatively. The results demonstrate that our method outperforms state-of-the-art approaches by more than 3.92% in mPA and 7.58% in mIoU.

show abstract

Section: Ablation Studymentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A multiple feature-maps interaction pyramid network for defect detection of steel surface

Zhao¹,

Zhao²,

He³

2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…In addition, there are some studies that try to address the data collection problem. Song et al [5] proposed a novel domain adaptive network that only needs a few labeled target samples and achieves better transfer from the source domain in the real industrial scenario. Lang et al [6] designed a novel virtual sample generation algorithm to solve the problem of insufficient defective samples.…”

Section: Surface Defect Recognitionmentioning

confidence: 99%

“…as the defect recognition on metal surfaces [5][6][7] and fabric surfaces [8,9]. Many methods employ deep learning-based models, which learn strong predictors from large-scale data sets, and they achieve great performance.…”

Section: Introductionmentioning

confidence: 99%

Cross-domain few-shot defect recognition for metal surfaces

Duan¹,

Song²,

Liu³

et al. 2022

Meas. Sci. Technol.

Self Cite

View full text Add to dashboard Cite

Defect recognition for metal surface in the industry has attracted more and more attention. However, the defect data scarcity brings a huge challenge for the defect recognition in the real industrial scenario. The traditional few-shot defect recognition method can address this problem when the train data and test data are collected from the same or similar metal surface. While the defect data from the similar metal surfaces is difficult to acquire to a certain extent. In this paper, we introduce a novel task setting which can achieve the few-shot defect recognition by transferring the knowledge across domains. This method consists of two levels: image-level and feature level. In the image-level, the meta-augmentation method is proposed to improve the recognition generalization in each meta-task by joint parameter updating from the original domain and augmented domain. In the feature-level, the class covariance guided feature perturbation method is proposed to perturb the feature distribution for enhancing the cross-domain generalization capability. Extension cross-domain experiments from the textured surface to metal surface show the superior performance of the proposed method compared with other mainstream methods.

show abstract

“…Studies that applied feature selection to improve the model achieved 72.2% mAP. In 2021, Cheng and Yu[41] used the DEA_RetinaNet model on metal surfaces. This model, which basically consists of ve parts, consists of feature extraction network, DE-block, FPN, adaptive spatial feature coupling (ASFF) module and prediction network modules.…”

mentioning

confidence: 99%

Comparative Analysis of Deep Learning Based Steel Surface Defect Detection under Different Lighting Conditions

AKIN

AYDIN²,

Aydın

2023

Preprint

View full text Add to dashboard Cite

The automatic inspection and detection of defects on flat steel surfaces is important for product manufacturers as well as for users of these products. Since industrial products are used in various fields such as transportation, energy production, food production, the inspection of these products is a difficult problem today. Traditional methods such as image processing or machine learning used to provide inspections give successful results in detecting sufficiently illuminated, strong contrast or obvious defects. In this study, flat steel surface defects, which is an industrial product, are discussed. The aim of the study is to test the robustness of deep learning methods under different illumination conditions and to determine their response. For this purpose, four popular YOLO object detection methods are used. Because of the different illuminations applied on the data set, the changes in the defect detection algorithms have been observed and the results are shared. Experimental results clearly demonstrate the effect of lighting on model success. In addition, the proposed approach presents an improvement in terms of both detection rate and frame rate compared to studies in the literature.

show abstract

Multiscale Adversarial and Weighted Gradient Domain Adaptive Network for Data Scarcity Surface Defect Detection

Cited by 17 publications

References 29 publications

A multiple feature-maps interaction pyramid network for defect detection of steel surface

A multiple feature-maps interaction pyramid network for defect detection of steel surface

Cross-domain few-shot defect recognition for metal surfaces

Comparative Analysis of Deep Learning Based Steel Surface Defect Detection under Different Lighting Conditions

Contact Info

Product

Resources

About