Low-pass U-Net: a segmentation method to improve strip steel defect detection

Liu, Bo; yang, bin; Zhao, Yelong

doi:10.1088/1361-6501/aca34a

Cited by 8 publications

(5 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Pratt et al [24] used the U-Net semantic segmentation model for defective image segmentation of solar photovoltaic modules, and the algorithm achieved a high recall rate of 84%, 69%, and 53% for inactive, cracked, and gridline defects, respectively. Liu et al [25] used the improved U-Net network to segment the surface defects of strip steel, and the experimental results showed that the Dice metrics, mIoU and Recall reached 0.903, 0.865, and 0.912, respectively. These methods demonstrate that semantic segmentation has achieved commendable results in various industrial domains, indicating its feasibility for application in the detection of WR surface defects.…”

Section: Referencesmentioning

confidence: 99%

Detection and segmentation of wire rope surface deficiency based on YOLOv8 and U-Net

Zhenbin,

Meishuai,

Xun

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

The presence of surface defects in wire ropes (WR) may lead to potential safety hazards and performance degradation, necessitating timely detection and repair. Hence, this paper proposes a method for detecting surface defects in WR based on the deep learning models YOLOv8s and U-Net, aiming to identify surface defects in real-time and extract defect data, thereby enhancing the efficiency of surface defect detection. Firstly, the ECA attention mechanism is incorporated into the YOLOv8 algorithm to enhance detection performance, achieving real-time localization and identification of surface defects in WR. Secondly, in order to obtain detailed defect data, the U-Net semantic segmentation algorithm is employed for morphological segmentation of defects, thereby obtaining the contour features of surface defects. Finally, in conjunction with OpenCV technology, the segmentation results of the defects are quantified to extract data, obtaining parameters such as the area and perimeter of the surface defects in the WR. Experimental results demonstrate that the improved YOLOv8-ECA model exhibits good accuracy and robustness, with the model’s mAP@0.5 reaching 84.78%, an increase of 1.13% compared to the base model, an accuracy rate of 90.70%, and an FPS of 65. The U-Net model can efficiently perform segmentation processing on surface defects of WR, with an mIOU of 83.54% and an mPA of 90.78%. This method can rapidly, accurately, and specifically detect surface defects in WR, which is of significant importance in preventing industrial production safety accidents.

show abstract

Section: Referencesmentioning

confidence: 99%

Detection and segmentation of wire rope surface deficiency based on YOLOv8 and U-Net

Zhenbin,

Meishuai,

Xun

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

show abstract

“…Experimental results on industrial datasets demonstrated successful metal defect detection under various conditions. In a similar vein, Liu et al 8 proposed a low-pass U-Net to enhance existing methods for detecting and segmenting surface defects in steel strips. By applying an adaptive variance Gaussian low-pass filter before downsampling in the encoder, aliasing is prevented, and high-frequency information is separated, leading to improved defect segmentation.…”

Section: Related Workmentioning

confidence: 99%

Steel Surface Defect Detection Based on Multi-Layer Fusion Networks

Li,

Liu,

Yin

et al. 2023

Preprint

View full text Add to dashboard Cite

Detecting defects on steel surfaces is crucial for ensuring product quality and production safety in industrial settings. Object detection using deep learning, particularly the YOLOv5 model, has become a widely adopted method for this purpose. However, the complex shapes of current steel surface defects pose challenges for precise detection, especially when using low-cost recognition devices with small resolution images. To address these challenges, we integrated the RepBi-PAN fusion network into YOLOv5, enhancing the detection capability for large targets in complex backgrounds. To mitigate issues related to the premature introduction of shallow features and decrease in Precision, we optimized the model structure by incorporating the DenseNet structure into the backbone for improved feature extraction. Additionally, we introduced the Normalized Attention Module (NAM) to enhance the detection capability for small targets. Experimental results demonstrate the effectiveness of the enhanced model, showing a 4.1% increase in mean average precision (mAP), a 3.2% improvement in precision, and a 2.4% enhancement in recall. The improved algorithm outperforms the original in complex backgrounds and recognizing small targets, addressing limitations of the Rep-Bi network. Compared to other YOLO algorithms, our approach achieves optimal values for recall and mAP while maintaining a smaller model size. In comparison with YOLOv8, the improved model surpasses all V8 models, being only 0.5% below the precision of the largest YOLOv8x model. Simultaneously, the improved model is smaller and has fewer parameters compared to all models in the YOLOv8 series, with slightly higher GFLOPs than the smaller v8 models.

show abstract

“…In addition, many researchers have extended the application of object detection methodologies to other fields such as airborne object detection, 3D detection, and medical detection. All of these works have contributed to the advancement of the defect detection sector [21][22][23][24].…”

Section: Introductionmentioning

confidence: 99%

A steel surface defect detection model based on graph neural networks

Pang,

Tan

2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

Steel is an indispensable raw material in the construction industry. To avert catastrophic events such as building collapse, it is essential to detect minute defects on steel surfaces during production. However, this has been a persistent challenge due to the minuscule and dense nature of these defects. To this end, we propose an efficient defect detector called Vision Grapher with Hadamard (ViGh) , which employs a novel attention mecha-nism (HDmA) to establish local-to-local relationships within an image and integrates global relationships by graph convolution. With the HDmA module, we can not only fuse information under the same field of view, but also under different fields of view, which significantly enhances the richness of the acquired features. In addition, com-pared to convolutional neural networks, graph neural networks can utilize the contextual information in the image more effectively and resulting in better performance. We eval-uate our model on the NEU-DET and GC-10 benchmark datasets, which encompass six and ten types of defects on the surfaces of hot-rolled and cold-rolled steel, and our mod-el achieves a mean Average Precision (mAP) of 79.04% and 66.93% on the two datasets, respectively. The results demonstrate that our model significantly improves the accuracy of defect detection compared to existing methods.

show abstract

Low-pass U-Net: a segmentation method to improve strip steel defect detection

Cited by 8 publications

References 25 publications

Detection and segmentation of wire rope surface deficiency based on YOLOv8 and U-Net

Detection and segmentation of wire rope surface deficiency based on YOLOv8 and U-Net

Steel Surface Defect Detection Based on Multi-Layer Fusion Networks

A steel surface defect detection model based on graph neural networks

Contact Info

Product

Resources

About