AMS-Net: Attention mechanism based multi-size dual light source network for surface roughness prediction

Zhang, Taohong; Guo, Xuxu; Fan, Shangwu; Li, Qianqian; Chen, Saian; Guo, Xueqiang

doi:10.1016/j.jmapro.2022.07.009

Cited by 14 publications

(6 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, the background at the welding site is usually cluttered, and the object occupies a small area in the entire image; thus, the object has a weak presence and is easily missed. In recent years, an attention mechanism module has been widely used in computer vision tasks and can enhance the extraction of useful features to improve model feature extraction 29 . This study improves the original YOLOv5 model by adding an attention mechanism.…”

Section: Methodsmentioning

confidence: 99%

A guidance system for robotic welding based on an improved YOLOv5 algorithm with a RealSense depth camera

Li,

Huang,

Xue

et al. 2023

Sci Rep

View full text Add to dashboard Cite

Vision-based automatic welding guidance technology plays an essential role in robotic welding. A laser vision sensor (LVS) relies on manual intervention to guide the robot when near the workpiece, which reduces the autonomy of the welding robot and productivity. To solve this problem, a robot welding guidance system based on an improved YOLOv5 algorithm with a RealSense Depth Camera was proposed. A coordinate attention (CA) module was embedded in the original YOLOv5 algorithm to improve the accuracy of weld groove detection. The center of the predicted frame of the weld groove in the pixel plane was combined with the depth information acquired by a RealSense depth camera to calculate the actual position of the weld groove. Subsequently, the robot was guided to approach and move over the workpiece. Then, the LVS was used to guide the welding torch installed at the end of the robot to move along the centerline of the weld groove and complete welding tasks. The feasibility of the proposed method was verified by experiments. The maximum error was 2.9 mm in guiding experiments conducted with a distance of 300 mm between the depth camera and the workpiece. The percentage error was within 2% in guidance experiments conducted with distances from 0.3 to 2 m. The system combines the advantages of the depth camera for accurate positioning within a large field and the LVS for high accuracy. Once the position of the weld groove of the workpiece to be welded has been determined, the LVS combined with the robot can easily track the weld groove and realize the welding operation without manual intervention.

show abstract

Section: Methodsmentioning

confidence: 99%

A guidance system for robotic welding based on an improved YOLOv5 algorithm with a RealSense depth camera

Li,

Huang,

Xue

et al. 2023

Sci Rep

View full text Add to dashboard Cite

show abstract

“…Fang et al proposed a roughness grade inspection method based on an improved model-agnostic meta-learning network and exhibited illumination robustness [9]. Zhang et al embedded channel and spatial attention mechanisms into the proposed multisize parallel framework and improved surface roughness prediction accuracy under different light sources [24]. Su et al established a recognition model based on deep transfer learning to identify the roughness grade of milled workpieces and obtained excellent cross-domain accuracy under different illuminations [25].…”

Section: Roughness Detection By Deep Learningmentioning

confidence: 99%

Machining surface roughness detection by adaptive deep fusion capsule network with low illumination and noise robustness

Huang,

Zhang,

Shao

et al. 2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

Surface roughness measurement is of great importance for monitoring machining quality. To address the performance degradation of deep models caused by low illumination and ambient noise, this study proposes a non-contact surface roughness detection method based on an adaptive deep fusion capsule network (ADFCNet) model. Firstly, principal competent analysis-based color image enhancement is employed to augment initial surface images collected from normal illumination. Then, consisting of a deep multi-model fusion for high-level feature representation and a capsule classifier for roughness recognition, the ADFCNet model is designed to detect roughness grades by using workpiece surface images. The key hyperparameters of the model are automatically determined by a genetic algorithm. Finally, laboratory and factory experiments under low illumination are carried out to validate the effectiveness and superiority of the proposed method. Experimental results demonstrate that the proposed method has strong low-illumination and noise robustness and generalization capability, indicating a better application prospect in actual scenarios.

show abstract

“…The deep learning-based surface roughness measurement method can effectively solve the interference factors brought by the human design of feature metrics and achieve high measurement accuracy. Zhang [8] extracted features from different light source environments by designing a two-branch network and then fused them to improve the grinding surface roughness recognition accuracy. Guo [9] fused features from different light sources from both space and channel to improve the recognition accuracy of grinding.…”

Section: Introductionmentioning

confidence: 99%

Recognition of grinding surface roughness grade based on adversarial domain adaptation under variable illumination

Yi,

Huang,

Shu

et al. 2024

Surf. Topogr.: Metrol. Prop.

View full text Add to dashboard Cite

Deep learning can realize the self-extraction of grinding surface features so that end-to-end roughness measurement can be realized. Still, due to the grinding surface texture being random, the features are weak, the self-extracted grinding surface features of the same surface under different lighting environments are different, and the training data and the test data when the lighting environments are inconsistent with the recognition of the measurement of the accuracy of the lower. To address these issues, this paper proposes an adversarial domain self-adaptation (NMDANN) based visual measurement method for grinding surface roughness under variable illumination. An improved residual network is used as a generator to extract more effective metastable features, and multi-head attention is introduced into the domain discriminator to enhance its domain adaptive capability. The experimental results show that the method can recognize different grades of roughness on the grinding surface under changing light environments, laying the foundation for online visual measurement of grinding surface roughness under variable light environments.

show abstract

AMS-Net: Attention mechanism based multi-size dual light source network for surface roughness prediction

Cited by 14 publications

References 19 publications

A guidance system for robotic welding based on an improved YOLOv5 algorithm with a RealSense depth camera

A guidance system for robotic welding based on an improved YOLOv5 algorithm with a RealSense depth camera

Machining surface roughness detection by adaptive deep fusion capsule network with low illumination and noise robustness

Recognition of grinding surface roughness grade based on adversarial domain adaptation under variable illumination

Contact Info

Product

Resources

About