Hformer: Hybrid convolutional neural network transformer network for fringe order prediction in phase unwrapping of fringe projection

Zhu, Xinjun; Han, Zhiqiang; Yuan, Mengkai; Guo, Qinghua; Wang, Hongyi; Song, Limei

doi:10.1117/1.oe.61.9.093107

Cited by 7 publications

(5 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, the rapid ascendancy of the model structure was particularly evident in the domain of natural language processing (NLP) following the breakthrough performance of bidirectional encoder representations from transformers (BERT) as demonstrated by Devlin et al 5 The dominance of BERT across a spectrum of NLP tasks validated the potency of the transformer architecture, solidifying its position as a cornerstone in the evolution of NLP methodologies. Inspired by its capabilities, researchers have found that the transformer structure also performs well in computer vision tasks, and transformers are playing powerful roles as emerging research frameworks during the more than a decade of dominance by convolutional neural networks (CNNs) in the vision domain 6 …”

Section: Deep Learning- and Transformer-related Researchmentioning

confidence: 99%

“…Inspired by its capabilities, researchers have found that the transformer structure also performs well in computer vision tasks, and transformers are playing powerful roles as emerging research frameworks during the more than a decade of dominance by convolutional neural networks (CNNs) in the vision domain. 6…”

Section: Deep Learning-and Transformer-related Researchmentioning

confidence: 99%

See 1 more Smart Citation

Transformer-based visual inspection algorithm for surface defects

Zhu,

Chang,

Liu

et al. 2023

Opt. Eng.

View full text Add to dashboard Cite

Industrial production often faces a variety of complex working conditions that lead to various defects, including Mura, on the surfaces of various industrial products. We propose a reconstruction network called RecTransformer, which is developed with a transformer for anomaly inpainting. RecTransformer is designed to effectively detect various types of surface defects despite using only a small number of defect samples. RecTransformer simplifies the defect detection problem to a patch-level image completion problem. Without using convolution, the given block image is processed by the transformer model to generate a defect-free reconstructed image. Herein, global semantic information is established, and an attention mechanism is built in the patch sequence, and the spatial information of the patches is determined by position encoding to complete the global image reconstruction process. With a limited number of defect samples as training data, the RecTransformer algorithm accurately reconstructs defects. It achieves an area under the receiver operating characteristic curve score of 97.6% for pixel-level segmentation on the testing dataset. Experiments conducted on a universal surface defect dataset demonstrate the effectiveness of the RecTransformer algorithm. RecTransformer can be adapted to detect various types of surface defects, including Mura in display devices, with only a small number of defect samples.

show abstract

Section: Deep Learning- and Transformer-related Researchmentioning

confidence: 99%

Section: Deep Learning-and Transformer-related Researchmentioning

confidence: 99%

Transformer-based visual inspection algorithm for surface defects

Zhu,

Chang,

Liu

et al. 2023

Opt. Eng.

View full text Add to dashboard Cite

show abstract

“…In addition, neural networks are also widely used in the field of 3D measurement of moving objects. [29][30][31] Qian et al 32 used geometric constraints and phase unwrapping supported by deep learning to obtain object phases from a single fringe and eliminate motion artifacts. Yu et al 33 used deep learning phase retrieval techniques for dynamic 3D measurement.…”

Section: Introductionmentioning

confidence: 99%

“…used CNN to analyze the 3D profiles of objects from a single fringe pattern. In addition, neural networks are also widely used in the field of 3D measurement of moving objects 29 – 31 Qian et al 32 .…”

Section: Introductionmentioning

confidence: 99%

Method for measuring the three-dimensional profile of moving objects based on neural network position transformation

Wu,

Di,

Guo

et al. 2024

Opt. Eng.

View full text Add to dashboard Cite

To eliminate motion-induced errors, a multi-step fitting position transformation method is proposed for measuring the three-dimensional (3D) profile of an object. The method employs a neural network to perform position transformation on the objects in the captured motion phase-shifting fringe projection images, aligning them to the same position, and then uses the phase-shifting method to obtain the 3D profile of the objects. Theoretically, the rigid body motion transformation matrix is used to deduce the change of points on the objects in the fringe pattern under different motion conditions, and the 3D profile measurement results of randomly moving objects are simulated and analyzed. Experimentally, a neural network model was trained by using 5000 sets of data obtained from experiments to eliminate motioninduced errors in various motion states. Simulation and experimental results show that this method significantly reduces the motion-induced errors that occur when measuring moving objects using the traditional phase-shifting method. It has broad application prospects in the field of real-time measurement of moving objects' 3D profiles.

show abstract

“…Zhu et al [26] have proposed a hybrid model 'Hformer' based on Cross Attention Transformer (CAT) and a type of CNN called High Resolution Network (HRNet) for phase unwrapping in fringe projection. In this model, input phase is fed to HRNet followed by Encoder Decoder network with CAT as building blocks.…”

Section: Introductionmentioning

confidence: 99%

Transformer based deep learning hybrid architecture for phase unwrapping

Bujagouni,

Pradhan

2024

Phys. Scr.

View full text Add to dashboard Cite

A deep learning Hybrid architecture for phase unwrapping has been proposed. The hybrid architecture is based on integration of Convolutional Neural Networks (CNN) with Vision Transformer. The performance of Hybrid architecture/network in phase unwrapping is compared against CNN based standard UNET network. Structural Similarity Index (SSIM) and Root Mean Square Error (RMSE) have been used as performance metrics to assess the performance of these deep learning networks for phase unwrapping. To train and test the networks, dataset with high mean Entropy has been generated using Gaussian filtering of random noise in Fourier plane. The Hybrid architecture is tested on test dataset and is found to have superior performance metrics against the UNET network. Their performance is also tested in noisy environment with various noise levels and Hybrid architecture demonstrated better anti-noise capability than UNET network. Hybrid architecture was successfully validated in real world scenario using experimental data from custom built Digital Holographic Microscope. With the advent of newer architectures and hardware, Deep learning networks can further improve the performance in solving inverse problems.

show abstract

Hformer: Hybrid convolutional neural network transformer network for fringe order prediction in phase unwrapping of fringe projection

Cited by 7 publications

References 32 publications

Transformer-based visual inspection algorithm for surface defects

Transformer-based visual inspection algorithm for surface defects

Method for measuring the three-dimensional profile of moving objects based on neural network position transformation

Transformer based deep learning hybrid architecture for phase unwrapping

Contact Info

Product

Resources

About