Single-frame digital stripe end-to-end phase analysis method based on deep learning

By constructing the InncepC-Unet network model, absolute phase prediction of a single frame fringe image was achieved, solving the problems of complex measurement process and error propagation in existing technologies, and improving the efficiency and accuracy of 3D measurement.

CN119006381AActive Publication Date: 2024-11-22XIAN UNIV OF POSTS & TELECOMM
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411021672.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-22
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

Existing technologies for 3D measurement using single-frame stripe images are complex, inefficient, and suffer from the problem of error forward propagation.

Method used

An end-to-end phase analysis method for single-frame digital stripes based on deep learning is adopted. By constructing the InncepC-Unet network model, the absolute phase prediction of single-frame stripe images is achieved by combining the encoding and decoding networks with skip connections.

Benefits of technology

It enables fast and accurate absolute phase calculation of single-frame stripe images, improves the efficiency and accuracy of 3D measurement, and promotes the development of 3D measurement towards intelligence.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006381A_ABST
    Figure CN119006381A_ABST
Patent Text Reader

Abstract

The invention provides a single-frame digital fringe end-to-end phase analysis method based on deep learning, and the method comprises the following steps: firstly building a fringe projection system based on digital twinning, and constructing a deformed fringe-absolute phase data set; then constructing an IncepC-Unet network model, dividing the deformed stripe-absolute phase data set into a test set and a training set according to a proportion, and training and testing the IncepC-Unet network model; and finally, performing absolute phase prediction on the single-frame deformed fringe image of the measured object by using the trained IncepC-Unet network model. According to the end-to-end phase analysis method for the single-frame digital fringes, a traditional cascade analysis process of absolute phase unwrapping is completely abandoned, end-to-end absolute phase calculation of the single-frame digital fringes is achieved, the efficiency and precision of phase analysis are improved, and three-dimensional measurement is promoted to develop towards the intelligent direction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of optical three-dimensional measurement, and in particular relates to a single-frame digital fringe end-to-end phase analysis method based on deep learning. Background Art

[0002] At present, three-dimensional measurement technology plays an important role in various fields. It is widely used in aviation manufacturing, smart medical care, intelligent identification, artificial intelligence and other fields. Fringe projection three-dimensional imaging has become the mainstream optical method in the current three-dimensional measurement technology due to its advantages of non-contact, high precision, full-field measurement, and high efficiency of point cloud reconstruction. However, traditional fringe projection three-dimensional imaging obtains absolute phase information by projecting multiple sets of grating fringes and combining phase shift method / time phase unwrapping method, which requires the process of fringe image acquisition, phase demodulation, and phase analysis, which inevitably prolongs the data acquisition time and has a great impact on the efficiency of three-dimensional information acquisition. The spatial domain method only needs one frame of fringe and its corresponding principal value phase to restore the absolute phase. The early algorithms for extracting the principal value phase include Fourier transform method (FT), windowed Fourier transform method (WFT), etc.; the methods for obtaining absolute phase mainly include: branch cutting method, quality map-guided method, unweighted FFT method, etc. Although there are many mature methods of phase resolution based on spatial domain method, their phase resolution needs to be affected by adjacent pixels, resulting in gradual error transmission. When facing targets with complex structures, discontinuities or isolated surface characteristics, it is easy to cause large phase resolution errors. The absolute phase is often accompanied by serious phenomena of "drawing" and "packet loss", and the reconstruction accuracy is still far inferior to the multiple fringe projection solution method.

[0003] Therefore, how to use a single-frame fringe image and a faster method to quickly, accurately and unambiguously obtain the absolute phase of the target is an urgent problem to be solved in the current field of fringe projection three-dimensional imaging.

[0004] In recent years, with the development of artificial intelligence technology, phase analysis methods based on deep learning technology have made significant breakthroughs. In terms of phase demodulation, a single-frame fringe analysis method based on deep learning uses two convolutional neural networks (CNN1 and CNN2) to learn and generate the sine part (M) and cosine part (D) required for demodulation phase from the collected original fringe image, and then substitutes the sine and cosine results into the inverse tangent function to calculate the wrapped phase. On this basis, multi-path CNN+ResNet networks are successively used to predict the M / D of the wrapped phase. Compared with traditional FT and WFT, this type of method can more accurately extract phase information using only one fringe image, especially for the surface of objects with rich details. In terms of phase unwrapping, there are mainly indirect unwrapping methods for predicting fringe order and direct unwrapping methods for predicting absolute phase. The former learns from one or more frames of wrapped phase, first infers or segments the fringe order image required for absolute phase, and then recovers the absolute phase information in combination with the wrapped phase; the latter learns from one frame of wrapped phase and directly generates absolute phase information.

[0005] Currently, in phase resolution, most deep learning methods only focus on a certain stage of phase resolution, phase demodulation or phase unfolding, or fuse the two using a multi-stage network. Such a measurement process is relatively complex and inefficient, and there is also the problem of forward error propagation. Summary of the invention

[0006] The purpose of the present invention is to solve the problems of complex measurement process, low efficiency and forward error propagation in the prior art of obtaining absolute phase using single-frame fringe images and phase resolution methods, and to provide a single-frame digital fringe end-to-end phase resolution method based on deep learning.

[0007] To achieve the above purpose, the technical solution provided by the present invention is:

[0008] A single-frame digital fringe end-to-end phase analysis method based on deep learning, comprising the following steps:

[0009] Step 1: Build a fringe projection system based on digital twins, use the phase shift method to collect the deformed fringe image, use the dual-frequency dephasing algorithm to calculate the absolute phase of the deformed fringe image, and construct a deformed fringe-absolute phase dataset;

[0010] Step 2: Construct the IncepC-Unet network model, including the encoding network, decoding network and skip connection;

[0011] Step 3: Divide the deformed fringe-absolute phase data set in step 1 into a training set and a test set according to a certain ratio, train the IncepC-Unet network model based on the training set, test the trained IncepC-Unet network model based on the test set, and optimize and adjust the relevant parameter settings of the IncepC-Unet network model based on the training and test results;

[0012] Step 4: Use the IncepC-Unet network model designed in step 3 to perform absolute phase prediction on the single-frame deformed fringe image of the object under test.

[0013] Furthermore, in step 1, the fringe projection system includes a camera, a projector, and an object to be measured, wherein the camera and the projector are placed horizontally, and the object to be measured is perpendicular to the projector.

[0014] Furthermore, in step 1, the step of constructing the deformed fringe-absolute phase data set includes:

[0015] Step 1.1: Use high-frequency and low-frequency sinusoidal stripes to render the object to be tested respectively, and then use a camera to collect and save the deformed stripe image, wherein the first frame of the high-frequency deformed stripe image is a single-frame deformed stripe image;

[0016] Step 1.2: For the high-frequency deformed fringe images other than the single-frame deformed fringe images saved in step 1.1, the absolute phases are calculated using a dual-frequency phase resolution algorithm and saved;

[0017] Step 1.3: Replace the object to be tested and repeat the operations of steps 1.1 to 1.2. Several single-frame deformed fringe images and absolute phases will eventually form a deformed fringe-absolute phase data set.

[0018] Furthermore, in step 2, the encoding network consists of five encoding layers, the first four encoding layers each consist of an Inception residual block, a coordinate attention block, and a 2×2 maximum pooling, and the fifth encoding layer consists of an Inception residual block;

[0019] The decoding network consists of four decoding layers, each of which consists of a transposed convolution, a concatenation block, and an Inception residual block;

[0020] Skip connections are used to fuse features between the encoding layer and the decoding layer.

[0021] Furthermore, the Inception residual block consists of a parallel layer structure of branch 0, branch 1, branch 2, branch 3 and branch 4; branch 0 consists of a convolution block filled with 0 and a batch normalization layer, branches 1 to 4 are all composed of convolution blocks filled with 1 and a batch normalization layer, and the four C / 4 channel outputs of branches 1 to 4 are connected to one C channel output, the dimension is specified to be 1, and the C channel output is added to the branch 0 output.

[0022] Furthermore, the Inception residual block in the encoding network enhances the multi-scale feature extraction capability of the IncepC-Unet network model through a multi-branch pathway; the coordinate attention block is used to extract the spatial dependency of the stripe image at different positions in the local feature map, and the coordinate attention block uses one-dimensional pooling in the X and Y directions respectively instead of the global two-dimensional pooling, so that the IncepC-Unet network model can learn the spatial distribution characteristics of the image more accurately and improve the IncepC-Unet network model's attention to the image in the effective target area; the 2×2 maximum pooling can halve the width and height of the extracted feature map, reduce the amount of calculation, and improve the training speed of the IncepC-Unet network model.

[0023] Furthermore, in step 3, the training steps of the IncepC-Unet network model include:

[0024] Step 3.1: Normalize the deformed fringe-absolute phase dataset generated in step 1, and scale the input and output of the IncepC-Unet network model. The input of the IncepC-Unet network model is a single-frame deformed fringe image.

[0025] Step 3.2: Set the relevant parameters for IncepC-Unet network model training;

[0026] Step 3.3: Input the predicted phase and the true phase into the loss function of the IncepC-Unet network model, and train the IncepC-Unet network model according to the result of the loss function until the result of the loss function converges.

[0027] Furthermore, in step 3, the deformed fringe-absolute phase dataset is randomly divided into a training set and a test set with a ratio of 9:1, wherein the training set is not visible in the test set.

[0028] Furthermore, in step 4, the step of using the IncepC-Unet network model to perform absolute phase prediction on a single-frame deformed fringe image includes:

[0029] Step 4.1: Input a single-frame deformed stripe image of the object to be tested into the IncepC-Unet network model;

[0030] Step 4.2: The input single-frame deformed stripe image is first subjected to local feature extraction by the encoding layer of the encoding network to obtain a feature map, and then the decoding layer of the decoding network performs layer-by-layer transposed convolution to obtain an intermediate feature map, and the intermediate feature map is spliced ​​with the feature map of the same size in the encoding network to obtain a spliced ​​feature map;

[0031] Step 4.3: Input the concatenated feature map obtained in step 4.2 into the Inception residual block of the decoding network, and obtain a feature map with a channel number of 1 through branch 0, which is the absolute phase corresponding to the deformed stripe image of the object under test.

[0032] The advantages of the present invention are:

[0033] The single-frame digital fringe end-to-end phase analysis method designed in the present invention completely abandons the traditional cascade analysis process of absolute phase expansion, realizes the end-to-end absolute phase calculation of single-frame digital fringe, improves the efficiency and accuracy of phase analysis, and promotes the development of three-dimensional measurement towards intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The features and advantages of the present invention will become more readily understood through the following description with reference to the accompanying drawings, which are not drawn to scale and in which some features are exaggerated or minimized to show details of particular components:

[0035] Figure 1 is a flow chart of a single-frame digital fringe end-to-end phase analysis method based on deep learning in an embodiment of the present invention;

[0036] Figure 2 A fringe projection system and a measurement scene in an embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of the structure of the IncepC-Unet network model in an embodiment of the present invention;

[0038] Figure 4 It is a schematic diagram of the structure of the Inception residual block in the IncepC-Unet network model in an embodiment of the present invention;

[0039] Figure 5 Schematic diagram of the coordinate attention block structure in the IncepC-Unet network model in an embodiment of the present invention;

[0040] Figure 6 3 is a binary phase error comparison diagram of the phase resolution method designed in the embodiment of the present invention and the other three phase resolution methods. DETAILED DESCRIPTION

[0041] The present invention will be described in detail below with reference to the accompanying drawings by means of exemplary embodiments of the present invention. It should be noted that the following detailed description of the present invention is only for the purpose of illustration, and is not intended to limit the present invention.

[0042] A single-frame digital fringe end-to-end phase resolution method based on deep learning, such as Figure 1 As shown, the following steps are included:

[0043] Step 1: Build a fringe projection system based on digital twins, use the phase shift method to collect the deformed fringe image, use the dual-frequency dephasing algorithm to calculate the absolute phase of the deformed fringe image, and construct a deformed fringe-absolute phase dataset.

[0044] Rich data sets provide support for the training of deep learning networks. The production of actual data sets is usually a time-consuming task, requiring a lot of time and resources to collect and annotate real data. To this end, the present invention selects 3D Max, a 3D modeling software, to build a virtual fringe projection system based on the actual measurement system. Figure 2 As shown, it includes a camera, a projector and an object to be measured. Specifically, the camera is a video camera, and the projector is composed of a spotlight. The camera and the projector are placed horizontally, and the object to be measured is perpendicular to the projector.

[0045] The actual measurement system parameters referenced by the fringe projection system are shown in Table 1.

[0046] Table 1 List of parameter settings of actual system projector and industrial camera

[0047]

[0048] The steps to construct the deformed fringe-absolute phase dataset include:

[0049] Step 1.1: Place the object to be measured at the measuring position, and use the projector to render the six-step phase-shifted fringe image sequences with fringe frequencies of 19 and 24 onto the surface of the object to be measured. Then, use the camera to collect twelve phase-shifted deformed fringe images and save them in *.bmp file format, where the first frame of the high-frequency deformed fringe image is a single-frame deformed fringe image. Convert the *.bmp file format of the single-frame deformed fringe image into a *.mat format file.

[0050] The nth deformed fringe image captured by the camera can be expressed as:

[0051]

[0052] Where n = 0, 1, 2, 3, ..., N-1, I n (x, y) represents the intensity of the deformed fringe image, A(x, y) represents the background intensity, B(x, y) represents the modulation degree, represents the main phase value, and 2πn / N represents the phase shift of the nth sinusoidal fringe.

[0053] Step 1.2: For the high-frequency deformed fringe images other than the single-frame deformed fringe images saved in step 1.1, use the dual-frequency phase analysis algorithm to calculate their absolute phases and save them in the *.mat file format.

[0054] The wrapped phase is calculated by the least square method using the above six-step phase-shifted fringe image, as shown in the following formula:

[0055]

[0056] The dual-frequency phase separation algorithm is performed on the wrapped phases of the above two different frequencies, as shown in the following formula:

[0057]

[0058] Among them, the two fringe frequencies are represented by f l (low frequency) and f h (high frequency), the main value phases corresponding to the high-frequency and low-frequency fringe images are and K h and K l Represent the number of high-frequency and low-frequency grating fringes k respectively l (x,y) and k h The set of integers (x,y), Ω represents k l (x,y)·f h -k h (x,y)·f l In the interval (-f h , f l ), round[x] means rounding to the nearest integer to x, and s represents the set count.

[0059] Step 1.3: Replace the object to be measured and repeat the operations of steps 1.1 to 1.2. Several single-frame deformed fringe images and absolute phases will eventually form a deformed fringe-absolute phase dataset in *.mat format.

[0060] In the fringe projection system, 150 3D models of any category are selected as the objects to be tested, and they are rotated, scaled, and combined to generate 1120 pairs of deformed fringe-absolute phase data sets.

[0061] Step 2: Construct the IncepC-Unet network model, such as Figure 3 As shown, it includes an encoding network, a decoding network, and a skip connection.

[0062] The encoding network consists of five encoding layers, the first four encoding layers each consist of an Inception residual block, a coordinate attention block, and a 2×2 maximum pooling with a stride of 2 for sampling. The fifth encoding layer consists of an Inception residual block.

[0063] The encoding network passes through five layers of Inception residual blocks and four layers of coordinate attention blocks in sequence. The Inception residual blocks enhance the multi-scale feature extraction capability of the IncepC-Unet network model through multi-branch paths. The coordinate attention block is used to extract the spatial dependence of the stripe image at different positions in the local feature map. The 2×2 maximum pooling can halve the width and height of the extracted feature map, aiming to reduce the amount of calculation and improve the training speed of the IncepC-Unet network model. The traditional global pooling operation aggregates the entire feature map. For example, the average pooling averages the eigenvalues ​​in the feature map, and the maximum pooling only retains the largest eigenvalue in the feature map. This operation loses some local feature information in the spatial range. In the phase prediction task, not only the overall information of the image is concerned, but more importantly, the continuity and monotonicity of the local phase. Therefore, in order to retain the position information in the feature map, the coordinate attention block designed by the present invention replaces the traditional global two-dimensional pooling with one-dimensional pooling in the X direction and the Y direction, which helps the IncepC-Unet network model to learn the spatial distribution characteristics of the image more accurately and improve the IncepC-Unet network model's attention to the image in the effective target area.

[0064] The decoding network consists of four decoding layers, each of which consists of a 3×3 transposed convolution with a stride of 2, a concatenation block, and an Inception residual block. The 3×3 transposed convolution is used to upsample the input feature map of the previous layer and restore the width and height of the feature map from bottom to top.

[0065] like Figure 4 As shown in the figure, the Inception residual block consists of a parallel layer structure consisting of branch 0, branch 1, branch 2, branch 3, and branch 4; branch 0 consists of a 1×1 convolution block filled with 0 and a batch normalization layer, branches 1 to 4 are all composed of a 3×3 convolution block filled with 1 and a batch normalization layer, and the four C / 4 channel outputs of branches 1 to 4 are connected to a C channel output, the dimension is specified to be 1, and the C channel output is added to the output of branch 0. The Inception residual block extracts feature information of different scales from four convolutional layers of different depths and performs feature fusion through splicing operations, which helps to improve the multi-scale feature extraction of the IncepC-Unet network model and improve the expression ability of the IncepC-Unet network model.

[0066] The jump connection after the Inception residual block of the encoding layer is used to concatenate the features of the encoding layer and the decoding layer. The concatenated feature map is input into the Inception residual block again to realize the processing of high-dimensional feature maps.

[0067] Step 3: Divide the deformed fringe-absolute phase data set in step 1 into a training set and a test set according to a certain ratio, train the IncepC-Unet network model based on the training set, and test the trained IncepC-Unet network model based on the test set. Based on the training and test results, optimize and adjust the relevant parameter settings of the IncepC-Unet network model.

[0068] The deformed fringe-absolute phase dataset is combined with 145 pairs of real data and randomly divided into a training set and a test set with a ratio of 9:1, where the training set is not visible in the test set.

[0069] The IncepC-Unet network model is trained using the above training set. The virtual system operating environment is set as shown in Table 2.

[0070] Table 2 Virtual system operating environment

[0071]

[0072] Specifically, the training steps of the IncepC-Unet network model include:

[0073] Step 3.1: Normalize the deformed fringe-absolute phase dataset generated in step 1, resize the input and output of the IncepC-Unet network model, use a single-frame deformed fringe image as input, and use the absolute phase as a label;

[0074] Step 3.2: Set the optimizer and parameters such as learning rate, batch size, and number of iterations for IncepC-Unet network model training, as shown in Table 3;

[0075] Table 3 List of configuration training options and parameters

[0076] Epoch batch size learning_rate weight decay momentum 300 8 0.01 0.0001 0.9

[0077] Step 3.3: Input the network predicted phase and the real phase into the loss function of the IncepC-Unet network model, and train the IncepC-Unet network model according to the result of the loss function until the result of the loss function converges. If the MAE function is used as the loss function, when the result of the loss function converges, the training cycle is 250 rounds.

[0078] The IncepC-Unet network model trained above was tested, and a comparative experiment was conducted on a group of David sculptures in the test set. The error of the phase resolution method designed by the present invention was compared with that of the traditional Unet, U2-net, Res-Unet and IncepC-Unet. Figure 6 As shown, Figure 6 (a) is the deformation fringes, Figure 6 (b)~ Figure 6 (e) shows the phase prediction results of the four algorithms of traditional Unet, U2-net, Res-Unet and the present invention, respectively. Figure 6 (f) is the real phase distribution, Figure 6 (g)~ Figure 6 (j) shows the binary phase error distribution between the traditional Unet, U2-net, Res-Unet and the four algorithms of the present invention and the true phase distribution. The more black pixels, the greater the error; conversely, the smaller the error. It can be seen from the figure that the phase analysis method designed by the present invention has small error and superior performance.

[0079] Step 4: Use the IncepC-Unet network model designed in step 3 to perform absolute phase prediction on the single-frame deformed fringe image of the object under test.

[0080] Specifically, the steps of using the IncepC-Unet network model to predict the absolute phase of a single-frame deformed fringe image include:

[0081] Step 4.1: Input a single-frame deformed stripe image of the object to be tested into the IncepC-Unet network model;

[0082] Step 4.2: The input single-frame deformed stripe image first passes through the five encoding layers of the encoding network to extract local features to obtain a feature map. Specifically: Figure 5 As shown in the figure, the coordinate attention block first performs one-dimensional pooling on the X and Y directions of the extracted feature map, and then splices them. After that, the 1×1 convolution block and nonlinear activation function of the Inception residual block are used to obtain the transition feature map. Finally, the feature map divided into the X and Y directions is multiplied with the original input single-frame deformed stripe image to obtain the final output feature map of the coordinate attention block. The decoding layer of the decoding network then performs layer-by-layer transposed convolution to obtain the intermediate feature map. Through the jump connection after the Inception residual block of the encoding layer, the intermediate feature map is spliced ​​with the feature map of the same size in the encoding network to obtain the spliced ​​feature map.

[0083] Step 4.3: Input the concatenated feature map obtained in step 4.2 into the Inception residual block of the decoding network, and obtain a feature map with a channel number of 1 through branch 0 output, which is the absolute phase corresponding to the deformed stripe image of the object under test.

[0084] Finally, it should be noted that the features mentioned and / or shown in the above description of the exemplary embodiments of the present invention may be combined in the same or similar manner into one or more other embodiments, combined with the features in other embodiments or substituted for the corresponding features in other implementations. The technical solutions obtained by these combinations or substitutions shall also be deemed to be included in the protection scope of the present invention.

Claims

1. A single-frame digital fringe end-to-end phase analysis method based on deep learning, characterized in that: The following steps are involved: Step 1: Build a fringe projection system based on digital twins, use the phase shift method to collect the deformed fringe image, use the dual-frequency dephasing algorithm to calculate the absolute phase of the deformed fringe image, and construct a deformed fringe-absolute phase dataset; Step 2: Construct the IncepC-Unet network model, including the encoding network, decoding network and skip connection; Step 3: Divide the deformed fringe-absolute phase data set in step 1 into a training set and a test set according to a certain ratio, train the IncepC-Unet network model based on the training set, and test the trained IncepC-Unet network model based on the test set; optimize and adjust the relevant parameter settings of the IncepC-Unet network model based on the training and test results; Step 4: Use the IncepC-Unet network model designed in step 3 to perform absolute phase prediction on the single-frame deformed fringe image of the object under test.

2. The single-frame digital fringe end-to-end phase analysis method according to claim 1, characterized in that: In step 1, the fringe projection system includes a camera, a projector, and an object to be measured, wherein the camera and the projector are placed horizontally, and the object to be measured is placed perpendicular to the projector.

3. The single-frame digital fringe end-to-end phase analysis method according to claim 2, characterized in that: In step 1, the steps of constructing the deformed fringe-absolute phase data set include: Step 1.1: Use high-frequency and low-frequency sinusoidal stripes to render the object to be tested respectively, and then use a camera to collect and save the deformed stripe image, wherein the first frame of the high-frequency deformed stripe image is a single-frame deformed stripe image; Step 1.2: For the high-frequency deformed fringe images other than the single-frame deformed fringe images saved in step 1.1, the absolute phases are calculated using a dual-frequency phase resolution algorithm and saved; Step 1.3: Replace the object to be tested and repeat the operations of steps 1.1 to 1.

2. Several single-frame deformed fringe images and absolute phases will eventually form a deformed fringe-absolute phase data set.

4. The single-frame digital fringe end-to-end phase analysis method according to claim 1, characterized in that: In step 2, the encoding network consists of five encoding layers. The first four encoding layers are composed of an Inception residual block, a coordinate attention block, and a 2×2 maximum pooling. The fifth encoding layer is composed of an Inception residual block. The decoding network consists of four decoding layers, each of which consists of a transposed convolution, a concatenation block, and an Inception residual block; Skip connections are used to fuse features between the encoding layer and the decoding layer.

5. The single-frame digital fringe end-to-end phase analysis method according to claim 4, characterized in that: The Inception residual block consists of a parallel layer structure of branch 0, branch 1, branch 2, branch 3 and branch 4; branch 0 consists of a convolution block filled with 0 and a batch normalization layer, branches 1 to 4 are all composed of convolution blocks filled with 1 and a batch normalization layer, and the four C / 4 channel outputs of branches 1 to 4 are connected to one C channel output, the dimension is specified to be 1, and the C channel output is added to the branch 0 output.

6. The single-frame digital fringe end-to-end phase analysis method according to claim 4, characterized in that: The Inception residual block in the encoding network enhances the multi-scale feature extraction capability of the IncepC-Unet network model through a multi-branch pathway; the coordinate attention block is used to extract the spatial dependency of the stripe image at different positions in the local feature map, and the coordinate attention block uses one-dimensional pooling in the X and Y directions respectively instead of global two-dimensional pooling, so that the IncepC-Unet network model can learn the spatial distribution characteristics of the image more accurately and improve the IncepC-Unet network model's attention to the image in the effective target area; the 2×2 maximum pooling can halve the width and height of the extracted feature map, reduce the amount of calculation, and improve the training speed of the IncepC-Unet network model.

7. The single-frame digital fringe end-to-end phase analysis method according to claim 3, characterized in that: In step 3, the training steps of the IncepC-Unet network model include: Step 3.1: Normalize the deformed fringe-absolute phase dataset generated in step 1, and scale the input and output of the IncepC-Unet network model. The input of the IncepC-Unet network model is a single-frame deformed fringe image. Step 3.2: Set the relevant parameters for IncepC-Unet network model training; Step 3.3: Input the predicted phase and the true phase into the loss function of the IncepC-Unet network model, and train the IncepC-Unet network model according to the result of the loss function until the result of the loss function converges.

8. The single-frame digital fringe end-to-end phase analysis method according to claim 3, characterized in that: In step 3, the deformed fringe-absolute phase dataset is randomly divided into a training set and a test set with a ratio of 9:1, where the training set is not visible in the test set.

9. The single-frame digital fringe end-to-end phase analysis method according to claim 5, characterized in that: In step 4, the steps of using the IncepC-Unet network model to perform absolute phase prediction on a single-frame deformed fringe image include: Step 4.1: Input a single-frame deformed stripe image of the object to be tested into the IncepC-Unet network model; Step 4.2: The input single-frame deformed stripe image first passes through the encoding layer of the encoding network to extract local features to obtain a feature map, and then the decoding layer of the decoding network performs layer-by-layer transposed convolution to obtain an intermediate feature map. The intermediate feature map is spliced ​​with the feature map of the same size in the encoding network to obtain a spliced ​​feature map; Step 4.3: Input the concatenated feature map obtained in step 4.2 into the Inception residual block of the decoding network, and obtain a feature map with a channel number of 1 through branch 0, which is the absolute phase corresponding to the deformed stripe image of the object under test.

Citation Information

Patent Citations

  • Single-frame double-frequency multiplexing fringe projection three-dimensional surface type measurement method based on deep learning

    CN114777677A

  • Three-dimensional reconstruction method and system for dynamic structured light, and computing device

    CN117132704A

  • Improved single-mode self-dephasing fringe projection three-dimensional measurement method based on deep learning

    CN117739861A

  • A deep learning-based temporal phase unwrapping method for fringe projection profilometry

    US20210356258A1