Phase unwrapping method and device based on deep learning, equipment and medium
Through the deep learning network U-Net framework establishing nonlinear mapping of wrapping phase and absolute phase in high noise environments, it solves the difficulty of precisely unwrapped by traditional phase dewrapping algorithms in the case of noise and phase discontinuity, and achieves high-quality Phase unwrap effect.
Patent Information
- Application Number
- CN202410956911.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-17
- Publication Date
- 2024-11-08
Smart Images

Figure CN118918023A_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present invention relate to the field of phase unwrapping technology, and in particular to a phase unwrapping method and apparatus, device and medium based on deep learning. Background Art
[0002] Phase unwrapping is an important signal processing problem in many applications, and its purpose is to recover the original phase from the wrapped phase. Traditional phase unwrapping algorithms are not robust to noise and cannot perform accurate phase unwrapping when the noise is high and the phase is discontinuous. Summary of the invention
[0003] The embodiments of the present invention provide a phase unwrapping method and apparatus, device and medium based on deep learning, which uses a deep learning network U-Net framework to perform phase unwrapping. By establishing a nonlinear mapping between the wrapped phase and the absolute phase, in a high-noise environment and in the case of phase discontinuity, a trained deep learning network model can be used to predict a high-quality absolute phase map in just one step, thereby completing accurate phase unwrapping.
[0004] In a first aspect, an embodiment of the present invention provides a phase unwrapping method based on deep learning, comprising:
[0005] Get the parcel phase map;
[0006] The wrapped phase image is input into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a dilated spatial convolution pooling pyramid module, and the encoder and the decoder include an Inception module.
[0007] In some embodiments, the training method of the deep learning network model includes:
[0008] Acquire a data set, the data set comprising a noise-wrapped phase image and a true absolute phase image;
[0009] Inputting the noise winding phase map into a deep learning network model, and outputting an absolute phase map corresponding to the noise winding phase map;
[0010] A loss function is calculated based on the true absolute phase image and the absolute phase image corresponding to the noise winding phase image;
[0011] The loss function performs parameter optimization to obtain a trained deep learning network model.
[0012] In some embodiments, acquiring the data set includes:
[0013] The data set is generated by random matrix amplification.
[0014] In some embodiments, the method of generating the data set by random matrix amplification includes:
[0015] Randomly generate a first matrix within a preset range;
[0016] The first matrix is enlarged to 128*128 by bicubic interpolation to obtain a true absolute phase map;
[0017] Normalizing the true absolute phase image to obtain a first absolute phase;
[0018] Adding Gaussian noise to the first absolute phase to obtain a second absolute phase;
[0019] Wrap the second absolute phase to [-π,π] to obtain the noise wrapped phase diagram;
[0020] The data set is generated based on the noise-wrapped phase map and the true absolute phase map.
[0021] In some embodiments, the encoder includes four downsampling operations, each downsampling operation includes an Inception module, the Inception module uses multiple different branches to process a feature map in parallel, and uses a residual structure to splice the different branches; the 1x1 convolution in the encoder is used to reduce the number of channels, and a batch normalization layer is used for regularization after each layer of convolution; the encoder also uses a jump connection to link the feature maps in the encoding process and the decoding process during the encoding process.
[0022] In some embodiments, the bottleneck layer of the deep learning network model uses the atrous spatial convolutional pooling pyramid module to perform multi-scale extraction of deep features.
[0023] In some embodiments, the decoder includes four upsampling operations. The feature map is input into the Inception module after upsampling, and then after 3x3 convolution and ReLU activation, an absolute phase map is output. In the last layer of the decoding process, the absolute phase maps output by each layer of convolution are spliced, and finally the final target absolute phase map is output through 1x1 convolution.
[0024] In a second aspect, an embodiment of the present invention further provides a phase unwrapping device based on deep learning, the device comprising:
[0025] An acquisition module, used for acquiring a parcel phase image;
[0026] A processing module is used to input the wrapped phase image into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a dilated spatial convolution pooling pyramid module, and the encoder and the decoder include an Incept ion module.
[0027] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the deep learning-based phase unwrapping method as described in the first aspect is implemented.
[0028] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the deep learning-based phase unwrapping method as described in the first aspect.
[0029] According to the phase unwrapping method and apparatus based on deep learning, the device and medium provided by the embodiment of the present invention, wherein the phase unwrapping method based on deep learning includes: obtaining a wrapped phase image; inputting the wrapped phase image into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a void space convolution pooling pyramid module, and the encoder and the decoder include an Inception module. Based on this, the embodiment of the present invention uses a deep learning network U-Net framework for phase unwrapping, and by establishing a nonlinear mapping between the wrapped phase and the absolute phase, in a high noise environment and in the case of phase discontinuity, the trained deep learning network model can predict a high-quality absolute phase image in just one step, thereby completing accurate phase unwrapping. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1A is a flow chart of a phase unwrapping method based on deep learning provided by an embodiment of the present invention;
[0031] Figure 1Bis a flowchart of a training method for a deep learning network model provided by an embodiment of the present invention;
[0032] Figure 2 This is a diagram showing a portion of a training set provided by an embodiment of the present invention;
[0033] Figure 3 is a structural diagram of a deep learning network model provided by an embodiment of the present invention;
[0034] Figure 4 is a schematic diagram of the structure of an Incept ion module provided by an embodiment of the present invention;
[0035] Figure 5 is a schematic diagram of the structure of a dilated spatial convolutional pooling pyramid module provided by an embodiment of the present invention;
[0036] Figure 6 is a comparative schematic diagram of a noise-free phase wrapping situation provided by an embodiment of the present invention;
[0037] Figure 7 It is a comparative schematic diagram of a noise phase wrapping case with a standard deviation of 0.1 provided by an embodiment of the present invention;
[0038] Figure 8 is a comparative schematic diagram of a noise phase wrapping case with a standard deviation of 0.2 provided by an embodiment of the present invention;
[0039] Fig. 9 is a comparative schematic diagram of a discontinuous phase wrapping situation provided by an embodiment of the present invention;
[0040] Fig.10 is a comparative schematic diagram of a real phase wrapping situation provided by an embodiment of the present invention;
[0041] Fig.11 is a schematic diagram of a phase unwrapping device based on deep learning provided by an embodiment of the present invention;
[0042] Fig.12 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0043] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0044] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification and claims and the following drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.
[0045] In the embodiments of the present invention, the words "further", "exemplarily" or "optionally" are used to indicate examples, illustrations or descriptions, and should not be interpreted as being more preferred or more advantageous than other embodiments or designs. The use of the words "further", "exemplarily" or "optionally" is intended to present related concepts in a specific way.
[0046] First, some terms involved in the present invention are analyzed:
[0047] Phase Unwrapping: Due to the periodic nature of the phase, when the phase of the signal changes by more than 2π (i.e. 360 degrees), a phase jump will occur, which will lead to loss or error of phase information. Phase unwrapping is a technique to solve this problem. It mathematically maps the phase value back to its original range, thereby restoring the continuity of the phase.
[0048] U-Net: is a deep convolutional neural network architecture named for its unique U-shaped design. It consists of two parts: a contracting encoding path (downsampling) and a symmetrical expansion path (upsampling). In the encoding path, the network gradually reduces the spatial dimension of the image while increasing the number of feature channels, which helps to extract more abstract features. In the decoding path, the network gradually restores the spatial dimension of the image while reducing the number of feature channels, which helps to restore the details of the image. A key feature of U-Net is the skip connections, which directly connect the outputs of different layers in the encoder to the corresponding layers of the decoder. This design allows the network to retain more positional information and details during upsampling, which is critical for accurate image reconstruction.
[0049] In order to more conveniently describe the working principle of the embodiment of the present invention later, an introduction to relevant technical scenarios is first given below.
[0050] The core of traditional phase unwrapping methods, such as path tracing and minimum norm methods, is to avoid the negative impact of invalid points in the wrapped phase map as much as possible. The quality of the unwrapping stage of the path tracing method depends on the optimal path selection, but it is difficult to determine the appropriate integration path when there is high noise or complex structure in the wrapped phase. The minimum norm method has noise immunity, but its large computational complexity makes this method unsuitable for real-time measurement. Therefore, the traditional phase unwrapping method is only applicable to normal situations where the noise is not too serious and the phase is continuous and not mixed. It should be pointed out that if a pixel is misclassified, the error of the pixel will be at least an integer multiple of 2π, which is unacceptable in practical applications.
[0051] Based on this, the present invention provides a phase unwrapping method and device, equipment and medium based on deep learning. Among them, the phase unwrapping method based on deep learning includes: obtaining a wrapped phase image; inputting the wrapped phase image into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a void space convolution pooling pyramid module, and the encoder and the decoder include an Inception module. Based on this, the embodiment of the present invention uses a deep learning network U-Net framework for phase unwrapping, and by establishing a nonlinear mapping between the wrapped phase and the absolute phase, in a high noise environment and in the case of phase discontinuity, the trained deep learning network model can predict a high-quality absolute phase image in just one step, thereby completing accurate phase unwrapping.
[0052] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.
[0053] like Figure 1A As shown, Figure 1A It is a flowchart of a deep learning-based phase unwrapping method provided by an embodiment of the present invention. The deep learning-based phase unwrapping method may include but is not limited to steps S101 to S102.
[0054] Step S101: obtaining a package phase map;
[0055] Step S102: Input the wrapped phase image into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a dilated spatial convolution pooling pyramid module, and the encoder and the decoder include an Inception module.
[0056] Obtain the wrapped phase image, input the wrapped phase image into the pre-trained deep learning network model for phase unwrapping, and the pre-trained deep learning network model can output the absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase. The deep learning network model adopts the U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a dilated spatial convolution pooling pyramid module, forming a U-shaped structure. This improved U-Net network structure can effectively extract and fuse multi-scale features, which helps to retain the edge and detail information in the image. The encoder and decoder contain the Inception module, which is a module that can increase the depth and width of the model and effectively reduce the number of parameters.
[0057] It can be understood that the present invention utilizes the deep learning network U-Net framework for phase unwrapping. By establishing a nonlinear mapping between the wrapped phase and the absolute phase, in a high noise environment and in the case of phase discontinuity, the trained deep learning network model can be used to predict a high-quality absolute phase map in just one step, thereby completing accurate phase unwrapping.
[0058] It can be understood that the present invention applies deep learning technology to perform phase unwrapping, which not only improves the performance of the deep network learning model in terms of anti-noise interference, generalization ability and robustness, but also can successfully achieve phase unwrapping with higher accuracy.
[0059] like Figure 1B As shown, the training method of the deep learning network model includes but is not limited to steps S201 to S204.
[0060] Step S201: Acquire a data set, where the data set includes a noise-wrapped phase map and a true absolute phase map;
[0061] Step S202: inputting the noise winding phase image into the deep learning network model, and outputting an absolute phase image corresponding to the noise winding phase image;
[0062] Step S203: Calculate a loss function according to the true absolute phase image and the absolute phase image corresponding to the noise entanglement phase image;
[0063] Step S204: Optimize the parameters of the loss function to obtain a trained deep learning network model.
[0064] It is understandable that, for data set generation, unlike the traditional phase unwrapping algorithm, the phase unwrapping algorithm based on deep learning needs to be trained with a large amount of data, but it is difficult to collect a large amount of data that wraps the phase and matches the unwrapped phase in actual measurement. Therefore, the present invention generates a data set by random matrix amplification. Some training sets in the data set are as follows Figure 2 shown.
[0065] First, a matrix A ranging from 2*2 to 8*8 is randomly generated, and then it is enlarged to 128*128 by bicubic interpolation to obtain the absolute phase. The absolute phase Φ(x,y) obtained by bicubic interpolation of matrix A is expressed as:
[0066]
[0067] Where A(i,j) is the value of the matrix A at point (i,j), and the weight w is calculated based on the weight function of the bicubic interpolation. The absolute phase Φ(x,y) is then normalized to between 10 and 40, where 50% of the values are distributed between 10 and 30, 20% of the values are distributed between 30 and 35, and 30% of the values are distributed between 35 and 40. After normalization, Gaussian noise with a standard deviation of 0.01 to 0.2 is added to the absolute phase. The absolute phase Φ(x,y) is wrapped in [-π,π] to obtain the wrapped phase Ψ(x,y) expressed as:
[0068]
[0069] It can be understood that the U-Net algorithm is characterized by a downsampling path called an encoder and an upsampling path called a decoder, which are connected to each other through skip connections. This structure enables U-Net to capture multi-scale information at the same time and perform fine image reconstruction in the decoder. At this time, in network training, the objective function to be solved can be expressed as:
[0070]
[0071] like Figure 3As shown in FIG. 1 , the deep learning network model proposed in the present invention uses an encoder and a decoder as the basic structure. The Inception module is used as the basic module in the deep learning network model. Figure 4 The deep learning network model contains a five-level Inception encoder, a five-level Inception decoder, a deep supervision feature fusion module, and an atrous spatial convolutional pyramid pooling (ASPP) module, forming a U-shaped structure. The ASPP module structure is shown in Figure 5 As shown in Figure 2. This improved U-Net network structure can effectively extract and fuse multi-scale features, which helps to preserve the edge and detail information in the image.
[0072] The encoder in the deep learning network model contains four downsampling operations, each of which contains an Inception module, which is a module that can increase the depth and width of the model and effectively reduce the number of parameters. It uses multiple different branches to process an image in parallel, and finally uses a residual structure to splice these different branches to ensure that the network still retains better performance in the deep layer. Among them, the role of the 1x1 convolution kernel is to reduce the number of channels and reduce the number of parameters of the model. After each layer of convolution, a batch normalization layer is used for regularization to avoid possible gradient disappearance and gradient explosion problems. In addition, skip connections are used in the encoding process to link the feature maps of the encoding process and the decoding process. Atrous spatial convolutional pyramid pooling (ASPP) module is used at the bottleneck layer of the network to extract deep features at multiple scales. The decoder contains four upsampling steps. After upsampling, the feature map is input into the Inception module, and then an absolute phase image is output after 3x3 convolution and ReLU activation. In the last layer of the decoding process, the absolute phase images output by each layer are spliced, and finally the final absolute phase image is output through 1x1 convolution.
[0073] In order to evaluate the performance of the deep learning network model at multiple scales, the deep learning network model is tested as follows, and all test results are comprehensively evaluated using RMSE, PSNR, and SSIM indicators.
[0074] exist Figure 6In the figure, (a) represents the noise-free wrapped phase image, i.e., the input of the deep learning network model, (b) represents the absolute phase image corresponding to (a), (c) and (d) represent the output and error image of the deep learning network model, respectively, and (e) and (f) represent the absolute phase image and error image of the output of U-Net, respectively.
[0075] exist Figure 7 In the figure, (a) represents the wrapped phase image with a noise standard deviation of 0.1, that is, the input of the deep learning network model, (b) represents the absolute phase image corresponding to (a), (c) and (d) represent the output and error image of the deep learning network model, respectively, and (e) and (f) represent the absolute phase image and error image of the output of U-Net, respectively.
[0076] exist Figure 8 In the figure, (a) represents the wrapped phase image with a noise standard deviation of 0.2, that is, the input of the deep learning network model, (b) represents the absolute phase image corresponding to (a), (c) and (d) represent the output and error image of the deep learning network model, respectively, and (e) and (f) represent the absolute phase image and error image of the output of U-Net, respectively.
[0077] exist Fig. 9 In the figure, (a) represents the discontinuous wrapped phase map, i.e., the input of the deep learning network model, (b) represents the absolute phase map corresponding to (a), (c) and (d) represent the output and error map of the deep learning network model, respectively, and (e) and (f) represent the absolute phase map and error map of the output of U-Net, respectively.
[0078] exist Fig.10 In the figure, (a) represents the true wrapped phase map, i.e., the input of the deep learning network model, (b) represents the absolute phase map corresponding to (a), (c) and (d) represent the output and error map of the deep learning network model, respectively, and (e) and (f) represent the absolute phase map and error map of the output of U-Net, respectively.
[0079] It should be noted that, in general, when the unwrapping error is less than 5%, it can be considered that the unwrapping is correct. Therefore, the error map is generated by calculating the relative error between the model output and the true value (gt). In the error map, any pixel point with a relative error exceeding 5% is marked as an error point.
[0080] From the data in Tables 1, 2, 3, 4, and 5 below, it can be seen that the deep learning network model proposed in the present invention is superior to U-Net in all evaluation index comparisons. The experimental results fully demonstrate the advantages of the deep learning network model proposed in the present invention in terms of model noise resistance, model generalization, and model robustness.
[0081] Table 1 Average PSNR (dB), RMSE, and SSIM of the noise-free phase wrapping model
[0082]
[0083] Table 2 Average PSNR (dB), RMSE, and SSIM of the noise phase wrapping model with a standard deviation of 0.1
[0084]
[0085] Table 3 Average PSNR (dB), RMSE, and SSIM of the noise phase wrapping model with a standard deviation of 0.2
[0086]
[0087] Table 4 Average PSNR (dB), RMSE, and SSIM of the discontinuous phase wrapping model
[0088]
[0089] Table 5 Average PSNR (dB), RMSE, and SSIM of the real phase wrapping model unwrapping
[0090]
[0091] Based on this, the present invention proposes a phase unwrapping method based on deep learning, which can solve the absolute phase image in one step by using a trained model. Compared with other types of deep learning methods, this deep learning network model shows superiority in terms of robustness in dealing with noise and phase unwrapping performance when dealing with complex situations such as phase discontinuity. This means that even in scenes with large noise or drastic phase changes, the deep learning network model can maintain high unwrapping accuracy and stability.
[0092] It should be noted that, in view of the limitations of current change detection technology, the present invention innovatively introduces a non-changing area information suppression module, which takes into account the influence of background context information, and the cross-attention mechanism therein pays more attention to the correlation between various regions. The present invention has significant innovation in the field of change detection technology and provides a more reliable and efficient solution for modern building monitoring and illegal construction inspection.
[0093] In addition, if Fig.11 As shown, one embodiment of the present invention further discloses a phase unwrapping device based on deep learning, the device comprising:
[0094] An acquisition module 110, used to acquire a package phase image;
[0095] The processing module 120 is used to input the wrapped phase image into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a void space convolution pooling pyramid module, and the encoder and the decoder include an Incept ion module.
[0096] The deep learning-based phase unwrapping device of an embodiment of the present invention is used to execute the deep learning-based phase unwrapping method in the above embodiment. Its specific processing process is the same as the deep learning-based phase unwrapping method in the above embodiment, and will not be repeated here.
[0097] In addition, if Fig.12 As shown, an embodiment of the present invention also discloses an electronic device, comprising: at least one processor 210; at least one memory 220, for storing at least one program; when the at least one program is executed by the at least one processor 210, a deep learning-based phase unwrapping method as in any of the previous embodiments is implemented.
[0098] In addition, an embodiment of the present invention further discloses a computer-readable storage medium, which stores computer-executable instructions for executing the deep learning-based phase unwrapping method as in any of the previous embodiments.
[0099] The system architecture and application scenarios described in the embodiments of the present invention are intended to more clearly illustrate the technical solutions of the embodiments of the present invention, and do not constitute a limitation on the technical solutions provided by the embodiments of the present invention. Those skilled in the art will appreciate that with the evolution of the system architecture and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present invention are equally applicable to similar technical problems.
[0100] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.
[0101] In hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.
[0102] The terms "component", "module", "system", etc. used in this specification are used to represent computer-related entities, hardware, firmware, a combination of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, or a computer. By way of illustration, both applications running on a computing device and a computing device can be components. One or more components may reside in a process or an execution thread, and a component may be located on a computer or distributed between two or more computers. In addition, these components may be executed from various computer-readable media having various data structures stored thereon. Components may communicate, for example, through local or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, or a network, such as the Internet interacting with other systems through signals).
Claims
1. A phase unwrapping method based on deep learning, comprising: Get the parcel phase map; The wrapped phase image is input into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a dilated spatial convolution pooling pyramid module, and the encoder and the decoder include an Inception module.
2. The method according to claim 1, characterized in that The training method of the deep learning network model includes: Acquire a data set, the data set comprising a noise-wrapped phase image and a true absolute phase image; Inputting the noise winding phase map into a deep learning network model, and outputting an absolute phase map corresponding to the noise winding phase map; A loss function is calculated based on the true absolute phase image and the absolute phase image corresponding to the noise winding phase image; The loss function performs parameter optimization to obtain a trained deep learning network model.
3. The method according to claim 2, characterized in that The acquiring of the data set comprises: The data set is generated by random matrix amplification.
4. The method according to claim 3, characterized in that The method of generating the data set by random matrix amplification includes: Randomly generate a first matrix within a preset range; The first matrix is enlarged to 128*128 by bicubic interpolation to obtain a true absolute phase map; Normalizing the true absolute phase image to obtain a first absolute phase; Adding Gaussian noise to the first absolute phase to obtain a second absolute phase; Wrap the second absolute phase to [-π,π] to obtain the noise wrapped phase diagram; The data set is generated based on the noise-wrapped phase map and the true absolute phase map.
5. The method according to claim 1, characterized in that The encoder includes four downsampling operations, each of which includes an Inception module. The Inception module uses multiple different branches to process a feature map in parallel, and uses a residual structure to splice the different branches. The 1x1 convolution in the encoder is used to reduce the dimension of the number of channels, and a batch normalization layer is used for regularization after each convolution layer. The encoder also uses a jump connection to link the feature maps in the encoding process and the decoding process during the encoding process.
6. The method according to claim 1, characterized in that The bottleneck layer of the deep learning network model uses the atrous spatial convolutional pooling pyramid module to perform multi-scale extraction of deep-level features.
7. The method according to claim 1, characterized in that The decoder includes four upsampling operations. The feature map is input into the Inception module after upsampling, and then after 3x3 convolution and ReLU activation, an absolute phase map is output. In the last layer of the decoding process, the absolute phase maps output by each layer of convolution are spliced, and finally the final target absolute phase map is output through 1x1 convolution.
8. A phase unwrapping device based on deep learning, characterized in that: The device comprises: An acquisition module, used for acquiring a parcel phase image; A processing module is used to input the wrapped phase image into a pre-trained deep learning network model for phase unwrapping, so that the deep network learning model outputs an absolute phase image corresponding to the wrapped phase image according to the nonlinear mapping relationship between the wrapped phase and the absolute phase, wherein the deep learning network model adopts a U-Net framework, and the deep learning network model includes an encoder, a decoder, a deep supervised feature fusion module and a dilated spatial convolution pooling pyramid module, and the encoder and the decoder include an Inception module.
9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the deep learning-based phase unwrapping method according to any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the deep learning-based phase unwrapping method as described in any one of claims 1 to 7.
Citation Information
Patent Citations
Two-dimensional phase unwrapping method based on deep learning semantic segmentation network
CN111043953A
Phase unwrapping method and device based on composite neural network
CN111797678A
High-noise environment phase diagram unwrapping method and system based on deep learning
CN117475172A
Systems and methods for phase unwrapping for dense MRI using deep learning
US20210267455A1
Cited By
Phase unwrapping method based on multi-scale pooling and expansion pyramid network
CN119294464A
Deep learning-based phase unwrapping method and apparatus, device, and medium
WO2026017065A1
