Complex amplitude reconstruction method based on frequency domain self-supervised complex valued neural network
The process of hologram data through the frequency domain self-supervised complex value neural network solves the problem of information capture and labeling data dependence in phase reconstruction, and achieves high-precision and efficient phase reconstruction, lowering the application threshold.
Patent Information
- Application Number
- CN202411218262.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-02
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-09-02
AI Technical Summary
The prior art is difficult to effectively capture and retain phase and amplitude information in holograms in phase reconstruction, especially when processing high noise and complex structure data, and relying on labeled data to train in low efficiency and high cost.
The complex amplitude reconstruction method based on the frequency domain self-supervised complex value neural network is adopted, and the hologram data is converted to the frequency domain through Fourier transform. The complex value convolution neural network is used to directly process the complex data, combining soft constraints and self-supervised learning. Reduce redundant information interference, highlight phase characteristics, and independently mine training labels from unlabeled data.
It improves the accuracy and efficiency of phase reconstruction, simplifies operation steps, reduces dependence on labeled data, reduces training costs, and significantly improves the accuracy and flexibility of phase reconstruction.
Smart Images

Figure CN118917362A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a hologram phase reconstruction method, and in particular to a complex amplitude reconstruction method based on a frequency domain self-supervised complex-valued neural network. Background Art
[0002] Coaxial holography is a holographic imaging technology based on the principle of coherent light interference. It mainly divides the coherent light source into two parts: one part is used as reference light to reach the detector directly, and the other part is used as object light to reach the detector after passing through the object to be measured. On the detector, the two beams of light meet and produce interference patterns, which record the amplitude and phase information of the object. This technology can simultaneously record the amplitude and phase data of the object in the same hologram, thereby achieving high-precision three-dimensional reconstruction. Compared with traditional imaging technology, modern holographic phase reconstruction combines high-resolution digital detectors and advanced image processing algorithms, which can achieve high-resolution three-dimensional imaging without the need for labeling or staining, and record and reconstruct the detailed physical characteristics of the object. In addition, the system design of holographic phase reconstruction is relatively simple and compact, without the need for complex optical path adjustment, and has good stability and easy operation. This makes the technology particularly suitable for real-time observation and non-destructive testing of dynamic processes, and has been widely used in many fields such as biomedical imaging, materials science and industrial testing. Through the development of modern technology, coaxial holograms have shown great potential in providing richer physical information and achieving higher-precision imaging, and have become an important tool in scientific research and application.
[0003] The application of complex-valued neural network (CVNN) in phase reconstruction achieves the goal of extracting and reconstructing high-quality phase information from holograms by combining optical imaging technology and machine learning. The core advantage of CVNN lies in its powerful nonlinear mapping capability and data-driven approach, which can overcome the limitations of traditional iterative algorithms in processing complex structures and high-noise data. Compared with real-valued neural network (RVNN), CVNN is based on complex expression and is more suitable for processing complex amplitude data containing amplitude and phase information, thus performing better in phase reconstruction tasks. Due to its ability to characterize the intrinsic relationship between the wavefront amplitude and phase components, CVNN can cope with the complexity of phase reconstruction and has strong robustness to noise, further improving the accuracy of reconstruction. Through these characteristics, CVNN demonstrates good generalization ability, adapts to different data sets and application scenarios, and provides richer and more efficient solutions in the field of phase reconstruction. Summary of the invention
[0004] The present invention provides a complex amplitude reconstruction method based on frequency domain self-supervised complex valued neural network. The phase reconstruction method starts from the frequency domain and explores the characteristic relationship between the hologram and the original phase in the frequency domain, and the complex valued neural network can directly process complex data. The method can accurately retain and reconstruct phase information, has higher representation capabilities, and can capture phase and amplitude changes in data. Thereby improving the reconstruction accuracy.
[0005] The technical solution of the present invention for simultaneously reconstructing the complex amplitude is:
[0006] A complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network includes the following steps:
[0007] (S1), based on the coaxial holography theory, a camera is selected to shoot multiple sets of pure phase coaxial holograms as network input, i.e., hologram A;
[0008] (S2), read the collected hologram A and perform Fourier transform to obtain spectrum B, which is used as input of the complex-valued convolutional neural network, and outputs the original object field spectrum C after the network;
[0009] (S3) After the inverse Fourier transform of the object field spectrum C, soft constraints are applied to its amplitude and it is forward propagated to the distance z. The soft constraints are to limit the amplitude value at each point to not be greater than the value specified by the absorption constraint. The range of this value is related to the object being recorded.
[0010] (S4), taking the intensity of the complex amplitude D of the diffraction field to obtain the reconstructed hologram E, and calculating its spectrum F;
[0011] (S5) Calculate the spatial domain loss term, that is, the similarity between hologram A and reconstructed hologram E, calculate the frequency domain loss term, that is, the similarity between spectrum B and spectrum F, assign the two weights and add them up as the loss function, back-propagate the gradient of the loss function value to update the parameters of the network model, and obtain the trained complex-valued neural network.
[0012] Preferably, in step (S1), the collected hologram is a pure phase hologram, and the amplitude of the original object wave is set to 1.
[0013] Preferably, in step (S1), holograms A are collected to produce a data set, the number of which is not less than 100.
[0014] Preferably, in step (S4), the forward propagation method includes an angular spectrum method, a Fresnel diffraction method, a convolution method, and the like.
[0015] Preferably, the complex-valued neural network has a complex-valued network structure with a U-shaped structure, including complex-valued convolution, complex-valued deconvolution, and complex-valued activation function, wherein the complex-valued convolution and deconvolution are respectively implemented by a convolution layer with a 3×3 convolution kernel and a deconvolution kernel with a 2×2 deconvolution kernel; the complex-valued activation function includes a sp lit type activation function and a fuly type activation function.
[0016] Preferably, the complex-valued neural network adopts a U-shaped structure of an encoder-decoder, and the depth of the encoder-decoder is more than 2 layers.
[0017] Preferably, the loss function of the complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network adopts a loss function for comparing image structure similarity, and the loss function is an SSIM loss function and an L1 norm.
[0018] Preferably, in step (S5), the reconstructed hologram E and spectrum F are also obtained by simulating the diffraction process of the collected hologram A after the network output, and the corresponding loss back propagation is calculated to guide the network training. The entire process of network training is a self-supervised learning process, and the method is self-supervised learning.
[0019] Compared with the prior art, the present invention has the following beneficial effects:
[0020] 1. The present invention introduces a frequency domain learning algorithm that integrates physical constraints, focusing on the accurate reconstruction of complex amplitude images. Starting from the frequency domain, this phase reconstruction method deeply explores the characteristic relationship between the hologram and the original phase in the frequency domain. By converting the hologram data into the frequency domain for analysis, this method can more effectively capture the frequency characteristics of the phase information and reveal the complex mapping relationship between the hologram and the phase. Taking advantage of the frequency domain, this method can reduce the interference of redundant information in the spatial domain, highlight the core characteristics of the phase, and thus improve the reconstruction accuracy. This method provides a new idea for phase reconstruction, which helps to better understand the relationship between the hologram and the original phase and achieve higher quality phase reconstruction.
[0021] 2. The present invention proposes a complex amplitude reconstruction method based on a frequency domain self-supervised complex-valued neural network, focusing on conveniently processing complex information in the frequency domain. This method directly operates the complex data of the hologram in the frequency domain, avoiding the complex process of traditional real-valued neural networks to decompose the complex numbers into real and imaginary parts for independent processing. Through this frequency domain processing method, the present invention not only simplifies the operation steps, but also fully retains the key phase and amplitude information in the hologram. By taking advantage of the frequency domain, the complex relationship between the hologram and the original phase can be captured and characterized more naturally, thereby significantly improving the accuracy and efficiency of phase reconstruction. By directly processing the complex information in the frequency domain, this method achieves accurate capture and efficient reconstruction of the phase, making the result closer to the true value, and providing a concise and powerful solution for the phase reconstruction task.
[0022] 3. The present invention proposes a complex amplitude reconstruction method based on a frequency-domain self-supervised complex-valued neural network, which can autonomously mine and generate effective training labels from large-scale unlabeled data, solving the problem of high-quality labeled data being difficult to obtain and costly in phase reconstruction tasks. Under the traditional supervised learning framework, model training is heavily dependent on accurately labeled data sets, while self-supervised learning is trained by mining the intrinsic structure and patterns in unlabeled data. The present invention greatly reduces the reliance on labeled data, which not only reduces the cost of data labeling, but also significantly improves training efficiency. This method lowers the threshold for phase reconstruction technology, making it easier to promote and apply, and shows greater flexibility and adaptability in actual operations. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 It is a flowchart of the complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network of the present invention.
[0024] Figure 2 It is an algorithm framework diagram of the complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network of the present invention. DETAILED DESCRIPTION
[0025] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.
[0026] See also Figure 1-Figure 2 The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network of the present invention comprises the following steps:
[0027] (S1), based on the coaxial holography theory, a camera is selected to shoot multiple sets of pure phase coaxial holograms as network input, i.e., hologram A;
[0028] (S2), read the collected hologram A and perform Fourier transform to obtain spectrum B, which is used as input of the complex-valued convolutional neural network, and outputs the original object field spectrum C after the network;
[0029] (S3) After the inverse Fourier transform of the object field spectrum C, soft constraints are applied to its amplitude and it is forward propagated to the distance z. The soft constraints are to limit the amplitude value at each point to not be greater than the value specified by the absorption constraint. The range of this value is related to the object being recorded;
[0030] (S4), taking the intensity of the complex amplitude D of the diffraction field to obtain the reconstructed hologram E, and calculating its spectrum F;
[0031] (S5) Calculate the spatial domain loss term, that is, the similarity between hologram A and reconstructed hologram E, calculate the frequency domain loss term, that is, the similarity between spectrum B and spectrum F, assign the two weights and add them up as the loss function, back-propagate the gradient of the loss function value to update the parameters of the network model, and obtain the trained complex-valued neural network.
[0032] See also Figure 1-Figure 2 In step (S4), the propagation method of the complex amplitude reconstruction method based on the frequency domain self-supervised complex-valued neural network includes angular spectrum method, Fresnel diffraction method, convolution method, etc. The angular spectrum method is used in this embodiment.
[0033] In step (S4), the value of z in this embodiment is 15 mm.
[0034] In step (S5), the optimization algorithm includes a structural similarity algorithm, an average error square sum algorithm, an error square sum algorithm, etc. In this embodiment, the structural similarity algorithm and the L1 norm are used.
[0035] The complex-valued neural network with an encoder-decoder structure adopted in the present invention has three layers and a maximum number of channels of 256 in this embodiment.
[0036] The above is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above content. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.
Claims
1. A complex amplitude reconstruction method based on a frequency domain self-supervised complex-valued neural network, comprising the following steps: (S1), based on the coaxial holography theory, a camera is selected to shoot multiple sets of pure phase coaxial holograms as network input, i.e., hologram A; (S2), read the collected hologram A and perform Fourier transform to obtain spectrum B, which is used as input of the complex-valued convolutional neural network, and outputs the original object field spectrum C after the network; (S3) After the inverse Fourier transform of the object field spectrum C, soft constraints are applied to its amplitude and it is forward propagated to the distance z. The soft constraints are to limit the amplitude value at each point to not be greater than the value specified by the absorption constraint. The range of this value is related to the object being recorded; (S4), taking the intensity of the complex amplitude D of the diffraction field to obtain the reconstructed hologram E, and calculating its spectrum F; (S5) Calculate the spatial domain loss term, that is, the similarity between hologram A and reconstructed hologram E, calculate the frequency domain loss term, that is, the similarity between spectrum B and spectrum F, assign the two weights and add them up as the loss function, back-propagate the gradient of the loss function value to update the parameters of the network model, and obtain the trained complex-valued neural network.
2. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 1, characterized in that: In step (S1), the collected hologram A is a pure phase hologram, and the amplitude of the original object wave is set to 1.
3. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 1, characterized in that: In step (S1), holograms A are collected to produce a data set, the number of which is no less than 100.
4. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 1, characterized in that: In step (S4), the diffraction forward propagation method includes angular spectrum method, Fresnel diffraction method, convolution method, etc.
5. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 1, characterized in that: The complex-valued neural network has a U-shaped complex-valued network structure, including complex-valued convolution, complex-valued deconvolution, and complex-valued activation function, wherein the complex-valued convolution and deconvolution are respectively implemented by a 3×3 convolution layer and a 2×2 deconvolution kernel; the complex-valued activation function includes a split activation function and a fully activation function.
6. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 5, characterized in that: The complex-valued neural network adopts a U-shaped structure of an encoder-decoder, and the depth of the encoder-decoder is more than 2 layers.
7. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 1, characterized in that: The complex amplitude reconstruction method of the frequency domain self-supervised complex-valued neural network adopts a loss function that can reflect the similarity of two images, and assigns appropriate weights to the spatial domain loss term and the frequency domain loss term. The loss function that reflects the similarity of the two images includes the L1 norm, the mean square error MSE function, and the structural similarity index SSIM function.
8. The complex amplitude reconstruction method based on frequency domain self-supervised complex-valued neural network according to claim 1, characterized in that: Self-supervised learning is adopted. In step (S4), the reconstructed hologram E and spectrum F are also obtained by simulating the diffraction process of the collected hologram A after passing through the network output, and the corresponding loss back propagation is calculated to guide the network training.
Citation Information
Patent Citations
Unsupervised deep learning holographic imaging method and device
CN114972924A
Convolutional neural network-based hologram reconstruction method and system
CN115690252A
Joint optimization image reconstruction method and system
CN116704070A
Image reconstruction method and system based on self-encoding neural network
CN117218149A
