Quantitative phase recovery system and method under uniform illumination condition

Abstract

The invention discloses a quantitative phase recovery system and method under a uniform illumination condition, and belongs to the field of quantitative phase recovery, and the system mainly comprises a laser, an attenuation sheet, a polaroid, a light beam uniform light device, a spatial light modulator, a beam splitter and a camera. The laser emitted by the laser device is subjected to power attenuation by the attenuation sheet and then polarized by the polarizer, and the polarized light is converted into a flat-topped light spot from a Gaussian light spot after passing through the light beam dodging device. The uniform light beams are modulated by the spatial light modulator and then reflected by the beam splitter, and light intensity distribution images of the uniform light beams are collected by the camera. A random two-dimensional speckle phase screen is generated by using a Monte Carlo method, and then a data set is acquired to train a neural network. And finally, inputting the light intensity distribution diagram into the trained neural network, outputting a corresponding phase diagram, and completing phase recovery. The system established by the invention solves the problem of phase distortion caused by intensity distortion and form degradation of far-field light spots. Meanwhile, the data set is high in universality and has the advantages of being simple in structure and real-time.

Description

Technical Field

[0001] The present invention relates to the field of quantitative phase retrieval, and particularly to a quantitative phase retrieval system and method under uniform illumination conditions. Background Art

[0002] The propagation characteristics of light include two aspects: amplitude and phase. Photoelectric detectors can directly capture the amplitude information of an object, but the phase information that characterizes physical properties such as the surface shape and refractive index of the object cannot be directly captured. Therefore, it is particularly important to retrieve the phase information from the light intensity information of the object. Currently, the existing quantitative phase retrieval techniques are mainly divided into two categories: interference-based and non-interference-based. Interference-based phase retrieval techniques complete phase retrieval through the interference effect of light waves, including digital holographic microscopy and coherent diffraction imaging techniques. Non-interference-based phase retrieval techniques complete phase retrieval through physical models or mathematical calculations, including the transport of intensity equation method, iterative phase retrieval algorithms, and wavefront sensing techniques.

[0003] In recent years, deep learning techniques have shown great potential in the field of quantitative phase retrieval. Especially when dealing with complex optical models, they have a faster processing speed compared to traditional iterative methods. Traditional quantitative phase retrieval techniques rely on simulation to generate light intensity distribution maps. Due to the insufficient authenticity of dataset acquisition and the pseudo-random effect of simulation, the retrieval effect is not good when actually training neural networks. The dataset used in traditional deep learning-based quantitative phase retrieval techniques for training neural networks should be consistent with the final input and output images of the system. For example, when the input and output are portrait information, the training set is also portrait information. When this network is used for other forms of image information, its retrieval effect is worse compared to portrait information and may not meet the requirements. Therefore, the generalization ability of traditional neural networks is not strong. At the same time, the scale of its dataset is usually large and its versatility is also poor. Chinese Patent CN119364185A, "A Phase Retrieval Method and System Combining the Transport of Intensity Equation and Angular Spectrum Iteration", completes phase retrieval after zero-padding processing and iteration of the collected images. Chinese Patent CN110455747A, "A Halo-Free White Light Phase Imaging Method and System Based on Deep Learning", trains a neural network with 7000 standard polystyrene microsphere phase images for collecting living biological cells, validates the network with 2000 sample images, and tests the network with 1000 sample images to obtain the best model and complete phase retrieval and imaging. The above methods have a large number of training sets, a long iteration time, and it is difficult to construct a general-purpose dataset. Therefore, how to construct a training dataset with strong versatility and then form a quantitative phase retrieval method is an urgent technical problem to be solved. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a quantitative phase recovery system and method under uniform illumination conditions, aiming to solve the problems of insufficient authenticity of the training data set used in the existing quantitative phase recovery methods, large number of training sets, long iteration time and poor generality.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a quantitative phase recovery system under uniform illumination conditions, including a laser, an attenuation sheet, a polarizer, a beam homogenizer, a beam splitter and a spatial light modulator sequentially placed at the same horizontal height, and a camera is arranged in the reflection direction of the split light of the beam splitter.

[0006] A quantitative phase recovery method under uniform illumination conditions includes the following steps:

[0007] Step 1: Generate a random two-dimensional speckle phase screen using the Monte Carlo method;

[0008] Step 2: Input the two-dimensional speckle phase screen into the spatial light modulator;

[0009] Step 3: Turn on the laser, the laser emits laser light, the light after power attenuation by the attenuation sheet is polarized by the polarizer, the polarized light after polarization is transformed from a Gaussian spot into a flat-top spot by the beam homogenizer, so that the beam becomes a uniform beam, the uniform beam enters the beam splitter, a part of the split light enters the spatial light modulator, the light carrying phase information after modulation is reflected back to the beam splitter, and is collected by the camera after being reflected by the beam splitter to obtain an intensity distribution map;

[0010] Step 4: Use the corresponding two-dimensional speckle phase screen and intensity distribution map as a data set to perform supervised training on the AttR2U-Net network to obtain the optimal AttR2U-Net network;

[0011] Step 5: Input any phase image such as a portrait into the trained optimal AttR2U-Net network and output its corresponding phase distribution map;

[0012] Step 6: Complete quantitative phase recovery.

[0013] A further improvement of the technical solution of the present invention lies in: the principle of using the Monte Carlo method to generate a random two-dimensional speckle phase screen in Step 1 is:

[0014] First, filter in the frequency domain spectrum, and calculate the Fourier transform pair F(k m ,k n ) corresponding to the rough phase screen through the power spectral density S(k m ,k n ), and finally perform inverse Fourier transform to obtain the two-dimensional speckle phase distribution f(x m ,y n), i.e.,

[0015]

[0016] wherein, the length of the random two-dimensional speckle phase screen in the x direction is L x , and the length in the y direction is L y , the interval between adjacent points is Δx and Δy, M and N are the number of discrete points, and based on the above formula, a random two-dimensional speckle phase screen can be obtained.

[0017] A further improvement of the technical solution of the present invention lies in: Step 4 is specifically: taking the intensity distribution maps of the signal light before and after propagation and the random two-dimensional speckle phase screen of n images as a data set, and then performing supervised training on the AttR2U-Net network. By continuously adjusting the weights w and biases b of the network, after iterating K times until the loss function converges, the system reaches the optimal effect, thereby obtaining the optimal neural network as the phase recovery system.

[0018] A further improvement of the technical solution of the present invention lies in: the number of data sets n ≥ 3000; the number of iteration times K ≥ 200.

[0019] Due to the adoption of the above technical solution, the technical progress achieved by the present invention is: By building an imaging system, a real data set can be obtained. At the same time, since the light emitted by the laser is usually a Gaussian spot, with high central light intensity and low edge intensity, using this non-uniform light illumination will cause the intensity distribution of the far-field spot to be uneven, and the intensity distortion of the Gaussian beam far-field spot will cause the training set to be distorted and the phase cannot be well restored. By introducing a light homogenizer, a flat-top spot can be obtained, and the flat-top spot has a uniform light intensity distribution, providing more stable conditions for phase recovery. Since the collected intensity distribution map is based on the condition of a uniform light beam, compared with the intensity distribution map generated by traditional simulation, the data set of this system is more real and has a better recovery effect when training the neural network. By using the phase screen generated based on the Monte Carlo method, the complexity is relatively high. After using the intensity distribution map-phase screen data set generated by this system and method to train the neural network, compared with the network trained by traditional methods, this network can process various information types and can achieve that when the input image is an arbitrary intensity distribution map of a person or an animal, etc., the output is its corresponding phase map. This method has a strong generalization ability for the training set, and there is a strong mapping relationship between the input intensity map and the output phase map, and it can realize the real-time recovery of image phase information. It overcomes the problems of uneven intensity distribution of the far-field spot caused by Gaussian light illumination and the difficulty in constructing a general data set, enhances the generalization between the input and output of the system, and speeds up the network iteration operation speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings;

[0021] Figure 1 It is a schematic structural diagram of the quantitative phase retrieval system of the present invention;

[0022] Figure 2 It is a schematic flowchart of the quantitative phase retrieval method of the present invention;

[0023] Among them, 1. Laser, 2. Attenuator, 3. Polarizer, 4. Beam homogenizer, 5. Beam splitter, 6. Spatial light modulator, 7. Camera. Detailed implementation manners

[0024] The following further elaborates on the present invention in conjunction with embodiments:

[0025] As Figure 1 shown, it is a schematic structural diagram of a quantitative phase retrieval system under uniform illumination conditions, including a laser 1, an attenuator 2, a polarizer 3, a beam homogenizer 4, a beam splitter 5, and a spatial light modulator 6 sequentially placed at the same horizontal height. A camera 7 is arranged in the light reflection direction of the beam splitter 5.

[0026] Based on the above imaging system, a quantitative phase retrieval method under uniform illumination conditions is implemented. The process is as Figure 2 shown, and the specific steps are as follows:

[0027] Step 1: Generate a random two-dimensional speckle phase screen using the Monte Carlo method; specifically, first perform filtering in the frequency domain spectrum, and calculate the Fourier transform pair F(k m , k n ) corresponding to the rough phase screen through the power spectral density S(k m , k n ), and finally perform an inverse Fourier transform to obtain the two-dimensional speckle phase distribution f(x m , y n ), that is

[0028]

[0029] Among them, the length of the random two-dimensional speckle phase screen in the x direction is L x , the length in the y direction is L y , the interval between adjacent points is Δx and Δy, and M and N are the number of discrete points. Based on the above formula, a random two-dimensional speckle phase screen can be obtained.

[0030] Step 2: Input the two-dimensional speckle phase screen into the spatial light modulator 5;

[0031] Step 3: Turn on the laser 1. The laser emits laser light. The light after power attenuation by the attenuation sheet 2 is polarized by the polarizer 3. The polarized light after polarization is transformed from a Gaussian spot into a flat-top spot by the beam homogenizer 4, so that the light beam becomes a uniform light beam. The uniform light beam is incident on the beam splitter 5, and multiple light beams are split. A part of the split light beams is directly emitted, and a part of the split light beams is incident on the spatial light modulator 6. The light carrying phase information after modulation is then reflected back to the beam splitter 5. The beam splitter is a reflective beam splitter. After being reflected by the beam splitter 5, it is collected by the camera 7 to obtain an intensity distribution map;

[0032] Step 4: Use the corresponding intensity distribution map and the random two-dimensional speckle phase screen as a data set to perform supervised training on the neural network. In this embodiment, the AttR2U-Net network is taken as an example. Through a large number of trainings of the AttR2U-Net network, the optimal AttR2U-Net network is finally obtained; the data set n≥3000; the number of iterations K≥200. Take n intensity distribution maps before and after the propagation of the signal light and the random two-dimensional speckle phase screen as a data set, and then perform supervised training on the AttR2U-Net network. By continuously adjusting the weights w and biases b of the network, after iterating K times until the loss function converges, the system reaches the optimal effect, so as to obtain the optimal quantitative phase retrieval system. Specifically in this embodiment: Take 2500 groups of intensity distribution maps and random two-dimensional speckle phase screens as a data set, and take the real phase screen as the detection standard for the phase reconstruction effect of the system. Then perform supervised training on the neural network, and continuously adjust the weights w and biases b of the AttR2U-Net network. After the system iterates 200 times, use 500 groups of test sets to test the system. It is found through testing that the phase difference between the reconstructed phase screen output by the system network model and the real phase screen is not obvious, and the structural similarity reaches 0.9880. Therefore, it is considered that the quantitative phase retrieval system reaches the optimal effect at this time.

[0033] Step 5: Input any phase picture such as a portrait into the trained optimal AttR2U-Net network, and output its corresponding phase distribution map;

[0034] Step 6: Complete quantitative phase retrieval.

[0035] The embodiments described above are only used to describe the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention should fall within the protection scope determined by the claims of the present invention.

Claims

1. A quantitative phase retrieval system under uniform illumination conditions, characterized in that: It includes a laser (1), an attenuation sheet (2), a polarizer (3), a beam homogenizer (4), a beam splitter (5) and a spatial light modulator (6) placed in sequence at the same horizontal height. A camera (7) is arranged in the reflection direction of the beam splitter (5).

2. A quantitative phase retrieval method under uniform illumination conditions, implemented based on the imaging system described in claim 1, characterized in that: It includes the following steps: Step 1: Generate a random two-dimensional speckle phase screen using the Monte Carlo method; Step 2: Input the two-dimensional speckle phase screen into the spatial light modulator (5); Step 3: Turn on the laser (1). The laser (1) emits laser light. The light after power attenuation by the attenuation sheet (2) is polarized by the polarizer (3). The polarized light after polarization is transformed from a Gaussian spot into a flat-top spot by the beam homogenizer (4), so that the beam becomes a uniform beam. The uniform beam enters the beam splitter (5). A part of the split light enters the spatial light modulator (6). The light carrying phase information after modulation is reflected back to the beam splitter (5) and collected by the camera (7) after being reflected by the beam splitter (5) to obtain an intensity distribution map; Step 4: Use the corresponding two-dimensional speckle phase screen and intensity distribution map as a data set to perform supervised training on the AttR2U-Net network to obtain the optimal AttR2U-Net network; Step 5: Input any phase image such as a portrait into the trained optimal AttR2U-Net network and output its corresponding phase distribution map; Step 6: Complete quantitative phase recovery.

3. A quantitative phase retrieval method under uniform illumination conditions according to claim 2, characterized in that: The principle of generating a random two-dimensional speckle phase screen using the Monte Carlo method in Step 1 is: First, filter in the frequency domain spectrum, and calculate the Fourier transform pair F(k m, k n ) corresponding to the rough phase screen through the power spectral density S(k m ,k n ). Finally, perform the inverse Fourier transform to obtain the two-dimensional speckle phase distribution f(x m ,y n ), that is Among them, the length of the random two-dimensional speckle phase screen in the x direction is L x , and the length in the y direction is L y , the interval between adjacent points is Δx and Δy, and M and N are the number of discrete points. Based on the above formula, a random two-dimensional speckle phase screen can be obtained.

4. A quantitative phase retrieval method under uniform illumination conditions according to claim 2, characterized in that: Step 4 is specifically: Use the intensity distribution maps of the signal light after propagation of n images and the random two-dimensional speckle phase screen as a data set. Then perform supervised training on the AttR2U-Net network. By continuously adjusting the weights w and biases b of the network, after iterating K times until the loss function converges to make the system reach the optimal effect, thus obtaining the optimal neural network as the phase recovery system.

5. A quantitative phase retrieval method under uniform illumination conditions according to claim 4, characterized in that: The data set n ≥ 3000; the number of iterations K ≥ 200.

Images