A single-photon multi-wavelength imaging system and method based on a physical driving model

Abstract

The application provides a single-photon multi-wavelength imaging system and method based on a physical driving model, and belongs to the technical field of single-pixel multi-wavelength imaging; the application solves the problems of being easily affected by noise and poor imaging quality of existing single-photon multi-wavelength imaging schemes in extreme environments; the application comprises a multi-wavelength modulation illumination module, a spatial modulation coding module and a detection imaging module; the multi-wavelength modulation illumination module uses signal light with different modulation repetition frequencies to illuminate a color target object and scatter the color target object to the spatial modulation coding module; the spatial modulation coding module spatially encodes the color target object and transmits the color target object to the detection imaging module; the detection imaging module captures the modulation signal light and transmits the modulation signal light to a computer; a physical driving model program is prewritten in the computer; the physical driving model comprises a measurement value denoising module and a physical driving image reconstruction module; and finally, a high-quality color target object reconstruction image is obtained; the application is applied to photon imaging.

Description

TECHNICAL FIELD

[0001] The application provides a single-photon multi-wavelength imaging system and method based on a physical driving model, and belongs to the technical field of single-pixel multi-wavelength imaging. BACKGROUND

[0002] Image information plays a crucial role in scientific research, industrial applications, medical diagnosis, and national defense and military fields. Currently, the main ways to obtain high-quality image information include optical cameras, infrared detection, and single-pixel imaging. Among them, single-pixel imaging technology uses only one detector to obtain the heat or other information of the target through time coding or space coding technology. This imaging method has low cost and low power consumption, and is suitable for imaging target objects in extreme environments. Single-pixel multi-wavelength imaging technology uses three basic color channels of red, green, and blue to record images, each channel corresponds to a monochrome image, and finally a color target object reconstruction image is generated by combining the three channels.

[0003] Single-pixel multi-wavelength imaging systems are mainly divided into time division multiplexing systems and frequency division multiplexing systems. The time division multiplexing system modulates light of different wavelengths into different time periods to irradiate the target, and then detects the echo signals of each time period. However, this system has high requirements for the pulse width and time stability of the light source, and in actual application, images of different depths may interfere with each other, thereby limiting its application range. In comparison, the frequency division multiplexing system marks light signals of different wavelengths as different repetition frequencies through intensity modulation, and uses fast Fourier transform to extract these echo signals in the frequency domain. Traditional single-pixel multi-wavelength imaging systems use photodetectors for data acquisition, resulting in low system sensitivity and poor imaging effect in low-light environments. Single-photon multi-wavelength imaging systems use single-photon detectors to collect multi-wavelength information, with high sensitivity and excellent low-light detection capability. However, existing single-photon multi-wavelength imaging technology still faces two major challenges in practical application. First, there is a lot of fluctuation noise. Due to the uncertainty and statistical fluctuations of photons, single-photon detectors will exhibit fluctuation phenomena when measuring photons multiple times. In extreme environments, this fluctuation noise will have a significant impact on image quality, limiting the performance of the imaging system. Second, there is a certain amount of background noise. Interference signals from non-target light sources in the environment are mixed with target light signals and are detected simultaneously, making it difficult to extract weak target signals. Therefore, it is necessary to design a single-photon multi-wavelength imaging system based on a physical driving model to effectively suppress fluctuation noise and background noise on the data source. SUMMARY

[0004] In order to solve the problems of being vulnerable to noise and poor imaging quality of existing single-photon multi-wavelength imaging schemes in extreme environments, the present application proposes a single-photon multi-wavelength imaging system and method based on a physical driving model. Compared with traditional denoising methods, the physical driving model proposed by the present application does not depend on a large number of sample learning, and can significantly improve the peak signal-to-noise ratio of the image by more than 10dB under the condition of an average of only 30 photons per pixel, and exhibits excellent robustness in various application scenarios.

[0005] The technical scheme adopted by the present application is as follows: a single-photon multi-wavelength imaging system based on a physical driving model, comprising a multi-wavelength modulation illumination module, a spatial modulation encoding module and a detection imaging module; the multi-wavelength modulation illumination module uses signal light with different modulation repetition frequencies to illuminate and scatter the color target object to the spatial modulation encoding module, the spatial modulation encoding module spatially encodes the color target object and transmits it to the detection imaging module, the detection imaging module captures the modulated signal light and transmits it to the computer, a physical driving model program is pre-written in the computer, the physical driving model comprises a measurement value denoising module and a physical driving image reconstruction module, and the measurement value denoising module denoises one-dimensional measurement values through a measurement value denoising network; the physical driving image reconstruction module restores the denoised one-dimensional photon counting sequence to a two-dimensional target image of three channels of red, green and blue, and finally merges it into a high-quality color target object reconstruction image.

[0006] The multi-wavelength modulation illumination module comprises a first wavelength laser, a beam splitter, a second wavelength laser, a third wavelength laser, an attenuator and a beam expander, the three wavelength lasers are modulated to different repetition frequencies, and the beam splitter is used to combine the two beams into one beam, the light intensity is adjusted by the attenuator and the collimated and expanded by the beam expander, and then the color target object is irradiated;

[0007] The first wavelength laser, the second wavelength laser and the third wavelength laser are respectively used to generate red, green and blue visible light, and the three lasers correspond to different modulation frequencies;

[0008] The beam splitter is provided with two, the first beam splitter is arranged between the first wavelength laser and the second wavelength laser, and the two beams are combined into one beam, and the second beam splitter is arranged between the third wavelength laser and the combined light of the first wavelength and the second wavelength, and the third wavelength and the first combined light are fused into one beam.

[0009] The spatial modulation encoding module comprises a first lens and a digital micromirror device, and the scattered light from the object is transmitted to the digital micromirror device through the first lens; the digital micromirror device preloads a set of spatial resolution 256*256 Hadamard masks to spatially encode and modulate the color target object;

[0010] The external trigger signal of the digital micromirror device is recorded by a time-correlated single photon counter.

[0011] The probe imaging module comprises a second lens, a single photon detector and a time-correlated single photon counter; the second lens converges the coded signal reflected by the digital micromirror device to the single photon detector, the response signal of the single photon detector is recorded by the time-correlated single photon counter and transmitted to the computer.

[0012] A single-photon multi-wavelength imaging method based on a physical driving model, using a single-photon multi-wavelength imaging system based on a physical driving model, comprising the following steps:

[0013] Step 1: A multi-wavelength modulation signal light source irradiates a color target and a spatial encoding modulation;

[0014] Step 2: Single-photon multi-wavelength modulation signal detection;

[0015] Step 3: Use the measurement value denoising network to remove the noise existing in the one-dimensional measurement value:

[0016] Step 3.1: Preprocess the original data obtained by the computer, and calculate the photon count value corresponding to each modulation mask respectively:

[0017]

[0018] In the formula, is the photon count value corresponding to the nth mask, is the nth Hadamard mask, T (x，y) represents the target object;

[0019] Step 3.2: Use the trained measurement value to denoise the one-dimensional photon count value:

[0020]

[0021] In the formula, is the photon count value containing noise, Cp is the denoised photon count value, D θ is a measurement value denoising network embedded with an occlusion attention mechanism;

[0022] Step 4: Use the physical driving image reconstruction module to iteratively reconstruct the two-dimensional image of the color target:

[0023] T * = R θ (DGI(C p ));

[0024] The constraint condition is:

[0025] In the formula, T *denotes the reconstructed image, R θ is a U-net neural network, DGI is a differential ghost imaging reconstruction algorithm, C p is the denoised photon counting value, P represents the Hadamard mask, and T represents the non-optimal reconstructed image in the iteration process.

[0026] Step 1 specifically comprises:

[0027] Step 1.1: The repetition frequency of the first wavelength laser, the repetition frequency of the second wavelength laser, and the repetition frequency of the third wavelength laser are modulated, the three different wavelength lasers are combined into one beam through a beam splitter, and the beam is collimated and expanded through a beam expander to irradiate the color target object;

[0028] Step 1.2: The scattered signal of the color target object is collected by the first lens and imaged to the modulation area of the digital micromirror device, and a set of Hadamard masks with a spatial resolution of 256*256 are loaded on the digital micromirror device to modulate the color target object.

[0029] Step 2 specifically comprises:

[0030] Step 2.1: The reflected light modulated by the digital micromirror device is converged to the single photon detector by the second lens;

[0031] Step 2.2: The response data of the single photon detector is recorded by the time-correlated single photon counter and transmitted to the computer.

[0032] The present application has the beneficial effects of the prior art:

[0033] 1. The present application designs a single-photon multi-wavelength imaging system based on a physical driving model, which directly processes the data collected in the imaging front end. Compared with using denoising algorithms in the imaging rear end to improve image quality, the present application can significantly improve the peak signal-to-noise ratio of the image by more than 10dB under the condition of an average of only 30 photons per pixel.

[0034] 2. The present application uses the denoised light intensity value as a constraint condition for the physical driving model, which only needs a small amount of prior knowledge to be applicable to any imaging scene, making the system highly robust and suitable for rapidly changing target application scenarios. DETAILED DESCRIPTION

[0035] The present application will be further described below in conjunction with the accompanying drawings:

[0036] Figure 1 is a structural schematic diagram of the system of the present application;

[0037] Figure 2 is a structural schematic diagram of the measurement value denoising network of the present application;

[0038] Figure 3 A physical drive image reconstruction network structure schematic diagram of the application;

[0039] Figure 4 A comparison chart of imaging results after denoising of different schemes;

[0040] Figure 5 A column chart of imaging quality after denoising of different schemes;

[0041] Figure 6 A single-photon multi-wavelength imaging experiment result chart for verifying the effectiveness of the application;

[0042] In the figure: 101 is a 450nm wavelength laser, 102 is a beam splitter, 103 is a 520nm wavelength laser, 104 is a 635nm wavelength laser, 105 is an attenuator, 106 is a beam expander, 107 is a color target object, 1081 is a first lens, 1082 is a second lens, 109 is a digital micromirror device, 110 is a single-photon detector, 111 is a time-correlated single-photon counter, and 112 is a computer. DETAILED DESCRIPTION

[0043] As Figures 1 to 6 shown, the application provides a single-photon multi-wavelength imaging system based on a physical drive model, which comprises a multi-wavelength modulation illumination module, a spatial modulation coding module, and a detection imaging module; the multi-wavelength modulation illumination module uses signal light with different modulation repetition frequencies to illuminate a color target object and scatter it to the spatial modulation coding module, the spatial modulation coding module spatially encodes the color target object and transmits it to the detection imaging module, the detection imaging module captures the modulated signal light and transmits it to a computer, a physical drive model program is pre-written in the computer, which comprises a measurement value denoising module and a physical drive image reconstruction module, in the measurement value denoising module, one-dimensional measurement values are denoised by a measurement value denoising network; the physical drive image reconstruction module restores the denoised one-dimensional photon counting sequence into a two-dimensional target image of three channels of red, green, and blue, and finally merges it into a high-quality color target object reconstruction image.

[0044] The multi-wavelength modulation illumination module comprises a 450 nm wavelength laser 101, a beam splitter 102, a 520 nm wavelength laser 103, a 635 nm wavelength laser 104, an attenuator 105 and a beam expander 106, and the color target 107 is illuminated by the above-mentioned devices; the specific process is that the three kinds of wavelength lasers are modulated into different repetition frequencies, and the beam splitter 102 is used to combine the light into one beam, the light intensity is adjusted by the attenuator 105, and the color target 107 is irradiated after collimation and expansion by the beam expander 106. More specifically, the beam splitter 102 is provided with two, the first beam splitter 102 is arranged between the 450 nm wavelength laser 101 and the 520 nm wavelength laser 103, the 450 nm wavelength and the 520 nm wavelength light are combined into one beam, and the second beam splitter 102 is arranged between the 635 nm wavelength laser 104 and the combined light of the 450 nm wavelength and the 520 nm wavelength, and the 635 nm wavelength and the first combined light are combined into one beam.

[0045] The spatial modulation coding module comprises a first lens 1081 and a digital micromirror device 109, and is used for spatially coding the optical image of the color target; the scattered light from the object is transmitted to the digital micromirror device 109 through the first lens 108; the digital micromirror device 109 is pre-loaded with a group of Hadamard masks with a spatial resolution of 256*256 to spatially code and modulate the color target 107, and the external trigger signal is recorded by the time-correlated single photon counter 111.

[0046] The detection imaging module uses a second lens 1082, a single photon detector 110, a time-correlated single photon counter 111 and a computer 112 to collect extremely weak signals; the coded signal reflected by the digital micromirror device 109 is converged to the single photon detector 110 by the second lens 1082, the response signal of the single photon detector 110 is recorded by the time-correlated single photon counter 111 and transmitted to the computer 112.

[0047] Based on the above-mentioned system, the application provides a single photon multi-wavelength imaging method based on a physical driving model, which comprises the following steps:

[0048] Step 1: The multi-wavelength modulation signal light source irradiates the color target and the spatial coding modulation, as shown in Figure 1

[0049] Step 1.1: The repetition frequency of the 450 nm wavelength laser 101 is modulated to 200 kHz, the repetition frequency of the 520 nm wavelength laser 103 is modulated to 230 kHz, and the repetition frequency of the 635 nm wavelength laser 104 is modulated to 260 kHz; the three kinds of wavelength lasers are modulated in repetition frequency, and the beam splitter 102 is used to combine the light into one beam, and the color target 107 is irradiated after collimation and expansion by the beam expander 106; ​

[0050] Step 1.2: Collect the scattering signal of the color target 107 with the first lens 1081 and image it to the modulation area of the digital micromirror device 109, load a set of Hadamard masks with spatial resolution of 256x256 to the digital micromirror device 109 to spatially encode and modulate the color target 107, and the external trigger signal is collected by the time-correlated single photon counter 111.

[0051] Step 2: Single-photon multi-wavelength modulation signal detection:

[0052] Step 2.1: Use the second lens 1082 to converge the reflected light modulated by the digital micromirror device 109 to the single-photon detector 110;

[0053] Step 2.2: Use the time-correlated single photon counter 111 to record the response data of the single-photon detector 110 and transmit it to the computer 112.

[0054] Step 3: Use the measurement value denoising network (such as Figure 2 shown) to remove the noise present in the one-dimensional measurement value:

[0055] Step 3.1: Preprocess the raw data obtained by the computer 112, and calculate the photon count value corresponding to each modulation mask respectively:

[0056]

[0057] wherein, is the photon count value corresponding to the nth mask, is the nth Hadamard mask, T (x，y) represents the target object;

[0058] Step 3.2: Use the trained measurement value to denoise the one-dimensional photon count value:

[0059]

[0060] wherein, is the photon count value containing noise, C p is the denoised photon count value, D θ is the measurement value denoising network embedded with the occlusion attention mechanism.

[0061] Step 4: Use the physical-driven image reconstruction module (such as Figure 3 shown) to iteratively reconstruct the two-dimensional image of the color target:

[0062] T * = R θ (DGI(C p ));

[0063] with the constraint that:

[0064] In the formula, T * R represents the reconstructed image. θ For U-net neural network, DGI is the differential ghost imaging reconstruction algorithm, C p The value represents the photon count after denoising, where P represents the Hadamard mask and T represents the non-optimal reconstructed image during the iteration process.

[0065] The effectiveness of this invention will be verified through experiments below:

[0066] To verify the feasibility of the physical driving model of the present invention, the target object was imaged under the conditions of a sampling time of 2ms, a photon count rate of 1Mcps, a duty cycle of 10%, and a compression ratio of 0.02, and the physical driving model was used for noise reduction. Figure 4 A comparison of the denoising results of different denoising algorithms on the reconstructed image of the target object. Figure 5 The peak signal-to-noise ratio (PSNR) of the reconstructed image of the target object is statistically shown for different denoising algorithms. The colored leaf was imaged under the conditions of a sampling time of 1.5 ms, a photon count rate of 1 Mcps, a duty cycle of 10%, and a compression ratio of 0.02, and denoising was performed using a physics-driven model. Figure 6 This is a diagram showing the results of a single-photon multi-wavelength imaging experiment on a colored leaf.

[0067] Through analysis Figures 4-6 We can conclude that:

[0068] 1) such as Figure 4 and Figure 5 As shown, the reconstructed image contains a large amount of noise, severely affecting image quality. After processing with different denoising schemes and calculating their peak signal-to-noise ratios (PSNR), the results show that the PSNR based on the physical driving model of this invention is improved by an average of more than 10 dB compared to other denoising algorithms. Specifically, the PSNR of the "moon" image after median filtering is 19.34 dB, the result of sparse three-dimensional transform domain collaborative filtering is 22.94 dB, while the PSNR of the image denoised using the physical driving model of this invention reaches 29.52 dB. Compared with median filtering and sparse three-dimensional transform domain collaborative filtering, the denoising results of the physical driving model proposed in this invention are improved by 10.18 dB and 6.58 dB, respectively, fully verifying the effective noise suppression capability of this invention during the imaging process.

[0069] 2) such as Figure 6As shown, the color leaf reconstruction image of the red, green and blue channels is denoised by using median filtering, sparse three-dimensional transform domain collaborative filtering image denoising and the physical driving model of the application respectively, and then is combined into the final color leaf reconstruction image. It can be concluded that the technical scheme of the application effectively suppresses various noises existing in single-photon multi-wavelength imaging by driving the neural network through the physical model and cooperating with the measurement value denoising network, and the effect is obviously better than other common denoising algorithms, and the image quality is significantly improved.

[0070] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, but not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A single-photon multi-wavelength imaging system based on a physical drive model, characterized by: The application relates to a multi-wavelength modulation illumination module, a spatial modulation coding module and a detection imaging module; the multi-wavelength modulation illumination module uses signal light with different modulation repetition frequencies to illuminate a color target object and scatter to the spatial modulation coding module; the spatial modulation coding module spatially encodes the color target object and transmits to the detection imaging module; the detection imaging module captures the modulation signal light and transmits to a computer (112); a physical drive model program is prewritten in the computer (112); the physical drive model comprises a measurement value denoising module and a physical drive image reconstruction module; the one-dimensional measurement value is denoised through a measurement value denoising network in the measurement value denoising module; the one-dimensional photon counting sequence after denoising is restored to a two-dimensional target image of three channels of red, green and blue; finally, a high-quality color target object reconstruction image is obtained. The multi-wavelength modulation illumination module comprises a first wavelength laser, a beam splitter (102), a second wavelength laser, a third wavelength laser, an attenuator (105) and a beam expander (106); the three kinds of wavelength lasers are modulated into different repetition frequencies; the beam splitter (102) is used for combining the two beams into one; the light intensity is adjusted through the attenuator (105) and collimated and expanded through the beam expander (106) to irradiate the color target object (107). The spatial modulation coding module comprises a first lens (1081) and a digital micromirror device (109); the scattered light from the object is transmitted to the digital micromirror device (109) through the first lens (1081); the digital micromirror device (109) is preloaded with a group of Hadamard masks with a spatial resolution of 256*256 to spatially encode and modulate the color target object (107). The detection imaging module comprises a second lens (1082), a single photon detector (110) and a time-correlated single photon counter (111); the coded signal reflected by the digital micromirror device (109) is converged to the single photon detector (110) through the second lens (1082); the response signal of the single photon detector (110) is recorded by the time-correlated single photon counter (111) and transmitted to the computer (112). The external trigger signal of the digital micromirror device (109) is recorded by the time-correlated single photon counter (111). The one-dimensional measurement value is denoised through a measurement value denoising network, and the steps are as follows: The original data obtained by the computer (112) is preprocessed, and the photon counting values corresponding to each modulation mask are calculated respectively: ； In the formula, is the first n The photon counting value corresponding to the mask, is the first n The Hadamard mask of the Zhang, Indicates the target object; Step 3.2: the one-dimensional photon counting value is denoised by using the trained measurement value: ； wherein is a photon count value including noise, C p is a photon count value after denoising, D θ is a measurement value denoising network embedded with an occlusion attention mechanism.

2. The single-photon multi-wavelength imaging system based on a physical driving model according to claim 1, wherein: The first wavelength laser, the second wavelength laser and the third wavelength laser are respectively used for generating red, green and blue visible light, and the three lasers correspond to different modulation frequencies. The beam splitter (102) is provided with two; the first beam splitter (102) is arranged between the first wavelength laser and the second wavelength laser, and the two beams are combined into one; the second beam splitter (102) is arranged between the third wavelength laser and the combined light of the first wavelength and the second wavelength, and the third wavelength and the combined light of the first time are fused into one.

3. A method of single-photon multiwavelength imaging based on a physical drive model, characterized by: The single-photon multi-wavelength imaging system based on a physical driving model according to claim 1 or 2, comprising the following steps: Step 1: a multi-wavelength modulated signal light source irradiates a color target and a spatial encoding modulation; Step 2: single-photon multi-wavelength modulated signal detection; Step 3: using a measurement value denoising network to remove noise existing in one-dimensional measurement values: Step 3.1: preprocessing the original data obtained by the computer (112), and calculating the photon count value corresponding to each modulation mask respectively: ； In the formula, is the first n The photon counting value corresponding to the mask, is the first n The Hadamard mask of the z-th, Indicates the target object; Step 3.2: using the trained measurement value to denoise the one-dimensional photon count value: ； In the formula, is a photon count value including noise, C p is a photon count value after denoising, D θ is a measurement value denoising network embedded with an occlusion attention mechanism; Step 4: using a physical driving image reconstruction module to iteratively reconstruct the two-dimensional image of the color target: ； with the proviso that: ; wherein T * represents a reconstructed image, R θ is a U-net neural network, DGI is a difference ptychographic reconstruction algorithm, C p is a denoised photon count value, P represents a Hadamard mask, T represents a non-optimal reconstructed image in an iteration process.

4. The method of claim 3, wherein the method is based on a physical driving model. Step 1 specifically comprises: Step 1.1: modulating the repetition frequency of the first wavelength laser, the repetition frequency of the second wavelength laser, and the repetition frequency of the third wavelength laser, and using a beam splitter (102) to combine the three different wavelength lasers into one beam, which is collimated and expanded by a beam expander (106) to irradiate the color target (107); Step 1.2: using the first lens (1081) to collect the scattered signal of the color target (107) and image it to the modulation area of the digital micromirror device (109), and loading a set of Hadamard masks with a spatial resolution of 256x256 on the digital micromirror device (109) to spatially encode and modulate the color target (107).

5. The method of claim 3, wherein: Step 2 specifically comprises: Step 2.1: using the second lens (1082) to converge the reflected light modulated by the digital micromirror device (109) to the single-photon detector (110); Step 2.2: using the time-dependent single-photon counter (111) to record the response data of the single-photon detector (110) and transmit it to the computer (112).

Images