Coding aperture joint optimization method and system of broadband compression calculation imaging system

Abstract

The invention discloses a coding aperture joint optimization method and system of a broadband compression calculation imaging system. According to the coding aperture joint optimization algorithm based on gradient descent, the image reconstruction quality of the two subsystems is improved at the same time by optimizing the pattern of the common coding aperture. A multi-target joint optimization function of a coding aperture is constructed based on a compressed sensing theory so as to minimize column coherence of sensing matrixes of the two subsystems. Reasonable constraint is carried out on the spatial range of a target solution by introducing a quadratic regular term, so that the quantization error of the optimization algorithm is reduced. In addition, coding apertures among different snapshots are optimized, complementation of the coding apertures is realized, and the spatial perception capability of the system is enhanced. By optimizing the pattern form of the coding aperture, high-precision image reconstruction of the target object in the visible light wave band and the short wave infrared wave band can be realized at the same time, so that the detection and identification capability of the target in a complex environment can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computational imaging, and particularly to a method and system for jointly optimizing an encoding aperture of a wide-band compressed computational imaging system. Background Art

[0002] Multispectral imaging technology senses two-dimensional spatial image data along multiple wavelengths to construct a three-dimensional spectral data cube of a target scene, and this technology has been applied in many fields such as remote sensing, biological imaging, and smart agriculture. Traditional scanning spectral imagers can obtain spectral images with high spatial resolution, but they have limitations such as long scanning time, high cost, and complex systems. In addition, visible light multispectral imaging can be achieved in a single snapshot using prisms and large array detectors, but there is a mutual restriction between spatial resolution and spectral resolution. These defective imaging architectures limit the further development of this technology. With the development of the compressed sensing theory, snapshot compressed computational spectral imaging technology has emerged and gradually plays an important role in this field. The coded aperture snapshot spectral imager is an architecture widely used in this technical field. This architecture only requires a small number of snapshot measurements, and a multispectral image within a given spectral range can be calculated through a reconstruction algorithm. At the same time, the compressed sensing theory has also been extended and applied in the field of infrared super-resolution imaging. The short-wave infrared super-resolution compressed computational imaging technology encodes and compresses images in the spatial dimension, and a high-resolution short-wave infrared image can be reconstructed using a low-resolution detector, which plays a crucial role in the field of national defense and military reconnaissance. The encoding aperture optimization method proposed by the present invention is based on a wide-band compressed computational imaging system composed of these two technologies.

[0003] In a wide-band compressed computational imaging system spanning "visible light - short-wave infrared", a group of encoding apertures is shared to perform spatial modulation on the target light field. Commonly used encoding methods include random encoding, Gaussian encoding, Bernoulli encoding, etc., but none of them satisfy the restricted isometry property in the compressed sensing theory, so none of them is an optimal encoding method. In addition, most of the current encoding aperture design methods are for specific compressed computational imaging systems and are not applicable to wide-band integrated imaging systems.

[0004] Therefore, the joint optimization design of the encoding aperture in a wide-band compressed computational imaging system is of great significance for simultaneously improving the imaging performance of subsystems, improving the target detection accuracy in complex environments, and improving the system integration performance, etc. However, for the problem that the existing encoding aperture design methods cannot simultaneously improve the imaging performance of the two subsystems in a wide-band compressed computational imaging system, the existing solutions have not been able to solve this problem. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for jointly optimizing the coded aperture of a wide-band compressive computational imaging system. By constructing an optimization function based on the compressive sensing theory to minimize the column coherence of the sensing matrices of two subsystems, a quadratic regularization term is introduced to reduce the quantization error of the optimization algorithm, and a complementary regularization term is introduced to enhance the spatial sensing ability of the system. This algorithm can simultaneously improve the image reconstruction performance of visible light multispectral imaging and short-wave infrared super-resolution imaging, and can further improve the detection and recognition accuracy of targets.

[0006] To achieve the above object, an embodiment of the present invention provides a method for jointly optimizing the coded aperture of a wide-band compressive computational imaging system, including the following steps:

[0007] Construct optimization functions for the coded aperture with respect to visible light multispectral imaging and short-wave infrared super-resolution imaging respectively according to the compressive sensing theory;

[0008] Construct a quadratic regularization term to constrain the solution space of the coded aperture;

[0009] Construct a complementary regularization term;

[0010] Integrate the constructed optimization functions and regularization terms to obtain a multi-objective joint optimization function for the coded aperture;

[0011] Decompose the multi-objective joint optimization function into two optimization functions representing visible light multispectral imaging and short-wave infrared super-resolution imaging;

[0012] Use the alternating iterative gradient descent algorithm to update the coded aperture value to obtain the optimal coded aperture.

[0013] Further, the optimization functions for the coded aperture with respect to visible light multispectral imaging and short-wave infrared super-resolution imaging are constructed respectively according to the compressive sensing theory, specifically:

[0014]

[0015] where Γ1 is the optimization function for visible light multispectral imaging, Γ2 is the optimization function for short-wave infrared super-resolution imaging, A1 is the sensing matrix of the visible light multispectral compressive computational imaging system, A2 is the sensing matrix of the short-wave infrared super-resolution compressive computational imaging system, I is the identity matrix, and ||·|| F is the Frobenius norm of the matrix.

[0016] Further, a quadratic regularization term to constrain the solution space of the coded aperture is constructed. Specifically, the quadratic regularization term is expressed as R Q :

[0017]

[0018] where \(K\) is the number of snapshots, \(I\) is the identity matrix, and \(I\) Λ is the matrix obtained by stacking multiple identity matrices \(K\) times in the column direction. \(C\) is the matrix obtained by stacking \(K\) different coding matrices in the column direction. The smaller the value of the quadratic regularization term, the closer the coding aperture value tends to 0 or 1.

[0019] Furthermore, a complementary regularization term is constructed. Specifically, the complementary regularization term \(R\) C is expressed as:

[0020]

[0021] where \(C\) is the matrix obtained by stacking \(K\) different coding matrices in the column direction, and \(I\) is the identity matrix.

[0022] Furthermore, the constructed optimization function and regularization term are integrated to obtain a multi-objective joint optimization function for the coding aperture. The multi-objective joint optimization function is expressed as:

[0023]

[0024] where \(\tau_1\) is the weight coefficient of the regularization term \(R\) Q and \(\tau_2\) is the weight coefficient of the regularization term \(R\) C .

[0025] Furthermore, the multi-objective joint optimization function is decomposed into two optimization functions representing visible light multispectral imaging and short-wave infrared super-resolution imaging, which are specifically expressed as:

[0026]

[0027] where \(\tau_1\), \(\tau_2\) and are weight coefficients, \(C\) and represent the coding matrices in the visible light imaging and short-wave infrared imaging systems respectively. The corresponding \(R\) Q , \(R\) C and are the regularization terms calculated using \(C\) and respectively.

[0028] Furthermore, the alternating iterative gradient descent algorithm is used to update the coding aperture value to obtain the optimal coding aperture. The specific steps are as follows:

[0029] Step (1): According to the first optimization function in the fifth step, calculate the gradient value of this optimization function with respect to the coding aperture \(C\), and then the coding aperture value can be updated once. The update formula is:

[0030] \(C\) n = \(C\) n-1 - s·(\(\nabla\Gamma_1+\tau_1\nabla R\) Q +\(\tau_2\nabla R\)C )(6)

[0031] where n is the number of iterations, s is the step size of gradient descent, ▽Γ1, ▽R Q and ▽R C are the gradient matrices of the optimization function and the regularization term with respect to the coded aperture;

[0032] Step (2): Extract the coded aperture value required by the short-wave infrared imaging system from the updated coded aperture value C Update the coded aperture value one more time using the second optimization function in the fifth step;

[0033] Step (3): Convert the coded aperture value optimized in step (2) to the C required in step (1), and continuously repeat steps (1) and (2) until the objective function value converges stably, and finally the optimal coded aperture value can be obtained.

[0034] Another embodiment of the present invention also provides a coded aperture joint optimization system for a wide-band compression computational imaging system, including the following modules:

[0035] An optimization function construction module, configured to respectively construct optimization functions of the coded aperture with respect to visible light multispectral imaging and short-wave infrared super-resolution imaging according to the compressive sensing theory;

[0036] A regularization term construction module, configured to construct a quadratic regularization term and a complementary regularization term for constraining the solution space of the coded aperture;

[0037] An integration module, configured to integrate the constructed optimization functions and regularization terms to obtain a multi-objective joint optimization function of the coded aperture;

[0038] A decomposition module, configured to decompose the multi-objective joint optimization function into two optimization functions representing visible light multispectral imaging and short-wave infrared super-resolution imaging;

[0039] An iteration module, configured to update the coded aperture value using the alternating iterative gradient descent algorithm to obtain the optimal coded aperture.

[0040] Furthermore, the optimization function construction module respectively constructs optimization functions of the coded aperture with respect to visible light multispectral imaging and short-wave infrared super-resolution imaging according to the compressive sensing theory, specifically:

[0041]

[0042] Among them, Γ1 is the optimization function for visible light multispectral imaging, Γ2 is the optimization function for short-wave infrared super-resolution imaging, A1 is the sensing matrix of the visible light multispectral compressive computational imaging system, A2 is the sensing matrix of the short-wave infrared super-resolution compressive computational imaging system, I is the identity matrix, and ||·|| F is the Frobenius norm of the matrix.

[0043] Furthermore, the regularization term construction module constructs a quadratic regularization term that constrains the solution space of the coded aperture. Specifically, the quadratic regularization term is expressed as R Q :

[0044]

[0045] where K is the number of snapshots, I is the identity matrix, and I Λ is the matrix obtained by stacking multiple identity matrices K times in the column direction, C is the matrix obtained by stacking K different coding matrices in the column direction, and the smaller the value of the quadratic regularization term, the more the coded aperture value tends to 0 or 1;

[0046] Construct a complementary regularization term. Specifically, the complementary regularization term R C is expressed as:

[0047]

[0048] where C is the matrix obtained by stacking K different coding matrices in the column direction, and I is the identity matrix;

[0049] The integration module integrates the constructed optimization function and regularization term to obtain a multi-objective joint optimization function for the coded aperture. The multi-objective joint optimization function is expressed as:

[0050]

[0051] where τ1 is the weight coefficient of the regularization term R Q and τ2 is the weight coefficient of the regularization term R C ;

[0052] The decomposition module decomposes the multi-objective joint optimization function into two optimization functions that characterize visible light multispectral imaging and short-wave infrared super-resolution imaging, specifically expressed as:

[0053]

[0054] where τ1, τ2, and are weight coefficients, C and respectively represent the coding matrices in the visible light imaging and short-wave infrared imaging systems, and the corresponding R Q , R C and respectively using C and the calculated regularization term;

[0055] Iteration module, using the alternating iterative gradient descent algorithm to update the coded aperture value to obtain the optimal coded aperture. The specific steps are as follows:

[0056] Step (1): According to the first optimization function in the fifth step, calculate the gradient value of this optimization function with respect to the coded aperture C, and then the coded aperture value can be updated once. The update formula is:

[0057] C n = C n-1 - s·(▽Γ1 + τ1▽R Q + τ2▽R C )(6)

[0058] where n is the number of iterations, s is the step size of gradient descent, ▽Γ1, ▽R Q and ▽R C are the gradient matrices of the optimization function and the regularization term with respect to the coded aperture;

[0059] Step (2): Extract the coded aperture value required by the short-wave infrared imaging system from the updated coded aperture value C and use the second optimization function in the fifth step to update the coded aperture value again;

[0060] Step (3): Convert the coded aperture value optimized in step (2) to the C required in step (1), and continuously repeat steps (1) and (2) until the objective function value converges stably. Finally, the optimal coded aperture value can be obtained.

[0061] Beneficial effects:

[0062] (1) A method for jointly optimizing the coded aperture of a wide-band compressive computational imaging system provided by the present invention improves the image reconstruction quality of visible light spectral imaging and short-wave infrared super-resolution imaging simultaneously by constructing a multi-objective joint optimization function of the coded aperture, which is beneficial to improving the integrated performance of the wide-band system and further improving the recognition and detection accuracy of targets in complex environments.

[0063] (2) The present invention constructs a multi-objective joint optimization function of the coded aperture to minimize the column coherence of the sensing matrices of the two subsystems, so that the sensing matrices as much as possible satisfy the restricted isometry property in the compressive sensing theory. In addition, a quadratic regularization term is introduced to reduce the quantization error of the optimization algorithm, and a complementary regularization term is introduced to enhance the spatial sensing ability of the system, further improving the algorithm performance. On this basis, the alternating gradient descent algorithm is used to optimize the coded aperture pattern, thereby improving the image reconstruction quality of the two subsystems simultaneously.

[0064] (3) The proposed coded aperture multi-objective joint optimization algorithm of the present invention can be further extended and applied to other high-performance integrated optical systems, and has universality. In addition to the purposes, features and advantages described above, the present invention also has other purposes, features and advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a schematic flow chart of the method of the present invention;

[0066] Figure 2 is a coded aperture map optimized by the coded aperture joint optimization method when the number of snapshots is 2, with a size of 1024×1024;

[0067] Figure 3 is a convergence curve graph of the loss function during the optimization process of a 1024×1024 coded aperture when the number of snapshots is 2;

[0068] Figure 4 is a comparison graph of the spectral image reconstruction results using a random coded aperture and an optimized coded aperture under different numbers of snapshots;

[0069] Figure 5 is a comparison graph of the spectral curves reconstructed using a random coded aperture and an optimized coded aperture at three different spatial positions; Figure 5 In (a) is a schematic diagram of the spatial positions of P1, P2 and P3; Figure 5 In (b)-(d) are the comparison results of the spectral curve reconstructions of P1, P2 and P3 respectively;

[0070] Figure 6 is a comparison graph of the reconstruction results of short-wave infrared high-resolution images; Figure 6 In (a) is the original short-wave infrared high-resolution image and the high-resolution image reconstructed using bilinear interpolation, Figure 6 In (b) is the comparison result of the short-wave infrared high-resolution image reconstruction results using random coding and optimized coding.. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0071] The present invention will be described in detail below with reference to the accompanying drawings and by way of examples.

[0072] An embodiment of the present invention provides a coded aperture joint optimization method for a wide-band compressed computational imaging system. The method first establishes a coded aperture multi-objective joint optimization function for visible light multi-spectral and short-wave infrared super-resolution imaging. While satisfying the limited isometry characteristic as much as possible, a quadratic regularization term is introduced to reduce the quantization error of the optimization algorithm, and a complementary regularization term is introduced to enhance the spatial perception capability of the system. Then, an alternating iterative gradient descent algorithm is used to alternately calculate and update the coded aperture value. After multiple rounds of iterations, the optimal coded aperture can be finally obtained, and the image reconstruction quality of the two subsystems can be improved simultaneously by using the optimal coded aperture.

[0073] like Figure 1 As shown, a coded aperture joint optimization method for a wide-band compressed computational imaging system, the specific steps are:

[0074] In the first step, based on the compressed sensing theory, we construct the optimization functions of the coded aperture for visible light multispectral imaging and shortwave infrared super-resolution imaging, which are expressed as:

[0075]

[0076] Among them, A1 is the perception matrix of the visible light multi-spectral compression computing imaging system, A2 is the perception matrix of the short-wave infrared super-resolution compression computing imaging system, I is the unit matrix, ||·|| F is the Frobenius norm of the matrix. The smaller the values ​​of these two optimization functions are, the more the perception matrix satisfies the finite isometry property, and the better the image reconstruction effect is.

[0077] The second step is to construct a quadratic regularization term that constrains the coded aperture solution space to reduce the quantization error of the optimization algorithm and make the coded aperture value tend to the actual physical coding value, that is, 0 or 1. The quadratic regularization term is expressed as:

[0078]

[0079] Where K is the number of snapshots, I is the identity matrix, and I Λ is a matrix obtained by superimposing multiple unit matrices K times in the column direction, and C is a matrix obtained by superimposing K different coding matrices in the column direction. The smaller the value of the quadratic regularization term, the closer the coded aperture value is to 0 or 1.

[0080] The third step is to construct complementary regularization terms to improve the reconstruction capability under multiple snapshots and enhance the spatial perception capability of the system. The complementary regularization term is expressed as:

[0081]

[0082] Among them, C is the matrix after stacking K different coding matrices in the column direction, and I is the identity matrix. The smaller the value of the complementary regularization term, the more complementary the coding apertures between different snapshots tend to be, and the stronger the spatial perception ability of the system.

[0083] Step 4: Integrate the optimization functions and regularization terms constructed in the first, second, and third steps to obtain the multi-objective joint optimization function of the coding aperture, which is expressed as:

[0084]

[0085] Among them, τ1 is the weight coefficient of the regularization term R Q , and τ2 is the weight coefficient of the regularization term R C .

[0086] Step 5: Since the dimensions of the coding matrices in the visible light multi-spectral compressive computational imaging system and the short-wave infrared super-resolution compressive computational imaging system are inconsistent in the case of multiple snapshots, the alternating iterative gradient descent algorithm is used to update the coding aperture value C. Decompose the joint optimization function in Step 4 into two optimization functions, which are respectively expressed as:

[0087]

[0088] Among them, τ1, τ2 and are weight coefficients, C and respectively represent the coding matrices in the visible light imaging and short-wave infrared imaging systems, and the corresponding R Q , R C and are the regularization terms calculated using C and respectively.

[0089] Step 6: It includes three steps. Step (1): According to the first optimization function in Step 5, calculate the gradient value of this optimization function with respect to the coding aperture C, and then update the coding aperture value once. The update formula is:

[0090] C n = C n-1 - s·(▽Γ1 + τ1▽R Q + τ2▽R C )(6)

[0091] Among them, n is the number of iterations, s is the step size of gradient descent, and ▽Γ1, ▽R Q and ▽R C are the gradient matrices of the optimization function and the regularization term with respect to the coding aperture. Step (2): Extract the coding aperture value required by the short-wave infrared imaging system from the updated coding aperture value CUpdate the coded aperture value one more time using the second optimization function in the fifth step. Step (3): Convert the coded aperture value optimized in step (2) to C required in step (1), and continuously repeat steps (1) and (2) until the objective function value converges stably. Finally, the optimal coded aperture value can be obtained.

[0092] In the first step, if the sensing matrices A1 and A2 can be extended as functions of the coding matrices, the optimization functions can be expressed as:

[0093]

[0094] where Ψ is the sparse transformation matrix, C and represent the coding matrices in the visible light imaging and short-wave infrared imaging systems respectively, D represents the dispersion matrix in the visible light spectral imaging system, and S represents the compression matrix in the short-wave infrared super-resolution imaging system.

[0095] In the first step, in order to convert the discrete optimization function into a continuous optimization function, a parameter transformation method was used to transform the coding matrix. The transformation method is as follows:

[0096] C = (cos(B) + 1) / 2, P = cos(B)(8)

[0097] where B is the transformed parameter matrix, B and C have the same dimension, and the parameter matrix B is optimized during optimization. After obtaining the optimal parameter matrix B through optimization, use the inverse transformation method to convert it into the coding matrix C. The inverse transformation method is expressed as:

[0098] C ij = U((cos(B ij ) + 1) / 2 - t c )(9)

[0099] where the U(t) function represents the Heaviside step function, t c is the set threshold. If t ≥ 0, then U(t) = 1; if t < 0, then U(t) = 0.

[0100] In the fourth step, the integrated coded aperture multi-objective joint optimization function is applicable to the case of multiple snapshots, that is, K ≥ 1.

[0101] In the fifth step, the coding matrices C and of the visible light multi-spectral compressive computational imaging system and the short-wave infrared super-resolution compressive computational imaging system have dimensions of KMNL×MNL and KMN×MN respectively, where M, N, and L represent the sizes of the spatial scene in the vertical dimension, horizontal dimension, and spectral dimension respectively.

[0102] In the sixth step, the initial value of the step size is set to 0.1, and the initial weight coefficients of the regularization terms are all set to 1.

[0103] In the sixth step, the encoding matrix C in the visible light multispectral imaging system and the encoding matrix in the short-wave infrared super-resolution imaging system can be converted to each other. In the case of a single snapshot (i.e., K = 1), C and have the following conversion relationship:

[0104]

[0105] where diag(·) means to stack the encoding matrix group in the diagonal direction.

[0106] In the sixth step, according to the parameter transformation method C = (cos(B) + 1) / 2 and P = cos(B), let G = D T D, then the gradient matrix ▽Γ in step (1) can be transformed into a form with respect to the parameter matrix B, and the calculation method is:

[0107]

[0108] In the sixth step, according to the parameter transformation method C = (cos(B) + 1) / 2 and P = cos(B), the gradient matrix ▽R Q in step (1) can be transformed into a form with respect to the parameter matrix B, and the calculation method is:

[0109]

[0110] In the sixth step, according to the parameter transformation method C = (cos(B) + 1) / 2 and P = cos(B), the gradient matrix ▽R C in step (1) can be transformed into a form with respect to the parameter matrix B, and the calculation method is:

[0111]

[0112] In the sixth step, after optimizing to obtain the optimal encoding aperture value, substitute this optimal encoding into the reconstruction algorithm, and according to the measurement values captured by the visible light multispectral imaging detector and the short-wave infrared super-resolution imaging detector, the multispectral image and the short-wave infrared high-resolution image within the given spectral range can be reconstructed.

[0113] Another embodiment of the present invention also provides an encoding aperture joint optimization system for a wide-band compression computational imaging system, which is characterized by including the following modules:

[0114] An optimization function construction module, configured to respectively construct optimization functions of the coded aperture for visible light multispectral imaging and short-wave infrared super-resolution imaging according to the compressed sensing theory;

[0115] A regularization term construction module, configured to construct a quadratic regularization term and a complementary regularization term that constrain the solution space of the coded aperture;

[0116] An integration module, configured to integrate the constructed optimization functions and regularization terms to obtain a multi-objective joint optimization function of the coded aperture;

[0117] A decomposition module, configured to decompose the multi-objective joint optimization function into two optimization functions representing visible light multispectral imaging and short-wave infrared super-resolution imaging;

[0118] An iteration module, configured to update the coded aperture value using the alternating iteration gradient descent algorithm to obtain an optimal coded aperture.

[0119] In an embodiment of the present invention, the optimization function construction module respectively constructs optimization functions of the coded aperture for visible light multispectral imaging and short-wave infrared super-resolution imaging according to the compressed sensing theory, specifically:

[0120]

[0121] Among them, Γ1 is the optimization function of visible light multispectral imaging, Γ2 is the optimization function of short-wave infrared super-resolution imaging, A1 is the sensing matrix of the visible light multispectral compressive computational imaging system, A2 is the sensing matrix of the short-wave infrared super-resolution compressive computational imaging system, I is the identity matrix, and ||·|| F is the Frobenius norm of the matrix.

[0122] The regularization term construction module constructs a quadratic regularization term that constrains the solution space of the coded aperture. Specifically, the quadratic regularization term is expressed as R Q :

[0123]

[0124] Among them, K is the number of snapshots, I is the identity matrix, and I Λ is the matrix obtained by stacking multiple identity matrices K times in the column direction, C is the matrix obtained by stacking K different coded matrices in the column direction. The smaller the value of the quadratic regularization term, the closer the coded aperture value tends to 0 or 1;

[0125] Construct a complementary regularization term. Specifically, the complementary regularization term R C is expressed as:

[0126]

[0127] Among them, C is the matrix after superimposing K different coding matrices in the column direction, and I is the identity matrix;

[0128] An integration module that integrates the constructed optimization function and the regularization term to obtain a multi-objective joint optimization function for the coded aperture. The multi-objective joint optimization function is expressed as:

[0129]

[0130] where τ1 is the weight coefficient of the regularization term R Q and τ2 is the weight coefficient of the regularization term R C ;

[0131] A decomposition module that decomposes the multi-objective joint optimization function into two optimization functions representing visible light multispectral imaging and short-wave infrared super-resolution imaging, specifically expressed as:

[0132]

[0133] where τ1, τ2 and are weight coefficients, C and respectively represent the coding matrices in the visible light imaging and short-wave infrared imaging systems, and the corresponding R Q , R C and are the regularization terms calculated using C and respectively;

[0134] An iteration module that uses the alternating iteration gradient descent algorithm to update the coded aperture value to obtain the optimal coded aperture. The specific steps are as follows:

[0135] Step (1): According to the first optimization function in the fifth step, calculate the gradient value of this optimization function with respect to the coded aperture C, and then the coded aperture value can be updated once. The update formula is:

[0136] C n = C n-1 - s·(▽Γ1 + τ1▽R Q + τ2▽R C ) (6)

[0137] where n is the number of iterations, s is the step size of gradient descent, and ▽Γ1, ▽R Q and ▽R C are the gradient matrices of the optimization function and the regularization term with respect to the coded aperture;

[0138] Step (2): Extract the coded aperture value required by the short-wave infrared imaging system from the updated coded aperture value C and use the second optimization function in the fifth step to update the coded aperture value again;

[0139] Step (3): Convert the coded aperture value optimized in step (2) to C required in step (1), and continuously repeat steps (1) and (2) until the objective function value converges stably. Finally, the optimal coded aperture value can be obtained.

[0140] Embodiment of the present invention:

[0141] When using the coded aperture joint optimization algorithm proposed by the present invention, the dimension of the coded aperture is consistent with the spatial dimension of the image used. The size of the original spectral image used is 1024×1024×6, that is, the spatial dimension of the image is 1024×1024, the number of spectral channels is 6, and the central wavelengths are 460nm, 510nm, 560nm, 610nm, 660nm, and 710nm respectively. The short-wave infrared image adopts the same spatial scene as the visible light multi-spectral image, the central wavelength is 2440nm, the spatial resolution is 1024×1024, and the compression ratio is 4×4, that is, the size of the low-resolution image captured by the short-wave infrared detector is 256×256. Under the above conditions, the dimension of the common coded aperture to be optimized is 1024×1024. When the number of snapshots K = 2, the two optimized coded apertures are as Figure 2 shown, and the convergence curve during the optimization process is as Figure 3 shown. It can be seen from Figure 2 that the two optimized coded aperture maps can complement each other at most positions. Figure 3 The green curve (legend: binarized) in

[0142] Figure 4 is the optimization function descent curve calculated using the quantized coding matrix; the red curve (legend: gray-scaled) is the optimization function descent curve calculated using the coding matrix before quantization. It can be seen that both curves can converge stably. Figure 5 shows the comparison of the quality of the visible light multi-spectral images reconstructed using the random coded aperture and the optimized coded aperture at different numbers of snapshots. Using the peak signal-to-noise ratio (PSNR) and the structural similarity (SSIM) as image evaluation metrics, they are marked in the upper right corner of the spectral image (PSNR / , SSIM). Compared with the random coded aperture, using the coded aperture joint optimization method proposed by the present invention can significantly improve the reconstruction quality of the spectral image. Figure 5 In Figure 5The comparison of the spectral curve reconstruction results of (b)-(d) in [reference] is for P1, P2, and P3 respectively. Different curve colors and shapes represent different snapshot number conditions and the encoding conditions used. Please refer to the legend. Compared with the random encoding aperture, the spectral curve reconstructed using the optimized encoding aperture is closer to the reference true value, and the spectral accuracy is higher. Figure 6 It is the comparison of the reconstruction results of short-wave infrared high-resolution images. Figure 6 In (a) of [reference], it is the original short-wave infrared high-resolution image and the high-resolution image reconstructed using bilinear interpolation. Figure 6 In (b) of [reference], it is the comparison of the results of the short-wave infrared high-resolution images reconstructed using random encoding and optimized encoding. Compared with the high-resolution image reconstructed using the traditional bilinear interpolation algorithm, the image quality of the image reconstructed by compressed coding calculation is higher. Compared with the random encoding aperture, using the optimized encoding aperture can further improve the image reconstruction quality, and with the increase of the number of snapshots, the improvement of the image quality is more obvious.

[0143] In summary, the above is only the preferred embodiment of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A coded aperture joint optimization method for a wideband compressed computational imaging system, characterized in that: The steps include: Based on the theory of compressed sensing, the optimization functions of coded aperture for visible light multispectral imaging and short-wave infrared super-resolution imaging are constructed respectively; Constructing the quadratic regularization term of the constraint coded aperture solution space; Construct complementary regularization terms; The constructed optimization function and regularization term are integrated to obtain the multi-objective joint optimization function of the coded aperture; Decomposing the multi-objective joint optimization function into two optimization functions that characterize visible light multispectral imaging and short-wave infrared super-resolution imaging; The coded aperture value is updated using an alternating iterative gradient descent algorithm to obtain the optimal coded aperture.

2. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 1, characterized in that: The optimization functions of coded aperture for visible light multi-spectral imaging and short-wave infrared super-resolution imaging are respectively constructed according to the compressed sensing theory, specifically: Among them, Γ1 is the optimization function of visible light multispectral imaging, Γ2 is the optimization function of short-wave infrared super-resolution imaging, A1 is the perception matrix of visible light multispectral compression computing imaging system, A2 is the perception matrix of short-wave infrared super-resolution compression computing imaging system, I is the unit matrix, ||·|| F is the Frobenius norm of the matrix.

3. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 1, characterized in that: The quadratic regularization term of the constrained coded aperture solution space is constructed. Specifically, the quadratic regularization term is expressed as R Q : Where K is the number of snapshots, I is the identity matrix, and I Λ is a matrix obtained by superimposing multiple unit matrices K times in the column direction, C is a matrix obtained by superimposing K different coding matrices in the column direction, and the smaller the value of the quadratic regularization term is, the closer the coding aperture value is to 0 or 1.

4. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 1, characterized in that: The complementary regularization term is constructed, specifically, the complementary regularization term R C It is expressed as: Where C is a matrix obtained by superimposing K different encoding matrices in the column direction, and I is a unit matrix.

5. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 1, characterized in that: The constructed optimization function and regularization term are integrated to obtain a multi-objective joint optimization function of the coded aperture. It is expressed as: Where τ1 is the regularization term R Q The weight coefficient, τ2 is the regular term R C The weight coefficient of .

6. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 5, characterized in that: The multi-objective joint optimization function is decomposed into two optimization functions that characterize visible light multispectral imaging and short-wave infrared super-resolution imaging, which is specifically expressed as follows: where τ1, τ2 and is the weight coefficient, C and Respectively represent the coding matrices in the visible light imaging and short-wave infrared imaging systems, and the corresponding R Q , R C as well as Use C and The calculated regularization term.

7. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 6, characterized in that: The method uses the alternating iterative gradient descent algorithm to update the coded aperture value to obtain the optimal coded aperture, specifically using the following steps: Step (1): According to the first optimization function in the fifth step, the gradient value of the optimization function to the coded aperture C is calculated, and the coded aperture value can be updated once. The update formula is: Where n is the number of iterations, s is the step size of gradient descent, as well as is the gradient matrix of the optimization function and the regularization term with respect to the coding aperture; Step (2): Extract the coded aperture value required by the short-wave infrared imaging system from the updated coded aperture value C Use the second optimization function in step 5 to update the coded aperture value again; Step (3): Optimize the coded aperture value obtained in step (2) Convert it to the C required by step (1), and continue to repeat steps (1) and (2) until the objective function value converges stably, and finally the optimal coded aperture value can be obtained.

8. A coded aperture joint optimization system for a wideband compressed computational imaging system, characterized in that: Includes the following modules: An optimization function construction module is used to construct optimization functions of coded aperture for visible light multispectral imaging and short-wave infrared super-resolution imaging respectively according to compressed sensing theory; A regularization term construction module, used to construct quadratic regularization terms and complementary regularization terms of the constrained coded aperture solution space; An integration module is used to integrate the constructed optimization function and regularization term to obtain a multi-objective joint optimization function of the coded aperture; A decomposition module, used for decomposing the multi-objective joint optimization function into two optimization functions representing visible light multi-spectral imaging and short-wave infrared super-resolution imaging; The iterative module is used to update the coded aperture value using an alternating iterative gradient descent algorithm to obtain an optimal coded aperture.

9. The coded aperture joint optimization system for a wideband compressed computational imaging system according to claim 1, characterized in that: The optimization function construction module constructs optimization functions of coded aperture for visible light multispectral imaging and short-wave infrared super-resolution imaging respectively according to the compressed sensing theory, specifically: Among them, Γ1 is the optimization function of visible light multispectral imaging, Γ2 is the optimization function of short-wave infrared super-resolution imaging, A1 is the perception matrix of visible light multispectral compression computing imaging system, A2 is the perception matrix of short-wave infrared super-resolution compression computing imaging system, I is the unit matrix, ||·|| F is the Frobenius norm of the matrix.

10. The coded aperture joint optimization method for a wideband compressed computational imaging system according to claim 1, characterized in that: The regularization term construction module constructs a quadratic regularization term of the constrained coded aperture solution space. Specifically, the quadratic regularization term is expressed as R Q : Where K is the number of snapshots, I is the identity matrix, and I Λ is a matrix obtained by superimposing multiple unit matrices K times in the column direction, C is a matrix obtained by superimposing K different coding matrices in the column direction, and the smaller the value of the quadratic regularization term is, the closer the coding aperture value is to 0 or 1; The complementary regularization term is constructed, specifically, the complementary regularization term R C It is expressed as: Where C is the matrix obtained by superimposing K different encoding matrices in the column direction, and I is the unit matrix; The integration module integrates the constructed optimization function and the regularization term to obtain a multi-objective joint optimization function of the coded aperture, and the multi-objective joint optimization function It is expressed as: Where τ1 is the regularization term R Q The weight coefficient, τ2 is the regular term R C The weight coefficient of The decomposition module decomposes the multi-objective joint optimization function into two optimization functions that characterize visible light multispectral imaging and short-wave infrared super-resolution imaging, which are specifically expressed as follows: where τ1, τ2 and is the weight coefficient, C and Respectively represent the coding matrices in the visible light imaging and short-wave infrared imaging systems, and the corresponding R Q , R C as well as Use C and The calculated regularization term; The iterative module uses an alternating iterative gradient descent algorithm to update the coded aperture value to obtain the optimal coded aperture, specifically using the following steps: Step (1): According to the first optimization function in the fifth step, the gradient value of the optimization function to the coded aperture C is calculated, and the coded aperture value can be updated once. The update formula is: Where n is the number of iterations, s is the step size of gradient descent, as well as is the gradient matrix of the optimization function and the regularization term with respect to the coding aperture; Step (2): Extract the coded aperture value required by the short-wave infrared imaging system from the updated coded aperture value C Use the second optimization function in step 5 to update the coded aperture value again; Step (3): Optimize the coded aperture value obtained in step (2) Convert it to the C required by step (1), and continue to repeat steps (1) and (2) until the objective function value converges stably, and finally the optimal coded aperture value can be obtained.

Images