Design method, device, storage medium and use of quantum dot spectral filter array

Abstract

This disclosure relates to the field of spectral imaging technology, including design methods, devices, storage media, and applications of quantum dot spectral filter arrays. The method involves: determining a predetermined number of target quantum dot materials from a quantum dot material library; obtaining a data mapping relationship between the second spectral property data of the target quantum dot materials and their design parameters; determining a simulated quantum dot spectral filter array corresponding to the target quantum dot materials based on the data mapping relationship, and establishing an imaging mathematical model based on the simulated quantum dot spectral filter array; processing the incident spectral data based on the imaging mathematical model to obtain simulated response results; obtaining reconstructed spectral data corresponding to the simulated response results; and iteratively training the imaging mathematical model using deep learning based on the differences between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship. This improves the reliability and efficiency of determining the design parameters.

Description

Technical Field

[0001] This disclosure relates to the field of spectral imaging technology, and in particular to a design method, apparatus, storage medium, and application of a quantum dot spectral filter array. Background Technology

[0002] Spectra, acting as the "fingerprint" of matter, allow for precise perception of the world. Spectral imaging technology, combining imaging and spectral sensing techniques, can simultaneously acquire spatial and spectral information of matter, thus obtaining high-dimensional spatial-spectral information. This enables spectral imaging technology to play a more significant role in fields such as medicine, remote sensing, food, and agriculture. To improve the portability of spectral imaging systems, quantum dot spectral imaging systems have emerged. The core component of a quantum dot spectral imaging system includes a quantum dot spectral filter array composed of various quantum dot materials. Because quantum dot materials possess highly continuously tunable absorption spectral characteristics across a wide wavelength range from deep violet to mid-infrared, quantum dot spectral filter arrays, compared to traditional narrowband filters, offer advantages such as high luminous flux and flexible design. Consequently, quantum dot spectral imaging systems based on this array offer advantages such as lightweight portability, real-time operation, stability, and customizability.

[0003] A typical design method for quantum dot spectral filter arrays includes: manually selecting the type and concentration of quantum dot materials from a quantum dot material library based on design experience; designing quantum dot materials based on the type and concentration; fabricating a quantum dot spectral filter array; and thus obtaining a quantum dot spectral imaging system.

[0004] However, because quantum dot materials have highly continuously tunable absorption spectral characteristics in a wide wavelength range from deep violet to mid-infrared, array design is very flexible, and theoretically there may be countless schemes to choose from. The design scheme selected based on experience may be unreliable and have low design efficiency. Summary of the Invention

[0005] In view of this, this disclosure proposes a design method, device, storage medium and application of quantum dot spectral filter array, which can solve the problems of low efficiency and poor reliability of manually determining design parameters.

[0006] According to one aspect of this disclosure, a method for designing a quantum dot spectral filter array is provided, the method comprising:

[0007] Based on the first spectral property data of various quantum dot materials stored in the quantum dot material library, a preset number of target quantum dot materials are determined;

[0008] Obtain the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material;

[0009] Based on the data mapping relationship, a simulated quantum dot spectral filter array corresponding to the target quantum dot material is determined, and an imaging mathematical model is established based on the simulated quantum dot spectral filter array; the imaging mathematical model is used to simulate the process of obtaining a response result after incident light passes through the quantum dot spectral filter array.

[0010] The incident spectral data is processed based on the imaging mathematical model to obtain the simulated response results;

[0011] Obtain the reconstructed spectral data corresponding to the simulation response results;

[0012] The imaging mathematical model is iteratively trained using deep learning based on the difference between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of actual target quantum dot spectral filter arrays.

[0013] In one possible implementation, the second spectral property data includes absorption spectral data; the design parameters include the concentration coefficients of the respective target quantum dot materials; accordingly,

[0014] The process of obtaining the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material includes:

[0015] Acquire reference absorption spectral data of each target quantum dot material at a preset reference concentration; wherein, the concentration coefficient is used to indicate the ratio between the current concentration and the reference concentration;

[0016] Determine the data mapping relationship between the absorption spectral data and the concentration coefficient and the reference absorption spectral data.

[0017] In one possible implementation, determining the data mapping relationship between the absorption spectral data and the concentration coefficient and the reference absorption spectral data includes:

[0018] Within a preset concentration range, the absorption spectral data is determined to be the product of the concentration coefficient and the reference absorption spectral data, thus obtaining the data mapping relationship.

[0019] In one possible implementation, determining the simulated quantum dot spectral filter array corresponding to the target quantum dot material based on the data mapping relationship, and establishing an imaging mathematical model based on the simulated quantum dot spectral filter array, includes:

[0020] The absorption spectral data represented by the data mapping relationship is converted into transmission spectral data based on the Lambert-Beer law;

[0021] Based on the transmission spectral data of each target quantum dot material, the simulated quantum dot spectral filter array is determined.

[0022] The imaging mathematical model is obtained by determining the light intensity at each pixel location of the image as the inner product between the incident spectral data and the transmitted spectral data corresponding to that pixel location.

[0023] In one possible implementation, the deep learning training method iteratively trains the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship, including:

[0024] The imaging mathematical model is iteratively trained based on the difference between the incident spectral data and the reconstructed spectral data using a weight update method, and the design parameters are updated during the training process to obtain the trained imaging mathematical model.

[0025] In one possible implementation, the first spectral property data includes transmission spectral data; accordingly,

[0026] The method determines a preset number of target quantum dot materials based on the first spectral property data corresponding to various quantum dot materials stored in the quantum dot material library, including:

[0027] Based on the transmission spectral data of various quantum dot materials in the quantum dot material library, the correlation between each quantum dot material and other quantum dot materials is determined, and a preset number of target quantum dot materials with the lowest correlation are obtained.

[0028] In one possible implementation, determining the correlation between each quantum dot material and other quantum dot materials based on the transmission spectral data of various quantum dot materials in the quantum dot material library, and obtaining a preset number of target quantum dot materials with the lowest correlation, includes:

[0029] A transmission spectrum matrix is ​​constructed based on the transmission spectrum data of various quantum dot materials. Each row of data in the transmission spectrum matrix represents the transmittance at the same wavelength, and each column of data represents the transmittance of the same quantum dot material at different wavelengths.

[0030] The transmission spectrum matrix is ​​subjected to permutation QR decomposition to obtain the permutation matrix P of the transmission spectrum matrix; wherein, the permutation matrix P is used to permutate the order of the column data in the transmission spectrum matrix, and the correlation of the column data after the order permutation is sorted from low to high.

[0031] Based on the sorting order of the column data in the permutation matrix P, the quantum dot materials corresponding to the first K columns of data are determined as the target quantum dot materials, where K is a preset number.

[0032] According to another aspect of this disclosure, a design apparatus for a quantum dot spectral filter array is provided, the apparatus comprising:

[0033] The material screening module is used to determine a preset number of target quantum dot materials based on the first spectral property data of various quantum dot materials stored in the quantum dot material library.

[0034] The parameter mapping module is used to obtain the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material;

[0035] The model building module is used to determine the simulated quantum dot spectral filter array corresponding to the target quantum dot material based on the data mapping relationship, so as to establish an imaging mathematical model based on the simulated quantum dot spectral filter array; the imaging mathematical model is used to simulate the process of obtaining the response result after the incident light passes through the quantum dot spectral filter array;

[0036] The simulation response module is used to process the incident spectral data based on the imaging mathematical model to obtain the simulation response results;

[0037] The spectral reconstruction module is used to acquire the reconstructed spectral data corresponding to the simulation response results;

[0038] The parameter acquisition module is used to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data using a deep learning training method, so as to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of the actual target quantum dot spectral filter array.

[0039] According to another aspect of this disclosure, a design apparatus for a quantum dot spectral filter array is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0040] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided that stores computer program instructions thereon, wherein the computer program instructions, when executed by a processor, implement the above-described method.

[0041] According to another aspect of this disclosure, a computer program product is provided, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0042] According to another aspect of this disclosure, a quantum dot spectral filter array obtained according to the above method is provided for use in spectral imaging or spatial spectral information acquisition.

[0043] By using first spectral property data of various quantum dot materials stored in a quantum dot material library, a predetermined number of target quantum dot materials are determined. The data mapping relationship between the second spectral property data of the target quantum dot materials and their design parameters is obtained. Based on this data mapping relationship, a simulated quantum dot spectral filter array corresponding to the target quantum dot materials is determined, and an imaging mathematical model is established based on this simulated quantum dot spectral filter array. The incident spectral data is processed based on the imaging mathematical model to obtain simulated response results. The reconstructed spectral data corresponding to the simulated response results is obtained. A deep learning training method is used to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data, updating the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of actual target quantum dot spectral filter arrays. This approach solves the problems of low efficiency and poor reliability in manually determining design parameters. By simulating the incident light processing process of the quantum dot spectral filter array, the optimal design parameters are automatically obtained, improving the reliability and efficiency of determining design parameters. The quantum dot spectral filter array designed using this method can be used for spectral imaging. In addition to its applications in fields such as medicine, remote sensing, food, and agriculture, it can also play an important role in acquiring spatial spectral information and thus be used in optical information systems.

[0044] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0045] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0046] Figure 1 A flowchart illustrating a design method for a quantum dot spectral filter array according to an embodiment of the present disclosure is shown.

[0047] Figure 2 A schematic diagram illustrating a concentration change process according to an embodiment of the present disclosure is shown;

[0048] Figure 3 A schematic diagram illustrating the design process of a quantum dot spectral filter array according to an embodiment of the present disclosure is shown.

[0049] Figure 4 A block diagram showing a design apparatus for a quantum dot spectral filter array according to an embodiment of the present disclosure;

[0050] Figure 5A block diagram of a design apparatus for a quantum dot spectral filter array according to another embodiment of the present disclosure is shown. Detailed Implementation

[0051] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0052] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0053] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0054] Traditional quantum dot spectral imaging systems employ imaging methods including, but not limited to, pushbroom spectral imaging and snapshot spectral imaging.

[0055] Pushbroom spectral imaging acquires spectral data for a single line during a single measurement. The detector or sample is then moved perpendicular to this line to complete the acquisition of the entire spectral image. Subsequently, a spectral reconstruction algorithm is used to obtain a multispectral image of the target. However, pushbroom spectral imaging systems have long imaging times, which cannot meet the requirements for rapid imaging.

[0056] Snapshot-style spectral imaging, by combining the response results obtained from a single exposure with spectral reconstruction algorithms, can quickly acquire multispectral images of the target in one go. However, snapshot-style spectral imaging requires highly complex spectral reconstruction algorithms, and the reconstructed spectral image data contains a certain error compared to the actual data, meaning that real-time performance is achieved at the expense of data acquisition accuracy.

[0057] Regardless of the imaging method used in a quantum dot spectral imaging system, there are numerous design approaches for the quantum dot material in the quantum dot spectral filter array. Finding a reliable and efficient design scheme for quantum dot materials is a pressing issue. This application proposes a design method for quantum dot spectral filter arrays. By using the design parameters of the quantum dot material as model variables, an imaging mathematical model is constructed to simulate the input light processing of the quantum dot spectral filter array. Based on the difference between the simulated imaging result and the input spectral image, the design parameters are optimized to obtain the optimal design scheme for the quantum dot spectral filter array. Optionally, if a spectral imaging system is designed according to the quantum dot spectral filter array obtained from this design scheme, the image acquired by the system will better match the input spectral image, resulting in higher accuracy. Therefore, this design method exhibits high reliability and determination efficiency. Applying this design method to a snapshot-type quantum dot spectral imaging system can ensure both image acquisition speed and accuracy, meeting the spectral imaging requirements that balance speed and accuracy.

[0058] It should be noted that the design method of the quantum dot spectral filter array provided in this application can also be applied to other types of quantum dot spectral imaging systems, such as pushbroom quantum dot spectral imaging systems. In addition, the design method of the quantum dot spectral filter array provided in this application can be applied not only to quantum dot spectral imaging systems, but also to other spectral imaging or spatial spectral information acquisition scenarios. This application does not limit the application scenarios of the method.

[0059] The design method of the quantum dot spectral filter array provided in this application will be described in detail below. Figure 1 A flowchart illustrating a design method for a quantum dot spectral filter array according to an embodiment of this disclosure is provided. This embodiment describes the method in an electronic device with computing capabilities, including but not limited to: a user terminal or a server. The user terminal includes, but is not limited to: computers, tablets, mobile phones, etc. This embodiment does not limit the implementation of the electronic device. Figure 1 As shown, the method includes:

[0060] Step 101: Based on the first spectral property data corresponding to various quantum dot materials stored in the quantum dot material library, determine a preset number of target quantum dot materials.

[0061] In this embodiment, the quantum dot material library is a database that stores at least the correspondence between first spectral property data and quantum dot material types. Different types of quantum dot materials correspond to different first spectral property data. Optionally, the quantum dot material library can be pre-stored in an electronic device; or sent to the electronic device by another device; or downloaded online by the electronic device. This embodiment does not limit the method by which the electronic device obtains the quantum dot material library.

[0062] The spectral property data (including the first spectral property data and the second spectral property data hereinafter) are used to indicate the ability of the quantum dot material to capture and utilize light energy based on its optical properties. Indicatively, the spectral property data includes, but is not limited to, at least one of the following: absorption spectral data (such as absorption spectral curves), transmission spectral data (such as absorption spectral curves), etc. This embodiment does not limit the implementation method of the spectral property data.

[0063] Taking transmission spectrum data as an example, based on the first spectral property data of various quantum dot materials stored in the quantum dot material library, a preset number of target quantum dot materials are determined. This includes: based on the transmission spectrum data of various quantum dot materials in the quantum dot material library, determining the correlation between each quantum dot material and other quantum dot materials, and obtaining the preset number of target quantum dot materials with the lowest correlation. In this embodiment, the correlation of the transmission spectrum data of a quantum dot material refers to the correlation between the transmission spectrum data of that quantum dot material and the transmission spectrum data of each other quantum dot material in the quantum dot material library. The weaker the correlation of the transmission spectrum data of quantum dot materials, the stronger the difference in the transmission spectrum data, the greater the difference in information carried after responding to incident light, and the easier it is to distinguish them.

[0064] In one example, the transmission spectrum data for each quantum dot material includes transmittance (or transmission spectrum curves) corresponding to different wavelengths. Accordingly, based on the transmission spectrum data of various quantum dot materials in the quantum dot material library, the correlation between each quantum dot material and other quantum dot materials is determined, and a preset number of target quantum dot materials with the lowest correlation are obtained. This includes: constructing a transmission spectrum matrix based on the transmission spectrum data of each quantum dot material, where each row of data in the transmission spectrum matrix represents the transmittance at the same wavelength, and each column of data represents the transmittance of the same quantum dot material at different wavelengths; performing a permutation QR decomposition on the transmission spectrum matrix to obtain a permutation matrix P; wherein, the permutation matrix P is used to permutate the order of the column data in the transmission spectrum matrix, and the correlation of the column data after the permutation is sorted from low to high; based on the sorting order of the column data in the permutation matrix P, the quantum dot materials corresponding to the first K columns of data are determined as target quantum dot materials, where K is a preset number.

[0065] The permutation QR decomposition is a matrix factorization technique that decomposes the transmission spectrum matrix A by multiplying it by the permutation matrix P into the product of an orthogonal matrix Q and an upper triangular matrix R. Since the importance of each column in the permutation matrix P is ranked from high to low, this importance can indicate the correlation between the columns. That is, the columns in the permutation matrix P can be considered as being ranked from low to high correlation, and this ranking of correlation corresponds to a ranking of importance. Therefore, the quantum dot materials corresponding to the first K columns are the target quantum dot materials with the lowest correlation.

[0066] In other embodiments, the correlation between each quantum dot material and other quantum dot materials can also be determined based on the Singular Value Decomposition (SVD) method. This embodiment does not limit the method of determining the correlation between quantum dot materials.

[0067] In other embodiments, the first spectral property data can also be absorption spectral data. Since the absorption spectral data and transmission spectral data of quantum dot materials satisfy the Lambert-Beer law, the absorption spectral data can be converted into transmission spectral data based on the Lambert-Beer law. Then, based on the transmission spectral data corresponding to the quantum dot materials, the correlation between each quantum dot material and other quantum dot materials can be determined in the manner described above, so as to obtain the target quantum dot materials with the lowest preset number of correlations.

[0068] Since the less correlated the transmission spectral data between quantum dot materials, the stronger the feature extraction capability of the quantum dot spectral filter array based on the quantum dot material for the input image information, in this embodiment, by determining a preset number of target quantum dot materials with the lowest correlation based on the transmission spectral data, the image information extraction effect of the quantum dot spectral filter array can be guaranteed.

[0069] In this embodiment, optionally, the preset quantity can be determined based on the number of image channels in the quantum dot imaging system to which the quantum dot spectral filter array is to be applied. For example, if the number of image channels in the quantum dot imaging system is 3, then each periodic unit in the quantum dot spectral filter array is a 3×3 quantum dot material matrix, that is, the preset quantity of target quantum dot material is 3×3=9. In this case, assuming the quantum dot material library includes N transmission spectrum curves, and N=161, and the preset quantity K=9, then the quantum dot materials corresponding to the 9 least relevant transmission spectrum curves are selected as target quantum dot materials to design the periodic units of the quantum dot spectral filter array. The quantum dot spectral filter array includes multiple periodic units arranged in a matrix, and each periodic unit includes a preset quantity of quantum dot material.

[0070] Step 102: Obtain the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material.

[0071] The type of the second spectral property data may be the same as or different from that of the first spectral property data. For example, the second spectral property data may be different from the first spectral property data, where the first spectral property data is transmission spectrum data and the second spectral property data is absorption spectrum data.

[0072] The design parameters of the target quantum dot material are used to indicate the manufacturing process parameters of the target quantum dot material. For example, the design parameters are used to indicate the concentration of the target quantum dot material.

[0073] Taking the second spectral property data as including absorption spectral data and the design parameters as including the concentration coefficients of each target quantum dot material as an example, the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material is obtained, including: obtaining the reference absorption spectral data of each target quantum dot material at a preset reference concentration; and determining the data mapping relationship between the absorption spectral data and the concentration coefficients and the reference absorption spectral data.

[0074] The concentration coefficient indicates the ratio between the current concentration and the reference concentration. The initial concentration coefficient is pre-stored in the electronic device. Since the initial concentration coefficient can be corrected later, its accuracy is not required in this embodiment. The current concentration refers to the simulated concentration value of the target quantum dot material in this calculation process. After determining the concentration coefficient, the current concentration can be determined by multiplying the concentration coefficient by the reference concentration.

[0075] Optionally, the quantum dot material library also includes absorption spectrum data of each quantum dot material at different concentrations. Based on this, reference absorption spectrum data of each target quantum dot material at a preset reference concentration is obtained, including: reading reference absorption spectrum data of the target quantum dot material at a reference concentration from the quantum dot material library.

[0076] In other embodiments, the reference absorption spectrum data of the target quantum dot material at the reference concentration may also be sent by other devices. This embodiment does not limit the method of obtaining the reference absorption spectrum data at the reference concentration.

[0077] Since the absorption spectrum data of quantum dot materials is linearly related to the concentration within a preset concentration range, in one example, to simplify the data mapping relationship, the data mapping relationship between the absorption spectrum data and the concentration coefficient and the reference absorption spectrum data is determined, including: within the preset concentration range, determining that the absorption spectrum data is the product of the concentration coefficient and the reference absorption spectrum data, thus obtaining the data mapping relationship.

[0078] Schematic, the data mapping relationship can be represented by the following formula:

[0079] A = K × I;

[0080]

[0081] Where I represents the reference absorption spectrum data of each target quantum dot material at the reference concentration, and Q represents the reference absorption spectrum data of each target quantum dot material. ij The matrix formed, where Q ij In this context, i represents the row index of the target quantum dot material in the simulated quantum dot spectral filter array, j represents the column index of the target quantum dot material in the simulated quantum dot spectral filter array, and Q represents the column index of the target quantum dot material. ij This is the reference absorption spectrum data of the target quantum dot material in the i-th row and j-th column of the simulated quantum dot spectral filter array; K represents the concentration coefficient k corresponding to each target quantum dot material. ij The matrix formed, where the concentration coefficient k ij For floating-point numbers; k ij In this array, i represents the row index of the target quantum dot material in the simulated quantum dot spectral filter array, j represents the column index of the target quantum dot material in the simulated quantum dot spectral filter array, and k represents the column index of the target quantum dot material. ij This is to simulate the concentration coefficient of the target quantum dot material in the i-th row and j-th column of a quantum dot spectral filter array. A represents a matrix composed of absorption spectral data, where any element a in A has a value of α. ij =Q ij ×k ij a ij In this context, 'i' represents the row index of the target quantum dot material in the simulated quantum dot spectral filter array, 'j' represents the column index of the target quantum dot material in the simulated quantum dot spectral filter array, and 'a' represents the column index of the target quantum dot material in the simulated quantum dot spectral filter array. ij This is to simulate the absorption spectrum data of the target quantum dot material in the i-th row and j-th column of a quantum dot spectral filter array. Here, i and j are integers within the range [1, n]; the value of n×n equals a preset quantity. In this embodiment, each periodic unit in the quantum dot spectral filter array is a square matrix as an example. In other embodiments, each periodic unit in the quantum dot spectral filter array may not be a square matrix. In this case, if the number of rows in the periodic unit is n and the number of columns is m, then the value of n×m equals a preset quantity.

[0082] Optionally, the reference concentration in this embodiment makes the reference absorption spectrum data the same for different target quantum dot materials, thereby simplifying the calculation of design parameters.

[0083] In other embodiments, the reference absorption spectrum data corresponding to different target quantum dot materials may also be different. This embodiment does not limit the selection method of reference concentration.

[0084] In other embodiments, the conversion relationship between absorption spectral data and concentration can be fitted in concentration ranges other than the concentration range to obtain a data mapping relationship; or, the conversion relationship between transmission spectral data and concentration can be fitted to obtain a data mapping relationship; this embodiment does not limit the implementation method of the data mapping relationship.

[0085] Step 103: Determine the simulated quantum dot spectral filter array corresponding to the target quantum dot material based on the data mapping relationship, and establish an imaging mathematical model based on the simulated quantum dot spectral filter array.

[0086] The physical structure principle of a quantum dot spectral filter array for processing incident light includes: after incident light passes through a quantum dot spectral filter array with a certain spatial distribution, a response result with a certain spatial distribution is obtained. This response result can be considered as the product and integration of a three-dimensional spectral image data cube L(x,y,λ) and the response function S(x,y,λ) of the two-dimensional quantum dot spectral filter array at corresponding positions. In this embodiment, an imaging mathematical model is used to simulate the process of incident light obtaining a response result after passing through the quantum dot spectral filter array.

[0087] Based on the above principles, the imaging mathematical model can be expressed by the following formula:

[0088] I(x,y)=∫S(x,y,λ)L(x,y,λ)dλ

[0089] Where L(x,y,λ) represents the light intensity at spatial coordinates (x,y) at different wavelengths λ; S(x,y,λ) represents the response of the target quantum dot material at spatial coordinates (x,y) to different wavelengths λ; and I(x,y) represents the light intensity at spatial coordinates (x,y) obtained after processing with a simulated quantum dot spectral filter array. S(x,y,λ) is the transmission spectrum data of the target quantum dot material.

[0090] According to the above imaging mathematical model, if the data mapping relationship includes the data mapping relationship between absorption spectral data and design parameters, then the simulated quantum dot spectral filter array corresponding to the target quantum dot material is determined based on the data mapping relationship. The imaging mathematical model is then established based on the simulated quantum dot spectral filter array, including: converting the absorption spectral data represented by the data mapping relationship into transmission spectral data based on the Lambert-Beer law; determining the simulated quantum dot spectral filter array based on the transmission spectral data of each target quantum dot material; and determining the light intensity of the image at each pixel position as the inner product between the incident spectral data and the transmission spectral data corresponding to the pixel position, thus obtaining the imaging mathematical model.

[0091] The inner product (also known as the dot product or dot product) is used to perform operations on two vectors in the real number field R to obtain a real number result.

[0092] For example: if the absorption spectrum data represented by the data mapping relationship is A = K × I as mentioned above; then based on the Lambert-Beer law, the transmission spectrum data can be obtained as follows:

[0093]

[0094] Right now,

[0095] Where T represents a matrix composed of transmission spectral data of various target quantum dot materials.

[0096] If the data mapping relationship includes the data mapping relationship between transmission spectral data and design parameters, then the simulated quantum dot spectral filter array corresponding to the target quantum dot material is determined based on the data mapping relationship. An imaging mathematical model is then established based on the simulated quantum dot spectral filter array, including: determining the simulated quantum dot spectral filter array based on the transmission spectral data represented by the data mapping relationship; determining the light intensity of the imaging image at each pixel position as the inner product between the incident spectral data and the transmission spectral data corresponding to the pixel position, thus obtaining the imaging mathematical model.

[0097] Determining a simulated quantum dot spectral filter array based on transmission spectral data refers to using a matrix composed of the transmission spectral data (such as T in the above text) corresponding to each target quantum dot material as a periodic unit, and combining each periodic unit into a simulated quantum dot spectral filter array. The simulated quantum dot spectral filter array is used to simulate the response of a real quantum dot spectral filter array to incident light.

[0098] By taking T as S(x,y,λ) in the imaging mathematical model above, we obtain the imaging mathematical model.

[0099] Step 104: Process the incident spectral data based on the imaging mathematical model to obtain the simulated response results.

[0100] The incident spectral data is used to simulate the incident spectral image of a quantum dot spectral filter array in a real-world scenario. This incident spectral data is a three-dimensional spectral image data cube, specifically L(x,y,λ) mentioned above. The incident spectral data can be understood as sample data in deep learning. There are multiple sets of incident spectral data, and the incident spectral data is pre-stored in an electronic device.

[0101] The simulated response results are obtained by simulating the response of a real quantum dot spectral filter array to incident light, as described in I(x,y) above. Each set of incident spectral data corresponds to one simulated response result.

[0102] Step 105: Obtain the reconstructed spectral data corresponding to the simulation response results.

[0103] Optionally, the electronic device uses a spectral reconstruction algorithm to process the simulated response results to obtain reconstructed spectral data. The spectral reconstruction algorithm is used to recover the measured incident spectrum based on the simulated response results. The spectral reconstruction algorithm can be a reconstruction algorithm based on a neural network (such as a convolutional neural network), or the TKVA algorithm, etc. This embodiment does not limit the implementation method of the spectral reconstruction algorithm.

[0104] In other embodiments, the reconstructed spectral data may also be obtained by the electronic device sending the simulation response result to other devices, and the other devices processing the simulation response result based on the spectral reconstruction algorithm; accordingly, the electronic device obtains the reconstructed spectral data generated by other devices. This embodiment does not limit the method of obtaining the reconstructed spectral data.

[0105] Step 106: Using deep learning training, an imaging mathematical model is iteratively trained based on the difference between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of the actual target quantum dot spectral filter array.

[0106] In the imaging mathematical model above, we first assume that the two-dimensional simulated quantum dot spectral filter array is composed of only a single target quantum dot material, represented by the response function S(λ). The response result can then be expressed as:

[0107] I(x,y)=∫S(λ)L(x,y,λ)dλ

[0108] In practical calculations, integration is generally replaced by discretized summation. When a continuous signal is sampled, the response can be approximated as follows:

[0109]

[0110] Where N represents the number of image channels in the imaging mathematical model; L(x,y,λ) k S(λ) represents the incident spectrum data; if S(λ) k ) through a 1*1 convolution kernel S c (λ k If ), then the above formula can be regarded as performing a 1*1 convolution operation on the incident spectral data of N image channels. The process of the convolution kernel performing the convolution operation on the incident spectral data is the inner product operation at the corresponding spatial positions, which can be written as:

[0111]

[0112] Based on this, if the number of sampling points for continuous signals in the imaging mathematical model (i.e., the number of each periodic unit in the simulated quantum dot spectral filter array) is the same as the number of image channels in the convolution operation, and the discretized response function S(λ) k ) and 1*1 convolution kernel weights S c (λ k Similarly, if the imaging mathematical model is equivalent to the convolutional model of a convolutional neural network, then the imaging mathematical model can be directly embedded into the training method of the convolutional neural network, and the concentration coefficient can be used as an updatable network parameter in the imaging mathematical model. The imaging mathematical model can then be trained using the training method of the convolutional neural network.

[0113] At this point, the imaging mathematical model can be trained directly using the training code of the convolutional neural network, without the need to set up an additional set of training code for the imaging mathematical model. This simplifies the difficulty of obtaining design parameters and improves the efficiency of obtaining design parameters.

[0114] It should be further explained that the simulated quantum dot spectral filter array is generally obtained by periodically arranging fixed-size periodic units in space. In this case, the convolution model differs slightly from the convolution calculation commonly used in computer vision. Specifically, the imaging mathematical model in this embodiment uses an n*n convolution kernel, which convolves with the incident spectral data with a stride of n, and the spatial summation is not performed after convolution. For example, a simulated quantum dot spectral filter array composed of 4*4 periodic units uses the transmission spectral data of the target quantum dot material to form a 4*4 convolution kernel, with a stride of 4, and the result obtained without spatial summation after convolution is the ideal simulated response result.

[0115] Based on the above equivalence principle, a deep learning training method is adopted to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data, so as to update the design parameters in the data mapping relationship. This includes: using a weight update method to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data, and updating the design parameters during the training process to obtain the trained imaging mathematical model.

[0116] For example, the data mapping relationship is the data mapping relationship between absorption spectrum data and concentration coefficient and reference absorption spectrum data. In this case, during the training of the imaging mathematical model, the reference absorption spectrum data remains unchanged, while the concentration coefficient is updated iteratively. After the model converges, the product of the matrix formed by the concentration coefficient and the reference concentration is the optimal concentration matrix.

[0117] For example, the concentration change process during network training can be referenced using gradient descent. Figure 2 As shown, according to Figure 2As can be seen, as the number of training epochs increases, the loss function continuously decreases, and the concentration matrix K converges to a stable value, thus obtaining the optimal concentration matrix.

[0118] Subsequently, based on the optimal concentration matrix, the concentration levels of each target quantum dot material can be obtained, allowing for the formulation of quantum dots of corresponding types and concentrations, thereby preparing the corresponding actual target quantum dot spectral filter array. Optionally, the target quantum dot spectral filter array can then be used to construct a quantum dot spectral imaging system or a spatial spectral information acquisition system, etc.

[0119] In summary, the quantum dot spectral filter array design method provided in this embodiment determines a preset number of target quantum dot materials based on the first spectral property data corresponding to various quantum dot materials stored in a quantum dot material library; obtains the data mapping relationship between the second spectral property data of the target quantum dot materials and the design parameters of the target quantum dot materials; determines the simulated quantum dot spectral filter array corresponding to the target quantum dot materials based on the data mapping relationship, and establishes an imaging mathematical model based on the simulated quantum dot spectral filter array; processes the incident spectral data based on the imaging mathematical model to obtain the simulated response result; obtains the reconstructed spectral data corresponding to the simulated response result; and uses a deep learning training method to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of the actual target quantum dot spectral filter array. This method can solve the problems of low efficiency and poor reliability of manually determining design parameters. By simulating the incident light processing process of the quantum dot spectral filter array, the optimal design parameters are automatically obtained, which can improve the reliability and efficiency of determining design parameters.

[0120] Furthermore, by making the imaging mathematical model equivalent to the convolutional model, the training method of the convolutional model can be used to train the imaging mathematical model. This allows electronic devices to directly use the training code of the convolutional neural network to train the imaging mathematical model without having to set up an additional set of training code for the imaging mathematical model. This simplifies the difficulty of obtaining design parameters and improves the efficiency of obtaining design parameters.

[0121] To better understand the design method of the quantum dot spectral filter array provided in the above embodiments, an example is given below for illustrative purposes. In this example, the design parameter is a concentration coefficient, the first spectral property data is transmission spectral data, and the second spectral property data is absorption spectral data. (Refer to...) Figure 3 The method includes the following steps:

[0122] After obtaining a predetermined quantity of target quantum dot material, matrix I is determined based on the reference absorption spectrum data of the target quantum dot material at a reference concentration. The concentration coefficient matrix K is initialized, and the matrix A corresponding to the absorption spectrum data of the target quantum dot material is determined as A = K × I. Based on the Lambert-Beer law, the matrix A corresponding to the absorption spectrum data is converted into the matrix corresponding to the transmission spectrum data. Using T as the response function in the imaging mathematical model, the following imaging mathematical model is obtained:

[0123]

[0124] Where N represents the number of image channels in the imaging mathematical model.

[0125] The input spectral data X is used as L(x,y,λ) in the imaging mathematical model. k Input the imaging mathematical model to obtain the simulated response result; use the spectral reconstruction algorithm to perform spectral reconstruction on the simulated response result to obtain the reconstructed spectral data Y; input the reconstructed spectral data Y and the input spectral data X into the loss function corresponding to deep learning, such as the mean squared error (MSE) function, to obtain the loss function value; based on the loss function value, use the gradient descent method to train the imaging mathematical model to update the value of matrix K in the imaging mathematical model until the number of training times reaches the preset number or the change in the loss function value is less than the preset change, at which point training stops.

[0126] At this point, the imaging mathematical model is trained using the gradient descent method. As the loss function value continuously decreases, the concentration coefficient matrix K eventually converges to a stable value. Then, the concentration matrix is ​​obtained by multiplying the concentration coefficient matrix K by the reference concentration. Each concentration matrix corresponds one-to-one with a specific target quantum dot material. Based on the type and concentration level of each target quantum dot material, an actual target quantum dot spectral filter array is configured, resulting in a quantum dot spectral imaging system with superior imaging performance.

[0127] Based on the above embodiments, this embodiment provides the use of the quantum dot spectral filter array obtained by the above method for spectral imaging or spatial spectral information acquisition. Specifically, after obtaining the quantum dot spectral filter array through the above embodiments, the quantum dot spectral filter array is used in a spectral imaging system or in a spatial spectral information acquisition device to process the incident light through the quantum dot spectral filter array, thereby obtaining a device with better processing effect.

[0128] Figure 4A block diagram of a design apparatus for a quantum dot spectral filter array according to an embodiment of the present disclosure is shown; the apparatus includes at least the following modules: a material screening module 410, a parameter mapping module 420, a model building module 430, a simulation response module 440, a spectral reconstruction module 450, and a parameter acquisition module 460.

[0129] The material screening module 410 is used to determine a preset number of target quantum dot materials based on the first spectral property data corresponding to various quantum dot materials stored in the quantum dot material library; the material library may vary depending on the scenario, for example, the quantum dot materials in the material library may also be different according to the different band requirements of different scenarios.

[0130] The parameter mapping module 420 is used to obtain the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material;

[0131] The model building module 430 is used to determine the simulated quantum dot spectral filter array corresponding to the target quantum dot material based on the data mapping relationship, so as to establish an imaging mathematical model based on the simulated quantum dot spectral filter array; the imaging mathematical model is used to simulate the process of obtaining the response result after the incident light passes through the quantum dot spectral filter array.

[0132] The simulation response module 440 is used to process the incident spectral data based on the imaging mathematical model to obtain the simulation response result;

[0133] The spectral reconstruction module 450 is used to acquire the reconstructed spectral data corresponding to the simulation response result;

[0134] The parameter acquisition module 460 is used to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data using a deep learning training method, so as to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of the actual target quantum dot spectral filter array.

[0135] For detailed descriptions, please refer to the above method embodiments.

[0136] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0137] This disclosure also proposes a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method. The computer-readable storage medium can be volatile or non-volatile.

[0138] This disclosure also proposes an electronic device, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0139] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0140] Figure 5 This is a block diagram illustrating a design device 1900 for a quantum dot spectral filter array according to an exemplary embodiment. For example, device 1900 can be provided as a server or terminal device. (Refer to...) Figure 5 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0141] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958 (I / O interface). Device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM macOS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0142] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.

[0143] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A design method for a quantum dot spectral filter array, characterized in that, The method includes: Based on the first spectral property data of various quantum dot materials stored in the quantum dot material library, a preset number of target quantum dot materials are determined; Obtain the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material; Based on the data mapping relationship, a simulated quantum dot spectral filter array corresponding to the target quantum dot material is determined, and an imaging mathematical model is established based on the simulated quantum dot spectral filter array; the imaging mathematical model is used to simulate the process of obtaining a response result after incident light passes through the quantum dot spectral filter array. The incident spectral data is processed based on the imaging mathematical model to obtain the simulated response results; Obtain the reconstructed spectral data corresponding to the simulation response results; The imaging mathematical model is iteratively trained using deep learning based on the difference between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of actual target quantum dot spectral filter arrays.

2. The method according to claim 1, characterized in that, The second spectral property data includes absorption spectral data; the design parameters include the concentration coefficients of each target quantum dot material; accordingly, The process of obtaining the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material includes: Acquire reference absorption spectral data of each target quantum dot material at a preset reference concentration; wherein, the concentration coefficient is used to indicate the ratio between the current concentration and the reference concentration; Determine the data mapping relationship between the absorption spectral data and the concentration coefficient and the reference absorption spectral data.

3. The method according to claim 2, characterized in that, Determining the data mapping relationship between the absorption spectral data and the concentration coefficient and the reference absorption spectral data includes: Within a preset concentration range, the absorption spectral data is determined to be the product of the concentration coefficient and the reference absorption spectral data, thus obtaining the data mapping relationship.

4. The method according to claim 2, characterized in that, The step of determining the simulated quantum dot spectral filter array corresponding to the target quantum dot material based on the data mapping relationship, and establishing an imaging mathematical model based on the simulated quantum dot spectral filter array, includes: The absorption spectral data represented by the data mapping relationship is converted into transmission spectral data based on the Lambert-Beer law; Based on the transmission spectral data of each target quantum dot material, the simulated quantum dot spectral filter array is determined. The imaging mathematical model is obtained by determining the light intensity at each pixel location of the image as the inner product between the incident spectral data and the transmitted spectral data corresponding to that pixel location.

5. The method according to claim 1, characterized in that, The method of using deep learning to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data to update the design parameters in the data mapping relationship includes: The imaging mathematical model is iteratively trained based on the difference between the incident spectral data and the reconstructed spectral data using a weight update method, and the design parameters are updated during the training process to obtain the trained imaging mathematical model.

6. The method according to claim 1, characterized in that, The first spectral property data includes transmission spectral data; correspondingly, The method determines a preset number of target quantum dot materials based on the first spectral property data corresponding to various quantum dot materials stored in the quantum dot material library, including: Based on the transmission spectral data of various quantum dot materials in the quantum dot material library, the correlation between each quantum dot material and other quantum dot materials is determined, and a preset number of target quantum dot materials with the lowest correlation are obtained.

7. The method according to claim 6, characterized in that, The process involves determining the correlation between each quantum dot material and other quantum dot materials based on the transmission spectral data of various quantum dot materials in the quantum dot material library, and obtaining a preset number of target quantum dot materials with the lowest correlation, including: A transmission spectrum matrix is ​​constructed based on the transmission spectrum data of various quantum dot materials. Each row of data in the transmission spectrum matrix represents the transmittance at the same wavelength, and each column of data represents the transmittance of the same quantum dot material at different wavelengths. The transmission spectrum matrix is ​​subjected to permutation QR decomposition to obtain the permutation matrix P of the transmission spectrum matrix; wherein, the permutation matrix P is used to permutate the order of the column data in the transmission spectrum matrix, and the correlation of the column data after the order permutation is sorted from low to high. Based on the sorting order of the column data in the permutation matrix P, the quantum dot materials corresponding to the first K columns of data are determined as the target quantum dot materials, where K is a preset number.

8. A design device for a quantum dot spectral filter array, characterized in that, The device includes: The material screening module is used to determine a preset number of target quantum dot materials based on the first spectral property data of various quantum dot materials stored in the quantum dot material library. The parameter mapping module is used to obtain the data mapping relationship between the second spectral property data of the target quantum dot material and the design parameters of the target quantum dot material; The model building module is used to determine the simulated quantum dot spectral filter array corresponding to the target quantum dot material based on the data mapping relationship, so as to establish an imaging mathematical model based on the simulated quantum dot spectral filter array; the imaging mathematical model is used to simulate the process of obtaining the response result after the incident light passes through the quantum dot spectral filter array; The simulation response module is used to process the incident spectral data based on the imaging mathematical model to obtain the simulation response results; The spectral reconstruction module is used to acquire the reconstructed spectral data corresponding to the simulation response results; The parameter acquisition module is used to iteratively train the imaging mathematical model based on the difference between the incident spectral data and the reconstructed spectral data using a deep learning training method, so as to update the design parameters in the data mapping relationship. The design parameters corresponding to the trained imaging mathematical model are used to guide the design of the actual target quantum dot spectral filter array.

9. A design device for a quantum dot spectral filter array, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the method of any one of claims 1 to 7 when executing instructions stored in the memory.

10. A non-volatile computer-readable storage medium storing computer program instructions thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

11. The use of a quantum dot spectral filter array obtained by the method of any one of claims 1-7 for spectral imaging or spatial spectral information acquisition.

Images