Simulation identification method and device for photoelectric array element parameters

Abstract

The invention discloses a simulation identification method and a simulation identification device for photoelectric array element parameters, which are used for obtaining various quality grade data of acquired images under a plurality of optical environment settings. According to various combinations of various optical parameters of the photoelectric array, a two-dimensional image sample with multiple channels is obtained through simulation. And constructing a deep simulation network with a parallel deep residual network structure. And training to obtain a deep simulation model. And through the trained deep simulation model, identifying the currently acquired image to obtain current simulation effect grade data. And according to the current simulation effect grade data, obtaining a set value which can be used for optical parameter adjustment of the photoelectric array. According to the method, a small convolution kernel multi-channel mode is adopted, the problem that a residual network avoids model degradation is solved, a double parallel parameter sharing mode is adopted, parameter calculation and depth iteration are reduced, the operation efficiency is improved, the time cost is saved, and visual operation is achieved.

Description

Technical Field

[0001] The present invention relates to the field of optoelectronic device signal processing, and specifically to a method and device for simulating and identifying parameters of optoelectronic array primitives. Background Art

[0002] The signal acquisition process of optoelectronic equipment involves many internal and external parameters. However, due to the uncertainty of the above parameters, the quality of the collected image signals varies greatly. The optoelectronic signal requires a lot of parameter adjustments during the equipment innovation process, which is time-consuming. Summary of the invention

[0003] The purpose of the present invention is to provide a method and device for simulating and identifying the parameters of an optoelectronic array element, thereby solving the problem of effectively and efficiently adjusting the parameters during the equipment experiment.

[0004] A first aspect of the present invention provides a method for simulating and identifying parameters of an optoelectronic array element, comprising: Based on the combination of multiple optical parameters of the optoelectronic array, multiple quality level data of the collected image under multiple optical environment settings are obtained. According to multiple combinations of multiple optical parameters of the optoelectronic array, two-dimensional image samples with multiple channels are simulated.

[0005] A deep simulation network with a parallel deep residual network structure is constructed. Two-dimensional image samples are used as input training data of the deep simulation network, and corresponding data of multiple quality levels are used as training target values ​​to train a deep simulation model.

[0006] The trained deep simulation model is used to identify the current captured image and obtain the current simulation effect level data. According to the current simulation effect level data, the setting values ​​that can be used for adjusting the optical parameters of the optoelectronic array are obtained.

[0007] In one embodiment of the simulation identification method of the photoelectric array element parameters of the present invention, the optical parameters of the various photoelectric arrays include: the refractive index, reflectivity, absorptivity, phase angle, distance from the center point, light intensity, light frequency, and chromaticity of the optical environment of the photoelectric element.

[0008] In another embodiment of the method for simulating and identifying the parameters of the optoelectronic array element of the present invention, the method comprises: Normalize multiple optical parameters. Use the Gram angle field to map the normalized multiple optical parameters to GADF and GASF into two-dimensional images of two channels; use the Markov transfer field to map the normalized multiple optical parameters to MTF into a two-dimensional image of a corresponding channel, thereby obtaining three channels of the two-dimensional image.

[0009] In another embodiment of the method for simulating and identifying the parameters of an optoelectronic array element of the present invention, a deep simulation network having a parallel deep residual network structure includes: An image input layer, two parallel residual network branches and a fully connected layer are arranged in sequence. Each residual network branch includes a residual module and a pooling layer arranged in sequence. The residual module includes a first residual module and a first pooling layer, a second residual module and a second pooling layer.

[0010] In another embodiment of the method for simulating and identifying the parameters of optoelectronic array primitives of the present invention, the network of the residual module is a deep residual network.

[0011] The structure of the first residual module includes multiple serial deep residual network units.

[0012] The deep residual network includes two convolutional layers, two normalization layers, and two activation function layers. The convolution kernel of the convolutional layer is 3*3 or 4*4.

[0013] The cross entropy is used as the loss function and the Adam adaptive moment estimation algorithm is used as the optimizer for training to obtain the best model.

[0014] A second aspect of the present invention provides a simulation and identification device for photoelectric array element parameters, comprising: A photoelectric array can obtain multiple quality level data of a captured image under multiple optical environment settings based on a combination of multiple optical parameters of the photoelectric array.

[0015] A training image generating unit is arranged to simulate and obtain a two-dimensional image sample with multiple channels according to multiple combinations of multiple optical parameters of the photoelectric array.

[0016] A training unit is provided for constructing a deep simulation network with a parallel deep residual network structure. A deep simulation model is obtained by training using two-dimensional image samples as input training data of the deep simulation network and corresponding data of multiple quality levels as training target values.

[0017] A recognition unit is configured to recognize the current captured image through a trained deep simulation model to obtain current simulation effect level data.

[0018] An adjustment control unit is configured to obtain setting values ​​that can be used for adjusting optical parameters of the optoelectronic array according to current simulation effect level data.

[0019] In one embodiment of the present invention, a photoelectric array is used as an optical environment to obtain an effect simulation device for collecting images, and the optical parameters of the various photoelectric arrays include: the refractive index, reflectivity, absorptivity, phase angle, distance from the center point, light intensity, light frequency, and light source chromaticity of the photoelectric elements.

[0020] In an embodiment of the present invention, in which a photoelectric array is used as an effect simulation device for obtaining a captured image in an optical environment, the training image generation unit is further configured to normalize a plurality of optical parameters.

[0021] The normalized multiple optical parameters are mapped to GADF and GASF as two-dimensional images of two channels through the Gram angle field; the normalized multiple optical parameters are mapped to MTF as a two-dimensional image of a corresponding channel through the Markov transfer field, thereby obtaining three channels of the two-dimensional image.

[0022] In one embodiment of the present invention, a photoelectric array is used as an effect simulation device for obtaining a captured image in an optical environment, and a deep simulation network having a parallel deep residual network structure includes: An image input layer, two parallel residual network branches and a fully connected layer are arranged in sequence. Each residual network branch includes a residual module and a pooling layer arranged in sequence. The residual module includes a first residual module and a first pooling layer, a second residual module and a second pooling layer.

[0023] In one embodiment of the present invention, a photoelectric array is used as an optical environment to obtain an effect simulation device for acquiring an image, and the network of the residual module is a deep residual network. The structure of the first residual module includes a plurality of serial deep residual network units. The deep residual network includes two convolutional layers, two normalization layers, and two activation function layers. The convolution kernel of the convolutional layer is 3*3 or 4*4.

[0024] The cross entropy is used as the loss function and the Adam adaptive moment estimation algorithm is used as the optimizer for training to obtain the best model.

[0025] The following will further explain the characteristics, technical features, advantages and implementation methods of the simulation identification method of optoelectronic array element parameters in a clear and understandable manner in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flow chart for illustrating a method for simulating and identifying parameters of a photoelectric array element in one embodiment of the present invention; Figure 2 is a schematic flow chart for illustrating a method for simulating and identifying parameters of a photoelectric array element in another embodiment of the present invention; Figure 3 It is a schematic diagram of the system architecture of a GAMK-resNet network in a simulation identification method of optoelectronic array primitive parameters in one embodiment of the present invention; Figure 4It is a schematic diagram of an interface after testing a simulation identification method of a photoelectric array element parameter in one embodiment of the present invention; Figure 5 It is a schematic diagram for illustrating the basic structure of a deep residual network unit in one embodiment of the present invention; Figure 6 It is a schematic diagram for explaining the Gram angle difference field conversion in one embodiment of the present invention; Figure 7 It is used to illustrate the Markov transition field transformation diagram in one embodiment of the present invention; Figure 8 It is a system composition diagram for illustrating a device for simulating the effect of acquiring an image using a photoelectric array as an optical environment in one embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to have a clearer understanding of the technical features, purposes and effects of the invention, the specific embodiments of the present invention are now described with reference to the accompanying drawings. The same reference numerals in the drawings represent components with the same structure or similar structures but the same functions.

[0028] In this article, "schematic" means "serving as an example, instance or illustration", and any diagram or implementation described as "schematic" in this article should not be interpreted as a more preferred or more advantageous technical solution. In order to make the drawings concise, each figure only schematically shows the parts related to the exemplary embodiment, and they do not represent the actual structure and true proportion of the product.

[0029] For the image signals that have been collected, the corresponding parameters are classified. According to past experimental experience, the experimental parameters are arranged in a one-dimensional array, and the sample parameter data are classified into multiple categories according to the level of image quality.

[0030] The present invention uses the optoelectronic device parameters of the collected images as historical data to perform feature extraction, and realizes the pre-determination of input parameters by constructing a neural network model. Combined with a visual front-end platform, it assists the experiment in achieving a device with good required effects.

[0031] The present invention aims to solve the problem that manual parameter adjustment in optoelectronic equipment experiments is time-consuming and inefficient. The present invention adopts an improved resNet (deep residual network) as a deep learning neural network classification tool to quickly predict a set of parameters.

[0032] The invention involves a neural network model that adopts a small convolution kernel multi-channel approach, multi-feature high-dimensional extraction of feature information, a residual network to avoid model degradation problems, and a parallel parameter sharing approach to replace the traditional serial multi-depth approach, with less parameter calculation and reduced depth iteration, thus improving computational efficiency and saving time costs.

[0033] In the invention, the one-dimensional device parameters with small correlation are mapped into three two-dimensional pseudo-color images, converted into spatial information, and the GAMK-resNet (Gram angle field (GASF / GADF) and Markov migration field (MTF) are constructed. The first Gram is GA, and the second MTF is MT, abbreviated as GAMK) network structure is used to process the pseudo-color image. The existing commonly used classification neural network resNet is in series form. Although the residual network is added to effectively ensure the problem of gradient disappearance, the operating efficiency decreases with the increase of depth. The parallel small convolution kernel in the present invention can optimize the performance in this aspect. Therefore, the combination of improved deep learning neural network and its application in the photoelectric parameter simulation system can improve the experimental efficiency and practicality.

[0034] Figure 1 It is a schematic flow chart for illustrating a method for simulating and identifying the parameters of a photoelectric array element in one embodiment of the present invention. Figure 1 As shown, the first aspect of the present invention provides a method for simulating and identifying parameters of an optoelectronic array element, which comprises: Step S101, through the photoelectric multi-phase polarization acquisition primitive array, based on the combination of multiple optical parameters of the photoelectric array, multiple quality level data of the collected images under multiple optical environment settings are obtained. In this step, the parameters that affect the experimental results are collected and formed into a one-dimensional list. The experimental parameters include multiple primitives in the photoelectric array, that is, the refractive index, reflectivity, absorptivity, phase angle, distance from the center point, light intensity, light frequency, light source chromaticity, probe gain cavity length, light sensing peripheral diameter, probe focal length, lens center deviation, detection aperture shape, indoor temperature, center wavelength, slit width, grating constant, Z distance, X distance, spacing, frequency, delay factor, incident wave , the incident wave , acquisition time, sensitivity factor, working voltage, curvature radius, aperture number, dispersion coefficient, chamfer, angle tolerance and other 32-bit parameters. The 32-bit one-dimensional key parameter values ​​summarized in previous experiments are manually divided into five categories according to the integrity and clarity of the imaging, namely excellent, good, medium, qualified, and unqualified, and folder labels are generated to store the corresponding five types of data.

[0035] The "photoelectric array" in the present invention is composed of a plurality of light-emitting and photosensitive elements arranged in an array. In actual implementation, the elements of the photoelectric array in the present invention can be directed toward or set as the backlight or area of ​​the image to be detected, thereby playing the role of collecting photoelectric parameters. The above-mentioned photoelectric array can be a light-emitting unit such as a diode, or it can be a light box structure. After the image is collected after illumination, the image is collected through a photosensitive element or a camera collection element. Various parameters can be obtained by collecting corresponding sensors. For example, the indoor temperature can be obtained by a temperature sensor, and the central wavelength can be obtained by a wavelength meter. Other parameters can be obtained by conversion.

[0036] Step S102, according to various combinations of various optical parameters of the photoelectric array, simulate and obtain a two-dimensional image sample with multiple channels. In this step, linear interpolation is performed on the data source to fit and interpolate the 32 key data, and the data is uniformly quantized and mapped to the [0,1] interval to facilitate subsequent experimental data processing.

[0037] Since many parameters are not correlated, features are difficult to extract, so the Gram angle field is used for transformation. The one-dimensional data is mapped into a two-dimensional image GADF, Gram angle summation and GASF through the Gram angle difference field, and the corresponding two-dimensional image MTF is generated through the Markov transfer field, a total of three channel images.

[0038] Step S103, constructing a deep simulation network with a parallel deep residual network structure. Using two-dimensional image samples as input training data of the deep simulation network, using corresponding multiple quality level data as training target values, and training to obtain a deep simulation model.

[0039] In this step, GAMK-ResNet is constructed, and the deep residual network is as follows Figure 3 As shown in the figure. This deep network model is based on the structure of resNet, including a two-dimensional image input module, two residual modules, two pooling layers, and finally a discriminant module output through the Softmax function, where each residual module has two in parallel. In order to extract the important features contained in the image matrix that are useful for classification, the GAMK-resNet model adopts a residual structure, each of which contains two convolutional layers, two BN layers, and two ReLU activation functions.

[0040] Use cross entropy as the loss function and the Adam algorithm (adaptive moment estimation algorithm) as the optimizer for training to obtain the best model. Use the trained model as the Python backend processing model, and use the front-end interface to import parameters and display prediction results. The overall process is as follows: Figure 2 shown.

[0041] Step S104: using the trained deep simulation model to identify the current captured image, obtain current simulation effect level data, and obtain setting values ​​that can be used to adjust the optical parameters of the optoelectronic array according to the current simulation effect level data.

[0042] The sorted sample data is batch processed and sent to the model for iterative training. The obtained model is used as the basis for back-end classification. The parameters are uploaded through the Web end, and the underlying Python is used to predict the classification results. The interface is as follows Figure 4 shown.

[0043] The beneficial effects of the present invention are: First, a small number of one-dimensional parameters are interpolated and filled with key points to map them into a two-dimensional image format that the resNet residual network is good at processing.

[0044] Secondly, the parameter signal is converted into GADF Gram angle summation and GASF through the Gram angle difference field, and the Markov transfer field is mapped into MTF, a total of three channels of two-dimensional signal data.

[0045] Afterwards, the traditional resNet was improved to construct a neural network combination mode, using a residual structure that reduces the serial connection direction, adding a parallel residual structure, and sharing weights, which greatly improved the operating efficiency.

[0046] Again, the trained model is applied to the visualization end to present a visualization operation device, which provides a low threshold and convenience for the operator.

[0047] In one embodiment of the simulation identification method of the photoelectric array element parameters of the present invention, the optical parameters of the various photoelectric arrays include: the refractive index, reflectivity, absorptivity, phase angle, distance from the center point, light intensity, light frequency, and chromaticity of the optical environment of the photoelectric element.

[0048] In another embodiment of the method for simulating and identifying the parameters of the optoelectronic array element of the present invention, the method comprises: In one embodiment of the method for simulating and identifying the parameters of the optoelectronic array element of the present invention, a plurality of optical parameters are normalized, and a one-dimensional parameter is Normalize it and assume that The normalized expression of experimental parameters is shown in formula 1: Formula 1 Mode In represents each one-dimensional parameter.

[0049] The normalized multiple optical parameters are mapped into two-dimensional images of two channels by using the Gram angle field. Figure 6 As shown, Figure 6After normalizing the parameters of the 32 key points, a two-dimensional image of the Gram angle difference field is mapped. The normalized multiple optical parameters are mapped to the MTF of a corresponding channel through the Markov transfer field, such as Figure 7 As shown, Figure 7 After the parameters of the 32 key points are normalized, they are mapped into a two-dimensional image of the Markov transition field, thereby obtaining three channels of the two-dimensional image.

[0050] The normalized parameters are converted into GADF and GASF two-dimensional pseudo-color images and Markov transfer field pseudo-color images through the Gram angular difference field. The definitions of the Gram sum angular field and the Gram differential angular field are shown in formulas 2 and 3: Formula 2 Formula 3 The matrix expressions of GASF and GADF are shown in formulas 4 and 5: Formula 4 Formula 5 The Markov transfer field pseudo-color image processing process discretizes the signal value into Equal parts, of which is the number of parameters, It is The number of equal divisions of the parameters, each Each corresponds to one , through the Markov chain calculation principle, the signal point is The transition probability between The matrix . yes The elements in follow The probability of an element in . The matrix is ​​shown in Formula 6: Formula 6 In order to fully display the detailed characteristics of the time series, it is expanded to MTF, as shown in Formula 7: Formula 7 in, yes The elements in The probability that an element in follows

[0051] The converted pseudo-color images are divided into five categories, namely excellent, good, medium, qualified, and poor. Two folders are set up as training set and test set, with a ratio of 8:2.

[0052] In another embodiment of the method for simulating and identifying the parameters of an optoelectronic array element of the present invention, a deep simulation network having a parallel deep residual network structure includes: An image input layer, two parallel residual network branches and a fully connected layer are arranged in sequence. Each residual network branch includes a residual module and a pooling layer arranged in sequence. The residual module includes a first residual module and a first pooling layer, a second residual module and a second pooling layer.

[0053] In another embodiment of the method for simulating the effect of acquiring an image by using the photoelectric array 101 as an optical environment of the present invention, the network of the residual module is a deep residual network.

[0054] The structure of the first residual module includes multiple serial deep residual network units. The deep residual network includes two convolutional layers, two normalization layers and two activation function layers. The convolution kernel of the convolutional layer is 3*3 or 4*4. Figure 5 Shown is the basic structure of a deep residual network unit.

[0055] In one embodiment of the present invention, GAMK-resNet is constructed. The deep network model is based on resNet. The basic units are as follows: Figure 3 As shown in the figure. The overall structure includes a two-dimensional image input module, two residual modules, two pooling layers, and finally a discriminant module output through the Softmax function, where each residual module is parallel to each other. In order to extract the important features contained in the image matrix that are useful for classification, the GAMK-resNet model adopts a residual structure. Each residual structure contains two convolutional layers, two BN layers, and two ReLU activation functions. The GAMK-resNet in this article uses the Adam algorithm as the optimizer for training, setting the learning rate to 0.001 and the exponential decay rate to , They are 0.9 and 0.999 respectively.

[0056] In the residual structure of this article, the ReLU module is used after the BN module. The ReLU activation function can prevent the gradient from disappearing during operation. The ReLU activation function is shown in Formula 8: Formula 8 By changing all negative values ​​to 0 and keeping the positive values ​​unchanged, the neurons in the neural network have sparse activation, and the gradient of the nonlinear function in the non-negative interval can be kept constant, thus effectively solving the problem of gradient disappearance.

[0057] After the network undergoes convolution operation feature extraction, the final result is output by the Softmax function, as shown in Formula 9: Formula 9 in, Refers to the data after feature extraction and the The correlation between the categories can be calculated by the Softmax function. Probability of category.

[0058] The loss function uses the cross entropy function. The cross entropy function is the distance between the actual output probability and the expected output probability. That is to say, the smaller the cross entropy value, the closer the two probability distributions are. The cross entropy function is shown in formula 10. When the number of model iterations is set to 100, it can be seen that the accuracy rate finally reaches about 90%, and the loss value is about 0.2. Formula 10 in, for The actual output, The model predicts The probability of each category is distributed as follows: The predicted output is is the cross entropy loss function.

[0059] like Figure 8 As shown, the second aspect of the present invention provides an effect simulation device for obtaining a captured image using a photoelectric array 101 as a light source, which includes: An optoelectronic array 101 can obtain various quality level data of the collected images under multiple optical environment settings based on the combination of multiple optical parameters of the optoelectronic array 101.

[0060] A training image generating unit 201 is configured to simulate and obtain a two-dimensional image sample with multiple channels according to various combinations of various optical parameters of the optoelectronic array 101 .

[0061] A training unit 301 is configured to construct a deep simulation network with a parallel deep residual network structure, use two-dimensional image samples as input training data of the deep simulation network, use corresponding multiple quality level data as training target values, and train to obtain a deep simulation model.

[0062] An identification unit 401 is configured to identify the current captured image through a trained deep simulation model to obtain current simulation effect level data.

[0063] An adjustment control unit 501 is configured to obtain setting values ​​that can be used to adjust the optical parameters of the optoelectronic array 101 according to the current simulation effect level data.

[0064] In one embodiment of the present invention, the photoelectric array 101 is used as an optical environment to obtain an effect simulation device for collecting images, and the multiple optical parameters of the photoelectric array 101 include: the refractive index, reflectivity, absorption rate, phase angle, distance from the center point, light intensity, light frequency, and light source chromaticity of the photoelectric element.

[0065] In an embodiment of the present invention, the photoelectric array 101 is used as an effect simulation device for obtaining a captured image in an optical environment, and the training image generation unit 201 is further configured to normalize a plurality of optical parameters.

[0066] The normalized multiple optical parameters are mapped to GADF and GASF as two-dimensional images of two channels through the Gram angle field; the normalized multiple optical parameters are mapped to MTF as a two-dimensional image of a corresponding channel through the Markov transfer field, thereby obtaining three channels of the two-dimensional image.

[0067] In one embodiment of the present invention, the photoelectric array 101 is used as an optical environment to obtain an effect simulation device for collecting images, and the deep simulation network with a parallel deep residual network structure includes: An image input layer, two parallel residual network branches and a fully connected layer are arranged in sequence. Each residual network branch includes a residual module and a pooling layer arranged in sequence. The residual module includes a first residual module and a first pooling layer, a second residual module and a second pooling layer.

[0068] In one embodiment of the present invention, the photoelectric array 101 is used as an optical environment to obtain an effect simulation device for collecting images, and the network of the residual module is a deep residual network. The structure of the first residual module includes a plurality of serial deep residual network units. The deep residual network includes two convolutional layers, two normalization layers, and two activation function layers. The convolution kernel of the convolutional layer is 3*3 or 4*4.

[0069] The present invention relates to a method and device for simulating and identifying the parameters of a photoelectric array element. In order to solve the problems that it is difficult to objectively and quantitatively evaluate the multiple parameters of a photoelectric element device in a photoelectric device device experiment, and that the actual experiment is time-consuming and inefficient, the present invention reduces the time cost, improves the experimental efficiency, and increases the predictability of the experiment by inventing the device.

[0070] In the present invention, the experimental parameters are mapped into a two-dimensional pseudo-color image as one-dimensional multiple key parameter signals, and mapped into three images using the Gram angle difference constant and the Markov transfer field, and a parallel two-channel neural network model is constructed. Through the residual network, the five types of experimental data are feature extracted and integrated, and a large number of samples are trained to obtain a training model, and a prediction classification function is provided for a set of data input by the experimenter. The invention involves a neural network model using a small convolution kernel multi-channel method, multi-feature high-dimensional feature information extraction, a residual network to avoid model degradation problems, and a dual parallel parameter sharing method to replace the traditional serial multi-depth method, less parameter calculation and reduced depth iteration, improved operation efficiency, and saved time cost. The invention is combined with a front-end browser to achieve visual operation.

[0071] It should be understood that although this specification is described according to various implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0072] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. Any equivalent implementation methods or changes that do not deviate from the technical spirit of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for simulating and identifying the parameters of an optoelectronic array element, characterized in that: It includes: Based on the combination of multiple optical parameters of the optoelectronic array, multiple quality level data of the collected images under multiple optical environment settings are obtained; According to various combinations of various optical parameters of the optoelectronic array, a two-dimensional image sample with multiple channels is simulated; Construct a deep simulation network with a parallel deep residual network structure; Using the two-dimensional image samples as input training data of the deep simulation network and using the corresponding multiple quality level data as training target values, training to obtain a deep simulation model; Using the trained deep simulation model, the current captured image is identified to obtain current simulation effect level data; According to the current simulation effect level data, setting values ​​that can be used for adjusting optical parameters of the optoelectronic array are obtained.

2. The method for simulating and identifying the parameters of an optoelectronic array element according to claim 1, characterized in that: The various optical parameters of the photoelectric array include: refractive index, reflectivity, absorptivity, phase angle, distance from the center point, light intensity, light frequency, and chromaticity of the optical environment of the photoelectric element.

3. The method for simulating and identifying the parameters of an optoelectronic array element according to claim 1, characterized in that: include: normalizing the plurality of optical parameters; The normalized multiple optical parameters are mapped to GADF and GASF as two-dimensional images of two channels through the Gram angle field; the normalized multiple optical parameters are mapped to MTF as a two-dimensional image of a corresponding channel through the Markov transfer field, thereby obtaining three channels of the two-dimensional image.

4. The method for simulating and identifying the parameters of an optoelectronic array element according to claim 1, characterized in that: The deep simulation network with a parallel deep residual network structure includes: The image input layer, two parallel residual network branches and a fully connected layer are set up in sequence; Each of the residual network branches includes a residual module and a pooling layer arranged in sequence; The residual module includes a first residual module and a first pooling layer, a second residual module and a second pooling layer.

5. The method for simulating and identifying the parameters of the optoelectronic array element according to claim 4, characterized in that: The network of the residual module is a deep residual network; The structure of the first residual module includes a plurality of serial deep residual network units; The deep residual network includes two convolutional layers, two normalization layers and two activation function layers; the convolution kernel of the convolutional layer is 3*3 or 4*4; The cross entropy is used as the loss function and the Adam adaptive moment estimation algorithm is used as the optimizer for training to obtain the best model.

6. A device for simulating and identifying the parameters of an optoelectronic array element, characterized in that: It includes: An optoelectronic array capable of obtaining multiple quality level data of a captured image under multiple optical environment settings based on a combination of multiple optical parameters of the optoelectronic array; a training image generating unit, which is configured to simulate and obtain a two-dimensional image sample with multiple channels according to multiple combinations of multiple optical parameters of the photoelectric array; A training unit configured to construct a deep simulation network having a parallel deep residual network structure; Using the two-dimensional image samples as input training data of the deep simulation network and using the corresponding multiple quality level data as training target values, training to obtain a deep simulation model; An identification unit, which is configured to identify the current captured image through the trained deep simulation model and obtain current simulation effect level data; An adjustment control unit is configured to obtain setting values ​​that can be used for adjusting the optical parameters of the optoelectronic array according to the current simulation effect level data.

7. The device for simulating and identifying the parameters of the optoelectronic array element according to claim 6, characterized in that: The various optical parameters of the photoelectric array include: refractive index, reflectivity, absorptivity, phase angle, distance from the center point, light intensity, light frequency, and light source chromaticity of the photoelectric element.

8. The device for simulating and identifying the parameters of the optoelectronic array element according to claim 6, characterized in that: The training image generating unit is further configured to normalize the multiple optical parameters; Mapping the normalized multiple optical parameters GADF and GASF into two channels of a two-dimensional image through a Gram angular field; The normalized multiple optical parameters are mapped to MTF as a channel of a corresponding two-dimensional image through a Markov transfer field, thereby obtaining three channels of the two-dimensional image.

9. The device for simulating and identifying the parameters of the optoelectronic array element according to claim 6, characterized in that: The deep simulation network with a parallel deep residual network structure includes: The image input layer, two parallel residual network branches and a fully connected layer are set up in sequence; Each of the residual network branches includes a residual module and a pooling layer arranged in sequence; The residual module includes a first residual module and a first pooling layer, a second residual module and a second pooling layer.

10. The device for simulating and identifying the parameters of the optoelectronic array element according to claim 9, characterized in that: The network of the residual module is a deep residual network; The structure of the first residual module includes a plurality of serial deep residual network units; The deep residual network includes two convolutional layers, two normalization layers and two activation function layers; the convolution kernel of the convolutional layer is 3*3 or 4*4; The cross entropy was used as the loss function and the Adam adaptive moment estimation algorithm was used as the optimizer for training to obtain the best model.

Images