Method for acquiring optical parameters of detection object and related device

Abstract

According to the method for obtaining the optical parameters of the detection object, the optical parameters of the detection object can be automatically generated based on the ncs spectral image of the detection object, and therefore the convenience from the ncs spectral image to optical parameter output is improved. The method provided by the embodiment of the invention comprises the following steps: acquiring an ncs spectral image of a detection object, wherein the ncs spectral image represents projection images of the spectral image of the detection object in three mutually orthogonal directions; multiple sets of optical parameter values corresponding to the ncs spectral image of the detection object are obtained, and each set of optical parameters comprises at least two types of optical parameters; inputting the plurality of groups of optical parameter values into a first neural network model to obtain a plurality of predicted ncs spectral images output by the first neural network model; determining a target ncs spectral image having the minimum error with the ncs spectral image of the detection object in the plurality of predicted ncs spectral images; a target optical parameter set corresponding to the target ncs spectral image is regarded as optical parameters of a detection object.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for acquiring optical parameters of a detection object and a related device. Background Art

[0002] When the detection device detects the detection object, the detection device can be used to perform optical imaging on the surface of the detection object, and the computer can perform image processing on the detected optical image and extract the detection image. The image data after the image processing can be stored in the computer. The above processes have been implemented in the detection device, and the spectral image of the detection object can be quickly displayed in the computer.

[0003] Generally, the spectral image of each detection object corresponds to a set of optical parameters of the detection object, such as the thickness, top width, bottom width, optical path angle and azimuth angle of the detection object. However, after obtaining the spectral image of each detection object, the prior art cannot automatically output the spectral image of the detection object to optical parameters. Summary of the invention

[0004] The embodiment of the present invention provides a method for obtaining optical parameters of a detection object, which can automatically generate the optical parameters of the detection object based on the NCS spectrum image of the detection object, thereby improving the convenience of outputting optical parameters from the NCS spectrum image.

[0005] A first aspect of an embodiment of the present application provides a method for obtaining optical parameters of a detection object, comprising:

[0006] Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0007] Acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters;

[0008] Inputting the plurality of sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0009] Determine, among the multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object;

[0010] The target optical parameter group corresponding to the target NCS spectrum image is regarded as the optical parameters of the detection object.

[0011] Preferably, determining a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images comprises:

[0012] A regression analysis is performed on the multiple predicted NCs spectrum images and the NCs spectrum image of the detection object to obtain a target NCs spectrum image with the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images.

[0013] Preferably, before inputting the plurality of sets of optical parameter values ​​into the first neural network model, the method further comprises:

[0014] Acquire a first training sample, wherein the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images;

[0015] Inputting a plurality of NCS spectral images of the plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0016] According to a preset loss function, calculating the loss between a plurality of NCs spectral images of a plurality of detection objects and the plurality of predicted NCs spectral images;

[0017] The first neural network model is trained according to the loss and the back propagation algorithm until the loss reaches a target loss.

[0018] Preferably, the first neural network model comprises: a convolution module, a normalization layer, a discard layer and a fully connected layer;

[0019] Inputting the plurality of sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model, including:

[0020] Inputting the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values;

[0021] Inputting the data features into the normalization layer to normalize the data features;

[0022] Inputting the normalized data features into the discard layer to desparse the normalized data features;

[0023] The de-sparse data features are input into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

[0024] A second aspect of an embodiment of the present application provides a method for obtaining optical parameters of a detection object, comprising:

[0025] Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0026] The NCS spectrum image is input into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into a set of optical parameters of the detection object.

[0027] Preferably, before inputting the NCS spectrum image into the second neural network model, the method further comprises:

[0028] Acquire second training sample data, wherein the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image;

[0029] The second training sample data is input into the second neural network model for training until the output of the second neural network model meets the target loss.

[0030] Preferably, inputting the second training sample data into the second neural network model for training includes:

[0031] Inputting a plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the second neural network model to obtain a plurality of sets of predicted optical parameters output by the second neural network model;

[0032] Calculating the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters according to a preset loss function;

[0033] The second neural network model is trained according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

[0034] Preferably, inputting the second training sample data into the second neural network model for training includes:

[0035] Deploy one of the second neural network models on each of the n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2;

[0036] Dividing the training sample data into n groups of sample data on average;

[0037] Using a set of sample data on each working node, updating the parameters of the second neural network model;

[0038] After updating the parameters of the second neural network model m times on each working node, n updated second neural network models are obtained, 5≤m≤15;

[0039] Calculate the average values ​​of the model parameters of the n updated second neural network models;

[0040] The average values ​​of the model parameters are used as the final model parameters of the second neural network model.

[0041] Preferably, the second neural network model includes a convolution module, a discard layer, a fully connected layer and a normalization layer;

[0042] Inputting the NCS spectrum image into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, including:

[0043] Inputting the ncs spectrum image into the convolution module to extract image features of the ncs spectrum image;

[0044] Inputting the image features into the dropout layer to perform de-sparseness on the image features;

[0045] Inputting the de-sparsed image features into the fully connected layer to fit the de-sparsed image features;

[0046] The fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0047] A third aspect of an embodiment of the present application provides a device for obtaining optical parameters of a detection object, including:

[0048] A first acquisition unit is used to acquire an NCs spectrum image of the detection object, wherein the NCs spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0049] The first acquisition unit is further used to acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters;

[0050] A first input-output unit is used to input the multiple sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0051] A first determining unit is used to determine, among the multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object;

[0052] The first determining unit is further configured to regard the target optical parameter group corresponding to the target NCS spectrum image as the optical parameter group of the detection object.

[0053] Preferably, the first determining unit is specifically configured to:

[0054] A regression analysis is performed on the multiple predicted NCs spectrum images and the NCs spectrum image of the detection object to obtain a target NCs spectrum image with the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images.

[0055] Preferably, the first acquiring unit is further used for:

[0056] Acquire a first training sample, wherein the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images;

[0057] The first input-output unit is further used to input the multiple NCS spectral images of the multiple detection objects and the multiple sets of optical parameters corresponding to the multiple NCS spectral images into the first neural network model to obtain the multiple predicted NCS spectral images output by the first neural network model;

[0058] Preferably, the device further comprises:

[0059] A calculation unit, used for calculating the loss between a plurality of NCs spectral images of a plurality of detection objects and the plurality of predicted NCs spectral images according to a preset loss function;

[0060] A training unit is used to train the first neural network model according to the loss and a back propagation algorithm until the loss reaches a target loss.

[0061] Preferably, the first neural network model comprises: a convolution module, a normalization layer, a discard layer and a fully connected layer;

[0062] The first input-output unit is specifically used for:

[0063] Inputting the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values;

[0064] Inputting the data features into the normalization layer to normalize the data features;

[0065] Inputting the normalized data features into the discard layer to desparse the normalized data features;

[0066] The de-sparse data features are input into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

[0067] A fourth aspect of an embodiment of the present application provides a device for obtaining optical parameters of a detection object, including:

[0068] A second acquisition unit is used to acquire an NCs spectrum image of the detection object, wherein the NCs spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0069] The second input-output unit is used to input the NCS spectrum image into the second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into the optical parameter group of the detection object.

[0070] Preferably, the second acquisition unit is further used for:

[0071] Acquire second training sample data, wherein the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image;

[0072] The device also includes:

[0073] A training unit is used to input the second training sample data into the second neural network model for training until the output of the second neural network model meets the target loss.

[0074] Preferably, the training unit is specifically used for:

[0075] Inputting a plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the second neural network model to obtain a plurality of sets of predicted optical parameters output by the second neural network model;

[0076] Calculating the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters according to a preset loss function;

[0077] The second neural network model is trained according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

[0078] Preferably, the training unit is specifically used for:

[0079] Deploy one of the second neural network models on each of the n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2;

[0080] Dividing the training sample data into n groups of sample data on average;

[0081] Using a set of sample data on each working node, updating the parameters of the second neural network model;

[0082] After updating the parameters of the second neural network model m times on each working node, n updated second neural network models are obtained, 5≤m≤15;

[0083] Calculate the average values ​​of the model parameters of the n updated second neural network models;

[0084] The average values ​​of the model parameters are used as the final model parameters of the second neural network model.

[0085] Preferably, the second neural network model includes a convolution module, a discard layer, a fully connected layer and a normalization layer;

[0086] The second input and output unit is specifically used for:

[0087] Inputting the ncs spectrum image into the convolution module to extract image features of the ncs spectrum image;

[0088] Inputting the image features into the dropout layer to perform de-sparseness on the image features;

[0089] Inputting the de-sparsed image features into the fully connected layer to fit the de-sparsed image features;

[0090] The fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0091] A fifth aspect of an embodiment of the present application provides a computer device, including a processor, which, when executing a computer program stored in a memory, is used to implement the method for obtaining optical parameters of a detection object provided in the first aspect or the second aspect of the embodiment of the present application.

[0092] A sixth aspect of an embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program stored in the memory is executed, the computer program is used to implement the method for obtaining optical parameters of a detection object provided in the first aspect or the second aspect of the embodiment of the present application.

[0093] It can be seen from the above technical solutions that the embodiments of the present invention have the following advantages:

[0094] In the embodiment of the present application, the optical parameters corresponding to the NCS spectrum image of the detection object are determined by combining a neural network model with a regression analysis method, thereby realizing the automatic output of the NCS spectrum image of the detection object to the optical parameters of the detection object. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 A schematic diagram of an embodiment of a method for obtaining optical parameters of a detection object in an embodiment of the present application;

[0096] Figure 2 for Figure 1 A refinement of step 103 in the embodiment;

[0097] Figure 3 This is a schematic diagram of the structure of the first neural network model in the embodiment of the present application;

[0098] Figure 4 This is a schematic diagram of an embodiment of training a first neural network model in an embodiment of the present application;

[0099] Figure 5 A schematic diagram of an embodiment of a method for obtaining optical parameters of a detection object in an embodiment of the present application;

[0100] Figure 6 For this application Figure 5 A refinement of step 502 in the embodiment;

[0101] Figure 7 This is a schematic diagram of the structure of the second neural network model in the embodiment of the present application;

[0102] Figure 8 This is a schematic diagram of an embodiment of training a second neural network model in an embodiment of the present application;

[0103] Fig. 9 This is another schematic diagram of an embodiment of training a second neural network model in an embodiment of the present application;

[0104] Fig.10 A schematic diagram of an embodiment of a device for obtaining optical parameters of a detection object in an embodiment of the present application;

[0105] Fig.11 Schematic diagram of another embodiment of the device for obtaining optical parameters of a detection object in an embodiment of the present application. DETAILED DESCRIPTION

[0106] The embodiment of the present invention provides a method for obtaining optical parameters of a detection object, which can automatically generate the optical parameters of the detection object based on the NCS spectrum image of the detection object, thereby improving the convenience of outputting optical parameters from the NCS spectrum image.

[0107] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0108] The terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0109] For ease of understanding, the following describes the method for obtaining the optical parameters of the detection object in the embodiment of the present application. Figure 1 , an embodiment of the method for obtaining optical parameters of a detection object in an embodiment of the present application includes:

[0110] 101. Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0111] The embodiment of the present application can obtain the NCS spectral image of the detection object through a detection device (such as a photodetector or a photomultiplier tube), wherein the NCS spectral image represents the projection images of the spectral image of the detection object in three mutually orthogonal directions. For example, when the detection object is a wafer, the NCS spectral image of the detection object represents the projection images of the spectral image of the wafer in the x-, y- and z-axis directions.

[0112] That is to say, the NCS spectrum image of the detection object includes at least three projection images.

[0113] 102. Acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters;

[0114] After obtaining the NCS spectrum image of the detection object, multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object are obtained, wherein each group of optical parameters includes at least two types of optical parameters, such as at least two of the thickness d, bottom width x1, top width x2 and azimuth angle θ of the detection object.

[0115] It should be noted here that the multiple groups of optical parameter values ​​here are not the multiple groups of optical parameter values ​​actually measured, but the multiple groups of optical parameter values ​​generated according to the NCS spectrum image and the generating function of the detection object. For example, the first group of optical parameter values ​​include {d=0.5, x1=1, x1=2, θ=30°}, the second group of optical parameter values ​​include {d=0.6, x1=2, x1=3, θ=40°}..... the mth group of optical parameter values ​​includes {d=0.9, x1=1, x1=3, θ=60°}, etc.

[0116] 103. Input the multiple sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectrum images output by the first neural network model;

[0117] After obtaining a plurality of sets of optical parameter values ​​corresponding to the NCs spectrum images of the detection object in step 102, the plurality of sets of optical parameter values ​​are input into the first neural network model to obtain a plurality of predicted NCs spectrum images output by the first neural network model.

[0118] The first neural network model here is used to generate an NCS spectrum image based on the optical parameter value. As for the specific model structure of the first neural network model and the process of generating the NCS spectrum image by the first neural network model, they will be described in the following embodiments and will not be repeated here.

[0119] 104. Determine, among the multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object;

[0120] After obtaining multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object is determined among the multiple predicted NCs spectrum images, that is, a target NCs spectrum image most similar to the NCs spectrum image of the detection object is determined.

[0121] Specifically, when the embodiment of the present application determines the target NCS spectral image that is most similar to the NCS spectral image of the detection object from multiple predicted NCS spectral images, the determination can be made based on a regression analysis algorithm, wherein the regression analysis algorithm includes a linear regression analysis algorithm, a ridge regression analysis algorithm, or a polynomial regression analysis algorithm, etc., and the specific type of the regression analysis algorithm is not limited here.

[0122] 105. Consider the target optical parameter group corresponding to the target NCS spectrum image as the optical parameters of the detection object.

[0123] After the target NCS spectrum image is obtained, the target optical parameter group corresponding to the target NCS spectrum image is regarded as the optical parameters of the detection object. For example, assuming that the target optical parameter group is the second group of optical parameters {d=0.6, x1=2, x1=3, θ=40°} in step 102, the optical parameters of this group are regarded as the optical parameters of the detection object.

[0124] In the embodiment of the present application, the optical parameters corresponding to the NCS spectrum image of the detection object are determined by combining a neural network model with a regression analysis method, thereby realizing the automatic output of the NCS spectrum image of the detection object to the optical parameters of the detection object.

[0125] based on Figure 1 The embodiment described above, step 103 is described in detail below. Figure 2 , Figure 2 The detailed steps of step 103 are as follows:

[0126] 201. Input the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values;

[0127] The first neural network model in the embodiment of the present application includes a convolution module, a normalization layer, a dropout layer and a fully connected layer. For ease of understanding, Figure 3 A structural diagram of the first neural network model is given.

[0128] Among them, after multiple sets of optical parameter values ​​are input into the convolution module, data features of the multiple sets of optical parameter values, such as data distribution, data type, etc., can be extracted.

[0129] 202. Input the data features to the normalization layer to normalize the data features;

[0130] In order to avoid data features that are too large or too small being ignored in the model output, a normalization layer is added to the first neural network model in the embodiment of the present application to normalize the data features, thereby ensuring that all data features can be taken into account in the model output.

[0131] 203. Input the normalized data features to the discard layer to desparse the normalized data features;

[0132] After obtaining the normalized data features, the data features are further input into a discard layer, wherein the function of the discard layer is to perform desparseness on the normalized data features.

[0133] Because the normalized data features include multiple zero values, and the zero value does not play an effective role in the calculation process, the embodiment of the present application de-sparsens the normalized data features, and the de-sparseness here means deleting the zero value in the normalized data features.

[0134] 204. Input the de-sparse data features into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

[0135] The user who obtains the de-sparse data features further inputs the de-sparse data features into the fully connected layer, because the fully connected layer can fit the de-sparse data features, that is, integrate the de-sparse data features to obtain multiple predicted NCS spectra.

[0136] The embodiment of the present application makes a detailed description of the structure of the first neural network model, and the first neural network model includes a normalization layer and a discard layer. On the one hand, the normalization layer can make the extracted data features be taken into account in the model output process, thereby improving the accuracy of the model output. On the other hand, the discard layer can delete invalid data in the normalized data features, reduce the amount of model calculations, and improve the efficiency of the model output.

[0137] based on Figure 1 In the embodiment described above, the training process of the first neural network model is described below. Figure 4 , an embodiment of the training process of the first neural network model in the embodiment of the present application includes:

[0138] 401. Obtain a first training sample, where the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images;

[0139] In order to train the first neural network model, it is first necessary to obtain a first training sample, wherein the first training sample includes a plurality of NCS spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCS spectral images.

[0140] If the multiple inspection objects include multiple wafers, the multiple NCS spectral images of the multiple inspection objects and the multiple sets of optical parameters corresponding to the multiple NCS spectral images include: multiple NCS spectral images of multiple wafers, and a set of optical parameters corresponding to the NCS spectral image of each wafer.

[0141] Assuming that m1 wafers are included, the first training sample data includes m1 NCS spectral images and m1 groups of optical parameters, and each NCS spectral image corresponds to a group of optical parameters, wherein the optical parameters in the same group include at least two types of optical parameters, such as at least two of thickness, bottom width, top width and azimuth.

[0142] 402. Input a plurality of NCS spectral images of the plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0143] A plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images are input into the first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model.

[0144] 403. Calculate the loss between the multiple NCs spectrum images of the multiple detection objects and the multiple predicted NCs spectrum images according to a preset loss function;

[0145] Then, according to a preset loss function, the loss between the multiple NCS spectral images of the multiple detection objects and the multiple predicted NCS spectral images is calculated, wherein the preset loss function includes L1, L2 or cross entropy loss function, etc., and the type of loss function is not specifically limited here.

[0146] 404. Train the first neural network model according to the loss and the back propagation algorithm until the loss reaches a target loss.

[0147] After the loss of the first neural network model is obtained, the parameters of the first neural network model are further trained according to the loss and back propagation algorithm until the loss meets the target loss.

[0148] Specifically, the process of training the first neural network model is the process of continuously updating the model parameters of the first neural network model. When updating the model parameters of the first neural network model, in order to prevent gradient explosion and improve the gradient convergence speed, the present application can also use the Adamw optimizer to optimize the learning rate of parameter update when updating the parameters of the first neural network model, which has at least the following advantages:

[0149] 1. Automatically adjust the learning rate of different parameters of the model to better adapt to the changes in the gradients of different parameters;

[0150] 2. The training speed is relatively faster, and it can also better handle sparse gradients;

[0151] 3. The Adamw optimizer performs L2 regularization on larger weight updates, which can further improve the training effect of the high model.

[0152] The structure of the Adamw optimizer itself is consistent with that described in the prior art and will not be described in detail here.

[0153] In an embodiment of the present application, the training process of the first neural network model is described in detail, and when updating the parameters of the first neural network model, the present application uses the Adamw optimizer to optimize the learning rate of the parameter update, so that the training speed of the first neural network model is faster and the training effect is better.

[0154] Apart from Figures 1 to 4 In addition to the above method for obtaining the optical parameters of the detection object, another method can be used to obtain the optical parameters of the detection object. Figure 5 In the embodiment of the present application, an embodiment of obtaining the optical parameters of the detection object includes:

[0155] 501. Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0156] The embodiment of the present application can obtain the NCS spectral image of the detection object through a detection device (such as a photodetector or a photomultiplier tube), wherein the NCS spectral image represents the projection images of the spectral image of the detection object in three mutually orthogonal directions. For example, when the detection object is a wafer, the NCS spectral image of the detection object represents the projection images of the spectral image of the wafer in the x-, y- and z-axis directions.

[0157] That is to say, the NCS spectrum image of the detection object includes at least three projection images.

[0158] 502. Input the NCS spectrum image into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into a set of optical parameters of the detection object.

[0159] Different from Figures 1 to 4The first neural network model in generates an NCS spectrum image of the detection object according to the optical parameters of the detection object. The second neural network model in the embodiment of the present application can directly generate the optical parameters of the detection object according to the NCS spectrum image of the detection object. Therefore, after obtaining the NCS spectrum image of the detection object, the embodiment of the present application inputs the spectrum image into the second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model.

[0160] As for the structure of the second neural network model and the process of how the second neural network model outputs the NCS spectrum image of the detection object into corresponding optical parameters, it will be described in the following embodiments and will not be repeated here.

[0161] In the embodiment of the present application, the NCS spectrum image of the detection object can be converted into corresponding optical parameters according to the second neural network model, thereby improving the convenience of obtaining the optical parameters of the detection object.

[0162] based on Figure 6 In the embodiment described above, step 502 is described in detail below. Figure 6 ,Book Figure 6 The detailed steps of step 502 are as follows:

[0163] 601. Input the NCS spectrum image into the convolution module to extract image features of the NCS spectrum image;

[0164] The second neural network model in this application includes a convolution module, a dropout layer, a fully connected layer, and a normalization layer. For ease of understanding, Figure 7 A structural diagram of the second neural network model is given.

[0165] Wherein, after the NCS spectrum image is input into the convolution module, the image features of the NCS spectrum image can be extracted through the convolution module.

[0166] 602. Input the image features to the discard layer to perform de-sparseness on the image features;

[0167] Because there may be some zero values ​​in the image features, and these zero values ​​are invalid data for the second neural network model, the application embodiment inputs the image features into the discard layer to perform de-sparseness on the image features, wherein the de-sparseness operation is to delete the zero values ​​in the image features.

[0168] 603. Inputting the de-sparsened image features into the fully connected layer to fit the de-sparsened image features;

[0169] After desparsening the image features, the desparsened image features are input into the fully connected layer to fit the desparsed image features, that is, the desparsed image features are integrated.

[0170] 604. Input the fitted image features into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0171] Finally, the fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0172] The embodiment of the present application makes a detailed description of the structure of the second neural network model, and also makes a detailed description of the process of the second neural network model outputting the optical parameters of the detection object. In addition, a discard layer is set in the second neural network model structure to delete invalid data in the data features, thereby reducing the amount of calculation of the model and improving the efficiency and convenience of the second neural network model in outputting the optical parameters of the detection object.

[0173] based on Figure 1 In the embodiment described above, before the NCS spectrum image of the detected object is input into the second neural network model, the second neural network model needs to be trained. The training process of the second neural network model is described below. Figure 8 , an embodiment of the training process of the second neural network model in the embodiment of the present application includes:

[0174] 801. Acquire second training sample data, where the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image;

[0175] Before training the second neural network model, it is necessary to obtain training sample data, which is different from Figures 1 to 4 The training sample data for the first neural network model in the present application embodiment is referred to as the training sample data for training the second neural network model as the second training sample data.

[0176] Specifically, the second training sample data in the embodiment of the present application includes: an NCS spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCS spectrum image.

[0177] 802. Input a plurality of NCS spectrum images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCS spectrum images into the second neural network model to obtain a plurality of groups of predicted optical parameters output by the second neural network model;

[0178] Multiple NCS spectral images of multiple detection objects and multiple sets of optical parameters corresponding to the multiple NCS spectral images are input into the second neural network model to obtain multiple sets of predicted optical parameters output by the second neural network model, wherein the second neural network model is used to generate corresponding optical parameters from the NCS spectral images of the detection objects.

[0179] The process of generating multiple sets of predicted optical parameters by the second neural network model is similar to Figure 6 The description in the embodiment is similar and will not be repeated here.

[0180] 803. Calculate the losses between the multiple groups of optical parameters of the multiple detection objects and the multiple groups of predicted optical parameters according to a preset loss function;

[0181] After obtaining the multiple sets of predicted optical parameters, the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters are calculated according to a preset loss function;

[0182] Among them, the preset loss functions include L1, L2 or cross entropy function, etc., and no specific limitation is made here.

[0183] 804. Train the second neural network model according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

[0184] After the loss is obtained, the second neural network model is further trained according to the loss and the back propagation algorithm until the loss of the second neural network model meets the target loss, wherein the process of training the second neural network model is the process of updating the model parameters of the second neural network model.

[0185] Specifically, when updating the parameters of the second neural network model, the Adamw optimizer may be used to optimize the learning rate of the parameter update. The Adamw optimizer has at least the following advantages:

[0186] 1. Automatically adjust the learning rate of different parameters of the model to better adapt to the changes in the gradients of different parameters;

[0187] 2. The training speed is relatively faster, and it can also better handle sparse gradients;

[0188] 3. The Adamw optimizer performs L2 regularization on larger weight updates, which can further improve the training effect of the high model.

[0189] The structure of the Adamw optimizer itself is consistent with that described in the prior art and will not be described in detail here.

[0190] In an embodiment of the present application, the training process of the second neural network model is described in detail, and when updating the parameters of the second neural network model, the present application uses the Adamw optimizer to optimize the learning rate of the parameter update, so that the second neural network model can be trained faster and with better training effect.

[0191] Further, based on Figure 8 In the above embodiment, when training the second neural network model, the second neural network model can also be trained in a data parallel manner. Fig. 9 , Fig. 9 Another embodiment of the second neural network model training process includes:

[0192] 901. Deploy one second neural network model on each of n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2;

[0193] In actual situations, in order to speed up the training of the second neural network model, the second neural network model can be deployed on multiple working nodes (one working node here corresponds to a processor or server), that is, a second neural network model is deployed on each working node, and the second neural network model is trained using part of the training samples on each working node.

[0194] For easier understanding, the following example is given:

[0195] Assuming n is 30, 30 second neural network models are deployed on 30 working nodes respectively, that is, one second neural network model is deployed on each working node, and the second neural network model on each working node uses the same initialization model parameters.

[0196] 902. Divide the training sample data into n groups of sample data on average;

[0197] Furthermore, the embodiment of the present application divides the training sample data into n groups of sample data. Assuming there are 15,000 training sample data and n is 30, the 15,000 training sample data are divided into 30 groups, wherein each group contains 500 sample data.

[0198] 903. Using a set of sample data on each working node, update the parameters of the second neural network model;

[0199] Furthermore, a set of sample data is used on each working node to update the parameters of the second neural network model.

[0200] For example, the 1st to 500th data are used to train the second neural network model deployed on the first working node, that is, the parameters of the second neural network model are updated.

[0201] 904. After updating the parameters of the second neural network model on each working node m times, n updated second neural network models are obtained, where 5≤m≤15;

[0202] Specifically, when multiple working nodes are used to train the second neural network model, if the parameter averaging method is used, the parameters of the n updated second neural network models are averaged. However, if the model parameters are updated once on each working node (that is, m=1), the average value of the model parameters in the n working nodes is calculated, which will cause large network data synchronization overhead and computing overhead, thereby bringing large additional overhead.

[0203] If m is large, the local parameters of each working node will be more diversified, which will lead to poor model effect after averaging.

[0204] Therefore, m in the embodiments of the present application is greater than or equal to 5 and less than or equal to 15.

[0205] 905. Calculate the average values ​​of the model parameters of the n updated second neural network models;

[0206] After the parameters of the second neural network model are updated m (5≤m≤15) times on each working node, the average value of the model parameters of the n updated corrected neural network models is further calculated, and step 906 is performed according to the average value of the model parameters.

[0207] 906. Using the average values ​​of the model parameters as final model parameters of the second neural network model.

[0208] After obtaining the average value of the model parameters of the second neural network model, the embodiment of the present application further uses the average value of the model parameters as the final model parameters of the second neural network model, thereby improving the training speed of the second neural network model and also improving the accuracy of the second neural network model.

[0209] The embodiment of the present application adopts a data parallel algorithm to train the same second neural network model on multiple different working nodes, thereby improving the training speed of the second neural network model while ensuring the accuracy of the second neural network model.

[0210] The above is a detailed description of the method for obtaining the optical parameters of the detection object in the embodiment of the present application. The following is a description of the device for obtaining the optical parameters of the detection object in the embodiment of the present application. Fig.10, an embodiment of the device for obtaining optical parameters of a detection object in the embodiment of the present application includes:

[0211] A first acquisition unit 1001 is used to acquire an NCs spectrum image of the detection object, wherein the NCs spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0212] The first acquisition unit 1001 is further used to acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters;

[0213] The first input-output unit 1002 is used to input the multiple sets of optical parameter values ​​into the first neural network model to obtain multiple predicted NCS spectrum images output by the first neural network model;

[0214] A first determining unit 1003 is used to determine, among the multiple predicted NCS spectrum images, a target NCS spectrum image having the smallest error with the NCS spectrum image of the detection object;

[0215] The first determining unit 1003 is further configured to regard the target optical parameter group corresponding to the target NCS spectrum image as the optical parameter group of the detection object.

[0216] Preferably, the first determining unit 1003 is specifically configured to:

[0217] A regression analysis is performed on the multiple predicted NCs spectrum images and the NCs spectrum image of the detection object to obtain a target NCs spectrum image with the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images.

[0218] Preferably, the first acquiring unit 1001 is further used for:

[0219] Acquire a first training sample, wherein the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images;

[0220] The first input-output unit 1001 is further used to input the multiple NCS spectral images of the multiple detection objects and the multiple sets of optical parameters corresponding to the multiple NCS spectral images into the first neural network model to obtain the multiple predicted NCS spectral images output by the first neural network model;

[0221] Preferably, the device further comprises:

[0222] The calculation unit 1004 is used to calculate the loss between the multiple NCs spectrum images of the multiple detection objects and the multiple predicted NCs spectrum images according to a preset loss function;

[0223] The first training unit 1005 is used to train the first neural network model according to the loss and the back propagation algorithm until the loss reaches the target loss.

[0224] Preferably, the first neural network model comprises: a convolution module, a normalization layer, a discard layer and a fully connected layer;

[0225] The first input-output unit 1002 is specifically used for:

[0226] Inputting the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values;

[0227] Inputting the data features into the normalization layer to normalize the data features;

[0228] Inputting the normalized data features into the discard layer to desparse the normalized data features;

[0229] The de-sparse data features are input into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

[0230] In the embodiment of the present application, through the first input-output unit 1002 and the first determination unit 1003, the neural network model is combined with the regression analysis method to determine the optical parameters corresponding to the NCS spectrum image of the detection object, thereby realizing the automatic output from the NCS spectrum image of the detection object to the optical parameters of the detection object.

[0231] The present application also provides a device for obtaining optical parameters of a detection object, see Fig.11 Another embodiment of a device for obtaining optical parameters of a detection object in an embodiment of the present application includes:

[0232] A second acquisition unit 1101 is used to acquire an NCs spectrum image of the detection object, wherein the NCs spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0233] The second input-output unit 1102 is used to input the NCS spectrum image into the second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into the optical parameter group of the detection object.

[0234] Preferably, the second acquiring unit 1101 is further used for:

[0235] Acquire second training sample data, wherein the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image;

[0236] The device also includes:

[0237] The second training unit 1103 is used to input the second training sample data into the second neural network model for training until the output of the second neural network model meets the target loss.

[0238] Preferably, the second training unit 1103 is specifically used for:

[0239] Inputting a plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the second neural network model to obtain a plurality of sets of predicted optical parameters output by the second neural network model;

[0240] Calculating the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters according to a preset loss function;

[0241] The second neural network model is trained according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

[0242] Preferably, the second training unit 1103 is specifically used for:

[0243] Deploy one of the second neural network models on each of the n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2;

[0244] Dividing the training sample data into n groups of sample data on average;

[0245] Using a set of sample data on each working node, updating the parameters of the second neural network model;

[0246] After updating the parameters of the second neural network model m times on each working node, n updated second neural network models are obtained, 5≤m≤15;

[0247] Calculate the average values ​​of the model parameters of the n updated second neural network models;

[0248] The average values ​​of the model parameters are used as the final model parameters of the second neural network model.

[0249] Preferably, the second neural network model includes a convolution module, a discard layer, a fully connected layer and a normalization layer;

[0250] The second input-output unit 1102 is specifically used for:

[0251] Inputting the ncs spectrum image into the convolution module to extract image features of the ncs spectrum image;

[0252] Inputting the image features into the dropout layer to perform de-sparseness on the image features;

[0253] Inputting the de-sparsed image features into the fully connected layer to fit the de-sparsed image features;

[0254] The fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0255] In this embodiment of the present application, the second input-output unit 1102 converts the NCS spectrum image of the detection object into corresponding optical parameters according to the second neural network model, thereby improving the convenience of obtaining the optical parameters of the detection object.

[0256] The above describes the device for obtaining the optical parameters of the detection object in the embodiment of the present invention from the perspective of modular functional entities. The following describes the computer device in the embodiment of the present invention from the perspective of hardware processing:

[0257] The computer device is used to realize the function of the device for obtaining the optical parameters of the detection object. In an embodiment of the present invention, an embodiment of the computer device includes:

[0258] Processor and memory;

[0259] The memory is used to store a computer program, and when the processor is used to execute the computer program stored in the memory, the following steps can be implemented:

[0260] Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0261] Acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters;

[0262] Inputting the plurality of sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0263] Determine, among the multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object;

[0264] The target optical parameter group corresponding to the target NCS spectrum image is regarded as the optical parameters of the detection object.

[0265] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0266] A regression analysis is performed on the multiple predicted NCs spectrum images and the NCs spectrum image of the detection object to obtain a target NCs spectrum image with the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images.

[0267] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0268] Acquire a first training sample, wherein the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images;

[0269] Inputting a plurality of NCS spectral images of the plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0270] According to a preset loss function, calculating the loss between a plurality of NCs spectral images of a plurality of detection objects and the plurality of predicted NCs spectral images;

[0271] The first neural network model is trained according to the loss and the back propagation algorithm until the loss reaches a target loss.

[0272] In some embodiments of the present invention, the first neural network model includes: a convolution module, a normalization layer, a dropout layer and a fully connected layer, and the processor can also be used to implement the following steps:

[0273] Inputting the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values;

[0274] Inputting the data features into the normalization layer to normalize the data features;

[0275] Inputting the normalized data features into the discard layer to desparse the normalized data features;

[0276] The de-sparse data features are input into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

[0277] The present application also provides a computer device, which is also used to implement the function of the device for obtaining optical parameters of the detection object. Another embodiment of the computer device in the present application includes:

[0278] Processor and memory;

[0279] The memory is used to store a computer program, and when the processor is used to execute the computer program stored in the memory, the following steps can be implemented:

[0280] Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0281] The NCS spectrum image is input into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into a set of optical parameters of the detection object.

[0282] In some embodiments of the present invention, before inputting the NCS spectrum image into the second neural network model, the processor may also be used to implement the following steps:

[0283] Acquire second training sample data, wherein the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image;

[0284] The second training sample data is input into the second neural network model for training until the output of the second neural network model meets the target loss.

[0285] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0286] Inputting a plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the second neural network model to obtain a plurality of sets of predicted optical parameters output by the second neural network model;

[0287] Calculating the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters according to a preset loss function;

[0288] The second neural network model is trained according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

[0289] In some embodiments of the present invention, the processor may also be used to implement the following steps:

[0290] Deploy one of the second neural network models on each of the n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2;

[0291] Dividing the training sample data into n groups of sample data on average;

[0292] Using a set of sample data on each working node, updating the parameters of the second neural network model;

[0293] After updating the parameters of the second neural network model m times on each working node, n updated second neural network models are obtained, 5≤m≤15;

[0294] Calculate the average values ​​of the model parameters of the n updated second neural network models;

[0295] The average values ​​of the model parameters are used as the final model parameters of the second neural network model.

[0296] In some embodiments of the present invention, the second neural network model includes a convolution module, a dropout layer, a fully connected layer, and a normalization layer; the processor may also be used to implement the following steps:

[0297] Inputting the ncs spectrum image into the convolution module to extract image features of the ncs spectrum image;

[0298] Inputting the image features into the dropout layer to perform de-sparseness on the image features;

[0299] Inputting the de-sparsed image features into the fully connected layer to fit the de-sparsed image features;

[0300] The fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0301] It can be understood that when the processor in the computer device described above executes the computer program, it can also implement the functions of the various units in the above-mentioned corresponding device embodiments, which will not be repeated here. Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the device for obtaining the optical parameters of the detection object. For example, the computer program can be divided into the various units in the above-mentioned device for obtaining the optical parameters of the detection object, and each unit can implement the specific functions described in the above-mentioned corresponding device for obtaining the optical parameters of the detection object.

[0302] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The computer device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that a processor and a memory are merely examples of computer devices and do not constitute a limitation of the computer device. The computer device may include more or fewer components, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, and the like.

[0303] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, and uses various interfaces and lines to connect various parts of the entire computer device.

[0304] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0305] The present invention also provides a computer-readable storage medium, which is used to implement the function of the device for obtaining the optical parameters of the detection object, and stores a computer program thereon. When the computer program is executed by a processor, the processor can be used to perform the following steps:

[0306] Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0307] Acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters;

[0308] Inputting the plurality of sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0309] Determine, among the multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object;

[0310] The target optical parameter group corresponding to the target NCS spectrum image is regarded as the optical parameters of the detection object.

[0311] In some embodiments of the present invention, when the computer program is executed by a processor, the processor may also be used to implement the following steps:

[0312] A regression analysis is performed on the multiple predicted NCs spectrum images and the NCs spectrum image of the detection object to obtain a target NCs spectrum image with the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images.

[0313] In some embodiments of the present invention, when the computer program is executed by a processor, the processor may also be used to implement the following steps:

[0314] Acquire a first training sample, wherein the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images;

[0315] Inputting a plurality of NCS spectral images of the plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model;

[0316] According to a preset loss function, calculating the loss between a plurality of NCs spectral images of a plurality of detection objects and the plurality of predicted NCs spectral images;

[0317] The first neural network model is trained according to the loss and the back propagation algorithm until the loss reaches a target loss.

[0318] In some embodiments of the present invention, the first neural network model includes: a convolution module, a normalization layer, a dropout layer and a fully connected layer. When the computer program is executed by a processor, the processor may also be used to implement the following steps:

[0319] Inputting the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values;

[0320] Inputting the data features into the normalization layer to normalize the data features;

[0321] Inputting the normalized data features into the discard layer to desparse the normalized data features;

[0322] The de-sparse data features are input into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

[0323] The present application also provides a computer-readable storage medium, which is also used to implement the function of the device for obtaining optical parameters of the detection object, and stores a computer program thereon. When the computer program is executed by a processor, the processor can be used to perform the following steps:

[0324] Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions;

[0325] The NCS spectrum image is input into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into a set of optical parameters of the detection object.

[0326] In some embodiments of the present invention, before the NCS spectrum image is input into the second neural network model, when the computer program is executed by the processor, the processor may also be used to implement the following steps:

[0327] Acquire second training sample data, wherein the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image;

[0328] The second training sample data is input into the second neural network model for training until the output of the second neural network model meets the target loss.

[0329] In some embodiments of the present invention, when the computer program is executed by a processor, the processor may also be used to implement the following steps:

[0330] Inputting a plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the second neural network model to obtain a plurality of sets of predicted optical parameters output by the second neural network model;

[0331] Calculating the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters according to a preset loss function;

[0332] The second neural network model is trained according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

[0333] In some embodiments of the present invention, when the computer program is executed by a processor, the processor may also be used to implement the following steps:

[0334] Deploy one of the second neural network models on each of the n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2;

[0335] Dividing the training sample data into n groups of sample data on average;

[0336] Using a set of sample data on each working node, updating the parameters of the second neural network model;

[0337] After updating the parameters of the second neural network model m times on each working node, n updated second neural network models are obtained, 5≤m≤15;

[0338] Calculate the average values ​​of the model parameters of the n updated second neural network models;

[0339] The average values ​​of the model parameters are used as the final model parameters of the second neural network model.

[0340] In some embodiments of the present invention, the second neural network model includes a convolution module, a dropout layer, a fully connected layer, and a normalization layer; when the computer program is executed by a processor, the processor may also be used to implement the following steps:

[0341] Inputting the ncs spectrum image into the convolution module to extract image features of the ncs spectrum image;

[0342] Inputting the image features into the dropout layer to perform de-sparseness on the image features;

[0343] Inputting the de-sparsed image features into the fully connected layer to fit the de-sparsed image features;

[0344] The fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

[0345] It is understandable that if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a corresponding computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned corresponding embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0346] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0347] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0348] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0349] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for obtaining optical parameters of a detection object, characterized in that: include: Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions; Acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters; Inputting the plurality of sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model; Determine, among the plurality of predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object; The target optical parameter group corresponding to the target NCS spectrum image is regarded as the optical parameter group of the detection object.

2. The method according to claim 1, characterized in that Determining a target NCS spectrum image having the smallest error with the NCS spectrum image of the detection object among the multiple predicted NCS spectrum images comprises: A regression analysis is performed on the multiple predicted NCs spectrum images and the NCs spectrum image of the detection object to obtain a target NCs spectrum image with the smallest error with the NCs spectrum image of the detection object among the multiple predicted NCs spectrum images.

3. The method according to claim 1, characterized in that Before inputting the plurality of preset values ​​into the first neural network model, the method further comprises: Acquire a first training sample, wherein the first training sample includes: a plurality of NCs spectral images of a plurality of detection objects and a plurality of groups of optical parameters corresponding to the plurality of NCs spectral images; Inputting a plurality of NCS spectral images of the plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model; According to a preset loss function, calculating the loss between a plurality of NCs spectral images of a plurality of detection objects and the plurality of predicted NCs spectral images; According to the loss and the back propagation algorithm, the parameters of the first neural network model are trained until the loss reaches the target loss.

4. The method according to claim 1, characterized in that: The first neural network model includes: a convolution module, a normalization layer, a discard layer and a fully connected layer; Inputting the plurality of sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model, including: Inputting the multiple sets of optical parameter values ​​into the convolution module to extract data features of the multiple sets of optical parameter values; Inputting the data features into the normalization layer to normalize the data features; Inputting the normalized data features into the discard layer to desparse the normalized data features; The de-sparse data features are input into the fully connected layer to fit the de-sparse data features to obtain the multiple predicted NCS spectral images.

5. A method for obtaining optical parameters of a detection object, characterized in that: include: Acquire an NCS spectrum image of the detection object, wherein the NCS spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions; The NCS spectrum image is input into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into a set of optical parameters of the detection object.

6. The method according to claim 1, characterized in that Before inputting the NCS spectrum image into the second neural network model, the method further comprises: Acquire second training sample data, wherein the second training sample data includes: an NCs spectrum image of each detection object among a plurality of detection objects and a set of optical parameters corresponding to the NCs spectrum image; The second training sample data is input into the second neural network model for training until the output of the second neural network model meets the target loss.

7. The method according to claim 6, characterized in that The step of inputting the second training sample data into the second neural network model for training includes: Inputting a plurality of NCS spectral images of a plurality of detection objects and a plurality of sets of optical parameters corresponding to the plurality of NCS spectral images into the second neural network model to obtain a plurality of sets of predicted optical parameters output by the second neural network model; Calculating the losses between the multiple sets of optical parameters of the multiple detection objects and the multiple sets of predicted optical parameters according to a preset loss function; The second neural network model is trained according to the loss and back propagation algorithm until the loss of the second neural network model meets the target loss.

8. The method according to claim 6, characterized in that Inputting the second training sample data into the second neural network model for training includes: Deploy one of the second neural network models on each of the n working nodes, and the initialization parameters of the second neural network models are the same, where n is greater than or equal to 2; Dividing the training sample data into n groups of sample data on average; Using a set of sample data on each working node, updating the parameters of the second neural network model; After updating the parameters of the second neural network model m times on each working node, n updated second neural network models are obtained, 5≤m≤15; Calculate the average values ​​of the model parameters of the n updated second neural network models; The average values ​​of the model parameters are used as the final model parameters of the second neural network model.

9. The method according to claim 6, characterized in that The second neural network model includes a convolution module, a dropout layer, a fully connected layer and a normalization layer; Inputting the NCS spectrum image into a second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, including: Inputting the ncs spectrum image into the convolution module to extract image features of the ncs spectrum image; Inputting the image features into the dropout layer to perform de-sparseness on the image features; Inputting the de-sparsed image features into the fully connected layer to fit the de-sparsed image features; The fitted image features are input into the normalization layer to obtain a set of optical parameters of the detection object output by the second neural network model.

10. A device for obtaining optical parameters of a detection object, characterized in that: include: A first acquisition unit is used to acquire an NCs spectrum image of the detection object, wherein the NCs spectrum image represents projection images of the spectrum image of the detection object in three mutually orthogonal directions; The first acquisition unit is further used to acquire multiple groups of optical parameter values ​​corresponding to the NCS spectrum image of the detection object, wherein each group of optical parameters includes at least two types of optical parameters; A first input-output unit is used to input the multiple sets of optical parameter values ​​into a first neural network model to obtain a plurality of predicted NCS spectral images output by the first neural network model; A first determining unit is used to determine, among the multiple predicted NCs spectrum images, a target NCs spectrum image having the smallest error with the NCs spectrum image of the detection object; The first determining unit is further configured to regard the target optical parameter group corresponding to the target NCS spectrum image as the optical parameter group of the detection object.

11. A device for obtaining optical parameters of a detection object, characterized in that: include: A second acquisition unit is used to acquire an NCs spectrum image of the detection object, wherein the NCs spectrum image includes projection images of the spectrum image of the detection object in three mutually orthogonal directions; The second input-output unit is used to input the NCS spectrum image into the second neural network model to obtain a set of optical parameters of the detection object output by the second neural network model, wherein the second neural network model is used to convert the NCS spectrum image of the detection object into the optical parameter group of the detection object.

12. A computer device comprising a processor, characterized in that: When executing the computer program stored in the memory, the processor is used to implement the method for obtaining optical parameters of the detection object as described in any one of claims 1 to 4, or the method for obtaining optical parameters of the detection object as described in any one of claims 5 to 9.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it is used to implement the method for obtaining optical parameters of a detection object as claimed in any one of claims 1 to 4, or the method for obtaining optical parameters of a detection object as claimed in any one of claims 5 to 9.

Images