Measurement value prediction method and device, computer equipment and readable storage medium

Abstract

The invention discloses a measurement value prediction method and device, computer equipment and a readable storage medium. The method comprises the steps of obtaining first time sequence data; performing preprocessing and feature extraction on the first time sequence data to obtain first feature data; and performing measurement value prediction on the first feature data to obtain a target measurement value. By adopting the method provided by the invention, real-time virtual full detection of the measurement value can be realized, and the prediction precision of the measurement value can be improved.

Description

Technical Field

[0001] The present application relates to the field of prediction technology, and in particular to a measurement value prediction method, device, computer equipment and readable storage medium. Background Art

[0002] The production process of LCD panels is complex and requires high precision. Measurement values ​​such as film thickness and threshold voltage (Vth) are important measurement indicators for evaluating whether the panel is qualified. During the production process of LCD panels, it is necessary to obtain the measurement values ​​of the LCD panels at the measurement points, so as to extract unqualified products or perform secondary repairs through the measurement values ​​to avoid the outflow of unqualified products. The existing method usually uses special equipment to perform random inspections on the panels in production. The random inspection method is not only inefficient but also cannot achieve full coverage measurement. Summary of the invention

[0003] The embodiments of the present application provide a measurement value prediction method, apparatus, computer equipment, and readable storage medium, which can realize real-time virtual full inspection of measurement values ​​and improve the prediction accuracy of measurement values.

[0004] The technical solution adopted by the present invention to solve the problem is as follows:

[0005] In one aspect, the present application provides a measurement value prediction method, comprising:

[0006] Obtaining first time series measurement value information;

[0007] Preprocessing and feature extracting the first time series measurement value information to obtain first feature data;

[0008] A measurement value prediction is performed on the first feature data to obtain a target measurement value.

[0009] In some implementation schemes of the present application, preprocessing and feature extraction are performed on the first time series measurement value information to obtain first feature data, including:

[0010] Performing multiple data optimization processes on the first time series measurement value information to obtain second time series measurement value information;

[0011] Feature extraction is performed on the second time series measurement value information to obtain first feature data.

[0012] In some implementation schemes of the present application, multiple data optimization processes are performed on the first time series measurement value information to obtain the second time series measurement value information, including:

[0013] Filter out data with a null value ratio greater than a preset ratio from the first time series measurement value information to obtain the first time series measurement value information after null value filtering;

[0014] Filling the first timing measurement value information after the null value filtering with null values ​​to obtain third timing measurement value information, where the third timing measurement value information includes third timing measurement value information of multiple production processes;

[0015] Performing outlier filtering on the third time series measurement value information of each production process to obtain fourth time series measurement value information;

[0016] The fourth time series measurement value information is differentially processed to obtain the second time series measurement value information.

[0017] In some implementation schemes of the present application, the first feature data includes frequency domain features, time domain features, and time-frequency domain features, and feature extraction is performed on the second time series measurement value information to obtain the first feature data, including:

[0018] Extracting frequency domain features from the second time series measurement value information to obtain frequency domain features;

[0019] Extracting time domain features from the second time series measurement value information to obtain time domain features;

[0020] The time-frequency domain features are extracted from the second time series measurement value information to obtain the time-frequency domain features.

[0021] In some implementation schemes of the present application, performing measurement value prediction on the first feature data to obtain a target measurement value includes:

[0022] Performing feature screening on the first feature data to obtain second feature data;

[0023] The second characteristic data is input into the measurement value prediction model, and the measurement value prediction is performed based on the second characteristic data by the measurement value prediction model to obtain a target measurement value.

[0024] In some embodiments of the present application, before performing measurement value prediction on the first feature data to obtain the target measurement value, the following steps are included:

[0025] Acquire a first training data set, wherein the first training data set includes fifth time series measurement value information of multiple sample panels and real measurement values ​​of multiple sample panels;

[0026] Preprocessing and feature extracting the fifth time series measurement value information of each sample panel to obtain third feature data of multiple sample panels;

[0027] The preset network model is trained and verified based on the third characteristic data and the actual measurement value to obtain a measurement value prediction model.

[0028] In some embodiments of the present application, the measurement value prediction model is used to predict the measurement values ​​of multiple measurement points, and the preset network model is trained and verified based on the third feature data and the actual measurement value to obtain the measurement value prediction model, including:

[0029] Performing feature screening on the third feature data to obtain fourth feature data;

[0030] Dividing the fourth characteristic data into a plurality of data groups, wherein the data ratios of the plurality of data groups are the same and the measurement points corresponding to each data group are the same as the measurement points corresponding to the fourth characteristic data;

[0031] Based on the multiple data groups, multiple second training data sets and multiple verification data sets are constructed, the multiple second training data sets correspond to the multiple verification data sets respectively, and the data in the verification data set corresponding to each second training data set is the data in other second training data sets;

[0032] Based on the fourth feature data, determine a training feature set, where the training feature set is a set consisting of feature data with the highest importance in the fourth feature data;

[0033] Based on the training feature set, multiple feature subsets are determined;

[0034] Based on the multiple feature subsets and the multiple second training data sets, determine multiple third training data sets corresponding to each second training data set;

[0035] The preset network model is trained and verified based on a plurality of third training data sets, a plurality of verification data sets and actual measurement values ​​to obtain a measurement value prediction model.

[0036] In a second aspect, an embodiment of the present invention further provides a measurement value prediction device, comprising:

[0037] A first acquisition unit, used to acquire first time series measurement value information;

[0038] A first extraction unit, used for preprocessing and feature extraction of the first time series measurement value information to obtain first feature data;

[0039] The measurement value prediction unit is used to perform measurement value prediction on the first feature data to obtain a target measurement value.

[0040] In a third aspect, the present application further provides a computer device, the computer device comprising:

[0041] one or more processors;

[0042] Memory; and

[0043] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement any measurement value prediction method in the first aspect.

[0044] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and the computer program is loaded by a processor to execute the steps in any one of the measurement value prediction methods in the first aspect.

[0045] The beneficial effects of the present invention are as follows: by preprocessing and extracting features of the first time series data, the features of each production process and the cross-features between multiple production processes can be extracted, the discrimination of the extracted features can be improved, and the measurement value prediction of the first feature data can be performed. By combining the features of each production process and the cross-features between multiple production processes to predict the measurement value, real-time virtual full inspection of the measurement value can be achieved while improving the prediction accuracy of the measurement value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 is a flow chart of a measurement value prediction method provided by an embodiment of the present invention;

[0048] Figure 2 is a flowchart of a specific embodiment of feature data classification and screening provided by an embodiment of the present invention;

[0049] Figure 3 This is a flowchart of a specific embodiment of the present invention for seeking the optimal parameters of a model using Bayesian sampling optimization;

[0050] Figure 4 is a principle block diagram of a measurement value prediction device provided by an embodiment of the present invention;

[0051] Figure 5 It is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0053] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, which are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, the terms "first", "second", "third" and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more features. In the description of the present application, "multiple" means two or more, and "several" means one or more, unless otherwise clearly and specifically defined.

[0054] In this application, the word "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described in this application as "exemplary" is not necessarily to be construed as being preferred or advantageous over other embodiments. The following description is given to enable any technician in the field to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present application can be implemented without using these specific details. In other instances, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present application with unnecessary details. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0055] It should be noted that since the method of the embodiment of the present application is executed in a computer device, the processing objects of each computer device exist in the form of data or information. For example, time is actually time information. It can be understood that if size, quantity, position, etc. are mentioned in subsequent embodiments, they are all corresponding data for processing by the computer device. The details will not be repeated here.

[0056] Measurement values ​​such as film thickness and threshold voltage (Vth) are important measurement indicators for evaluating whether a panel is qualified. During the production process of liquid crystal panels, it is necessary to obtain the measurement values ​​of the liquid crystal panels to determine whether the produced liquid crystal panels are qualified. The inventors have found through research that the traditional method usually uses special equipment to perform random sampling and measurement on the panels in production. The random sampling method is not only inefficient but also cannot achieve full coverage measurement. In order to achieve real-time virtual full inspection to improve the efficiency of the production line, there are existing methods that use the single variable data of a certain process in EDA to construct features and then model them, and then use the constructed model to predict the measurement values ​​of one or more measurement points. This method uses the single variable data of a certain process to predict the measurement value, and the prediction accuracy is low.

[0057] Based on this, in an embodiment of the present application, first time series data is acquired; the first time series data is preprocessed and feature extracted to obtain first feature data; measurement value prediction is performed on the first feature data to obtain a target measurement value, thereby realizing real-time virtual full inspection of the measurement value, and by combining the characteristics of each production process and the cross-features between multiple production processes to perform measurement value prediction, the prediction accuracy of the measurement value can be improved.

[0058] The content of this application is further illustrated below through the description of embodiments in conjunction with the accompanying drawings.

[0059] This embodiment provides a measurement value prediction method, such as Figure 1 As shown in , the method includes:

[0060] Step S10: Obtain first timing measurement value information.

[0061] The production process of LCD panels is complex and requires high precision. During the production process of LCD panels, one or more measurement points need to be set up. The quality of the produced LCD panels can be evaluated by obtaining the film thickness, threshold voltage (Vth) and other measurement values ​​at one or more measurement points. The panel is a LCD panel that needs to predict the measurement values ​​of the measurement points.

[0062] During the production of a panel, multiple processes will be passed over time, such as chemical vapor deposition (CVD), WET, and DRY, etc. The equipment parameters of each process are different and change over time. The first timing measurement value information is the equipment parameter timing data of the panel in multiple production processes. For example, the first timing measurement value information shown in Table 1 includes a timestamp, multiple production processes (Prehea, NH3, PV2-1, PL1, PL2), and the equipment parameter timing data of each production process (parameter 1 to parameter 9).

[0063] Table 1 The first time series data of a single panel in multiple production processes

[0064]

[0065]

[0066] Taking into account that the production process of liquid crystal panels involves multiple production processes, and the equipment parameters of each production process are different and change over time, when predicting the measurement value of the panel, this embodiment obtains the first timing measurement value information of the panel in multiple production processes, and combines the equipment parameter timing data of multiple production processes to perform predictive analysis of the panel measurement value, which can improve the prediction accuracy of the panel measurement value.

[0067] Step S20: preprocess and extract features of the first time series measurement value information to obtain first feature data.

[0068] The first feature data is a feature set of each production process extracted from the first time series measurement value information according to the production process, and the feature extraction includes but is not limited to time domain feature extraction, frequency domain feature extraction, time-frequency domain feature extraction, etc. For example, when the feature extraction includes time domain feature extraction, frequency domain feature extraction, and time-frequency domain feature extraction, feature extraction is performed on the first time series measurement value information shown in Table 1 to obtain the first feature data shown in Table 2, that is, after the first time series measurement value information in Table 1 is preprocessed and feature extracted, each production process corresponds to a time domain feature, a frequency domain feature, and a time-frequency domain feature. This embodiment extracts features from the first time series measurement value information in a manner of distinguishing production processes, thereby improving the discrimination of the extracted feature data, and performing measurement value prediction based on the first feature data can further improve the prediction accuracy of the panel measurement value.

[0069] Table 2 First characteristic data

[0070]

[0071] In a specific implementation, step S20 includes:

[0072] Step S21, performing multiple data optimization processes on the first time series measurement value information to obtain second time series measurement value information;

[0073] Step S22: extract features from the second time series measurement value information to obtain first feature data.

[0074] In order to improve the prediction accuracy of the measurement value, before extracting the features of the first time series measurement value information of each production process, this embodiment performs multiple data optimization processes on the first time series measurement value information of each production process, and the second time series measurement value information is the first time series measurement value information after the multiple data optimization processes. The multiple data optimization processes include but are not limited to data filling processes, data screening processes, and differential processes.

[0075] In a specific implementation, step S21 includes:

[0076] Step S211, filtering out data having a null value ratio greater than a preset ratio from the first time series measurement value information, to obtain the first time series measurement value information after the null value filtering;

[0077] Step S212, performing null-filling on the first timing measurement value information after the null value filtering to obtain third timing measurement value information, where the third timing measurement value information includes third timing measurement value information of a plurality of production processes;

[0078] Step S213, filtering the third time series measurement value information of each production process for abnormal values ​​to obtain fourth time series measurement value information;

[0079] Step S214: perform differential processing on the fourth timing measurement value information to obtain second timing measurement value information.

[0080] In a specific implementation, multiple data optimization processes include data filling process, data screening process and differential process. Accordingly, the steps of performing multiple data optimization processes on the first time series measurement value information specifically include: performing data filling process on the first time series measurement value information to obtain third time series measurement value information; performing data screening process on the third time series measurement value information to obtain fourth time series measurement value information; performing differential process on the fourth time series measurement value information to obtain second time series measurement value information.

[0081] In a specific implementation, the data filling process includes null value filtering and null value filling. Accordingly, the step of performing data filling process on the first time series measurement value information may specifically include: performing null value filtering on the first time series measurement value information, filtering out data with a null value ratio greater than a preset ratio from the first time series measurement value information, obtaining the first time series measurement value information after null value filtering, and then performing null value filling on the first time series measurement value information after null value filtering to obtain the third time series measurement value information. For example, filtering out data with a null value ratio greater than 70%, and then performing mean-based filling on the null values ​​in the first time series measurement value information after null value filtering.

[0082] Data screening is the process of removing abnormal values ​​from the third time series measurement value information. An abnormal value refers to data whose values ​​in the third time series measurement value information are obviously deviated from other values. Putting abnormal values ​​into the third time series measurement value information without removal will affect the prediction of the panel measurement value. Therefore, after filtering and filling the first time series measurement value information of each production process with null values, this embodiment further filters the third time series measurement value information of each production process to remove abnormal values ​​in the third time series measurement value information. Among them, the box plot method can be used to filter the abnormal values ​​of the third time series measurement value information of each production process. The result of identifying abnormal values ​​by the box plot method is more objective, and it has certain advantages in identifying abnormal values. In the box plot method, outliers are usually defined as values ​​less than QL-1.5IQR or greater than QU+1.5IQR. QL is called the lower quartile, indicating that one-quarter of all observations are smaller than it. QU is called the upper quartile, indicating that one-quarter of all observations are larger than it. IQR is called the interquartile range, which is the difference between the upper quartile QU and the lower quartile QL, and contains half of all observations.

[0083] The role of differential processing is to reduce the irregular fluctuations between data and make its fluctuation curve smoother. After obtaining the fourth time series measurement value information, this embodiment further performs differential processing on the fourth time series measurement value information. Among them, the step of performing differential processing on the fourth time series measurement value information may specifically include: obtaining the initial value of each production process, subtracting the initial value of each production process from the fourth time series measurement value information of each production process, and obtaining the second time series measurement value information of each production process. For example, continuing to refer to Table 1, the initial value of parameter 1 in the Prehea process is 25. After differential processing, the value of parameter 1 in the Prehea process at a timestamp of 250 becomes 1.2.

[0084] In a specific implementation, the first feature data includes frequency domain features, time domain features, and time-frequency domain features, and step S22 includes:

[0085] Step S221, extracting time domain features from the second time series measurement value information to obtain time domain features;

[0086] Step S222, extracting frequency domain features from the second time series measurement value information to obtain frequency domain features;

[0087] Step S223: extract time-frequency domain features from the second time series measurement value information to obtain time-frequency domain features.

[0088] In a specific implementation, the first feature data includes frequency domain features, time domain features and time-frequency domain features, as shown in Table 3, the types of time domain features include maximum value, minimum value, peak-to-peak value, variance, skewness, kurtosis, waveform factor, margin factor, root amplitude; the types of frequency domain features include: frequency mean, frequency variance, skew index, kurtosis index, frequency root mean square, peak-to-peak value, kurtosis factor, centroid frequency, standard deviation coefficient; the types of time-frequency domain features include: envelope spectrum mean, envelope spectrum variance, envelope spectrum centroid, wavelet packet decomposition add energy, wavelet packet decomposition aad energy, wavelet packet decomposition ada energy, EMD decomposition IMF0 energy, EMD decomposition IMF1 energy, EMD decomposition IMF2 energy.

[0089] Table 3. Time domain, frequency domain and time-frequency domain feature types

[0090]

[0091] Correspondingly, the step of extracting features from the second time series measurement value information may specifically include: extracting time domain features from the second time series measurement value information to obtain time domain features of the panel to be tested in multiple production processes, extracting frequency domain features from the second time series measurement value information to obtain frequency domain features of the panel to be tested in multiple production processes, and extracting time and frequency domain features from the second time series measurement value information to obtain time and frequency domain features of the panel to be tested in multiple production processes. For example, envelope spectrum analysis, wavelet packet decomposition, and EMD decomposition are performed on the second time series measurement value information to obtain the envelope spectrum mean, envelope spectrum variance, envelope spectrum centroid, wavelet packet decomposition add energy, wavelet packet decomposition aad energy, wavelet packet decomposition ada energy, EMD decomposition IMF0 energy, EMD decomposition IMF1 energy, and EMD decomposition IMF2 energy of the panel to be tested in multiple production processes.

[0092] Step S30: predict the measurement value of the first feature data to obtain a target measurement value.

[0093] The target measurement value is the measurement value of one or more measurement points of the panel predicted based on the first characteristic data. For example, the target measurement value can be the film thickness and / or voltage threshold (Vth) of the panel at multiple measurement points predicted based on the first characteristic data. Table 4 shows the measurement values ​​of panel TESTID6C1899BB at multiple measurement points.

[0094] Table 4 Measurement values ​​of single panel at multiple measurement points

[0095]

[0096]

[0097] Since the first characteristic data is obtained by extracting features from the first time series measurement value information according to the production process, the first characteristic data includes not only the features of each production process (time domain features and frequency domain features), but also the cross-features between multiple production processes (time and frequency domain features). By predicting the measurement values ​​of the panel to be measured based on the first characteristic data, not only can real-time virtual full inspection of the measurement values ​​be achieved, but also the accuracy of measurement value prediction can be improved.

[0098] In a specific implementation, step S30 includes:

[0099] Step S31, performing feature screening on the first feature data to obtain second feature data;

[0100] Step S32: input the second characteristic data into the measurement value prediction model, and perform measurement value prediction based on the second characteristic data through the measurement value prediction model to obtain a target measurement value.

[0101] The measurement value prediction model is a pre-trained model for predicting the measurement value of the first feature data, and the second feature data is the feature data obtained after feature screening of the first feature data. In order to improve the generalization and prediction accuracy of the measurement value prediction model, when generating the measurement value prediction model, this embodiment will screen out features with good distribution consistency and high importance to form a training feature set to train the preset network model, and the features contained in the second feature data are consistent with the features contained in the training feature set. For example, the training feature set contains Prehea time domain features, Prehea frequency domain features, and PL2 time and frequency domain features. When feature screening is performed on the first feature data, the data contained in the Prehea time domain features, Prehea frequency domain features, and PL2 time and frequency domain features are screened out from the first feature data as the second feature data.

[0102] The input of the measurement value prediction model is the second characteristic data, and the output of the measurement value prediction model is the measurement value of one or more measurement points of the panel. When the present embodiment needs to predict the measurement value of the panel, the first characteristic data can be feature screened to obtain the second characteristic data, and then the second characteristic data can be input into the measurement value prediction model. The measurement value prediction model is used to predict the measurement value based on the second characteristic data to obtain the target measurement value of the panel. By predicting the measurement value of the panel through the measurement value prediction model, real-time virtual full inspection and real-time monitoring and interception of abnormalities can be realized for the panel, which can effectively help the factory improve product quality, shorten product cycle, and bring significant economic benefits.

[0103] In addition, when predicting the measurement values ​​of multiple measurement points, the existing method needs to model the multiple measurement points separately, which reduces the generalization and prediction performance of the model, and cannot perform real-time prediction and model update. Compared with the prior art, this embodiment predicts the measurement values ​​of multiple measurement points through a measurement value prediction model, that is, predicts the measurement values ​​of multiple measurement points through one model, which can improve the generalization and prediction performance of the model, and can perform real-time measurement value prediction and model update.

[0104] In a specific implementation, before step S30, the following steps are included:

[0105] Step S40, obtaining a first training data set, wherein the first training data set includes fifth time series measurement value information of multiple sample panels and real measurement values ​​of multiple sample panels;

[0106] Step S50, preprocessing and feature extraction of the fifth time series measurement value information of each sample panel to obtain third feature data of multiple sample panels;

[0107] Step S60: training and verifying the preset network model based on the third characteristic data and the actual measurement value to obtain a measurement value prediction model.

[0108] In the aforementioned steps, it is mentioned that in step S30, the measurement value prediction model is used to predict the measurement value of the first feature data. Therefore, before the measurement value prediction is performed on the first feature data in step S30, the preset network model needs to be trained to obtain the measurement value prediction model. The first training data set is a pre-set data set for training the preset network model, and the first training data set includes the fifth time series measurement value information of multiple sample panels in multiple production processes and the actual measurement values ​​of multiple sample panels.

[0109] The fifth timing measurement value information is the equipment parameter timing data of the sample panel under multiple production processes. The data content included in the fifth timing measurement value information is similar to the data content included in the first timing measurement value information. For details, please refer to the discussion of the aforementioned first timing measurement value information, and this application will not repeat it here.

[0110] The true measurement value is the actual measurement value of the sample panel at the measurement point, which can be obtained by measuring the measurement points of the sample panel using special equipment during the production process of the sample panel. The true measurement value is used as a label in the training process of the preset network model, and can be used to evaluate whether the training of the preset network model has achieved the expected effect. It should be noted that when the measurement value prediction model is used to predict the measurement value of a measurement point of the panel to be measured, the true measurement value can be the actual measurement value of a measurement point of the sample panel; when the measurement value prediction model is used to predict the measurement value of multiple measurement points of the panel to be measured, the true measurement value needs to include the actual measurement values ​​of multiple measurement points of the sample panel.

[0111] The third feature data is a feature set of multiple sample panels in multiple production processes extracted from the fifth time series measurement value information according to the production process, wherein the third feature data includes M rows and N columns, N represents the number of features, M represents the number of sample panels, each row corresponds to feature data of a sample panel, and each column corresponds to a feature. For example, when feature extraction includes time domain feature extraction, frequency domain feature extraction, and time-frequency domain feature extraction, the third feature data is as shown in Table 5, and each row corresponds to feature data of a sample panel.

[0112] Table 5 Second characteristic data

[0113]

[0114]

[0115] In order to train a measurement value prediction model that can predict the measurement value of the first characteristic data, after obtaining the first training data set, this embodiment preprocesses and extracts features of the fifth time series measurement value information of each sample panel in each production process to obtain the third characteristic data of multiple sample panels in multiple production processes, and then trains and verifies the preset network model based on the third characteristic data and the actual measurement value to obtain the measurement value prediction model.

[0116] The steps of preprocessing and feature extracting the fifth timing measurement value information of each sample panel in each production process may specifically include: preprocessing the fifth timing measurement value information of each sample panel in each production process to obtain the preprocessed fifth timing measurement value information, and extracting features from the preprocessed fifth timing measurement value information to obtain third feature data of multiple sample panels in multiple production processes.

[0117] Among them, the step of preprocessing the fifth timing measurement value information is the same as the step of preprocessing the first timing measurement value information, that is, preprocessing the fifth timing measurement value information includes performing multiple data optimization processes on the fifth timing measurement value information. For details, please refer to the discussion of the aforementioned step S21, and this application will not repeat them here.

[0118] Furthermore, the step of performing feature extraction on the preprocessed fifth time series measurement value information is the same as the step of performing feature extraction on the second time series measurement value information, that is, performing feature extraction on the preprocessed fifth time series measurement value information includes performing time domain, frequency domain and time-frequency domain feature extraction on the preprocessed fifth time series measurement value information. For details, please refer to the discussion of the aforementioned step S22, and the present application will not repeat them here.

[0119] In a specific implementation, the measurement value prediction model is used to predict the measurement values ​​of multiple measurement points, and step S60 includes:

[0120] Step S61, performing feature screening on the third feature data to obtain fourth feature data;

[0121] Step S62, dividing the fourth characteristic data into a plurality of data groups, wherein the data ratios of the plurality of data groups are the same and the measurement points corresponding to each data group are the same as the measurement points corresponding to the fourth characteristic data;

[0122] Step S63: construct multiple second training data sets and multiple verification data sets based on the multiple data sets, wherein the multiple second training data sets correspond to the multiple verification data sets respectively, and the data in the verification data set corresponding to each second training data set is the data in other second training data sets;

[0123] Step S64: training and verifying the preset network model based on the plurality of second training data sets, the plurality of verification data sets and the actual measurement values ​​to obtain a measurement value prediction model.

[0124] In order to reduce the model generalization problem caused by the multi-linear problem and improve the stability of the subsequent model prediction effect, when the preset network model is trained and verified based on the third feature data and the real measurement value, the third feature data is first feature-screened to obtain the fourth feature data. The step of feature-screening the third feature data may specifically include: filtering the third feature data to obtain the filtered third feature data, and deleting the filtered third feature data to obtain the fourth feature data.

[0125] As can be seen from the above Table 5, the third feature data includes M rows and N columns, each row corresponds to the feature data of a sample panel, and each column corresponds to a feature. The step of filtering the third feature data may specifically include: calculating the variance of each column of feature data in the third feature data, removing feature columns whose variance is less than a preset variance threshold from the third feature data, then calculating the correlation (linear and nonlinear) between the features of each column, retaining only one of the feature columns in the third feature data whose correlation is higher than a preset correlation threshold, and finally calculating the variance inflation factor (VIF) of each column of feature data, deleting feature columns whose variance inflation factor is greater than the preset inflation factor threshold, and obtaining the filtered third feature data. For example, deleting feature columns with VIF>10 to obtain the filtered third feature data.

[0126] Furthermore, feature deletion of the filtered third feature data refers to deleting features with poor distribution consistency in the filtered third feature data. In order to measure the distribution consistency of the filtered third feature data, the present embodiment pre-sets a test data set, and sets the first true label of each data in the third feature data and the second true label of each data in the test data set. When feature deletion of the filtered third feature data is performed, the KS test is first used to perform a statistical distribution consistency test on the features of the filtered third feature data and the test data set, and the feature columns with larger P values ​​in the filtered third feature data are deleted. Then, the third feature data and the test data set are input into a pre-constructed classifier, and the classifier outputs the first predicted label of each data in the third feature data and the second predicted label of each data in the test data set. Based on the first predicted label, the second predicted label, the first true label and the second true label, the classification accuracy of each feature is determined, and features with a classification accuracy greater than a preset accuracy threshold are deleted from the third feature data.

[0127] The classifier can be constructed using a tree model based on a single feature in the third feature data, and the first true label, the second true label and the accuracy threshold can be set as needed. Figure 2 As shown, the first true label can be set to 0, the second true label can be set to 1, and the accuracy threshold can be set to 0.9. When the classification accuracy is greater than 0.9, it is considered that the distribution consistency of the feature in the third feature data and the test data set is poor, and the feature column corresponding to the feature is deleted from the third feature data.

[0128] Multiple second training data sets and multiple verification data sets are data sets constructed based on multiple data sets, and multiple second training data sets correspond to multiple verification data sets respectively, and the data in the verification data set corresponding to each second training data set is the data in other second training data sets. For example, the second training data set A, the second training data set B and the second training data set correspond to the verification data set A, the verification data set B and the verification data set C respectively, the second training data set A includes data set B and data set C, the verification data set A includes data set A, the second training data set B includes data set A and data set C, the verification data set B includes data set B, the second training data set C includes data set A and data set B, and the verification data set C includes data set C.

[0129] In the process of machine learning modeling, the common practice is to divide the data into a training set and a test set. Since the test set is independent of the training data and does not participate in the training at all, overfitting problems often occur during the training process. In order to solve this technical problem, the present embodiment divides the fourth characteristic data into k data groups, and then uses kn data groups as training data sets, and uses another n data groups as verification data sets, thereby obtaining multiple second training data sets and multiple verification data sets, where n is a positive integer. For example, the fourth characteristic data is divided into five data groups, and one of the data groups is used as a verification data set, and the other data groups are used as second training data sets to obtain five second training data sets and five verification data sets.

[0130] When the measurement value prediction model is used to predict the measurement values ​​of multiple measurement points, that is, when a single model is used to predict the measurement values ​​of multiple measurement points, considering that the data between different points corresponding to all sample panels may have different means and variances, if the fourth characteristic data is directly and randomly divided into multiple data groups, data imbalance problems may occur, resulting in poor model effect and generalization. In order to solve this technical problem, when the fourth characteristic data is divided into multiple data groups, the data ratios of the multiple data groups are the same and the measurement points corresponding to each data group are the same as the measurement points corresponding to the fourth characteristic data, even if each verification data contains equal proportions of data for all points.

[0131] It should be noted that, in the aforementioned steps, the third characteristic data includes M rows and N columns, and the fourth characteristic data also includes multiple rows and multiple columns. The fourth characteristic data described in this embodiment is divided into multiple data groups, which means dividing the fourth characteristic data into multiple data groups by row. For example, the fourth characteristic data includes 100 rows, and each 20 rows is a group, and the fourth characteristic data is divided into five groups.

[0132] In addition, each row in the fourth characteristic data corresponds to a sample panel, and each sample panel includes measurement points. The measurement points corresponding to the fourth characteristic data refer to the measurement points included in the sample panel corresponding to the fourth characteristic data, and the measurement points corresponding to each data group refer to the measurement points included in the sample panel corresponding to each data group. For example, the sample panel corresponding to the fourth characteristic data includes measurement points A, B, and C, and the sample panel corresponding to the data group includes measurement points A, B, and C. Then the measurement points corresponding to the data group are the same as the measurement points corresponding to the fourth characteristic data.

[0133] In a specific implementation, step S64 includes:

[0134] Step S641: Determine a training feature set based on the fourth feature data, where the training feature set is a set of feature data with a higher degree of importance in the fourth feature data;

[0135] Step S642: Determine multiple feature subsets based on the training feature set;

[0136] Step S643: determining multiple third training data sets corresponding to each second training data set based on multiple feature subsets and multiple second training data sets;

[0137] Step S644: training and verifying the preset network model based on the plurality of third training data sets, the plurality of verification data sets and the actual measurement values ​​to obtain a measurement value prediction model.

[0138] The training feature set is a set of feature data with the highest importance in the fourth feature data, and each feature subset in the multiple feature subsets is a feature data set composed of part of the feature data in the training feature set. For example, the training feature set includes five features A, B, C, D, and E, and the multiple feature subsets include (AB CD), (AC DE), (AB CE), (BC DE), and (ABDE). The multiple third training data sets corresponding to each second training data set are data sets after the intersection of the multiple feature subsets and each second training data set is taken. For example, the second training data set A includes the first 80 rows of data of the fourth feature data, and the feature subset A includes (ABCD). Then the third training data set determined based on the second training data set A and the feature subset A is the first 80 rows of data of features A, B, C, and D in the fourth feature data.

[0139] In order to avoid the risk of overfitting caused by a single feature exposure during the model training process, when the preset network model is trained based on multiple second training data sets, the present embodiment first determines the training feature set based on the fourth feature data, then determines multiple feature subsets based on the training feature set, and then determines multiple third training data sets corresponding to each second training data set based on the multiple feature subsets and the multiple second training data sets. Finally, the preset network model is iteratively trained based on the multiple third training data sets, which can avoid the risk of overfitting caused by a single feature exposure during the model training process.

[0140] In a specific implementation, when determining the training feature set based on the fourth feature data, a regression tree can be constructed based on the fourth feature data using a tree model, and then the optimal tree model parameters are obtained based on a network search, and then the importance coefficient of each feature in the fourth feature data is determined based on the tree model, and based on the importance coefficient, the training feature set can be determined from the fourth feature data. For example, the data ranked top 50% in importance can be selected from the fourth feature data to form a training feature set.

[0141] Furthermore, when multiple feature subsets are determined based on the training feature set, some features can be selected from the training feature set as training features each time and other feature value data can be set to 0, thereby obtaining multiple feature subsets. For example, the training feature set includes five features A, B, C, D, and E. (ABCD), (ACDE), (A BC E), (BCDE), and (AB DE) are selected from the training feature set as training features, and zero feature value data is set to 0, thereby obtaining five feature subsets.

[0142] The preset network model includes multiple ones. When the preset network model is trained and verified based on multiple third training data sets, multiple verification data sets and real measurement values, for each preset network model, the preset network model is iteratively trained using multiple third training data sets corresponding to the first second training data set to obtain a pre-trained network model corresponding to the first second training data set, and then the preset network model is iteratively trained using multiple third training data sets corresponding to the second second training data set to obtain a pre-trained network model corresponding to the second second training data set, and so on, the preset network model is iteratively trained using multiple third training data sets corresponding to the Rth second training data set to obtain a pre-trained network model corresponding to the Rth second training data set.

[0143] After obtaining multiple pre-trained network models corresponding to multiple second training data sets, input the verification data set corresponding to each second training data set into the pre-trained network model corresponding to the second training data set, obtain the predicted measurement value of each sample panel in the verification data set, and determine the index value of each pre-trained network model based on the predicted measurement value and the true measurement value. Then, each preset network model can obtain R index values, where R is the number of second training data sets. Then, based on the R index values, determine the target network model from multiple preset network models, and train the target network model based on the fourth feature data and the true measurement value. The trained target network model is the measurement value prediction model. For example, the index value of each pre-trained network model is determined based on the predicted measurement value, the true measurement value and the mean absolute percentage error (MAPE) formula, and then the average value of the R index values ​​of each preset network model is determined as the target index value of the preset network model, and the network model with the smallest target index value is selected from multiple preset network models as the target network model.

[0144] Due to the drift of machine parameters or the running time process, the measurement value prediction model needs to be updated regularly. In order to improve the update speed of the measurement value prediction model and ensure the real-time performance of the measurement value prediction, this embodiment uses Bayesian sampling optimization to find the best model parameters during the process of updating the measurement value prediction model. Specifically, the process of Bayesian sampling optimization to find the best model parameters is as follows: Figure 3 As shown, firstly, an initial sampling point is randomly generated and the function value is calculated based on the initial sampling point, then Gaussian process regression is performed based on the function value, then the sampling function is calculated and the next sampling point is determined according to the maximum value, the function value is calculated based on the next sampling point, and it is determined whether the termination condition is met. When the termination condition is not met, the step of performing Gaussian process regression based on the function value is continued until the termination condition is met, thereby obtaining the optimal parameters and function.

[0145] In order to better implement the measurement value prediction method in the embodiment of the present application, based on the measurement value prediction method, the embodiment of the present application also provides a measurement value prediction device, such as Figure 4 As shown, the measurement value prediction device includes:

[0146] A first acquisition unit 610, configured to acquire first timing measurement value information;

[0147] A first extraction unit 620, configured to preprocess and extract features from the first time series measurement value information to obtain first feature data;

[0148] The measurement value prediction unit 630 is used to perform measurement value prediction on the first feature data to obtain a target measurement value.

[0149] In an embodiment of the present application, the first measurement value information is preprocessed and feature extracted, so that the characteristics of each production process and the cross-features between multiple production processes can be extracted, the discrimination of the extracted features can be improved, and the measurement value prediction is performed on the first feature data. By combining the characteristics of each production process and the cross-features between multiple production processes to predict the measurement value, real-time virtual full inspection of the measurement value can be achieved and the prediction accuracy of the measurement value can be improved.

[0150] In some embodiments of the present application, the first extraction unit 620 is specifically used to:

[0151] Performing multiple data optimization processes on the first time series measurement value information to obtain second time series measurement value information;

[0152] Feature extraction is performed on the second time series measurement value information to obtain first feature data.

[0153] In some embodiments of the present application, the first extraction unit 620 is further configured to:

[0154] Filter out data with a null value ratio greater than a preset ratio from the first time series measurement value information to obtain the first time series measurement value information after null value filtering;

[0155] Filling the first timing measurement value information after the null value filtering with null values ​​to obtain third timing measurement value information, where the third timing measurement value information includes third timing measurement value information of multiple production processes;

[0156] Performing outlier filtering on the third time series measurement value information of each production process to obtain fourth time series measurement value information;

[0157] The fourth time series measurement value information is differentially processed to obtain the second time series measurement value information.

[0158] In some embodiments of the present application, the first feature data includes frequency domain features, time domain features and time-frequency domain features, and the first extraction unit 620 is further specifically used for:

[0159] Extracting frequency domain features from the second time series measurement value information to obtain frequency domain features;

[0160] Extracting time domain features from the second time series measurement value information to obtain time domain features;

[0161] The time-frequency domain features are extracted from the second time series measurement value information to obtain the time-frequency domain features.

[0162] In some embodiments of the present application, the measurement value prediction unit 630 is specifically used to:

[0163] Performing feature screening on the first feature data to obtain second feature data;

[0164] The second characteristic data is input into the measurement value prediction model, and the measurement value prediction is performed based on the second characteristic data by the measurement value prediction model to obtain a target measurement value.

[0165] In some embodiments of the present application, the measurement value prediction device further includes:

[0166] A second acquisition unit is used to acquire a first training data set, where the first training data set includes fifth time series measurement value information of multiple sample panels and real measurement values ​​of multiple sample panels;

[0167] A second extraction unit, used for preprocessing and feature extraction of the fifth time series measurement value information of each sample panel to obtain third feature data of multiple sample panels;

[0168] The model training unit is used to train and verify the preset network model based on the third characteristic data and the actual measurement value to obtain a measurement value prediction model.

[0169] In some embodiments of the present application, the measurement value prediction model is used to predict the measurement values ​​of multiple measurement points, and the model training unit is specifically used to:

[0170] Performing feature screening on the third feature data to obtain fourth feature data;

[0171] Dividing the fourth characteristic data into a plurality of data groups, wherein the data ratios of the plurality of data groups are the same and the measurement points corresponding to each data group are the same as the measurement points corresponding to the fourth characteristic data;

[0172] Based on the multiple data groups, multiple second training data sets and multiple verification data sets are constructed, the multiple second training data sets correspond to the multiple verification data sets respectively, and the data in the verification data set corresponding to each second training data set is the data in other second training data sets;

[0173] Based on the fourth feature data, determine a training feature set, where the training feature set is a set consisting of feature data with the highest importance in the fourth feature data;

[0174] Based on the training feature set, multiple feature subsets are determined;

[0175] Based on the multiple feature subsets and the multiple second training data sets, determine multiple third training data sets corresponding to each second training data set;

[0176] The preset network model is trained and verified based on a plurality of third training data sets, a plurality of verification data sets and actual measurement values ​​to obtain a measurement value prediction model.

[0177] The embodiment of the present application further provides a computer device, which integrates any one of the measurement value prediction devices provided in the embodiment of the present application, and the computer device includes:

[0178] one or more processors;

[0179] Memory; and

[0180] One or more applications, wherein the one or more applications are stored in the memory and are configured to execute, by the processor, the steps of the measurement value prediction method in any of the above measurement value prediction method embodiments.

[0181] The present application also provides a computer device that integrates any of the measurement value prediction devices provided in the present application. Figure 5 As shown, it shows a schematic diagram of the structure of the computer device involved in the embodiment of the present application, specifically:

[0182] The computer device may include one or more processing core processors 801, one or more computer-readable storage media memories 802, a power supply 803, an input unit 804 and other components. Those skilled in the art will appreciate that Figure 5 The computer device structure shown in the figure does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. Among them:

[0183] The processor 801 is the control center of the computer device. It uses various interfaces and lines to connect various parts of the entire computer device. By running or executing software programs and / or modules stored in the memory 802 and calling data stored in the memory 802, it executes various functions of the computer device and processes data, thereby monitoring the computer device as a whole. Optionally, the processor 801 may include one or more processing cores; preferably, the processor 801 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 801.

[0184] The memory 802 can be used to store software programs and modules. The processor 801 executes various functional applications and data processing by running the software programs and modules stored in the memory 802. The memory 802 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 802 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 802 may also include a memory controller to provide the processor 801 with access to the memory 802.

[0185] The computer device also includes a power supply 803 for supplying power to each component. Preferably, the power supply 803 can be logically connected to the processor 801 through a power management system, so that the power management system can manage charging, discharging, power consumption and other functions. The power supply 803 can also include one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, power status indicators and other arbitrary components.

[0186] The computer device may further include an input unit 804, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0187] Although not shown, the computer device may further include a display unit, etc., which will not be described in detail herein. Specifically in this embodiment, the processor 801 in the computer device will load the executable files corresponding to the processes of one or more application programs into the memory 802 according to the following instructions, and the processor 801 will run the application programs stored in the memory 802, thereby realizing various functions, as follows:

[0188] Obtaining first time series measurement value information;

[0189] Preprocessing and feature extraction of the first time series measurement value information to obtain first feature data;

[0190] A measurement value prediction is performed on the first feature data to obtain a target measurement value.

[0191] A person of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0192] To this end, an embodiment of the present application provides a computer-readable storage medium, which may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc. A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in any of the measurement value prediction methods provided in the embodiments of the present application. For example, the computer program is loaded by a processor to execute the following steps:

[0193] Obtaining first time series measurement value information;

[0194] Preprocessing and feature extraction of the first time series measurement value information to obtain first feature data;

[0195] A measurement value prediction is performed on the first feature data to obtain a target measurement value.

[0196] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the detailed description of other embodiments above, and will not be repeated here.

[0197] In specific implementation, the above units or structures can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units or structures can refer to the previous method embodiments, which will not be repeated here.

[0198] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0199] The above is a detailed introduction to a measurement value prediction method, device, computer equipment and readable storage medium provided in an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A measurement value prediction method, characterized in that: include: Obtaining first time series measurement value information; Preprocessing and feature extracting the first time series measurement value information to obtain first feature data; A measurement value prediction is performed on the first characteristic data to obtain a target measurement value.

2. The method according to claim 1, characterized in that The preprocessing and feature extraction of the first time series measurement value information to obtain first feature data includes: Performing multiple data optimization processes on the first time series measurement value information to obtain second time series measurement value information; Feature extraction is performed on the second time series measurement value information to obtain first feature data.

3. The method according to claim 2, characterized in that The performing multiple data optimization processes on the first time series measurement value information to obtain second time series measurement value information includes: Filter out data having a null value ratio greater than a preset ratio from the first time series measurement value information to obtain the first time series measurement value information after null value filtering; Filling the first timing measurement value information after the null value filtering with null values ​​to obtain third timing measurement value information, wherein the third timing measurement value information includes third timing measurement value information of a plurality of production processes; Performing outlier filtering on the third time series measurement value information of each production process to obtain fourth time series measurement value information; Perform differential processing on the fourth timing measurement value information to obtain second timing measurement value information.

4. The method according to claim 2, characterized in that: The first feature data includes time domain features, frequency domain features and time-frequency domain features; The step of extracting features from the second time series measurement value information to obtain first feature data includes: Performing time domain feature extraction on the second time series measurement value information to obtain a time domain feature; Performing frequency domain feature extraction on the second time series measurement value information to obtain frequency domain features; Perform time-frequency domain feature extraction on the second time series measurement value information to obtain time-frequency domain features.

5. The method according to claim 1, characterized in that The step of predicting the measurement value of the first feature data to obtain a target measurement value includes: Performing feature screening on the first feature data to obtain second feature data; The second feature data is input into a measurement value prediction model, and the measurement value prediction is performed based on the second feature data by the measurement value prediction model to obtain a target measurement value.

6. The method according to claim 5, characterized in that Before the measurement value prediction of the first feature data is performed to obtain the target measurement value, the method includes: Acquire a first training data set, wherein the first training data set includes fifth time series measurement value information of a plurality of sample panels and a plurality of real measurement values ​​of the sample panels; Preprocessing and feature extraction are performed on the fifth time series measurement value information of each of the sample panels to obtain third feature data of a plurality of the sample panels; The preset network model is trained and verified based on the third characteristic data and the actual measurement value to obtain the measurement value prediction model.

7. The method according to claim 6, characterized in that The measurement value prediction model is used to predict the measurement values ​​of multiple measurement points, and the preset network model is trained and verified based on the third feature data and the actual measurement value to obtain the measurement value prediction model, including: Performing feature screening on the third feature data to obtain fourth feature data; Dividing the fourth characteristic data into a plurality of data groups, wherein the data ratios of the plurality of data groups are the same and the measurement points corresponding to each of the data groups are the same as the measurement points corresponding to the fourth characteristic data; Based on the plurality of data groups, constructing a plurality of second training data sets and a plurality of validation data sets, wherein the plurality of second training data sets correspond to the plurality of validation data sets respectively, and the data in the validation data set corresponding to each second training data set is the data in other second training data sets; Based on the fourth feature data, determine a training feature set, where the training feature set is a set consisting of feature data with higher importance in the fourth feature data; Based on the training feature set, determining a plurality of feature subsets; Based on the plurality of feature subsets and the plurality of the second training data sets, determining a plurality of third training data sets corresponding to each of the second training data sets; The preset network model is trained and verified based on the plurality of the third training data sets, the plurality of the verification data sets and the actual measurement values ​​to obtain the measurement value prediction model.

8. A measurement value prediction device, characterized in that: include: A first acquisition unit, used to acquire first time series measurement value information; A first extraction unit, configured to preprocess and extract features from the first time series measurement value information to obtain first feature data; The measurement value prediction unit is used to perform measurement value prediction on the first feature data to obtain a target measurement value.

9. A computer device, characterized in that: The computer device comprises: one or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the measurement value prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps in the measurement value prediction method according to any one of claims 1 to 7.

Images