Intelligent prediction method, system, device and medium for color masterbatch combination

Abstract

The present application relates to the technical field of industrial automation, and discloses an intelligent prediction method, system, device and medium for color masterbatch combinations, the method comprising: obtaining a color feature vector of a color to be predicted; inputting the color feature vector into a pre-trained multi-label prediction network for processing to obtain the appearance probabilities of multiple candidate color masterbatches, and constructing a candidate color masterbatch set according to the appearance probabilities; constructing a plurality of color masterbatch combination candidate subsets based on the candidate color masterbatch set, and verifying the combination consistency of each color masterbatch combination candidate subset through a pre-trained neural network color mixing decoder to select a color masterbatch combination feasible subset from each color masterbatch combination candidate subset according to the verification result; and screening the color masterbatch combination feasible subset through a counterfactual necessity checking mechanism to obtain the final color masterbatch combination of the color to be predicted. Without a physical model, high-precision, strong-robustness and interpretable color formula prediction is achieved.

Description

Technical Field

[0001] This invention relates to the field of industrial automation technology, and in particular to an intelligent prediction method, system, device and medium for color masterbatch combination. Background Technology

[0002] In industrial production processes such as coatings, inks, plastics, textiles, and automotive painting, color formulation prediction directly affects product consistency, raw material costs, and production efficiency.

[0003] Currently, academia and industry are beginning to try using machine learning and deep learning methods to achieve direct mapping from color to recipe. These methods collect a large amount of historical recipe data and use neural networks or regression models to learn the relationship from color features to color masterbatch ratios. However, these methods still cannot guarantee the mixability and physical consistency of predicted combinations when dealing with multi-label, strongly sparse, and extremely low-fault-tolerance color recipe prediction tasks, making it difficult to meet the actual requirements of industry.

[0004] Therefore, how to achieve high-precision, robust, and interpretable color formula prediction without the need for a physical model has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides an intelligent prediction method, system, device, and medium for color masterbatch combinations, solving the problem of how to achieve high-precision, robust, and interpretable color formula prediction without the need for a physical model.

[0006] To address the aforementioned technical problems, the first aspect of this invention provides an intelligent prediction method for color master combinations, comprising: Obtain the color feature vector of the color to be predicted; The color feature vector is input into a pre-trained multi-label prediction network for processing to obtain the occurrence probabilities of various candidate color mothers, and a candidate color mother set is constructed based on each occurrence probability. Based on the candidate color master set, several candidate subsets of color master combinations are constructed, and the combination consistency of each candidate subset of color master combinations is verified by a pre-trained neural network color mixing decoder, so as to select a feasible subset of color master combinations from each candidate subset of color master combinations according to the verification results. The feasible subset of the color numerator combinations is screened by a counterfactual necessity test mechanism to obtain the final color numerator combination of the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color numerator contained in the feasible subset of the color numerator combinations.

[0007] As one preferred embodiment, the step of constructing a candidate color set based on the occurrence probabilities of each color includes: Obtain the negative sample probability distribution of the multi-label prediction network on the validation set, and determine the conservative threshold for each candidate colorimeter based on the negative sample probability distribution; The candidate color master set is obtained by combining candidate color masters whose probability of occurrence is greater than their own conservative threshold.

[0008] As one preferred embodiment, the step of performing combination consistency verification on each of the candidate subsets of color numeral combinations using a pre-trained neural network color mixing decoder, and selecting a feasible subset of color numeral combinations from each of the candidate subsets of color numeral combinations based on the verification results, includes: Each candidate subset of the color master combination is input into the neural network color mixing decoder to simulate the color mixing process, and the reconstructed color vector is output. Calculate the color difference between the reconstructed color vector and the color feature vector, and based on the color difference, introduce an uncertainty weighted residual evaluation mechanism to determine the weighted residual value of each candidate subset of the color master combination; If the weighted residual value of any candidate subset of the color master combination is not greater than a preset residual threshold, then the candidate subset of the color master combination is determined to have passed the combination consistency verification, and the candidate subset of the color master combination is taken as the feasible subset of the color master combination.

[0009] As one preferred embodiment, the step of determining the weighted residual value of each candidate subset of color master combinations based on the color difference using an uncertainty-weighted residual evaluation mechanism includes: The confidence level of the color feature vector in each feature dimension is determined based on a pre-trained uncertainty prediction model. The squares of each confidence level are used as diagonal elements to construct the uncertainty weighting matrix of the color feature vector; Obtain the optimal ratio of each colorant in the candidate subset of colorant combinations, and determine the weighted residual value of each candidate subset of colorant combinations by combining the color difference and the uncertainty weighting matrix.

[0010] As one preferred embodiment, obtaining the optimal ratio of each colorant in the candidate subset of colorant combinations includes: Based on the color difference of the candidate subset of color master combinations and the uncertainty weighting matrix, a weighted error term is constructed; Based on the sum of the L1 norms of the non-zero ratios of all colorants in the candidate subset of colorant combinations, a sparse regularization term is constructed. A weighted residual function is constructed based on the weighted error term and the sparse regularization term. The objective function is to minimize the weighted residual function. The projected gradient descent method is used to solve the objective function to obtain the optimal ratio of each color numeral in the candidate subset of color numeral combinations.

[0011] As one preferred embodiment, the step of filtering the feasible subset of the color master combinations through a counterfactual necessity test mechanism to obtain the final color master combination of the color to be predicted includes: Counterfactual tests are performed on each of the color numerators contained in the feasible subset of the color numerator combination to obtain the counterfactual contribution of each color numerator in the corresponding feasible subset of the color numerator combination; For each feasible subset of the color master combinations, color masters whose counterfactual contribution is greater than the necessity threshold are retained to obtain a refined set of color masters as the final color master combination of the color to be predicted.

[0012] As one preferred embodiment, the step of performing counterfactual tests on each of the color numerals contained in the feasible subset of color numeral combinations to obtain the counterfactual contribution of each color numeral in the corresponding feasible subset of color numeral combinations includes: Take any one of the colorants contained in the feasible subset of the colorant combination as the base colorant and remove it to form a new subset; The weighted residual value of the new subset is calculated using the neural network color mixing decoder and the uncertainty prediction model; Based on the weighted residual values ​​of the new subset and the weighted residual values ​​of the feasible subsets of the color numerator combination, the counterfactual contribution of the basic color numerator in the feasible subsets of the color numerator combination is determined; The basic color master is updated, and the counterfactual contribution determination step is iteratively performed based on the updated basic color master until the counterfactual contribution of all color masters in the feasible subset of the color master combinations is obtained.

[0013] A second aspect of the present invention provides an intelligent prediction system for color masterbatch combinations, comprising: The data acquisition module is used to acquire the color feature vector of the color to be predicted; The set construction module is used to input the color feature vector into a pre-trained multi-label prediction network for processing, obtain the occurrence probabilities of multiple candidate color mothers, and construct a set of candidate color mothers based on each occurrence probability; The combination verification module is used to construct several candidate subsets of color numeral combinations based on the candidate color numeral set, and to perform combination consistency verification on each candidate subset of color numeral combinations through a pre-trained neural network color mixing decoder, so as to select a feasible subset of color numeral combinations from each candidate subset of color numeral combinations according to the verification results. The causal refinement module is used to filter the feasible subset of the color numeral combination through a counterfactual necessity test mechanism to obtain the final color numeral combination of the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color numeral contained in the feasible subset of the color numeral combination.

[0014] A third aspect of the present invention provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the intelligent prediction method for color master combination as described above.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium comprising a stored computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the intelligent prediction method for color master combination as described above.

[0016] Compared with the prior art, the beneficial effects of the embodiments of the present invention are as follows: By constructing a candidate color masterbatch set to avoid omissions, a foundation is provided for subsequent verification. A neural network color mixing decoder is introduced, employing a purely data-driven approach to replace complex optical models. This learns the nonlinear mixing relationship between color masterbatch combinations and color features, and verifies the consistency of the combined color masters, ensuring that the output combinations are physically mixable and reproducible, thus addressing the shortcomings of traditional models that are numerically reasonable but practically unusable. A counterfactual necessity test mechanism is introduced to determine the necessity of the color masters contained in the verified set, providing a causal basis for the selection of each color masterbatch and meeting the requirements of strong sparsity and low fault tolerance in industrial color matching. This achieves high-precision prediction while maintaining robustness and interpretability. The purely data-driven approach eliminates the need to measure color masterbatch optical constants and physical models, enabling high-precision, robust, and interpretable color formulation prediction. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an intelligent prediction method for color masterbatch combination provided in a certain embodiment of the present invention; Figure 2 This is a structural diagram of an intelligent prediction system for color masterbatch combination provided in a certain embodiment of the present invention; Figure 3 This is a structural diagram of an electronic device provided in a certain embodiment of the present invention; Figure label: Among them, 10 is the data acquisition module; 20 is the set construction module; 30 is the combination verification module; 40 is the causal refinement module; 5000 is the electronic device; 5001 is the processor; 5002 is the bus; 5003 is the memory; and 5004 is the transceiver. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings and examples. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In this invention description, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In this invention description, unless otherwise stated, "a plurality of" means two or more. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is merely for describing specific embodiments and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0022] In the early days of color science and the coatings industry, both academia and industry widely adopted color mixing models based on physical optics theory, such as the Kubelka–Munk (K–M) dual-constant model, the Saunderson correction model, and the improved Kubelka–Munk–Spencer model. These models predict the spectral reflectance after mixing by describing the absorption and scattering behavior of the coating, and have been the main theoretical basis for color formulation calculations since the 20th century.

[0023] However, such models require prior measurement of the optical constants (absorption coefficient K and scattering coefficient S) of each colorant and assume uniform medium, uniform light scattering, and linear superposition. In real-world production, non-ideal factors such as uneven particle distribution, large differences in coating thickness, and variations in lighting conditions significantly increase model errors. Furthermore, the K-M model cannot effectively handle complex systems such as multi-layer coatings, transparent pigments, or fluorescent pigments, limiting its application in modern automated color matching systems.

[0024] In recent years, with the development of artificial intelligence technology, researchers have begun to explore solutions to the color recipe prediction problem based on purely data-driven machine learning and deep learning methods, including: Regression-based methods: These methods use models such as Multilayer Perceptron (MLP), Support Vector Regression (SVR), and Random Forest (RF) to directly fit the mapping relationship between color features and color proportions. While these methods can overcome dependence on physical parameters, they still fall under single-output or independent multi-output numerical regression and cannot express the combinatorial dependencies between color proportions.

[0025] Multi-label classification methods treat each color identifier as an independent label, outputting probabilities using a sigmoid function, and training with cross-entropy or Focal Loss. These models can partially capture sparsity, but because they ignore the structural relationships and physical constraints between color identifiers, they often result in logically inconsistent or physically unmixable combinations.

[0026] Generative model attempts: Researchers such as Nippon Paint have explored using variational autoencoders (VAEs) or generative adversarial networks (GANs) to learn the distribution mapping from color to recipe, attempting to improve diversity through generative mechanisms. However, generative models lack explicit combinatorial verification mechanisms, resulting in poor interpretability and robustness of the output recipes, making them difficult to directly apply to industrial production.

[0027] Indirect learning based on spectral-to-RGB mapping: Some studies simplify the task to predicting intermediate variables in the RGB or Lab color space and then inferring the color master combination. However, this two-step mapping suffers from information loss and is sensitive to spectral resolution, resulting in insufficient accuracy.

[0028] Therefore, although existing research has made some progress in automated formulation, the following common problems still exist overall: Lack of combinatorial consistency verification mechanism: Most models only focus on single-label probability prediction and do not verify whether the predicted color combination can actually generate the target color. This causes the model's output to appear "reasonable" numerically, but fail to produce the target color physically or perceptually.

[0029] Unable to handle strong sparsity and extremely low fault tolerance requirements: In color formulation tasks, the true labels are extremely sparse and the fault tolerance requirements are extremely high. Traditional neural networks rely on average loss optimization, often pursuing overall accuracy rather than "combinatorial perfect accuracy," making them difficult to apply to quality standards in industrial production.

[0030] Sensitive to noise and weak generalization ability: The spectral features collected by the colorimeter have significantly different noise levels in different channels. Traditional models treat all dimensions the same, which leads to high noise dimensions interfering with the training process; the model has poor transferability between different instruments, light sources or batches.

[0031] Insufficient interpretability and controllability: Existing AI models typically output probabilities or regression values, lacking an explanation of the "logical relationship between colorants" and failing to provide a basis for the "necessity of ingredients," thus not meeting the human-machine collaboration and traceability requirements of industrial formulations.

[0032] Based on this, such as Figure 1 As shown, the first aspect of the present invention provides an intelligent prediction method for color master combinations, comprising: S1. Obtain the color feature vector of the color to be predicted; specifically, use a colorimeter to collect data on the color to be predicted, obtaining spectral or numerical features to construct the color feature vector. and Where N represents the feature dimension of the colorimeter output, and each component x i Reflectivity or absorptivity at a specific wavelength.

[0033] S2. Input the color feature vector into a pre-trained multi-label prediction network for processing to obtain the occurrence probabilities of various candidate color mothers, and construct a candidate color mother set based on each occurrence probability; Specifically, the system contains M basic colorants (pigments or dyes), denoted as: In the formula, It is a specific color masterbatch.

[0034] Sparse set of target formula: A target color A is made up of only a small amount of colorants. Define the true colorant set: In the formula, For sparsity, this set is the output that the model needs to predict.

[0035] Multi-label output vector: Mapped to 0–1 label vectors: In the formula, This indicates that the j-th color masterbatch is used.

[0036] The multi-label prediction network used in this invention is a two-layer fully connected neural network. It takes color feature vectors as input, maps the input, and outputs the occurrence probabilities of various candidate color numerals. This process is expressed by the following formula: In the formula, High-dimensional abstract features; , It is a non-linear activation function (usually ReLU or LeakyReLU) used to capture the complex non-linear relationship between color features and the presence of colorants; , The weight matrix and bias of the first fully connected neural network are used to map the input features from N dimensions to a high-dimensional abstract space. The Sigmoid function is used to apply the Sigmoid function to the mapping result of each colorimetric and output the probability of the existence of that colorimetric. w j , The weight vector and bias of the second fully connected neural network are used to map the high-dimensional abstract feature h to a dimension that matches the number of colorants M. Let be the probability of the j-th candidate color.

[0037] In one embodiment, constructing a candidate color set based on each of the occurrence probabilities includes: Obtain the negative sample probability distribution of the multi-label prediction network on the validation set, and determine the conservative threshold for each candidate colorimeter based on the negative sample probability distribution; The candidate color master set is obtained by combining candidate color masters whose probability of occurrence is greater than their own conservative threshold.

[0038] Specifically, to ensure that no true color numerators are missed, this invention sets a conservative threshold for label j of each candidate color numerator: The negative sample probability distribution of the multi-label prediction network on the validation set is obtained, which is the probability distribution data of samples where color numerator j was not used. Quantiles are extracted from this data to obtain the conservative threshold for each candidate color numerator. This process is expressed by the following formula: In the formula, Let be the conservative threshold for the j-th color pigment; Let be the negative sample probability distribution corresponding to the j-th color denominator, and let be the set of predicted probabilities of the j-th color denominator on all negative samples (i.e., all y in the validation set). j p corresponding to the sample with =0 j value); It is a quantile function. , which means that the probability that the value of the random variable X does not exceed that point is exactly . The threshold, in other words, is what satisfies The minimum t, 0.05 is preferred.

[0039] Then, candidate color numerators with a probability greater than their own conservative threshold are combined to obtain the candidate color numerator set. This process is represented by the following formula: In the formula, The candidate color set contains real color and a small number of candidates, which serve as input for subsequent verification. The number of candidates is usually between 10 and 20.

[0040] This invention sets a conservative threshold based on the negative sample distribution to ensure that the false negative rate of the true color is extremely low, that is, "it is better to select more than to miss". It can filter out low probability predictions caused by model noise or overfitting, making the candidate set more reliable and adapting to the distribution differences of different color colors. The threshold of each color color is calculated independently to adapt to the differences in the prediction probability distribution of different color colors.

[0041] S3. Based on the candidate color master set, construct several candidate subsets of color master combinations, and perform combination consistency verification on each candidate subset of color master combinations using a pre-trained neural network color mixing decoder, so as to select a feasible subset of color master combinations from each candidate subset of color master combinations according to the verification results; specifically, the present invention selects a subset of color master combinations from the candidate set. Construct several candidate subsets of color master combinations Each candidate subset of color combinations represents a potential color combination. These subsets can be generated using two methods: greedy search (starting with the color combination with the highest probability and gradually expanding) and beam search (concurrently retaining several optimal combination paths). Other methods can also be used; for details, please refer to the application of the algorithm in existing technologies, which will not be elaborated upon here.

[0042] In one embodiment, the step of performing combination consistency verification on each of the candidate subsets of color numeral combinations using a pre-trained neural network color mixing decoder, and selecting a feasible subset of color numeral combinations from each of the candidate subsets of color numeral combinations based on the verification results, includes: Each candidate subset of the color master combination is input into the neural network color mixing decoder to simulate the color mixing process, and the reconstructed color vector is output. Calculate the color difference between the reconstructed color vector and the color feature vector, and based on the color difference, introduce an uncertainty weighted residual evaluation mechanism to determine the weighted residual value of each candidate subset of the color master combination; If the weighted residual value of any candidate subset of the color master combination is not greater than a preset residual threshold, then the candidate subset of the color master combination is determined to have passed the combination consistency verification, and the candidate subset of the color master combination is taken as the feasible subset of the color master combination.

[0043] Specifically, traditional multi-label neural networks only output the existence probability of each colorimetric independently, lacking a consistent judgment on whether the overall combination can generate the target color. This invention introduces a neural network color mixing decoder for the first time, using a data-driven approach to learn the non-linear mixing relationship between colorimetric combinations and color features. The constructed candidate subsets of color master combinations are input into a neural network color mixing decoder to simulate the color mixing process, and the reconstructed color vector is output. This process is represented by the following formula: In the formula, Candidate subsets of color combination S The reconstructed color vector obtained by mixing is the spectral / numerical feature of the color that should theoretically be generated after the candidate subsets of the color master combination are mixed according to the ratio. It is used to directly compare with the color feature vector x of the color to be predicted. Let w be the formula ratio vector. j ≥0 indicates the non-negative weight of the j-th colorant (which needs to be solved through optimization); Here, θ is the mapping function of the multilayer perceptron (MLP), and θ is the parameter of the neural network color mixing decoder. Let be the learnable embedding vector of the j-th color matrix; h is the embedding dimension, which is learned through training to represent the optical properties of the color matrix.

[0044] The neural network color mixing decoder includes a color master embedding layer, a weighted summation layer, an MLP mapping layer, and an output layer; the color master embedding layer is for each color master c. j Assign a learnable embedding vector The weighted summation layer calculates the color embedding vectors in the candidate subset of color combinations according to the weighted summation layer w. j Weighted summation is performed for values ​​≥0 to simulate the process in actual color mixing where "the proportion of the color source determines the color contribution." The MLP mapping layer consists of 2-3 fully connected layers (with parameter θ) with non-linear activation (such as ReLU), mapping the weighted summed embedding vector to an N-dimensional space consistent with the target color features. The output layer directly outputs the reconstructed color feature vector (with the same dimension as the color to be predicted, which can be directly used for residual calculation). Then, the color difference between the reconstructed color vector and the color feature vector of the color to be predicted is calculated. .

[0045] Traditional color matching may rely solely on probability of occurrence, but high-probability color masterbatch combinations may not accurately reproduce the target color after actual mixing. This invention, by simulating the real color mixing process, shifts the verification from "chemical probability" to "physical feasibility," ensuring that the final selected color masterbatch combination can theoretically produce a result highly consistent with the target color, greatly enhancing the practical value and reliability of the solution.

[0046] In one embodiment, the step of determining the weighted residual value of each candidate subset of color master combinations based on the color difference using an uncertainty-weighted residual evaluation mechanism includes: The confidence level of the color feature vector in each feature dimension is determined based on a pre-trained uncertainty prediction model. The squares of each confidence level are used as diagonal elements to construct the uncertainty weighting matrix of the color feature vector; Obtain the optimal ratio of each colorant in the candidate subset of colorant combinations, and determine the weighted residual value of each candidate subset of colorant combinations by combining the color difference and the uncertainty weighting matrix.

[0047] Because the spectral characteristics acquired by the colorimeter exhibit significant noise differences across different dimensions, directly applying equally weighted Euclidean distances can easily lead to misclassification due to distortion caused by high noise dimensions. To address the noise differences across different wavelength dimensions, this invention introduces an uncertainty-weighted residual evaluation mechanism to calculate the weighted residual values ​​of each candidate subset of color master combinations: The color feature vector is input into the uncertainty prediction model for prediction, and the confidence scores of each feature dimension are obtained. The uncertainty prediction model is a lightweight fully connected network that shares the previous layer abstract features with the multi-label prediction network. This includes a shared feature layer (reusing the second-layer activation output h (i.e., the abstract feature vector) of the multi-label prediction network to avoid redundant feature extraction and reduce the number of parameters and computational cost) and a fully connected layer (the parameters of which are the weight matrix W). σ and bias b σ The shared feature h is mapped to a dimension that matches the feature dimension N, the activation layer (the Softplus function is applied to the output of the fully connected layer to ensure that the uncertainty of the output is greater than 0), and the output layer (the uncertainty of each dimension is output, which is directly used to construct the uncertainty weighting matrix).

[0048] The squares of the confidence scores from the model output are used as diagonal elements, and all other elements are set to 0 to construct an uncertainty weighted matrix for the color feature vector. This matrix is ​​represented by the following formula: In the formula, It is an uncertainty weighting matrix; σ i ( x ) is the model's predicted first i The uncertainty of the 3D color feature vector, that is, the color feature vector in the 3D color feature vector. i Confidence level of dimensionality .

[0049] This invention introduces an uncertainty-weighted residual evaluation mechanism to calculate an uncertainty-weighted matrix based on feature dimension confidence, making the evaluation more in line with human visual perception and the actual capabilities of the model, thereby making more accurate decisions.

[0050] In one embodiment, obtaining the optimal ratio of each colorant in the candidate subset of colorant combinations includes: Based on the color difference of the candidate subset of color master combinations and the uncertainty weighting matrix, a weighted error term is constructed; Based on the sum of the L1 norms of the non-zero ratios of all colorants in the candidate subset of colorant combinations, a sparse regularization term is constructed. A weighted residual function is constructed based on the weighted error term and the sparse regularization term. The objective function is to minimize the weighted residual function. The projected gradient descent method is used to solve the objective function to obtain the optimal ratio of each color numeral in the candidate subset of color numeral combinations.

[0051] This invention utilizes the inverse of the uncertainty weighting matrix. The weighted error term is obtained by weighting the color difference between the color vector features and the candidate subsets of color numeral combinations. It can be used to quantify the difference between the new combination color mixing result and the color to be predicted; then, the sum of the L1 norms of the non-zero ratios of all colorants in the candidate subset of colorant combinations is calculated, and the sum is multiplied by the sparsity regularization coefficient to obtain the sparsity regularization term. , The sparse regularization coefficient controls the number of non-zero weights. Its value is determined based on historical formula data and can be adjusted according to actual conditions. This invention constrains the sparsity of the proportions through the sparse regularization term, and by penalizing the total weights of non-zero proportions, it ensures that only a few non-zero proportions of colorants are retained in the optimization result. Finally, combining the weighted error term and the sparse regularization term yields the weighted residual function, also known as the consistency residual function, which is expressed by the following formula: In the formula, Candidate subsets of color combination The weighted residual value; w is the addition ratio of the j-th color pigment, which is the current ratio vector to be solved, and it is non-zero and non-negative (as a constraint).

[0052] This yields the weighted residual values ​​of all candidate subsets of color numeral combinations. Finally, the objective function is minimized using the weighted residual function, and projected gradient descent (PGD) is used to solve the objective function. The allocation ratio is then updated using unconstrained gradient descent (calculating the objective function for each w). j The gradient is used to adjust w along the negative gradient direction to reduce error; then the updated ratio is projected onto the non-negative region (if a certain w... j <0, forced to be 0 to satisfy physical constraints); repeat iterations until convergence (e.g., the gradient norm is less than a preset threshold, or the number of iterations reaches the upper limit); thus, the optimal ratio of each color numeral in the candidate subset of color numeral combinations is obtained. Alternatively, the non-negative Lasso algorithm can be used to solve the objective function; it should be noted that this invention does not improve the algorithm used, and its application in existing technologies can be referred to to understand the implementation process of this invention, which will not be elaborated upon here.

[0053] By substituting the optimal ratio of each colorant in the calculated candidate subset of colorant combinations, the color difference corresponding to the candidate subset, and the uncertainty weighting matrix into the weighted residual function, the weighted residual value of each candidate subset of colorant combinations can be obtained. This invention achieves noise adaptive weighting of the residuals, resulting in a greater contribution from reliable dimensions and a reduced impact from unreliable dimensions; it realizes dimensional noise adaptation, improving the robustness of the model under equipment fluctuations and illumination deviations.

[0054] For each candidate subset of color master combinations, if its weighted residual value is not greater than the preset residual threshold (preferably 0.1, which can be determined according to the model training process or adjusted according to the actual situation), it means that the subset has passed the combination consistency verification. It is considered that the combination can successfully reconstruct the color to be predicted and can be used as a feasible subset of color master combinations for the next step of refinement; otherwise, the combination cannot reconstruct the color to be predicted and will be eliminated. Through this mechanism, the combination consistency verification of the neural network color mixing decoder only retains those color master combinations that are truly physically feasible.

[0055] This invention uses a neural network color mixing decoder and an uncertainty-weighted residual evaluation mechanism to verify the physical consistency of the initially screened color masterbatch combinations. This process is fully automated, replacing the traditional methods that rely on expert experience or require a large number of physical sampling tests, which significantly improves the development efficiency of color formulas and reduces time and material costs.

[0056] S4. The feasible subset of the color numerator combinations is screened through a counterfactual necessity test mechanism to obtain the final color numerator combination of the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color numerator contained in the feasible subset of the color numerator combinations. In one embodiment, step S4 includes: Counterfactual tests are performed on each of the color numerators contained in the feasible subset of the color numerator combination to obtain the counterfactual contribution of each color numerator in the corresponding feasible subset of the color numerator combination; For each feasible subset of the color master combinations, color masters whose counterfactual contribution is greater than the necessity threshold are retained to obtain a refined set of color masters as the final color master combination of the color to be predicted.

[0057] Specifically, since the consistency verification stage may still retain multiple approximately correct combinations, in order to further improve accuracy and interpretability, this invention introduces a counterfactual necessity test mechanism to determine the irreplaceability of each color masterbatch in the combination.

[0058] In one embodiment, the step of performing counterfactual tests on each of the colorants contained in the feasible subset of the colorant combinations to obtain the counterfactual contribution of each colorant in the corresponding feasible subset of the colorant combinations includes: Take any one of the colorants contained in the feasible subset of the colorant combination as the base colorant and remove it to form a new subset; The weighted residual value of the new subset is calculated using the neural network color mixing decoder and the uncertainty prediction model; Based on the weighted residual values ​​of the new subset and the weighted residual values ​​of the feasible subsets of the color numerator combination, the counterfactual contribution of the basic color numerator in the feasible subsets of the color numerator combination is determined; The basic color master is updated, and the counterfactual contribution determination step is iteratively performed based on the updated basic color master until the counterfactual contribution of all color masters in the feasible subset of the color master combinations is obtained.

[0059] For any feasible subset of color numeral combinations that passes the combinatorial consistency verification, any numeral contained within it is taken as the basic numeral and removed, leaving a new subset of numerals. The weighted residual value of this new subset is calculated using a neural network color mixing decoder and an uncertainty prediction model. The difference between the newly calculated weighted residual value and the original weighted residual value yields the counterfactual contribution of the basic numerals in this feasible subset of color numeral combinations. This contribution can also be used to characterize the necessity of the basic numerals in this feasible subset. The basic numerals in this subset are updated, and the counterfactual contribution determination steps are iteratively executed based on the updated basic numerals until all numerals in the subset have been iterated over. This yields the counterfactual contribution of all numerals in the feasible subset of color numeral combinations. Based on this principle, the counterfactual contribution of all numerals in all feasible subsets of color numeral combinations can be obtained.

[0060] This invention establishes a causal relationship between color masterbatches and color mixing effects through a removal-comparison counterfactual logic. That is, by simulating the change in color mixing results after removing a certain color masterbatch, it determines whether the color masterbatch is a key component necessary for generating the target color, rather than simply relying on the probability of the color masterbatch's presence. The irreplaceability of the color masterbatch is quantified by the amount of residual change, which can accurately quantify the actual contribution of each color masterbatch in the overall color matching, avoiding rough judgments based on intuition or experience, and providing data support for formula optimization.

[0061] For each feasible subset of color numerator combinations, retain the color numerators whose counterfactual contribution is greater than the necessity threshold (preferably 0.03, which can be determined according to the model training process or adjusted according to the actual situation) to obtain the refined color numerator set. That is, remove the color numerators in the subset whose counterfactual contribution is not greater than the necessity threshold (these color numerators have little impact on the color mixing result and can be removed).

[0062] If there is only one feasible subset of the color masterbatch combination, then the refined color masterbatch set is the final color masterbatch combination for the color to be predicted. If there are multiple feasible subsets of the color masterbatch combination, the refined color masterbatch set of the feasible subset with the smallest "residual after refining" and "fewest color masterbatch quantities" (meeting the needs of reducing the types of raw materials and lowering costs in industrial production) is selected as the final color masterbatch combination for the color to be predicted. This invention introduces causal reasoning into the color formulation prediction task, achieves automatic interpretability of the combination results through counterfactual testing, provides a basis for the necessity of components for each prediction result, and improves verification efficiency.

[0063] It should be noted that the multi-label prediction network, neural network color mixing decoder, and uncertainty prediction model used in this invention are all pre-trained, and their specific training process is as follows: Obtain the training dataset, which includes at least three thousand historical color feature vectors and their corresponding real color numeral combination labels, real color features, and color numeral ratios. Perform feature normalization on the color feature vectors in the training dataset to eliminate the influence of dimensions. Divide the dataset into a training set (for parameter updates), a validation set (for hyperparameter selection), and a test set (for performance verification) in a 7:2:1 ratio.

[0064] A color numerator combination prediction model is constructed, which includes three sub-networks: a multi-label prediction network, a neural network color mixing decoder, and an uncertainty prediction model. The multi-label prediction network is used to output the existence probability of each color numerator, the neural network color mixing decoder is used to simulate the mapping from color numerator combination to color features, and the uncertainty prediction module is used to estimate the uncertainty of each dimension of color features to construct an uncertainty weighting matrix. The weights and biases of the sub-networks contained in the color numerator combination prediction model are initialized using He normal or Xavier normal methods to initialize the color numerator combination prediction model.

[0065] The training set is input into the initialized color numerator combination prediction model, allowing its constituent sub-networks to process the set and obtain corresponding output results. A joint loss function is used during the training of the color numerator combination prediction model, which is obtained by weighted summation of the multi-label classification loss (weight 1), the consistency reconstruction loss (weight preferably 0.4), and the uncertainty regularization term (weight preferably 0.2). The multi-label classification loss is used to optimize the output probability of the candidate color numerator selection network. To ensure the model can identify the presence of each color chromatic atom, the binary cross entropy (BCE) function can be used, which is expressed by the following formula: In the formula, This is the multi-label classification loss, used to constrain the model's multi-label discrimination accuracy at the sample level.

[0066] The consistency reconstruction loss is used to optimize the reconstruction capability of the neural network color mixing decoder, enabling it to correctly simulate the true color master combination of color A. Corresponding color features Defined as a minimization problem with uncertainty weighting, it is expressed by the following equation: In the formula, The consistency reconstruction loss measures whether the model can reconstruct the input color features based on the true color master combination. The smaller the residual, the more the decoder conforms to the actual color mixing rules.

[0067] The uncertainty regularization term is used to stabilize the feature dimension noise estimation σx of the network prediction, preventing the model from overconfidence or collapse. It adopts the Gaussian negative log-likelihood form and is expressed by the following equation: In the formula, This is an uncertainty regularization term used to assign lower weights to high-noise dimensions and enhance supervision of low-noise dimensions, thereby achieving adaptive weighted learning and improving model stability and robustness. The corresponding components for model reconstruction; To suppress overconfidence; To weight the residuals with confidence levels.

[0068] The joint loss function used in this invention embodies the model's "triple learning objective": through Learn the probabilistic mapping from color features to colorimetric labels; through Learn the consistency between the generation of color features and color masterbatch combinations; through By learning the reliability weights for each dimension, the model's robustness is enhanced. Therefore, the entire training process not only teaches the model which colorants to use, but also enables it to understand whether these colorants can produce a color and to self-correct based on the noise level.

[0069] Using the Adam optimizer (learning rate) The gradient is calculated through backpropagation, and all sub-network parameters are updated. The above training process is repeated, and the combined perfect accuracy is evaluated with a validation set every 10 rounds. If there is no performance improvement for 20 consecutive rounds, training is stopped and the final model parameters are saved.

[0070] This invention addresses the problem of achieving high-precision, robust, and interpretable color formula prediction without the need for a physical model. It proposes an intelligent prediction method for color masterbatch combinations. This method constructs a candidate color masterbatch set to avoid omissions, providing a foundation for subsequent verification. By introducing a neural network color mixing decoder, a purely data-driven approach replaces complex optical models to learn the nonlinear mixing relationship between color masterbatch combinations and color features. It also verifies the consistency of the combined color masters, ensuring that the output combinations are physically mixable and reproducible, thus overcoming the shortcomings of traditional models that are numerically reasonable but practically unusable. A counterfactual necessity test mechanism is introduced to determine the necessity of the color masters included in the verified set, providing a causal basis for the selection of each color masterbatch and meeting the requirements of strong sparsity and low fault tolerance in industrial color matching. This achieves both high-precision prediction and robustness and interpretability. By adopting a purely data-driven approach, it eliminates the need to measure color masterbatch optical constants and physical models, thus realizing high-precision, robust, and interpretable color formula prediction.

[0071] It should be noted that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0072] In another embodiment, such as Figure 2 As shown, a second aspect of the present invention provides an intelligent prediction system for color masterbatch combinations, comprising: Data acquisition module 10 is used to acquire the color feature vector of the color to be predicted; The set construction module 20 is used to input the color feature vector into a pre-trained multi-label prediction network for processing, obtain the occurrence probabilities of multiple candidate color mothers, and construct a set of candidate color mothers based on each occurrence probability; The combination verification module 30 is used to construct several candidate subsets of color numeral combinations based on the candidate color numeral set, and to perform combination consistency verification on each candidate subset of color numeral combinations through a pre-trained neural network color mixing decoder, so as to select a feasible subset of color numeral combinations from each candidate subset of color numeral combinations according to the verification results. The causal refinement module 40 is used to filter the feasible subset of the color numeral combination through a counterfactual necessity test mechanism to obtain the final color numeral combination of the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color numeral contained in the feasible subset of the color numeral combination.

[0073] It should be noted that each module in the aforementioned intelligent prediction system for color master combinations can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module. For specific limitations regarding the intelligent prediction system for color master combinations, please refer to the limitations of the intelligent prediction method for color master combinations described above; both have the same function and role, and will not be repeated here.

[0074] A third aspect of the present invention provides an electronic device comprising: Processor, memory, and bus; The bus is used to connect the processor and the memory; The memory is used to store operation instructions; The processor is configured to execute instructions by calling the operation instructions, causing the processor to perform operations corresponding to an intelligent prediction method for color master combination as shown in the first aspect of the present invention.

[0075] In one alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3 The illustrated electronic device 5000 includes a processor 5001 and a memory 5003. The processor 5001 and the memory 5003 are connected, for example, via a bus 5002. Optionally, the electronic device 5000 may also include a transceiver 5004. It should be noted that in practical applications, the transceiver 5004 is not limited to one type, and the structure of this electronic device 5000 does not constitute a limitation on the embodiments of the present invention.

[0076] Processor 5001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 5001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0077] Bus 5002 may include a path for transmitting information between the aforementioned components. Bus 5002 may be a PCI bus or an EISA bus, etc. Bus 5002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0078] The memory 5003 may be a ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0079] The memory 5003 is used to store application code that executes the present invention, and its execution is controlled by the processor 5001. The processor 5001 is used to execute the application code stored in the memory 5003 to implement the content shown in any of the foregoing method embodiments.

[0080] Among them, electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers.

[0081] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements an intelligent prediction method for color master combination as shown in the first aspect of the present invention.

[0082] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the foregoing method embodiments.

[0083] Furthermore, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0084] In summary, this invention relates to the field of industrial automation technology and discloses an intelligent prediction method, system, device, and medium for color master combination. The method includes acquiring the color feature vector of the color to be predicted; inputting the color feature vector into a pre-trained multi-label prediction network for processing to obtain the occurrence probabilities of multiple candidate color masters, and constructing a candidate color master set based on each occurrence probability; constructing several candidate subsets of color master combinations based on the candidate color master set, and performing combination consistency verification on each candidate subset of color master combinations through a pre-trained neural network color mixing decoder, so as to select a feasible subset of color master combinations from each candidate subset of color master combinations according to the verification results; filtering the feasible subset of color master combinations through a counterfactual necessity test mechanism to obtain the final color master combination of the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color master contained in the feasible subset of color master combinations; achieving high-precision, robust, and interpretable color formula prediction without the need for a physical model.

[0085] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0086] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for intelligent prediction of color master combinations, characterized in that, include: Obtain the color feature vector of the color to be predicted; The color feature vector is input into a pre-trained multi-label prediction network for processing to obtain the occurrence probabilities of various candidate color mothers, and a candidate color mother set is constructed based on each occurrence probability. Based on the candidate color master set, several candidate subsets of color master combinations are constructed, and the combination consistency of each candidate subset of color master combinations is verified by a pre-trained neural network color mixing decoder, so as to select a feasible subset of color master combinations from each candidate subset of color master combinations according to the verification results. The final color numerator combination for the color to be predicted is obtained by screening the feasible subset of color numerator combinations through a counterfactual necessity test mechanism. This includes: performing counterfactual tests on each color numerator contained in the feasible subset of color numerator combinations to obtain the counterfactual contribution of each color numerator in the corresponding feasible subset of color numerator combinations; for each feasible subset of color numerator combinations, retaining the color numerators with a counterfactual contribution greater than a necessity threshold to obtain a refined set of color numerators as the final color numerator combination for the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color numerator contained in the feasible subset of color numerator combinations. The step of verifying the combinatorial consistency of each candidate subset of color numeral combinations using a pre-trained neural network color mixing decoder, and selecting a feasible subset of color numeral combinations from each candidate subset of color numeral combinations based on the verification results, includes: Each candidate subset of the color master combination is input into the neural network color mixing decoder to simulate the color mixing process, and the reconstructed color vector is output. Calculate the color difference between the reconstructed color vector and the color feature vector, and based on the color difference, introduce an uncertainty weighted residual evaluation mechanism to determine the weighted residual value of each candidate subset of the color master combination; If the weighted residual value of any candidate subset of the color master combination is not greater than the preset residual threshold, then the candidate subset of the color master combination is determined to have passed the combination consistency verification, and the candidate subset of the color master combination is regarded as the feasible subset of the color master combination. The step of determining the weighted residual value of each candidate subset of color master combinations based on the color difference includes: The confidence level of the color feature vector in each feature dimension is determined based on a pre-trained uncertainty prediction model. The squares of each confidence level are used as diagonal elements to construct the uncertainty weighting matrix of the color feature vector; Obtain the optimal ratio of each colorant in the candidate subset of colorant combinations, and determine the weighted residual value of each candidate subset of colorant combinations by combining the color difference and the uncertainty weighting matrix; The step of performing counterfactual tests on each of the color numerators contained in the feasible subset of the color numerator combinations to obtain the counterfactual contribution of each color numerator in the corresponding feasible subset of the color numerator combinations includes: Take any one of the colorants contained in the feasible subset of the colorant combination as the base colorant and remove it to form a new subset; The weighted residual value of the new subset is calculated using the neural network color mixing decoder and the uncertainty prediction model; Based on the weighted residual values ​​of the new subset and the weighted residual values ​​of the feasible subsets of the color numerator combination, the counterfactual contribution of the basic color numerator in the feasible subsets of the color numerator combination is determined; The basic color master is updated, and the counterfactual contribution determination step is iteratively performed based on the updated basic color master until the counterfactual contribution of all color masters in the feasible subset of the color master combinations is obtained.

2. The intelligent prediction method for color masterbatch combination according to claim 1, characterized in that, The step of constructing a candidate color set based on the occurrence probabilities of each color includes: Obtain the negative sample probability distribution of the multi-label prediction network on the validation set, and determine the conservative threshold for each candidate colorimeter based on the negative sample probability distribution; The candidate color master set is obtained by combining candidate color masters whose probability of occurrence is greater than their own conservative threshold.

3. The intelligent prediction method for color masterbatch combination according to claim 1, characterized in that, Obtaining the optimal ratio of each colorant in the candidate subset of colorant combinations includes: Based on the color difference of the candidate subset of color master combinations and the uncertainty weighting matrix, a weighted error term is constructed; Based on the sum of the L1 norms of the non-zero ratios of all colorants in the candidate subset of colorant combinations, a sparse regularization term is constructed. A weighted residual function is constructed based on the weighted error term and the sparse regularization term. The objective function is to minimize the weighted residual function. The projected gradient descent method is used to solve the objective function to obtain the optimal ratio of each color numeral in the candidate subset of color numeral combinations.

4. An intelligent prediction system for color masterbatch combinations, characterized in that, include: The data acquisition module is used to acquire the color feature vector of the color to be predicted; The set construction module is used to input the color feature vector into a pre-trained multi-label prediction network for processing, obtain the occurrence probabilities of multiple candidate color mothers, and construct a set of candidate color mothers based on each occurrence probability; The combination verification module is used to construct several candidate subsets of color numeral combinations based on the candidate color numeral set, and to perform combination consistency verification on each candidate subset of color numeral combinations through a pre-trained neural network color mixing decoder, so as to select a feasible subset of color numeral combinations from each candidate subset of color numeral combinations according to the verification results. The causal refinement module is used to filter the feasible subset of color numeral combinations through a counterfactual necessity test mechanism to obtain the final color numeral combination of the color to be predicted. This includes: performing counterfactual tests on each color numeral contained in the feasible subset of color numeral combinations to obtain the counterfactual contribution of each color numeral in the corresponding feasible subset; for each feasible subset of color numeral combinations, retaining color numerals with a counterfactual contribution greater than a necessity threshold to obtain a refined color numeral set as the final color numeral combination of the color to be predicted; the counterfactual necessity test mechanism is based on counterfactual reasoning to determine the necessity of the existence of each color numeral contained in the feasible subset of color numeral combinations. The step of verifying the combinatorial consistency of each candidate subset of color numeral combinations using a pre-trained neural network color mixing decoder, and selecting a feasible subset of color numeral combinations from each candidate subset of color numeral combinations based on the verification results, includes: Each candidate subset of the color master combination is input into the neural network color mixing decoder to simulate the color mixing process, and the reconstructed color vector is output. Calculate the color difference between the reconstructed color vector and the color feature vector, and based on the color difference, introduce an uncertainty weighted residual evaluation mechanism to determine the weighted residual value of each candidate subset of the color master combination; If the weighted residual value of any candidate subset of the color master combination is not greater than the preset residual threshold, then the candidate subset of the color master combination is determined to have passed the combination consistency verification, and the candidate subset of the color master combination is regarded as the feasible subset of the color master combination. The step of determining the weighted residual value of each candidate subset of color master combinations based on the color difference includes: The confidence level of the color feature vector in each feature dimension is determined based on a pre-trained uncertainty prediction model. The squares of each confidence level are used as diagonal elements to construct the uncertainty weighting matrix of the color feature vector; Obtain the optimal ratio of each colorant in the candidate subset of colorant combinations, and determine the weighted residual value of each candidate subset of colorant combinations by combining the color difference and the uncertainty weighting matrix; The step of performing counterfactual tests on each of the color numerators contained in the feasible subset of the color numerator combinations to obtain the counterfactual contribution of each color numerator in the corresponding feasible subset of the color numerator combinations includes: Take any one of the colorants contained in the feasible subset of the colorant combination as the base colorant and remove it to form a new subset; The weighted residual value of the new subset is calculated using the neural network color mixing decoder and the uncertainty prediction model; Based on the weighted residual values ​​of the new subset and the weighted residual values ​​of the feasible subsets of the color numerator combination, the counterfactual contribution of the basic color numerator in the feasible subsets of the color numerator combination is determined; The basic color master is updated, and the counterfactual contribution determination step is iteratively performed based on the updated basic color master until the counterfactual contribution of all color masters in the feasible subset of the color master combinations is obtained.

5. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the intelligent prediction method for color master combinations as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the intelligent prediction method for color master combination as described in any one of claims 1 to 3.

Images