Active learning type defect detection sample labeling method inspired by quantum

Abstract

The invention relates to the technical field of sample labeling, in particular to a quantum inspired active learning type defect detection sample labeling method which can reduce the manual labeling amount and the labeling cost. Specifically, sample representativeness and uncertainty are balanced through unified process design, firstly, a center cluster set is obtained based on Hamiltonian operator clustering, secondly, the quantum entanglement degree is calculated, a representative sample set is screened, residual unlabeled sample sets are processed in combination with virtual time evolution and a Schrodinger equation, and a to-be-labeled sample set is obtained through a fusion result; a budget scene does not need to be distinguished, sample selection can be adaptively optimized, the labeling efficiency is improved, and the problem of strategy simplification is effectively solved; in addition, samples are mapped into quantum states through self-supervised learning, clustering precision is optimized based on Hamiltonian operators, quantum entanglement is utilized to quantify sample association, uncertainty is captured by means of virtual time evolution and a Schrodinger equation, the quality of labeled samples can be improved, and the problem of representation limitation of a classical framework is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sample labeling, in particular to a quantum-inspired active learning type defect detection sample labeling method. BACKGROUND

[0002] In the field of organic light-emitting diode (OLED) defect detection, a large number of detection samples will be generated during the production process of OLED products. If all samples are manually labeled, it will consume a lot of time, manpower and cost. Therefore, active learning is a key technology to solve this problem. Active learning can train a high-performance defect detection model by labeling a small number of key samples.

[0003] In the prior art, Guy Hacohen et al. published a paper entitled "Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets" in the 2022 International Conference on Machine Learning (ICML). Two classic active learning sampling strategies are proposed: in the low labeling budget scenario, the most uncertain samples of the model are selected for labeling to quickly improve the performance of the model; in the high labeling budget scenario, samples that best represent the data distribution structure are selected to optimize the generalization ability of the model. The existing technology realizes active learning through clustering and density sampling methods, and reveals the key dependence relationship between strategy effect and budget size.

[0004] However, the existing technology has significant defects. On the one hand, the strategy used has a single feature, using independent strategies in high and low budget scenarios, lacking a unified theoretical framework to balance the representativeness and uncertainty of samples, and unable to realize adaptive adjustment of the strategy. In practical applications, not only do uncertainty samples need to quickly start model training in the low budget stage, but a certain amount of representative samples are also needed to ensure data coverage. In the high budget stage, representative samples are still needed to further improve the accuracy of the model while consolidating the coverage of representative samples. The existing technology cannot take both into account, resulting in selected samples being either too conservative or too aggressive.

[0005] On the other hand, the existing technology is completely based on the classical machine learning framework and fails to take advantage of the superposition of quantum states, quantum entanglement and other characteristics in quantum mechanics, making it impossible to more finely represent the complex relationships and uncertainties between samples. The core cause of the above defects is that traditional active learning methods have limited ability to represent the internal structure of data, and lack a unified theoretical system that can simultaneously handle the two key factors of representativeness and uncertainty, resulting in the labeling efficiency and model performance being difficult to achieve ideal results when facing a large number of OLED detection samples.

[0006] Therefore, the existing technology still needs to be improved and improved. SUMMARY

[0007] In order to overcome the deficiencies of the prior art, the purpose of the present application is to provide a quantum-inspired active learning type defect detection sample labeling method, which breaks through the limitations of the prior art from the aspects of strategy framework and representation ability, and provides a more efficient and more accurate solution for OLED defect detection sample labeling.

[0008] The first aspect of the present application provides a quantum-inspired active learning type defect detection sample labeling method, comprising: obtaining a defect detection original sample set, performing quantum state encoding on the defect detection original sample set through self-supervised learning to obtain an initial labeled sample set and an initial unlabeled sample set, and setting an iteration hyperparameter; based on a predefined Hamiltonian operator, clustering the initial unlabeled sample set to obtain a center cluster set and a cluster sample set; calculating the quantum entanglement degree of each unlabeled sample in the cluster sample set and the center cluster set, and performing screening processing on the cluster sample set based on the quantum entanglement degree to obtain a representative sample set and a remaining unlabeled sample set; based on virtual time evolution and the Schrödinger equation, performing screening processing on the remaining unlabeled sample set, fusing the screening result and the representative sample set, and obtaining a to-be-labeled sample set; based on a preset defect classification standard, performing labeling processing on the to-be-labeled sample set to obtain a labeled sample set; based on the labeled sample set, updating the initial labeled sample set and the initial unlabeled sample set to obtain an updated labeled sample set and an updated unlabeled sample set, and replacing the initial unlabeled sample set with the updated unlabeled sample set, and returning to perform the clustering of the initial unlabeled sample set based on the Hamiltonian operator to obtain the center cluster set and the cluster sample set; based on the set iteration hyperparameter, judging whether an iteration stop condition is met, and if the iteration stop condition is met, outputting the updated labeled sample set as a final labeled set.

[0009] Optionally, in the first implementation manner of the first aspect, the obtaining of the defect detection original sample set, the quantum state encoding of the defect detection original sample set by the self-supervised learning, the obtaining of the initial labeled sample set and the initial unlabeled sample set, and the setting of the iteration hyperparameters include: obtaining the defect detection original sample set, the defect detection original sample set including a plurality of original samples; inputting each original sample into a self-supervised learning function, extracting a sample feature by a simCLR architecture, and mapping each original sample to a quantum state, the quantum state satisfying a normalization condition; integrating quantum states of all original samples to obtain a quantum state set, and initializing a labeled sample set and an unlabeled sample set to obtain an initial labeled sample set and an initial unlabeled sample set, the initial labeled sample set being empty, and the initial unlabeled sample set being the quantum state set; and setting iteration hyperparameters, the iteration hyperparameters being associated with a size of the defect detection original sample set, and the iteration hyperparameters including a total iteration round, a number of labeled samples per round, a clustering convergence threshold, and a maximum number of clustering iterations.

[0010] Optionally, in the second implementation manner of the first aspect, the clustering of the initial unlabeled sample set based on the predefined Hamiltonian operator to obtain the center cluster set and the cluster sample set includes: calculating a sample selection number corresponding to the current iteration round according to the current iteration round and the number of labeled samples per round; performing a random selection operation on the initial unlabeled sample set to construct an initial cluster center set, the initial cluster center set including a plurality of initial cluster centers, and the number of initial cluster centers being consistent with the sample selection number; performing iterative optimization clustering on the initial unlabeled sample set based on the predefined Hamiltonian operator and in combination with the initial cluster center set; and determining whether a clustering iteration stop condition is met based on the clustering convergence threshold and the maximum number of clustering iterations, and if the clustering iteration stop condition is met, outputting the center cluster set and the cluster sample set.

[0011] Optionally, in the third implementation manner of the first aspect, the calculation of the quantum entanglement degree of each unlabeled sample in the cluster sample set and the center cluster set, and the screening of the cluster sample set based on the quantum entanglement degree to obtain the representative sample set and the remaining unlabeled sample set include: obtaining a preset sorting criterion, performing a sorting operation on the unlabeled samples in the cluster sample set based on the preset sorting criterion to obtain a cluster sorted sample set; calculating, by using a von Neumann entropy, the quantum entanglement degree of each unlabeled sample in the cluster sorted sample set and the center cluster set; and performing a screening operation on the cluster sorted sample set based on the quantum entanglement degree to construct the representative sample set and the remaining unlabeled sample set, the number of unlabeled samples in the representative sample set being less than the number of labeled samples per round.

[0012] Optionally, in a fourth implementation form of the first aspect of the present application, the filtering processing of the remaining unlabeled sample set based on the virtual time evolution and the Schrodinger equation, and fusing the filtering result and the representative sample set to obtain the labeled sample set, comprises: calculating the quantum entanglement degree of each unlabeled sample in the remaining unlabeled sample set and the center cluster set through the von Neumann entropy; performing a sorting operation on the remaining unlabeled sample set based on the quantum entanglement degree to obtain an unlabeled sorted sample set; calculating the discard probability of each unlabeled sample in the unlabeled sorted sample set according to the Boltzmann distribution based on a preset inverse temperature parameter; performing a filtering operation on the unlabeled sorted sample set based on the discard probability, and fusing the filtering result and the representative sample set to obtain the labeled sample set.

[0013] Optionally, in a fifth implementation form of the first aspect of the present application, the labeling processing of the labeled sample set based on the preset defect classification standard to obtain the labeled sample set comprises: obtaining original OLED imaging data corresponding to the labeled sample set, and integrating the labeled sample set and the original OLED imaging data corresponding thereto to obtain expert labeling data; obtaining labeling information based on the expert labeling data, wherein the labeling information is determined based on a preset defect classification standard, and the labeling information is a defect category label of each unlabeled sample in the labeled sample set; and integrating the labeling information and the labeled sample set to obtain the labeled sample set.

[0014] Optionally, in a sixth implementation form of the first aspect of the present application, after the updating labeled sample set is output as the final labeled set, the method further comprises: constructing an OLED defect detection model to be trained, and iteratively training the to-be-trained model based on the final labeled set to obtain a trained OLED defect detection model; and inputting the updating unlabeled sample set into the trained OLED defect detection model to obtain a defect detection result corresponding to the updating unlabeled sample set.

[0015] In the technical solution of the application, the sample representativeness and uncertainty are balanced through unified process design. Firstly, a central cluster set is obtained based on Hamilton operator clustering. Secondly, a representative sample set is screened by calculating the quantum entanglement degree. The remaining unlabeled sample set is processed by combining virtual time evolution and Schrodinger equation. The result is fused to obtain a sample set to be labeled. Without distinguishing the budget scenario, the sample selection can be adaptively optimized, the labeling efficiency is improved, and the problem of single strategy is effectively solved. In addition, the sample is mapped to a quantum state through self-supervised learning. The clustering accuracy is optimized based on the Hamilton operator. The sample correlation is quantified by using the quantum entanglement. The uncertainty is captured by means of virtual time evolution and Schrodinger equation. The quality of the labeled sample can be improved, and the representation limitation of the classical framework can be solved. The technical solution is suitable for the OLED massive sample scenario. The amount of manual labeling can be reduced, the labeling cost can be reduced, the accuracy and generalization ability of the subsequent training OLED defect detection model are stronger. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A logic flow chart of a quantum-inspired active learning type defect detection sample labeling method provided for an embodiment of the application. DETAILED DESCRIPTION

[0017] The application provides a quantum-inspired active learning type defect detection sample labeling method. In the application, the terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the application and the above drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the term "comprising" or "having" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process or method including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes or methods.

[0018] The application discloses a quantum-inspired active learning type defect detection sample labeling method. For the sake of understanding, the specific process of the embodiment of the application is described below. Please refer to Figure 1 One embodiment of the quantum-inspired active learning type defect detection sample labeling method in the embodiment of the application includes: 101, obtaining a defect detection original sample set, performing quantum state coding on the defect detection original sample set through self-supervised learning to obtain an initial labeled sample set and an initial unlabeled sample set, and setting an iteration hyperparameter; In the embodiment, firstly, a large amount of defect detection original samples generated in an OLED production process are acquired, the defect detection original samples include defect samples and non-defect samples, and the defect detection original samples exist in the form of high-resolution imaging data; subsequently, a self-supervised learning technology is used to perform quantum state coding on the defect detection original samples, and the essence is to map high-dimensional imaging features of the defect detection original samples to low-dimensional quantum states with quantum mechanical characteristics, so as to finely depict complex relationships of the samples; finally, based on data scale and labeling requirements, key hyperparameters that guarantee iterative and orderly progress are set, and initialization of a labeled set is completed.

[0019] 102. Clustering the initial unlabeled sample set based on a predefined Hamiltonian operator to obtain a central cluster set and a sample set of each cluster; In the embodiment, the natural division of the unlabeled samples is realized by simulating the central force field gathering characteristics in quantum mechanics; the predefined Hamiltonian operator is an energy operator in quantum mechanics, and the core role thereof is to quantify the correlation strength between the samples and the cluster center; the clustering process is guided by the defined Hamiltonian operator, samples with similar features in the quantum state space are gathered into clusters, and finally a stable central cluster set (core quantum state of each cluster) and a sample subset corresponding to each cluster are output, thereby providing grouping basis for subsequent sample screening.

[0020] 103. Calculating a quantum entanglement degree of each unlabeled sample in the sample set of each cluster and the central cluster set, and performing screening processing on the sample set of each cluster based on the quantum entanglement degree to obtain a representative sample set and a remaining unlabeled sample set; In the embodiment, the quantum entanglement degree is a physical quantity representing the correlation strength between particles in quantum mechanics, and in this step, it is converted into a correlation measure between the sample and the central cluster set by using von Neumann entropy; the higher the entanglement degree of the sample, the more capable the sample is of representing the characteristics of the cluster to which the sample belongs and the overall data distribution; based on the quantum entanglement degree, samples are selected from the clusters to form a representative sample set covering the core distribution area of the data, and the remaining samples that are not selected constitute a remaining unlabeled sample set.

[0021] 104. Performing screening processing on the remaining unlabeled sample set based on virtual time evolution and the Schrödinger equation, and fusing the screening result and the representative sample set to obtain a sample set to be labeled; In the embodiment, the virtual time evolution characteristic of the Schrödinger equation can simulate the uncertainty distribution of the sample; by substituting the quantum entanglement degree into the Schrödinger equation as a Hamiltonian, the discard probability of the sample is derived; the lower the entanglement degree (i.e., the deviation from the central cluster) of the sample, the higher the uncertainty of the sample, and the greater the probability of being retained; the uncertainty samples selected are fused with the representative sample set to form a sample set to be labeled, which has both data coverage and model optimization value.

[0022] 105、based on the preset defect classification standard, the sample set to be labeled is labeled to obtain a labeled sample set; In the embodiment, the preset defect classification standard is a standardized defect judgment rule formulated for OLED products, including category definitions such as no defect, bright spot defect, dark spot defect, line defect, and surface defect. A technical personnel (expert) with professional knowledge performs artificial judgment according to the preset defect classification standard and in combination with the original OLED imaging data corresponding to the sample set to be labeled, and finally forms a labeled sample set containing the association relationship between sample quantum state and defect label.

[0023] 106、based on the labeled sample set, the initial labeled sample set and the initial unlabeled sample set are updated to obtain an updated labeled sample set and an updated unlabeled sample set, and the updated unlabeled sample set is used to replace the initial unlabeled sample set, and the step of clustering based on the Hamiltonian operator is executed on the initial unlabeled sample set to obtain a center cluster set and a cluster sample set; In the embodiment, the labeled sample set is added to the initial labeled sample set to complete the expansion of the labeled sample. Meanwhile, the labeled sample is removed from the unlabeled sample set, and the range of the unlabeled sample set is updated. Then, the updated unlabeled sample set is used as the input of the next iteration, and a new cycle is started in the clustering step, and the quality of the labeled sample set is gradually optimized through multiple iterations. The core logic of the iteration cycle is to optimize the clustering and sampling strategy based on the labeling results of the last round in each round, so that the samples selected in the subsequent screening are more suitable for the model training requirements.

[0024] 107、based on the set iteration hyperparameter, it is judged whether the iteration stop condition is met, if yes, the updated labeled sample set is output as a final labeled set; In the embodiment, the iteration stop condition is that the current iteration round reaches the total iteration round , or the unlabeled sample set is empty, that is, all original samples are labeled. When any condition is met, the iteration cycle is stopped, and the updated labeled sample set at this time has covered the core features of the data distribution and the key samples required for the optimization of the OLED defect detection model, and can be used as the final labeled set for the training of the subsequent OLED defect detection model. The final labeled set includes key information such as sample quantum state, defect label, and original sample index, and provides complete data support for the training of the OLED defect detection model.

[0025] The quantum-inspired active learning type defect detection sample labeling method disclosed by the application balances sample representativeness and uncertainty through a unified process design. Firstly, a center cluster set is obtained based on a Hamiltonian operator clustering. Secondly, a representative sample set is screened by calculating a quantum entanglement degree. The remaining unlabeled sample set is processed by combining virtual time evolution and a Schrodinger equation. The result is fused to obtain a sample set to be labeled. Without distinguishing budget scenarios, the sample selection can be adaptively optimized, the labeling efficiency is improved, and the problem of single strategy is effectively solved. In addition, by mapping the sample to a quantum state through self-supervised learning, the clustering accuracy is optimized based on a Hamiltonian operator, the sample correlation is quantified by using a quantum entanglement, and the uncertainty is captured by means of virtual time evolution and a Schrodinger equation, so that the quality of the labeled sample can be improved, and the representation limitation of the classical framework can be solved. The technical solution is suitable for the OLED massive sample scenario, can reduce the amount of artificial labeling and reduce the labeling cost, and makes the accuracy and generalization ability of the subsequent training OLED defect detection model stronger.

[0026] Further, in the embodiment of the application, the obtaining of the defect detection original sample set, the quantum state coding of the defect detection original sample set through self-supervised learning, the obtaining of the initial labeled sample set and the initial unlabeled sample set, and the setting of the iteration hyperparameters are as follows. 201. Obtain a defect detection original sample set, wherein the defect detection original sample set comprises a plurality of original samples. In this embodiment, the defect detection original sample set is derived from a detection device on an OLED production line and contains samples generated by OLED products under different production links and different working conditions, which not only covers various known defect samples (such as bright spots, dark spots, line defects, etc.), but also contains defect-free samples. The original samples are stored in a digital imaging data format to ensure the integrity and identifiability of the sample features.

[0027] 202. Input each original sample into a self-supervised learning function, extract sample features through a simCLR architecture, and map each original sample to a quantum state, wherein the quantum state satisfies a normalization condition. In this embodiment, it is assumed that the defect detection original sample set comprises N samples, each sample Encoding as a quantum state through self-supervised learning , wherein Input each original sample into a self-supervised learning function, wherein the self-supervised learning function is as follows: ; wherein the parameters of the self-supervised learning function need to be optimized through unsupervised pre-training, and the pre-training process aims to maximize the contrast loss of the samples, so that the quantum states of similar samples are closer and the quantum states of different samples are more distant. The dimension of the quantum state The sample feature complexity preset value is usually 64, 128 or 256; the dimension selection needs to balance the feature representation ability and the calculation efficiency; The quantum state encoding needs to use the simCLR self-supervised learning architecture, and a tanh activation function is connected to the last layer of the network to ensure that the mapped quantum state satisfies the normalization condition The simCLR architecture is a self-supervised learning framework based on contrastive learning, which has strong feature extraction ability and can automatically learn the intrinsic feature representation of the sample; by using self-supervised learning based on the simCLR architecture for quantum state encoding, the dependence on pre-training data labeled by artificial labeling is reduced, and at the same time, the deep features of the sample can be automatically learned, which is more accurate and has stronger generalization ability than traditional manual feature extraction methods; in addition, by connecting a tanh activation function to the last layer of the simCLR architecture, the extracted features are nonlinearly transformed and normalized to ensure that the output quantum state satisfies , The inner product operation of the quantum state; through normalization processing, the foundation is laid for subsequent quantum entanglement degree calculation and Hamiltonian operator clustering.

[0028] 203、Integrate all original sample quantum states to obtain a quantum state set, and initialize the labeled sample set and the unlabeled sample set to obtain an initial labeled sample set and an initial unlabeled sample set, wherein the initial labeled sample set is empty, and the initial unlabeled sample set is the quantum state set; In this embodiment, all sample quantum states are integrated into a unified quantum state set, which facilitates batch processing of subsequent clustering and screening operations; in the initial state, since no artificial labeling has been performed, the labeled sample set is set as an empty set, and the combination of all sample quantum states is included in the unlabeled sample set, ensuring that the sample selection in the initial stage covers the entire data distribution; let the current iteration round be , let be the labeled sample set in the th iteration, and let be the unlabeled sample set in the th iteration.

[0029] 204、Set iteration hyperparameters, which are associated with the size of the defect detection original sample set, and include the total iteration round, the number of labeled samples per round, the clustering convergence threshold, and the maximum number of clustering iterations; In this embodiment, the value of the iteration hyperparameter directly affects the annotation efficiency and the quality of the label set. It needs to be positively correlated with the size of the original defect detection sample set. That is, the larger the original defect detection sample set, the more iterations can be performed. The number of labeled samples per round can be adjusted according to the annotation cost. For example, when the original defect detection sample set includes 10,000 original samples, the number of labeled samples per round can be set. =20, Total number of iterations =10; By designing the correlation between hyperparameters and sample set size, the method has good adaptability and can be adapted to OLED defect detection sample sets of different sizes, thus improving the practicality and versatility of the method. The total number of iteration rounds The total number of labeled samples is determined, and the number of labeled samples in each round is also determined. To control the workload of a single round of annotation, the clustering convergence threshold and the maximum number of clustering iterations ensure the stability and efficiency of the clustering process. The clustering convergence threshold is generally set to a value of [value missing]. The maximum number of clustering iterations is typically set to 50 to 200 to prevent the clustering process from getting stuck in an infinite loop.

[0030] Furthermore, in this embodiment of the invention, the clustering of the initial unlabeled sample set based on the predefined Hamiltonian operator to obtain a set of central clusters and sample sets of each cluster includes: 301. Based on the current iteration round and the number of labeled samples per round, calculate the number of samples selected corresponding to the current iteration round; In this embodiment, the number of samples selected is the first... The number of clusters in each iteration needs to be dynamically adapted to the iteration process, that is, the cluster size in each iteration needs to be dynamically adjusted; The number of clusters in the round of iteration is , For the current iteration round, The number of labeled samples per round; by increasing the number of samples per round. The sample division is gradually refined by using clusters. In the early stage of iteration, due to the small number of labeled samples and clusters, the focus is on capturing the macroscopic distribution of the data. In the later stage of iteration, due to the increase in labeled samples and clusters, the focus is on mining the microscopic features of the data, so that the clustering results are more in line with the progressive needs of model training.

[0031] 302. Randomly select samples from the initial unlabeled sample set to construct an initial cluster center set, wherein the initial cluster center set includes multiple initial cluster centers, and the number of initial cluster centers is consistent with the number of samples selected; In the embodiment, a uniform random sampling method is adopted to select quantum states equal to the number of clusters from the unlabeled sample set as initial cluster centers, to ensure that the distribution of the initial cluster centers in the quantum state space has randomness and uniformity, to avoid clustering result bias caused by the initial cluster centers concentrated in a local area, and to lay a fair initial foundation for subsequent iterative optimization clustering.

[0032] 303、Based on the predefined Hamiltonian operator, the initial cluster center set is combined to perform iterative optimization clustering on the initial unlabeled sample set; In the embodiment, the iterative optimization clustering based on the predefined Hamiltonian operator combines quantum mechanics principles and clustering algorithms organically, realizes dynamic allocation of samples and iterative update of cluster centers through energy expectation values, and can more accurately capture the complex correlation of samples in the quantum state space compared with traditional clustering algorithms (such as K-Means), and the clustering result is more consistent with the internal distribution structure of the data; the clustering process needs to go through the iteration cycles of initial cluster center random selection, sample allocation, and cluster center update until the convergence condition is met. Predefined Hamiltonian operator For the first iteration, the Hamiltonian operator of the first cluster is the key operator for quantifying the energy state of the sample in the cluster, and for each unlabeled sample , the fidelity is calculated: ; Wherein, is the fidelity between the first sample and the first cluster center in the first iteration, which is used to measure the similarity between the sample and the cluster center. is the initial cluster center quantum state of the first cluster in the first iteration, is the cluster index, and the value range is 1 to ; The fidelity is negatively correlated with the expected energy value : ; Wherein, is the energy expectation value of the first sample to the first cluster in the first iteration, which is used for sample cluster attribution determination. During the iteration process, each sample is first assigned to the cluster with the smallest expected energy value (i.e., the highest fidelity). Then, the cluster center is updated based on the mean quantum state of all unlabeled samples within the cluster. This process is repeated until the clustering converges. During the sample assignment process, it is necessary to traverse each unlabeled sample and all initial cluster centers to calculate the corresponding expected energy value to ensure the accuracy of the sample assignment. make Indicates the first Assigned to clusters in round iteration The sample set, for each cluster Calculate the new cluster center

[0033] ; After the cluster center is updated, it needs to be normalized, that is... ; Ensure that the updated cluster center still satisfies the quantum state normalization condition; in each round of iterative optimization, the fidelity and energy expectation of the unlabeled sample and the cluster center need to be recalculated to achieve dynamic adjustment of the cluster structure.

[0034] 304. Based on the clustering convergence threshold and the maximum number of clustering iterations, determine whether the clustering iteration stopping condition is met. If it is met, output the set of central clusters and the sample set of each cluster. In this embodiment, the clustering iteration stops when the mean of the expected energy values ​​of all unlabeled samples is less than a preset clustering convergence threshold, or when the number of iterations reaches a preset maximum number of clustering iterations. When either condition is met, it indicates that the cluster structure has become stable, and further iterations cannot significantly improve the clustering quality. At this point, the final set of central clusters (updated cluster center quantum states) and the sample set corresponding to each cluster are output. The random selection of the initial cluster centers and the control of the clustering convergence threshold avoid bias and instability in the clustering results, ensuring the reliability and efficiency of the clustering process, and providing a high-quality grouping basis for subsequent representative sample selection.

[0035] Further, in this embodiment of the invention, the step of calculating the degree of quantum entanglement between each unlabeled sample in each cluster sample set and the central cluster set, and then filtering each cluster sample set based on the degree of quantum entanglement to obtain a representative sample set and a remaining unlabeled sample set, includes: 401. Obtain the preset sorting criteria, and sort the unlabeled samples in each cluster sample set based on the preset sorting criteria to obtain the sorted sample set of each cluster. In this embodiment, the screening priority of each cluster is determined through a sorting operation. The preset sorting criteria are divided into primary criteria and secondary criteria. The primary criterion is "whether the cluster contains labeled samples", and clusters containing labeled samples are sorted first. The secondary criterion is "the number of samples in the cluster". When the primary criteria are the same, clusters with more samples are sorted first. Through the preset sorting criteria, it is ensured that subsequent sample screening focuses on high-value clusters.

[0036] 402. Calculate the degree of quantum entanglement between each unlabeled sample in each cluster sorted sample set and the central cluster set using von Neumann entropy; In this embodiment, the degree of quantum entanglement is the core indicator for measuring the representativeness of a sample. The higher the correlation strength, the more representative the sample is of its cluster and the overall data distribution. By calculating the degree of quantum entanglement based on von Neumann entropy, the entanglement characteristics in quantum mechanics can be transformed into a quantitative indicator of sample representativeness. Compared with traditional representativeness measures such as density and distance, it can more accurately capture the complex correlation of samples in quantum state space and make the representativeness judgment more accurate. The calculation process requires first constructing each sample density matrix : ; in, For the first In the round of iteration, the remaining unlabeled sample set is the first... The density matrix of each sample is used to fuse the association information between the sample and all cluster centers; For the first Round iterations use Hamiltonian operators to cluster the set of central clusters. For the first The quantum states of the remaining unlabeled samples, For the central cluster set The Middle The central quantum state of each cluster; based on Constructing the density matrix To ensure the timeliness of related information; Then, solve for the density matrix. eigenvalues , density matrix No. 1 eigenvalue, , density matrix The total number of eigenvalues ​​is equal to the dimension of the quantum state. ; Will Substitute into the von Neumann entropy formula to calculate the degree of entanglement : ; in, For the first In the first iteration The degree of entanglement between unlabeled samples and the central cluster is calculated; the degree of entanglement is calculated based on eigenvalues ​​to ensure the rigor and accuracy of the calculation results. The larger the sample size, the more representative the sample.

[0037] 403. Based on the degree of quantum entanglement, a screening operation is performed on the sorted sample sets of each cluster to construct a representative sample set and a remaining unlabeled sample set, wherein the number of unlabeled samples in the representative sample set is less than the number of labeled samples in each round; In this embodiment, the most representative samples are selected from the high-priority clusters. The selection operation starts from the first cluster after sorting and selects the samples with the highest degree of quantum entanglement from each cluster in turn, until the number of selected samples is less than the number of labeled samples in each round. The selected samples constitute the representative sample set, and the unselected samples constitute the remaining unlabeled sample set, reserving space for subsequent screening of uncertain samples; only one representative sample is selected for each cluster to ensure that the representative sample set covers different clusters and improves data coverage.

[0038] Furthermore, in this embodiment of the invention, the step of filtering the remaining unlabeled sample set based on imaginary time evolution and the Schrödinger equation, and fusing the filtering results with the representative sample set to obtain the sample set to be labeled, includes: 501. Calculate the degree of quantum entanglement between each unlabeled sample in the remaining unlabeled sample set and the central cluster set using von Neumann entropy; In this embodiment, the calculation method for the degree of quantum entanglement between each unlabeled sample and the central cluster set in the remaining unlabeled sample set is consistent with step 402, ensuring the consistency and rigor of representativeness and uncertainty determination, making the mathematical basis of dual-objective screening coherent, and avoiding screening result conflicts caused by different measurement methods.

[0039] 502. Based on the degree of quantum entanglement, sort the remaining unlabeled sample set to obtain an unlabeled sorted sample set; In this embodiment, the sorting logic is "sorting by the degree of quantum entanglement from smallest to largest". Since the smaller the degree of entanglement, the higher the uncertainty of the sample, that is, the sample deviates from the core distribution, the samples with the smaller degree of entanglement are sorted first, which makes it easier to prioritize the screening of high uncertainty samples in the future and improve the marginal contribution of sample labeling to model optimization.

[0040] 503. Based on the preset inverse temperature parameter, calculate the discard probability of each unlabeled sample in the unlabeled sorted sample set according to the Boltzmann distribution; In this embodiment, to simulate the distribution characteristics of sample uncertainty, an imaginary time form of the Schrödinger equation is introduced. Imaginary time evolution can transform the probability distribution of quantum states into a stable distribution of the lowest energy state, which can adapt to the requirement of screening samples with low entanglement (high uncertainty). The imaginary time form of the Schrödinger equation is: ; in, This is the imaginary time parameter, used to simulate the probability distribution evolution of quantum states; virtual time The quantum state at a given moment; Let Hamiltonian be the degree of quantum entanglement of the samples. This equation is used to quantify the correlation between the energy state of a sample and uncertainty. Its core function is to describe the evolution of the quantum state with imaginary time, and its form determines the trend of subsequent probability amplitude changes. The solution to the Schrödinger equation is: ; in, The initial quantum state, corresponding to the initial quantum state of the sample. It is an exponential operator used to describe the evolution of a quantum state over imaginary time; Based on the solution to the Schrödinger equation, the evolved quantum state expression can be obtained, and the relationship between the evolved probability amplitude and the initial probability amplitude is as follows: ; in, virtual time After evolution, the first The quantum state probability amplitude of each sample; At the initial moment, the first The probability amplitude of each sample; the exponent term is Since the probability amplitude of a quantum state is the core intermediate quantity for measuring the likelihood of a sample being retained, and the rejection probability is positively correlated with the degree of entanglement, meaning that the more representative the sample, the more likely it is to be rejected; therefore, the Hamiltonian is set to be the negative value of the sample's degree of entanglement, so that the exponential term of the evolved probability amplitude is positively correlated with the degree of entanglement and the imaginary time, meaning that the higher the degree of entanglement of a sample, the faster its probability amplitude grows with the evolution of the imaginary time, laying the foundation for the formation of a high rejection probability for highly entangled samples in the future; The squared modulus of the probability amplitude represents the initial probability of the sample. However, the differences in probability amplitudes among different samples need to be normalized to form a mathematically sound probability distribution. Normalization is achieved by summing the squared moduli of the probability amplitudes of all samples, ensuring that the sum of the probabilities in the final distribution is 1. After normalization, the Boltzmann distribution is obtained. ; in, , Z is the partition function, Z= , for ensuring that the sum of the discard probability meets the probability distribution requirements; is any sample in the unlabeled sample set ; In order to further strengthen the sensitivity control of the discard probability to the entanglement degree, the inverse temperature parameter is introduced to convert the normalized probability distribution into a Boltzmann distribution; the value of the inverse temperature parameter is dynamically adjusted according to the current iteration round , and through the dynamic adjustment design of the inverse temperature parameter, the adaptive of the discard probability to the iteration process is realized, which can switch the screening focus at different budget stages without manual intervention, solving the problem of manual adjustment of the strategy in the prior art, and improving the automation and adaptability of the method; in the early iteration, the inverse temperature parameter is small, the sensitivity of the discard probability to the entanglement degree is low, and the representative samples are preferentially retained; in the later iteration, the inverse temperature parameter is large, the discard probability of high entanglement samples (low uncertainty) is high, and the uncertainty samples are preferentially retained; The discard probability is calculated through the Boltzmann distribution, ensuring the rationality and normativity of the probability distribution; the discard probability is positively correlated with the entanglement degree, so the calculation method of the discard probability of each sample is: ; By taking the Hamiltonian as , and obtaining in the virtual time evolution ; Among them, the partition function Z is used as a normalization factor to ensure that the discard probability is in the interval [0, 1]; the exponential term retains the correlation between the entanglement degree and , so that the discard probability of high entanglement samples is always higher than that of low entanglement samples, and finally the goal of accurately selecting uncertainty samples through quantum mechanical evolution rules is achieved.

[0041] 504、Based on the discard probability, the unlabeled ranked sample set is screened, and the screening result and the representative sample set are fused to obtain a to-be-labeled sample set; In this embodiment, the screening operation is realized by generating a uniformly distributed random number , if (the random number is greater than the discard probability), the unlabeled sample is retained; the screened uncertainty samples are fused with the foregoing representative sample set, and if the number of samples after fusion is insufficient Then, the remaining unlabeled samples are randomly supplemented to ensure that the number of the sample set to be labeled is exactly the number of samples to be labeled in each round The moderate randomness introduced by the random number judgment improves the diversity of sample screening. The fusion strategy of representative samples and uncertain samples ensures that the sample set to be labeled has data coverage and model optimization value.

[0042] Further, in the embodiment of the present application, the sample set to be labeled is labeled based on the preset defect classification standard to obtain a labeled sample set, which comprises: 601, obtaining original OLED imaging data corresponding to the sample set to be labeled, integrating the sample set to be labeled and the original OLED imaging data corresponding thereto to obtain expert-labeled data; In this embodiment, the sample set to be labeled contains the quantum state of unlabeled samples, and expert labeling needs to be based on the physical defect features of unlabeled samples, so the original OLED imaging data corresponding to each sample to be labeled is obtained through sample index association; the quantum state and the original imaging data are integrated to form expert-labeled data, so that experts can view the quantum state features and original physical features of the sample at the same time, avoid labeling errors caused by incomplete information, and improve the labeling accuracy.

[0043] 602, obtaining labeling information based on the expert-labeled data, the labeling information being determined based on the preset defect classification standard, and the labeling information being a defect category label of each unlabeled sample in the sample set to be labeled; In this embodiment, experts need to have professional knowledge of OLED defect detection and be familiar with the preset defect classification standard, which includes the definition, features and determination threshold of categories such as no defect, bright spot defect, dark spot defect, line defect and surface defect; by viewing the original OLED imaging data in the expert-labeled data, combining their own professional knowledge and the preset defect classification standard, a unique defect category label is determined for each sample to form the labeling information.

[0044] 603, integrating the labeling information and the sample set to be labeled to obtain the labeled sample set; In this embodiment, the defect category label in the labeling information is associated and integrated with the quantum state corresponding in the sample set to be labeled, and the quantum state of each sample corresponds to a unique defect label to form the labeled sample set; the standard sample set is the core data of updating the labeled sample set, and provides labeled training data for subsequent OLED defect detection model training.

[0045] Further, in the embodiment of the present application, after outputting the updated labeled sample set as the final labeled set, the method further comprises: 701. constructing an OLED defect detection model to be trained, and iteratively training the to-be-trained model based on the final labeled set to obtain a trained OLED defect detection model; In this embodiment, the to-be-trained OLED defect detection model can be a deep learning model suitable for image classification tasks, such as a convolutional neural network (CNN), a Transformer, a ResNet, etc. The structure of the OLED defect detection model needs to be designed according to the sample feature complexity and the detection accuracy requirement. In the training process, the final labeled set is used as the training data, the sample quantum state is used as the input, and the defect class label is used as the output. The model parameters of the to-be-trained OLED defect detection model are iteratively optimized through a backpropagation algorithm until the detection accuracy of the to-be-trained OLED defect detection model on the validation set reaches a preset threshold (e.g., 95%), and a trained model is obtained. The loss function used in the training process is the cross-entropy loss function, and the optimizer used is the Adam optimizer. The loss value of the model and the accuracy of the validation set need to be recorded during the training process to facilitate the analysis of the model training effect. If the model overfits, regularization techniques can be used for optimization.

[0046] In this embodiment, based on the model training of the final labeled set, the key samples selected by active learning are fully utilized, and a high-performance OLED defect detection model can be trained with only a small amount of labeled samples, greatly reducing the labeling cost and solving the problem of difficulty in labeling a large number of samples in OLED defect detection.

[0047] 702. inputting the updated unlabeled sample set into the trained OLED defect detection model to obtain a defect detection result corresponding to the updated unlabeled sample set; In this embodiment, the updated unlabeled sample set is the remaining unlabeled sample after the iteration is terminated, and the number of the updated unlabeled sample set usually accounts for the vast majority of the original defect detection sample set. The quantum state of the updated unlabeled sample set is input into the trained OLED defect detection model. The OLED defect detection model outputs the defect class prediction result of each sample by learning the mapping relationship between the quantum state and the defect class, to complete the automatic detection of OLED defects. The defect class prediction result includes sample index, predicted defect class, predicted confidence, etc. The automatic inference process realizes the rapid detection of a large number of unlabeled samples, and the detection efficiency is much higher than manual detection, meeting the large-scale and real-time detection requirements on the OLED production line, and improving the production efficiency and quality control level.

[0048] Finally, it should be noted that the above only describes the preferred examples of the present application, and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art will appreciate that modifications can be made to the technical solutions described in the foregoing embodiments, or some of the technical features thereof can be replaced equivalently, without departing from the spirit and principle of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A quantum-inspired active learning based sample labeling method for defect detection, characterized in that, The method comprises the following steps: obtain a defect detection original sample set, encode quantum states of the defect detection original sample set through self-supervised learning, obtain an initial labeled sample set and an initial unlabeled sample set, and set iteration hyperparameters; based on a predefined Hamiltonian operator, cluster the initial unlabeled sample set to obtain a center cluster set and a cluster sample set; calculate the quantum entanglement degree of each unlabeled sample in the cluster sample set and the center cluster set, and filter the cluster sample set based on the quantum entanglement degree to obtain a representative sample set and a remaining unlabeled sample set; based on virtual time evolution and Schrödinger equation, filter the remaining unlabeled sample set, fuse the filtering result and the representative sample set, and obtain a sample set to be labeled; based on a preset defect classification standard, label the sample set to be labeled to obtain a labeled sample set; update the initial labeled sample set and the initial unlabeled sample set based on the labeled sample set to obtain an updated labeled sample set and an updated unlabeled sample set, replace the initial unlabeled sample set with the updated unlabeled sample set, and return to perform the clustering of the initial unlabeled sample set based on the Hamiltonian operator to obtain the center cluster set and the cluster sample set; based on the set iteration hyperparameters, determine whether the iteration stop condition is met, and if the iteration stop condition is met, output the updated labeled sample set as a final labeled set. 2.The quantum-inspired active learning based defect detection sample labeling method of claim 1, wherein, The method comprises the following steps: obtain a defect detection original sample set, encode quantum states of the defect detection original sample set through self-supervised learning, obtain an initial labeled sample set and an initial unlabeled sample set, and set iteration hyperparameters; obtain a defect detection original sample set, wherein the defect detection original sample set comprises a plurality of original samples; input each original sample into a self-supervised learning function, extract sample features through a simCLR architecture, and map each original sample to a quantum state, wherein the quantum state satisfies a normalization condition; integrate the quantum states of all original samples to obtain a quantum state set, and initialize a labeled sample set and an unlabeled sample set to obtain an initial labeled sample set and an initial unlabeled sample set, wherein the initial labeled sample set is empty, and the initial unlabeled sample set is the quantum state set; 3.The quantum-inspired active learning based defect detection sample labeling method of claim 2, wherein, set iteration hyperparameters, wherein the iteration hyperparameters are associated with the size of the defect detection original sample set, and the iteration hyperparameters comprise total iteration rounds, the number of labeled samples per round, a clustering convergence threshold, and the maximum number of clustering iterations. The method comprises the following steps: based on the current iteration round and the number of labeled samples per round, calculate the number of sample selections corresponding to the current iteration round; perform a random selection operation on the initial unlabeled sample set to construct an initial cluster center set, wherein the initial cluster center set comprises a plurality of initial cluster centers, and the number of initial cluster centers is consistent with the number of sample selections; based on a predefined Hamiltonian operator, combine the initial cluster center set, and iteratively optimize and cluster the initial unlabeled sample set. Determine whether a clustering iteration stop condition is met based on the clustering convergence threshold and the maximum number of clustering iterations, and if the condition is met, output the center cluster set and the cluster sample set.

4. The quantum-inspired active learning-based defect detection sample labeling method of claim 2, wherein, The quantum entanglement degree of each unlabeled sample in the cluster sample set and the center cluster set is calculated, and the cluster sample set is filtered based on the quantum entanglement degree to obtain a representative sample set and a remaining unlabeled sample set, including: Obtain a preset sorting criterion, sort the unlabeled samples in the cluster sample set based on the preset sorting criterion, and obtain a cluster sorted sample set; Calculate the quantum entanglement degree of each unlabeled sample in the cluster sorted sample set and the center cluster set through von Neumann entropy; Based on the quantum entanglement degree, the cluster sorted sample set is filtered to construct a representative sample set and a remaining unlabeled sample set, and the number of unlabeled samples in the representative sample set is less than the number of labeled samples in each round.

5. The quantum-inspired active learning-based defect detection sample labeling method of claim 1, wherein, The remaining unlabeled sample set is filtered based on the virtual time evolution and the Schrödinger equation, and the filtering result and the representative sample set are fused to obtain a sample set to be labeled, including: Calculate the quantum entanglement degree of each unlabeled sample in the remaining unlabeled sample set and the center cluster set through von Neumann entropy; Based on the quantum entanglement degree, the remaining unlabeled sample set is sorted to obtain an unlabeled sorted sample set; Based on the preset inverse temperature parameter, calculate the discard probability of each unlabeled sample in the unlabeled sorted sample set according to the Boltzmann distribution; Based on the discard probability, the unlabeled sorted sample set is filtered, and the filtering result and the representative sample set are fused to obtain a sample set to be labeled.

6. The quantum-inspired active learning-based defect detection sample labeling method of claim 1, wherein, The sample set to be labeled is labeled based on the preset defect classification standard to obtain a labeled sample set, including: Obtain the original OLED imaging data corresponding to the sample set to be labeled, integrate the sample set to be labeled and its corresponding original OLED imaging data to obtain expert labeling data to be labeled; Obtain labeling information based on the expert labeling data to be labeled, the labeling information is determined based on the preset defect classification standard, and the labeling information is the defect category label of each unlabeled sample in the sample set to be labeled; Integrate the labeling information and the sample set to be labeled to obtain the labeled sample set.

7. The quantum-inspired active learning-based defect detection sample labeling method of claim 1, wherein, After outputting the updated labeled sample set as the final labeled set, it further includes: Construct an OLED defect detection model to be trained, and iteratively train the to-be-trained model based on the final labeled set to obtain a trained OLED defect detection model; Input the updated unlabeled sample set into the trained OLED defect detection model to obtain a defect detection result corresponding to the updated unlabeled sample set.

Images