Defect identification method, system, device, medium and product based on hierarchical decision

Abstract

The application provides a defect identification method, system, device, medium and product based on hierarchical decision, wherein the defect identification method based on hierarchical decision comprises: acquiring multiple criteria from different detection conditions; determining and executing a first to-be-executed criterion based on a preset decision priority; determining and executing a next to-be-executed criterion based on an execution result of the executed criterion and a state transition rule; repeating the steps of determining and executing the next to-be-executed criterion until a preset termination condition is met; and determining a defect identification result based on the execution result of the executed criterion after the preset termination condition is met. The above scheme reconstructs the defect identification from the traditional terminal state parallel judgment without execution sequence to the ordered decision process with cognitive sequence and decision logic by constructing the hierarchical decision mechanism based on the preset decision priority and the state transition rule, so that the multi-source criterion can be activated and gradually decided according to its technical characteristics and detection sensitivity.

Description

Technical Field

[0001] This application relates to the field of electronic digital data processing technology, and more specifically, to a defect identification method, system, device, medium, and product based on hierarchical decision-making. Background Technology

[0002] In the field of semiconductor manufacturing and industrial defect detection, to improve the comprehensiveness and accuracy of detection, multi-view imaging (such as top view and side view) or multi-condition imaging (such as high voltage and low voltage) techniques are often used to acquire diverse image data of the object under test. For image data from different views or different imaging conditions, existing technologies typically adopt a parallel processing mode, that is, the defect detection algorithm is executed independently for each image data, and then the output results of each detection node are fused in parallel (such as weighted summation, voting decision, or Boolean logic combination based on AND / OR / NOT), and the defect judgment is directly given based on the final fused result.

[0003] However, existing technologies typically fuse defect criteria from different detection conditions in parallel, and obtain the final judgment result at once through weighted summation, voting decision, or Boolean logic combination. This solidifies the defect identification process into a final state judgment without execution order, resulting in unnecessary detection being performed even when key criteria have clearly indicated the defect status, causing a waste of computing resources and introducing noise interference. Summary of the Invention

[0004] The purpose of this application is to provide a defect identification method, system, device, medium, and product based on hierarchical decision-making to solve the above-mentioned problems.

[0005] In a first aspect, embodiments of this application provide a defect identification method based on hierarchical decision-making. The method includes: acquiring multiple criteria from different detection conditions; determining and executing the first criterion to be executed among the multiple criteria based on a preset decision priority; determining and executing the next criterion to be executed based on the execution result of the executed criteria and state transition rules; repeating the steps of determining and executing the next criterion to be executed until a preset termination condition is met; and determining the defect identification result based on the execution result of the executed criteria after the preset termination condition is met.

[0006] In the implementation of the above scheme, by constructing a hierarchical decision-making mechanism based on preset adjudication priorities and state transition rules, defect identification is reconstructed from the traditional parallel final state judgment without execution order to an ordered decision-making process with cognitive order and adjudication logic. This enables multi-source criteria to be activated and adjudicated step by step according to their technical characteristics and detection sensitivity, improving the engineering interpretability and controllability of defect judgment. On the other hand, by introducing a dynamic criterion determination mechanism driven by state transition rules, the causal dependency between criterion executions is made explicit, so that the activation of subsequent criteria strictly depends on the execution results of previous criteria. This avoids the problem of irreconcilable evidence conflicts caused by independent calculation of each criterion in traditional parallel fusion, and enhances the robustness of multi-source evidence coordination adjudication. Furthermore, by iteratively executing the criterion determination and execution steps until the adjudication structure meets the preset termination conditions, a formal framework is provided for embedding complex decision-making logics such as short-circuit termination, evidence accumulation, and rebuttal. This allows subsequent unnecessary detections to be terminated in advance when the key criteria are sufficient to determine the result, reducing the computational load and reducing the interference of noisy evidence on the final adjudication.

[0007] In one implementation of the first aspect, the method further includes: constructing a state set including an initial state, an intermediate evidence accumulation state, and a termination state; wherein the state transition rule is used to trigger a transition from the initial state or the current intermediate evidence accumulation state to the next intermediate evidence accumulation state based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or to trigger entering the termination state when the preset termination condition is met.

[0008] In the implementation of the above scheme, by constructing a state set including an initial state, intermediate evidence accumulation state, and a termination state, the hierarchical decision-making process is formalized into a state machine model with a clear state space and transition conditions. This provides an implementable structured framework for the ordered adjudication mechanism, enhancing the feasibility and verifiability of the technical solution. On the other hand, by explicitly introducing the intermediate evidence accumulation state to represent the gradual accumulation process of defective evidence, the evidence state after each criterion execution has independent identification and storage capabilities, improving the observability of the adjudication process and the auditability of intermediate states. Furthermore, by triggering a transition mechanism from the initial state or the current intermediate evidence accumulation state to the next state based on the execution result of the executed criterion, the criterion execution sequence strictly follows the state dependency relationship, avoiding the arbitrariness of the execution order and ensuring the coherence and consistency of the hierarchical decision-making logic.

[0009] In one implementation of the first aspect, the method further includes: constructing a decision tree including a root node, internal criterion nodes, and leaf nodes; wherein the root node corresponds to the first criterion to be executed; the internal criterion nodes correspond to the next criterion to be executed; the leaf nodes correspond to the preset termination condition; and the state transition rule is used to determine a branch path from the root node or the internal criterion node to the next internal criterion node based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or, when the preset termination condition is met, to enter the leaf node.

[0010] In the implementation of the above scheme, by constructing a decision tree including root nodes, internal criterion nodes, and leaf nodes, the hierarchical decision-making process is formalized into a tree structure with hierarchical topological relationships. This provides an intuitive branch path expression framework for the ordered adjudication mechanism, enhancing the visualization and interpretability of the decision logic. On the other hand, by assigning the root node to the first criterion to be executed, the internal criterion nodes to the next criterion to be executed, and the leaf nodes to the preset termination conditions, a direct mapping relationship is established between the decision tree nodes and the criterion execution and termination conditions. This improves the structural clarity and node management convenience of the above-mentioned defect identification method based on hierarchical decision-making during implementation. Furthermore, by determining the branch path from the root node or internal criterion node to the next internal criterion node based on the execution result of the executed criterion, or by directly entering the leaf node, conditional branching and dynamic path selection between criters are realized. This allows the decision tree to adaptively adjust the adjudication path according to the actual detection results, enhancing the adaptability and flexibility of defect identification to different evidence scenarios.

[0011] In one implementation of the first aspect, the preset termination condition includes a preset short-circuit condition; After determining and executing the first criterion to be executed from among the multiple criteria based on a preset decision priority, the method further includes: determining whether the execution result of the executed criterion satisfies a preset short-circuit condition; wherein the preset short-circuit condition includes at least one of the following: a key criterion short-circuit condition, a threshold short-circuit condition, an evidence accumulation short-circuit condition, a consistency short-circuit condition, and a rebuttal evidence short-circuit condition; when the execution result of the executed criterion satisfies the preset short-circuit condition, it is determined that the preset termination condition is satisfied; when the execution result of the executed criterion does not satisfy the preset short-circuit condition, it is determined that the preset termination condition is not satisfied, and based on the execution result of the executed criterion and the state transition rule, the next criterion to be executed is determined and executed.

[0012] In implementing the above scheme, by introducing diverse pre-defined short-circuit conditions, including key criterion short-circuit, threshold short-circuit, evidence accumulation short-circuit, consistency short-circuit, and rebuttal short-circuit, the adjudication process can be terminated early based on decisive evidence from different dimensions, avoiding the waste of computational resources in performing unnecessary detections when key criteria have fully indicated the defect status. On the other hand, by identifying the defect determination result of the pre-defined decisive criteria through the key criterion short-circuit condition, and by identifying the quantitative indicator compliance status through the threshold short-circuit, evidence accumulation short-circuit, and consistency short-circuit conditions, the process can be terminated immediately when high-sensitivity criteria or accumulated evidence confirms the defect, shortening the decision path and improving the defect response speed. Furthermore, by identifying strong negative evidence through the rebuttal short-circuit condition, the forward detection process can be terminated in a timely manner when the defect hypothesis is clearly negated, avoiding the continued advancement of invalid calculations in scenarios where no defects have been determined.

[0013] In one implementation of the first aspect, determining and executing the next pending criterion based on the execution result of the executed criterion and the state transition rule includes: determining a target rule in the state transition rule based on the execution result of the executed criterion; wherein the state transition rule includes at least one of the following: sequence triggering rule, condition triggering rule, threshold triggering rule, evidence accumulation triggering rule, consistency triggering rule, conflict triggering rule, short-circuit triggering rule, evidence of contradiction triggering rule, and re-examination triggering rule; and determining and executing the next pending criterion based on the target rule.

[0014] In the implementation of the above scheme, a diversified set of state transition rules is constructed, including sequential triggering, conditional triggering, threshold triggering, evidence accumulation triggering, consistency triggering, conflict triggering, short-circuit triggering, evidence of contradiction triggering, and re-examination triggering. The target rule is dynamically selected based on the execution result of the executed criteria, so that the transition mechanism between criteria can adapt to the evidence characteristics and adjudication requirements under different detection scenarios, thereby improving the flexibility of hierarchical decision-making and scenario adaptability.

[0015] In one implementation of the first aspect, determining the target rule among the state transition rules based on the execution result of the executed criteria includes: when the execution result of the executed criteria satisfies multiple state transition rules, determining the target rule among the multiple state transition rules based on a preset rule priority.

[0016] In the implementation of the above scheme, by introducing a preset rule priority for target rule arbitration when the execution result of the executed criterion satisfies multiple state transition rules, the problem of uncertainty in the adjudication path caused by multiple triggering condition competition conflicts is solved, ensuring the determinism and uniqueness of state transition. On the other hand, by using the preset rule priority, a clear adjudication order standard is provided for multiple triggering rules that are satisfied at the same time, avoiding logical confusion or decision deadlock caused by concurrent activation of multiple rules, and ensuring the coherent execution of the hierarchical decision-making process. Furthermore, the explicit priority arbitration mechanism makes the rule selection process in complex scenarios predictable and interpretable, facilitating engineering debugging and rule optimization.

[0017] In one implementation of the first aspect, the execution result of the executed criterion includes at least one of the following: Boolean decision type, discrete state type, scalar metric type, vector feature type, structured object type, interval constraint type, probability confidence type, and symbolic state type.

[0018] In the implementation of the above scheme, by supporting various criterion output forms such as Boolean judgment type, discrete state type, scalar metric type, vector feature type, structured object type, interval constraint type, probability confidence type, and symbolic state type, the hierarchical decision framework can be compatible with heterogeneous data types output by different detection algorithms, improving the adaptability and versatility of the technical solution to multi-source detection conditions. On the other hand, by supporting diverse data types, a single criterion can express structured evidence information ranging from simple binary logic to complex multidimensional features, enhancing the information dimension and expressive power of defect judgment. Furthermore, by matching different types of execution results with state transition rules, the triggering mechanism can be flexibly executed based on various judgment logics such as Boolean judgment, threshold comparison, probability evaluation, or symbolic state, improving the adaptation accuracy and execution flexibility of the adjudication rules.

[0019] In one implementation of the first aspect, the defect identification result includes at least one of the following: defect category information, confidence level information, evidence summary information, decision path information, anomaly level information, spatial region information, and consistency status information.

[0020] In the implementation of the above scheme, by encompassing defect category information, confidence level information, evidence summary information, decision path information, anomaly level information, spatial area information, and consistency status information in the defect identification results, a multi-dimensional adjudication status output is constructed. This ensures that the final judgment not only includes the conclusion of whether a defect exists, but also carries the judgment basis, process traceability, and risk assessment information, thereby enhancing the engineering practical value and decision support capability of defect judgment. On the other hand, through defect category information and anomaly level information, the output results can directly guide subsequent defect classification and processing and risk response strategy formulation, enhancing the pertinence of defect management. Furthermore, by recording decision path information and consistency status information, the adjudication process possesses complete audit traceability and evidence conflict visualization capabilities, facilitating process backtracking and algorithm debugging.

[0021] In one implementation of the first aspect, the method further includes: when a preset degradation constraint condition is met, performing parallel execution on all the criteria, and determining the defect identification result based on the execution results of all the criteria.

[0022] In the implementation of the above scheme, a degradation mechanism is introduced to execute all criteria in parallel when preset degradation constraints are met and to determine the defect identification result based on the execution results of all criteria. This enables the hierarchical decision-making method to maintain its orderly decision-making capability while being compatible with traditional parallel fusion modes, thus enhancing the adaptability of the technical solution to different application scenarios and performance requirements. On the other hand, in simple detection scenarios or production environments with high real-time requirements, the timing overhead caused by orderly decision-making is avoided by triggering degradation constraints to switch to parallel execution mode, thereby improving the processing efficiency and response speed of defect identification. Furthermore, the degradation mechanism achieves backward compatibility with existing parallel decision-making technologies, enabling traditional defect identification schemes based on Boolean logic or weighted fusion to smoothly introduce hierarchical decision-making capabilities while maintaining the existing architecture, reducing the cost of technology migration and transformation.

[0023] In one implementation of the first aspect, the different detection conditions include top-view imaging conditions and side-view imaging conditions; The plurality of criteria include a side-view structure criterion corresponding to the side-view imaging condition and a top-view criterion corresponding to the top-view imaging condition; wherein, the side-view structure criterion is used to evaluate the consistency of interlayer structure, and the top-view criterion is used to assist in the determination of defect causes. The step of determining and executing the first criterion to be executed among multiple criteria based on a preset decision priority includes: determining the side view structure criterion as the first criterion to be executed and executing it; The process of determining and executing the next pending criterion based on the execution result and state transition rules of the executed criterion includes: when the execution result of the side view structure criterion indicates that the structure is consistent, the top view criterion is determined as the next pending criterion and executed.

[0024] In the implementation of the above scheme, by using the side-view structure criterion as the first criterion to evaluate the interlayer structural consistency, and activating the top-view criterion for cause-aid judgment when the structure is consistent, a hierarchical adjudication process of first structure and then cause, which conforms to the cognitive order of semiconductor manufacturing process, is constructed. This avoids the waste of computational resources by performing unnecessary top-view analysis when structural defects have been confirmed. On the other hand, by prioritizing the execution of the side-view structure criterion to evaluate the interlayer structural consistency, the high sensitivity of the side view to the structural continuity of interlayer contours, sidewall morphology, and depth direction is utilized to achieve rapid identification and early interception of structural defects. Furthermore, by activating the top-view criterion only when the side-view structure criterion indicates structural consistency, the top-view analysis focuses on candidate samples with normal structures for cause-aid judgment, improving the accuracy and effectiveness of cause attribution analysis.

[0025] In one implementation of the first aspect, the different detection conditions include low-voltage imaging conditions and high-voltage imaging conditions; The plurality of criteria include a low-voltage criterion corresponding to the low-voltage imaging condition and a high-voltage criterion corresponding to the high-voltage imaging condition; wherein, the low-voltage criterion is used for on-site defect trigger detection, and the high-voltage criterion is used for defect classification and previous station source determination. The step of determining and executing the first criterion to be executed among multiple criteria based on a preset decision priority includes: determining the low voltage criterion as the first criterion to be executed and executing it. The step of determining and executing the next pending criterion based on the execution result of the executed criterion and the state transition rule includes: when the execution result of the low voltage criterion is consistent, determining the high voltage criterion as the next pending criterion and executing it.

[0026] In the implementation of the above scheme, by using the low-voltage criterion as the first execution criterion for triggering defect detection at the current station, and activating the high-voltage criterion for defect classification and previous station source determination when the execution results are consistent, a layered adjudication process that conforms to the cognitive order of voltage comparison detection process is constructed, which is to first determine the current station and then the previous layer. This avoids the waste of computational resources in performing previous layer penetration analysis when the defect has been confirmed at the current station. On the other hand, by prioritizing the execution of the low-voltage criterion and utilizing the non-penetrating characteristic of low-voltage imaging to sensitively detect defects at the current station, the rapid discovery and early interception of real defect samples in the current layer are achieved, improving the response speed and detection efficiency of defects at the current station. Furthermore, by activating the high-voltage criterion only when the low-voltage criterion indicates consistency, the high-voltage penetration analysis focuses on normal candidate samples at the current station to trace back the problems left over from the previous layer, achieving accurate classification of defect sources and inter-layer attribution.

[0027] Secondly, embodiments of this application provide a defect identification system based on hierarchical decision-making, comprising a criterion acquisition module, a hierarchical decision-making module, and a result output module connected in sequence, wherein: The criterion acquisition module is used to acquire multiple criteria from different detection conditions; The hierarchical decision module is used to determine and execute the first criterion to be executed among multiple criteria based on a preset decision priority, and to determine and execute the next criterion to be executed based on the execution result and state transition rules of the executed criteria, and to repeat the determination and execution steps of the next criterion to be executed until a preset termination condition is met. The result output module is used to determine the defect identification result based on the execution result of the executed criteria after the preset termination condition is met.

[0028] Thirdly, embodiments of this application provide an electronic device, including: a processor, a memory, and a communication bus, wherein the processor and the memory communicate with each other through the communication bus; the memory stores computer program instructions that can be executed by the processor, and the computer program instructions are read and executed by the processor to perform the method provided in the first aspect or any possible implementation of the first aspect.

[0029] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the method provided in the first aspect or any possible implementation thereof.

[0030] Fifthly, embodiments of this application provide a computer program product, the computer program product including a computer program, which, when executed by a processor, implements the method provided by the first aspect or any possible implementation of the first aspect.

[0031] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims and drawings. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 A flowchart illustrating an existing defect identification scheme provided in an embodiment of this application; Figure 2 A flowchart illustrating the defect identification method based on hierarchical decision-making provided in this application embodiment; Figure 3 A schematic diagram of the adjudication process for an ordered adjudication scenario, provided in the top and side views for embodiments of this application; Figure 4 A schematic diagram of the decision-making process for a stratified determination scenario of high voltage and low voltage conditions provided in an embodiment of this application; Figure 5 A schematic diagram of the architecture of a defect identification system based on hierarchical decision-making provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0034] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0035] like Figure 1As shown, existing defect identification schemes mainly adopt model-driven or differential-driven technical approaches. For image data from single or multiple views and different imaging conditions, one or more detection algorithms are used for parallel analysis. After receiving multi-view or multi-condition input, features are extracted through parallel detection models or differential calculations, and the detection results from each channel are directly sent to the fusion module. All criteria are calculated synchronously and uniformly adjudicated without temporal dependence. Then, the detection results from different views or different imaging conditions are fused in parallel. Fusion methods include weighted summation, voting decision, or Boolean logic combination, and the fusion result is used as the final defect determination basis.

[0036] In addition, some schemes also introduce reference images or template images for difference or similarity calculation, but this reference information is only used as a one-time calculation input, and its result directly participates in the final judgment. It is not used to build a decision process with execution order and intermediate states.

[0037] The aforementioned scheme treats the defect identification process as a one-time parallel decision-making problem, where all criteria are calculated synchronously and decided uniformly under conditions of no temporal dependence. This fails to express the need for layer-by-layer decision-making based on evidence sensitivity and process understanding in engineering practice. Due to the lack of explicit modeling of intermediate criterion states, the existing scheme struggles to support complex decision structures such as short-circuit termination, disproven evidence, and multi-path decision-making. This results in unnecessary detections being performed even when key evidence is sufficient to determine the defect status, leading to wasted computational resources and the introduction of noise interference.

[0038] In view of this, this application provides a defect identification method based on hierarchical decision-making. This method reconstructs defect identification from the traditional parallel final state judgment without execution order to an ordered decision-making process with cognitive order and judgment logic by constructing a hierarchical decision-making mechanism based on preset adjudication priority and state transition rules. This enables multi-source criteria to be activated and adjudicated step by step according to their technical characteristics and detection sensitivity, improving the engineering interpretability and controllability of defect judgment. On the other hand, by introducing a dynamic criterion determination mechanism driven by state transition rules, the causal dependency between criteria execution is made explicit, so that the activation of subsequent criteria strictly depends on the execution result of the preceding criteria. This avoids the problem of irreconcilable evidence conflicts caused by independent calculation of each criterion in traditional parallel fusion, and enhances the robustness of multi-source evidence coordination adjudication. Furthermore, by iteratively executing the criterion determination and execution steps until the adjudication structure meets the preset termination conditions, a formal framework is provided for embedding complex decision-making logics such as short-circuit termination, evidence accumulation, and rebuttal. This allows subsequent unnecessary detections to be terminated in advance when the key criteria are sufficient to determine the result, reducing the computational load and reducing the interference of noisy evidence on the final adjudication.

[0039] Please see Figure 2The illustrated flowchart illustrates a hierarchical decision-based defect identification method provided in this application embodiment. This hierarchical decision-based defect identification method can be applied to electronic devices, which may include physical devices such as servers, PCs, tablets, or smartphones, or virtual devices such as virtual machines or containers. The electronic device can be a single device, a combination of multiple devices, or a cluster of a large number of devices. The aforementioned hierarchical decision-based defect identification method may include: Step S110: Obtain multiple criteria from different detection conditions.

[0040] The aforementioned different detection conditions refer to imaging configurations used in the defect identification process that produce differentiated defect feature manifestations. These differences can stem from variations in imaging geometry or imaging physical mechanisms. These differentiated settings allow the same object to exhibit complementary defect-sensitive features under different conditions, thus providing a multi-source evidence basis for stratified decision-making. In terms of imaging geometry, detection conditions can include imaging configurations with different spatial perspectives. For example, top-view imaging acquires surface planar morphology information from a vertical perspective, side-view imaging acquires interlayer structural depth information from an oblique perspective, and multi-angle surround imaging acquires omnidirectional sidewall morphology information from a circumferentially distributed perspective. In terms of imaging physical mechanisms, detection conditions can include imaging configurations with different energy levels or different spectral characteristics. For example, low-voltage electron beam imaging acquires surface defect information by limiting penetration depth, high-voltage electron beam imaging acquires deep structural information by enhancing penetration capability, visible light imaging acquires surface reflection characteristics through specific wavelengths, and infrared imaging acquires subsurface defect information through thermal radiation characteristics. In terms of imaging time sequence, detection conditions can also include imaging configurations at different process stages or different time points. For example: pre-etching imaging conditions are used to obtain the state before pattern transfer, post-etching imaging conditions are used to obtain the morphology after pattern transfer, pre-film formation imaging conditions are used to obtain the substrate state, and post-film formation imaging conditions are used to obtain the integrity information of the capping layer.

[0041] Criteria are structured evidence representations output by detection nodes after performing defect detection calculations. They originate from detection algorithms or models configured for different detection conditions and are used to carry the identification results of defect features under specific perspectives or physical conditions. Each criterion corresponds to a specific detection execution unit. This unit extracts defect-related feature information based on input image or signal data through preset detection logic and encapsulates this information into a standardized output format that can be processed by subsequent decision logic. The core function of a criterion is to characterize the structural anomalies, consistency disruptions, or other identifiable features of potential defects. Its output reflects the judgment on whether a defect exists or its degree under specific conditions. Criteria generated by different detection conditions have complementary representational capabilities. For example, criteria generated by side-view imaging focus on evaluating interlayer structural consistency, criteria generated by top-view imaging focus on describing surface planar morphology, criteria generated by low-voltage imaging focus on capturing surface-sensitive defects, and criteria generated by high-voltage imaging focus on analyzing deep penetrating structures.

[0042] In the above schemes, criteria can be obtained through a model-driven approach. For example, image data collected under different detection conditions can be input into a pre-trained detection model, and the feature extraction and defect identification calculations can be performed using the representation capabilities of deep learning networks for complex visual patterns, generating structured criteria that include defect confidence, abnormal region coordinates, and type encoding. Criteria can also be obtained through a difference-driven approach. For example, pixel-level difference operations or feature-level similarity calculations can be performed between the image to be inspected and a reference image under corresponding detection conditions, generating scalar metric criteria based on the difference measure to determine the degree of integrity breach or structural deviation. Criteria can also be obtained through feature engineering. For example, dedicated geometric structure, texture statistics, or frequency domain feature extraction algorithms can be designed for the physical imaging characteristics of different detection conditions, and discrete state-type or Boolean decision-type criteria with clear physical meaning can be generated based on preset thresholds or classification rules. In addition, criteria can be obtained through multi-source fusion and adaptive methods. For example, multiple acquisition methods can be dynamically combined according to the characteristics of detection conditions. High-sensitivity differential detection can be used for rapid screening under low voltage conditions, and deep models can be used for fine classification under high voltage conditions. Alternatively, edge detection algorithms can be used to extract interlayer contour features under side-view conditions, thereby obtaining complementary structured criteria in different detection dimensions.

[0043] Step S120: Based on the preset decision priority, determine the first decision to be executed among multiple decisions and execute it.

[0044] The aforementioned preset decision priority is a ranking criterion used to select the first execution node among multiple candidate criteria. This criterion is pre-set based on the sensitivity of each criterion to defect features, imaging physical characteristics, or process cognition order under different detection conditions. Its aim is to prioritize the criteria with the highest defect identification sensitivity or that best reflects key structural information in the decision-making process. The preset decision priority can be a static definition based on expert knowledge, where domain technicians set the priority order among criteria based on process experience. For example, criteria evaluating interlayer structural consistency can be prioritized over criteria analyzing surface morphology causes, or criteria under non-penetrating imaging conditions can be prioritized over criteria under penetrating imaging conditions. The preset decision priority can also be a dynamic optimization based on historical data statistics. By analyzing the detection performance of past defect samples, the detection rate and response speed indicators of each criterion under different defect types are extracted, and the priority order among criteria is adjusted accordingly. The preset decision priority can also be an adaptive form based on machine learning models. By training classification or ranking models, the mapping relationship between criterion combinations and defect identification accuracy is learned, automatically generating a criterion execution priority sequence that maximizes detection performance.

[0045] An optional implementation of step S120 above, which determines the first criterion to be executed, includes: first, obtaining the priority attributes pre-assigned to each criterion, which characterize the sensitivity of the criterion to defect features or the importance of process cognition under different detection conditions; then, traversing the priority attributes of multiple candidate criters and comparing and ranking them, and determining the criterion with the highest priority attribute as the first criterion to be executed. In scenarios based on static definitions of expert knowledge, priority attributes are stored in the form of discrete levels or sequence numbers, and the determination process directly selects the criterion identified as the highest level or the lowest number; in scenarios based on dynamic optimization of data statistics or machine learning, priority attributes exist in the form of continuous weight values ​​or probability distributions, and the determination process calculates the comprehensive priority score of each criterion and selects the one with the largest score; after the determination is completed, the detection node corresponding to the criterion is activated to perform defect detection calculations, and its structured execution results are obtained as the input basis for subsequent state transitions.

[0046] Execution criteria refer to the process of activating a specific detection node and running its defect identification logic to obtain structured output results. This process transforms the criteria from static configuration to dynamic evidence data. Criterion execution methods include: full execution, which involves independently running the entire algorithm flow of the detection node, from the original image input through feature extraction, pattern matching or difference operations to generating structured criterion output; incremental execution, which reuses some feature data based on existing intermediate calculation results, executing only subsequent inference or judgment steps to accelerate criterion generation; conditional execution, which selectively executes some sub-modules of the detection node based on the preceding state or context information, skipping unnecessary computational branches in specific scenarios to reduce execution overhead; and parallel pre-execution, which preprocesses lower-priority candidate criteria in the background while determining the first criterion to be executed, directly obtaining the pre-execution results when activation is needed to shorten the state. Step S130: Based on the execution results of the executed criteria and the state transition rules, determine and execute the next criterion to be executed.

[0047] Optionally, the execution result of the above-mentioned executed criteria includes at least one of the following: Boolean decision type, discrete state type, scalar metric type, vector feature type, structured object type, interval constraint type, probability confidence type, and symbolic state type.

[0048] The execution result of the executed criteria is a structured output of the detection node after identifying the defect features under specific detection conditions. Its form is diverse to adapt to the output characteristics of different detection algorithms and the processing requirements of subsequent adjudication logic. The form of the execution result may include at least one of the following: (1) Boolean judgment type execution result represents whether a certain defect condition is met in binary logic form. The output is a logical value of 1 or 0 or a status indicator of Pass and Fail. It is suitable for threshold judgment scenarios with clear pass / fail standards. (2) Discrete state type execution result describes the category or anomaly level of the object with a finite discrete state set. The output is an enumeration value such as {Normal, Weak_Anomaly, Strong_Anomaly} or {TypeA, TypeB, Unknown}. It is suitable for scenarios that need to grade and evaluate the severity of defects. (3) Scalar measurement type execution result represents the degree of anomaly, similarity or deviation with real number field numerical quantification. The output is a continuous variable such as anomaly_score, similarity_score or deviation_value. It is suitable for scenarios that need to accurately measure the intensity of defects. (4) Vector feature type execution result uses multi-dimensional feature vectors to jointly describe multiple related indicators, and the output is a real number vector of [f1,f2,...,fn]. It is suitable for scenarios that require comprehensive multi-dimensional features for joint judgment. (5) Structured object type execution result uses a composite data structure containing multiple fields to carry rich criterion information. The fields may include regional coordinates, confidence level, anomaly type code and additional metadata. It is suitable for scenarios that require the transmission of complex context information. (6) Interval constraint type execution result uses numerical interval relationship or threshold constraint to characterize whether the indicator is within the allowable range or the abnormal range. The output form is value∈[lower,upper] or value>threshold. It is suitable for fast screening scenarios based on boundary checks. (7) Probability confidence type execution result uses probability value or confidence level in the interval of 0 to 1 to characterize the possibility that a certain defect hypothesis is true. The output is a normalized value of probability or confidence. It is suitable for uncertainty reasoning or scenarios that require risk quantification. (8) Symbolic state type execution results use symbolic identifiers to describe the logical state or decision stage of evidence. The output is discrete symbols such as Evidence_Accumulated, Evidence_Contradicted or Evidence_Insufficient, which are suitable for expressing the abstract logical state in the process of evidence accumulation.

[0049] The execution results of the above different forms can all be used as judgment conditions in the state transition rules. Through Boolean logic operations, threshold comparison, interval inclusion checks or symbol matching mechanisms, the decision to transition from the current state to the next state or to enter the termination state is triggered, so that the hierarchical decision framework can be compatible with heterogeneous data types and execute unified decision logic.

[0050] Optionally, step S130 above includes: determining a target rule in the state transition rules based on the execution result of the executed criteria; wherein the state transition rules include at least one of the following: sequence triggering rule, condition triggering rule, threshold triggering rule, evidence accumulation triggering rule, consistency triggering rule, conflict triggering rule, short-circuit triggering rule, evidence of contradiction triggering rule, and re-examination triggering rule; and determining and executing the next criterion to be executed based on the target rule.

[0051] The aforementioned state transition rules are a set of transition logics defined within a hierarchical decision structure. They determine the direction of the adjudication process based on the output characteristics of executed criteria. These rules operate between state nodes in a state machine or between branch nodes in a decision tree, establishing causal dependencies from preceding criteria to subsequent criteria. State transition rules can employ various triggering mechanisms, each corresponding to different evidence processing logic and state transition conditions, enabling the adjudication process to adapt to various detection scenarios, from simple sequential execution to complex evidence fusion. Target rules are specific rule instances activated from the set of state transition rules in the current adjudication stage to guide the determination of subsequent criteria. Their determination process is based on the matching verification between the execution results of executed criteria and various triggering rules. When the output of an executed criterion satisfies the preset conditions of a triggering rule, that rule is identified as the target rule, becoming the sole basis for transition in the current state, ensuring the determinism and predictability of the adjudication path.

[0052] In its formal representation, state transition can be abstracted as a function S×C→S, where S is the state space and C is the set of criterion variables. This set of rules contains multiple triggering mechanisms, each corresponding to different evidence processing logic and state transition conditions. These triggering mechanisms include: (1) is a state transition mechanism that determines the next criterion to be executed based on a predefined fixed execution sequence. This mechanism is independent of the specific output content or logical judgment result of the executed criterion, and only activates subsequent nodes in the order of the pre-arranged criterion identifiers to ensure that the adjudication process follows the preset process steps or detection logic to advance linearly; its abstract expression is C i →C i+1 , where C i This indicates that the current judgment has been executed, C. i+1This indicates the next criterion to be executed immediately in the sequence. By eliminating the need for conditional judgments on the output of previous criteria, this rule simplifies the state transition logic and is suitable for pipeline scenarios with clear process step dependencies and where each stage of detection must be fully executed, thus ensuring the integrity and determinism of the detection sequence.

[0053] (2) Condition-triggered rules are a state transition mechanism that determines the subsequent decision path based on the logical matching relationship between the output state of the executed criterion and the preset condition expression. This mechanism compares the criterion output value with a specific condition using Boolean judgment or discrete state comparison. When the preset condition is met, it activates the corresponding next criterion node to be executed or transitions to a specified state. Its abstract expression is if condition(C i then goto S j , where C i This indicates that the current criterion has been executed. `condition(·)` represents the logical decision function output based on the criterion. S j This indicates the target state or the corresponding next criterion node to be executed when the conditions are met. By introducing branching logic, this rule enables the adjudication process to dynamically select subsequent detection paths based on the previous detection results. It is suitable for adaptive adjudication scenarios that require differentiated processing based on the characteristics of evidence, thereby enhancing the flexibility and adaptability of the decision-making structure.

[0054] (3) Threshold triggering rule is a mechanism that triggers state transition by comparing the scalar metric output of an executed criterion with the value of the metric. This mechanism is achieved by comparing the value of the currently executed criterion C. i Quantized output value metric(C) i The metric is compared with a preset threshold to determine the numerical relationship, and a transition to the target state is activated when a specific inequality is satisfied; its abstract expression is if metric(C i )≥threshold then goto S j (or if metric(C) i )≤thresholdthen goto S j (Of course, less than and greater than symbols can also be used), where metric(·) represents the function that extracts the output of the scalar metric of the criterion. This rule achieves precise trigger control through numerical comparison and is suitable for scenarios that require precise state transitions based on quantified indicators.

[0055] (4) The evidence accumulation triggering rule is a mechanism that uses weighted accumulation or logical aggregation of the outputs of multiple executed criteria to trigger a state transition. This mechanism calculates the set of executed criteria {C1, C2, ..., C...} nThe weighted cumulative value or aggregation function value of} is compared with a preset decision threshold T, and the transition to the target state is activated when the cumulative condition is met; its abstract expression is if Aggregate(C1,C2,...,C n )≥T then goto S j Where T represents the preset adjudication threshold. This rule achieves collaborative adjudication of weak evidence from multiple sources through progressive evidence fusion, and is suitable for progressive decision-making scenarios that require the integration of outputs from multiple detection nodes to form a definitive conclusion.

[0056] (5) The consistency triggering rule is a mechanism that triggers a state transition based on the consistency evaluation among multiple executed criteria outputs. This mechanism calculates the set of executed criteria {C1,C2,...,C...} from different detection conditions or different views. n The consistency metric between C1, C2, ..., C3 is used to activate the transition to the target state when the consistency criterion is met; its abstract expression is if Consistency(C1, C2, ..., C3) = ... n =True then goto S j (Or it can be expressed as: ifConsistency(C1,C2,...,C...) n )≥kthengotoS j (where k is a preset consistency threshold). This rule forms a high-confidence consensus decision through collaborative verification of multi-source evidence, and is suitable for collaborative verification scenarios that require confirmation of the consistency of defect assumptions across views or detection nodes.

[0057] (6) The conflict triggering rule is a mechanism that detects inconsistencies or contradictions among multiple executed criteria to trigger an exception handling state transition. This mechanism evaluates the degree of logical divergence or numerical conflict intensity of the outputs of different criteria in the set of executed criteria, and activates a transition to an additional verification criterion or conflict resolution branch when a significant conflict is detected; its abstract expression is if Conflict(C i C j =True then goto S j , where C i and C j This rule indicates that there are two conflicting executed criteria. By identifying contradictory evidence, the rule initiates additional verification or arbitration procedures and is applicable to scenarios involving the handling of anomalies and conflict resolution when logical inconsistencies arise from multiple sources of evidence.

[0058] (7) The short-circuit triggering rule is a mechanism that triggers direct entry into the termination state when the output of the executed criterion is sufficient to independently determine the final defect identification result. This mechanism determines the current executed criterion C by judging the current executed criterion C. iWhether the decisive condition is met, if it is met, immediately activate the transition to the termination state and terminate the execution of subsequent criteria; its abstract expression is: ifdecisive_criterion(C i =True then goto S terminal , among which, S terminal This indicates a termination state node. The `decisive_criterion(·)` function is used to determine whether the output of the criterion constitutes decisive evidence. This rule avoids unnecessary detection when the defect status is clear by identifying the deterministic output of key criteria. It is suitable for efficient adjudication scenarios that require fast response or avoid redundant calculations.

[0059] (8) The proof-of-contrast triggering rule is a mechanism that triggers a state transition or termination of the ruling when the output of the executed criterion contradicts the current defect hypothesis with strong negative evidence. This mechanism evaluates whether the output of the currently executed criterion Ci satisfies the negation condition, and activates the transition to the proof-of-contrast verification state or the termination state when strong negative evidence is detected; its abstract expression is if NegationEvidence(C i =True then goto S j Here, NegationEvidence(·) is a decision function used to determine whether to activate the rebuttal verification state. This rule enables the rapid elimination of flawed hypotheses or the timely correction of the adjudication state by identifying contradictory evidence. It is suitable for scenarios that require the rapid rejection of erroneous hypotheses or the initiation of reverse verification.

[0060] (9) The re-examination trigger rule is a state transition mechanism that triggers a re-verification or supplementary detection process when the output of the executed criterion indicates insufficient evidence or uncertain results. This mechanism evaluates the currently executed criterion C. i The output of the function determines whether the evidence is insufficient. When an evidence gap is detected or the confidence level does not meet the adjudication standard, the function activates the transition to the re-verification state or the supplementary detection criterion node. Its abstract expression is if Evidence_insufficient(Ci) then gotoSj, where Evidence_insufficient(·) is the decision function used to determine whether to activate the re-verification state. This rule fills the evidence gap by initiating a review or additional detection of the previous detection. It is suitable for supplementary verification scenarios when the evidence accumulation is insufficient or the confidence level of the result is low, ensuring that the final adjudication is based on sufficient and reliable evidence.

[0061] In addition to being predefined by domain experts based on process experience, the aforementioned state transition rules can also be automatically generated and optimized through data-driven or knowledge discovery methods. These include, but are not limited to, extraction based on statistical analysis of historical defect samples, learning criterion thresholds and transition conditions from historical data through distribution statistics and anomaly detection methods; rule learning based on machine learning models, automatically extracting condition-action form state transition rules from labeled data through decision tree learning, association rule mining, or rule learning algorithms; rule extraction based on complex model distillation, transforming the implicit decision-making behavior of deep neural networks into interpretable state transition structures through decision tree distillation or rule extraction techniques; and automatic rule optimization based on feedback iteration, automatically adjusting transition thresholds and rule weights within the existing rule framework using reinforcement learning or parameter optimization methods, enabling state transition rules to continuously and adaptively update with process evolution or data accumulation, thereby enhancing the generalization ability and long-term adaptability of the hierarchical decision framework.

[0062] When determining and executing the next criterion based on the target rule, we can first parse the subsequent node identifiers or branch path information defined in the target rule, and then select the candidate criterion set C={C1,C2,...,C...}. n The system retrieves the corresponding criterion from the list as the next criterion to be executed, then activates the detection node corresponding to the criterion to perform defect detection calculations, and obtains the execution result as the input for a new round of state transition, thereby completing the iterative evolution of the hierarchical decision-making process.

[0063] Optionally, the above-mentioned determination of the target rule in the state transition rules based on the execution result of the executed criteria includes: when the execution result of the executed criteria satisfies multiple state transition rules, determining the target rule among multiple state transition rules based on the preset rule priority.

[0064] The aforementioned preset rule priority is a predefined sorting criterion used to adjudicate conflicting rules within the state transition rule set. When the execution result of an executed criterion simultaneously satisfies the triggering conditions of multiple state transition rules, these rules form a concurrently activated competing state. The preset rule priority assigns a unique priority order to multiple concurrently activated state transition rules based on the importance and reliability level of various triggering mechanisms or a preset arbitration strategy. This uniquely determines the target rule from multiple satisfying rules in complex evidence scenarios, avoiding uncertainty in the adjudication path or logical deadlock.

[0065] The preset rule priorities can be set using a static sorting method based on rule type. For example, short-circuit trigger rules can be prioritized over sequential trigger rules, and consistency trigger rules can be prioritized over evidence accumulation trigger rules, thus executing higher-priority rules when both are simultaneously met. Alternatively, a dynamic adjustment method based on criterion confidence can be used. Different priority weights are dynamically assigned to associated state transition rules based on the confidence level or reliability of the executed criterion's output results; higher confidence levels correspond to higher rule priorities. Finally, a scenario-adaptive preset strategy can be employed, pre-configuring different rule priority sequences for specific defect types or detection stages, and loading the corresponding priority configuration to adapt to the current detection requirements when entering a specific scenario.

[0066] When satisfying multiple state transition rules, in addition to pre-setting rule priorities, a competitive adjudication based on evidence strength can be adopted. This involves quantitatively comparing the degree of satisfaction or confidence level of each rule's triggering conditions, and determining the rule with the highest evidence support as the target rule. The degree of satisfaction can be measured by calculating the deviation of the criterion output from the rule triggering threshold, the probability value, or the cumulative weight. Alternatively, an adjudication based on historical performance statistics can be used. By querying historical performance indicators such as the success rate, false positive rate, or false negative rate of each state transition rule in similar past detection scenarios, the rule with the best historical performance can be determined as the target rule to maximize detection reliability. Finally, an adjudication based on computational resource overhead can be used. By evaluating the computational complexity, time overhead, or resource consumption level of each candidate rule's corresponding subsequent criteria, the rule with the lowest resource overhead can be determined as the target rule, prioritizing the most efficient adjudication path when multiple conditions are met. In addition, a risk-conservative decision-making approach can be adopted. By setting a decision-making strategy that prioritizes safety or defect sensitivity, when multiple rules point to continuing testing or terminating the decision, the conservative path that can minimize missed or false detections is selected first, and the rule that conforms to the current risk preference is determined as the target rule.

[0067] Step S140: Repeat step S130 until the preset termination condition is met.

[0068] The aforementioned preset termination condition is a critical criterion used in the hierarchical decision-making process to determine whether the currently accumulated evidence sufficiently supports the final defect identification conclusion. It operates between the criterion determination and execution steps in the iterative execution, serving as a boundary condition for the transition from the evidence accumulation process to the output of the adjudication result. When the execution result of the executed criterion is evaluated and meets the evidence sufficiency standard represented by this condition, the adjudication process determines that the current state meets the termination requirements, triggers entry into the termination state, and outputs the defect identification result. At the same time, it terminates the activation and calculation of subsequent pending criteria, ensuring that the hierarchical decision-making process converges in a timely manner when the evidence is sufficient to determine whether a defect exists or has a clear cause, avoiding the continuation of unnecessary detection calculations when key adjudication information is complete.

[0069] Optionally, the aforementioned preset termination conditions include preset short-circuit conditions; After determining and executing the first criterion to be executed among multiple criteria based on a preset decision priority, the aforementioned defect identification method based on hierarchical decision-making further includes: determining whether the execution result of the executed criterion satisfies a preset short-circuit condition; wherein the preset short-circuit condition includes at least one of the following: a key criterion short-circuit condition, a threshold short-circuit condition, an evidence accumulation short-circuit condition, a consistency short-circuit condition, and a rebuttal evidence short-circuit condition; when the execution result of the executed criterion satisfies the preset short-circuit condition, it is determined that the preset termination condition is satisfied; when the execution result of the executed criterion does not satisfy the preset short-circuit condition, it is determined that the preset termination condition is not satisfied, and based on the execution result of the executed criterion and the state transition rule, the next criterion to be executed is determined and executed.

[0070] Optionally, the aforementioned preset termination conditions include preset short-circuit conditions; Following step S130 above, the aforementioned defect identification method based on hierarchical decision-making further includes: determining whether the execution result of the executed criterion satisfies a preset short-circuit condition; wherein the preset short-circuit condition includes at least one of the following: a key criterion short-circuit condition, a threshold short-circuit condition, an evidence accumulation short-circuit condition, a consistency short-circuit condition, and a rebuttal short-circuit condition; when the execution result of the executed criterion satisfies the preset short-circuit condition, it is determined that the preset termination condition is satisfied; when the execution result of the executed criterion does not satisfy the preset short-circuit condition, it is determined that the preset termination condition is not satisfied, and based on the execution result of the executed criterion and the state transition rule, the next criterion to be executed is determined and executed.

[0071] The aforementioned preset short-circuit conditions are a subset of preset termination conditions. They represent the critical judgment criteria for early termination of the execution of subsequent criteria when a single decisive piece of evidence or accumulated evidence meets the sufficiency standard. The core mechanism is that when the output of the executed criteria is sufficient to independently determine the defect identification conclusion, it immediately triggers the termination state and outputs the adjudication result without waiting for the calculation of the remaining criteria to be completed, thereby avoiding redundant calculations and reducing interference from noisy evidence.

[0072] The aforementioned key criterion short-circuit condition is an early termination mechanism based on a preset set of decisive criteria. When criterion C has been executed... i When an item belongs to the preset set DecisiveCriteria and its execution result meets the defect determination criteria, it is determined that the current evidence is sufficient to independently support the final ruling, and the process immediately triggers the termination state S. terminal The transfer; its expression form is if(C i ∈DecisiveCriteria)∧(Output(C i =Defective)then goto S terminal Or equivalently represented as ifdecisive_criterion(C i =True then terminate, where DecisiveCriteria represents a pre-identified set of key criteria with high decision weights. This condition immediately converges the decision process by identifying the deterministic output of the key criteria. It is applicable to scenarios where a specific detection node is given high confidence decision-making power and its output can directly determine the defect status, thereby avoiding unnecessary subsequent detection when the structural defect is clear.

[0073] The aforementioned threshold short-circuit condition is a mechanism that triggers early termination by performing a strong threshold comparison on the scalar metric output of an executed criterion. This occurs when the executed criterion C... i metric(C) output i When the threshold strong_threshold is exceeded, it is determined that the current evidence is sufficient to independently support the defect ruling, and the process immediately triggers the termination state S. terminal The transfer; its expression form is if metric(C i )≥strong_threshold then goto S terminal This condition enables immediate termination of decisions based on quantitative indicators through numerical comparison. It is applicable to scenarios where scalar measurements have clearly exceeded the normal allowable range and are sufficient to independently determine the existence of defects, thereby avoiding unnecessary subsequent detections when key quantitative indicators have clearly indicated the defect status.

[0074] The aforementioned evidence cumulative short-circuit condition is a mechanism for early termination based on the weighted cumulative value of multiple executed criteria. When the set of executed criteria {C1, C2, ..., C...}... n When the weighted cumulative value of} reaches or exceeds the preset adjudication threshold, it is determined that the current accumulated evidence is sufficient to independently support the final adjudication, and the process immediately triggers the termination state S. terminal The transfer; its expression is if ∑w i Ci ≥decision_threshold then goto S terminal , where w i represents the weight coefficient of the i-th executed criterion, and decision_threshold represents the preset decision threshold parameter. This condition achieves immediate termination of decision when there is sufficient accumulated evidence through the progressive fusion of weak evidence from multiple sources. It is suitable for scenarios that require the comprehensive output of multiple detection nodes to form a deterministic conclusion, thereby avoiding the mechanical execution of the remaining criteria when the accumulated evidence has sufficiently supported the defect determination.

[0075] The aforementioned consistency short-circuit condition is a mechanism for early termination based on a consistency measure among the outputs of multiple executed criteria. When the set of executed criteria {C1, C2, ..., C...}... n When the consistency metric value of} reaches or exceeds the preset consistency threshold k, it is determined that the multi-source evidence has formed a high degree of consensus and is sufficient to independently support the final ruling, and the process immediately triggers the termination state S. terminal The transfer; its expression form is if Consistency(C1,C2,…,C…). n )≥k then goto S terminal This condition enables immediate termination of the decision when collaborative verification passes by identifying highly consistent evidence across detection nodes. It is applicable to scenarios where strong consensus is formed by multiple views or multiple conditions of evidence to confirm defective hypotheses, thereby avoiding the need to perform subsequent redundant verification when the consistency of evidence has been fully established.

[0076] The aforementioned short-circuit condition for rebuttal is an early termination mechanism based on the detection of strong negative evidence, when criterion C has been executed. i When the output indication provides strong negative evidence that contradicts the current defect hypothesis, it is determined that the current evidence is sufficient to exclude the defect hypothesis or establish a non-defect state, and a move to the termination state S is immediately triggered. terminal The transfer; its expression form is if NegationEvidence(C i =True then goto S terminal Alternatively, it can be expressed as if Evidence_negate→terminate, where NegationEvidence(·) represents the decision function that determines whether the output of the criterion constitutes strong negative evidence. This condition enables the rapid elimination of defective hypotheses or the timely termination of the positive detection process by identifying contradictory evidence. It is suitable for scenarios where the executed criteria clearly indicate a normal state or strongly deny the existence of defects, thereby avoiding the execution of subsequent unnecessary detection calculations when the defective hypothesis has been refuted by contradiction.

[0077] In the above scheme, after the first pending criterion is executed, it is immediately determined whether its execution result satisfies any of the aforementioned preset short-circuit conditions. If it is determined to be satisfied, the preset termination condition is confirmed, and the system directly enters the termination state. Based on the output of the currently executed criterion, the defect identification result is determined, and the execution of the remaining criteria is terminated. If it is determined not to be satisfied, the preset termination condition is confirmed not to be satisfied, and the system continues to determine and execute the next pending criterion based on the execution result of the executed criterion and the state transition rules, thus maintaining the iterative advancement of the hierarchical decision-making process.

[0078] It is understandable that, in addition to the preset short-circuit condition, the preset termination condition may also include a termination condition based on complete traversal, that is, when all necessary criteria in the preset execution sequence have been executed and no early termination mechanism has been triggered, it is determined that the evidence collection is sufficient and the final defect identification result is determined based on the complete evidence chain; it may also include a forced convergence condition based on the maximum number of execution steps, which prevents infinite loops or over-computation caused by abnormal evidence combinations or state transition logic defects by preset the maximum iteration depth or the maximum number of criteria executed in the adjudication process; it may also include a termination condition based on external intervention instructions, which immediately terminates the current hierarchical decision-making process and outputs an intermediate adjudication state based on the executed criteria when a manual adjudication instruction or a forced stop signal from the upper-level control system is received; it may also include a termination condition based on evidence accumulation saturation, which determines that a reliable adjudication can be completed without introducing additional criteria when the output of the executed criteria has reached the preset decision completeness index after fusion calculation.

[0079] Optionally, the above-mentioned defect identification method based on hierarchical decision-making further includes: constructing a set of states including an initial state, an intermediate evidence accumulation state, and a termination state; wherein, the state transition rule is used to trigger a transition from the initial state or the current intermediate evidence accumulation state to the next intermediate evidence accumulation state based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or to trigger entering the termination state when a preset termination condition is met.

[0080] A state machine is a formal modeling framework for finite state automata. By defining a finite set of states and the transition conditions between states, the defect identification process is abstracted into a directed transition process between discrete states. In formal expression, a state machine can be defined as a quintuple (S, C, δ, S0, S...). terminal ), where S represents a finite set of states, C represents the input (corresponding to the set of criterion variables), δ: S×C→S represents the state transition function, S0∈S represents the initial state, S terminal S represents a subset of terminating states. This framework discretizes continuous defect detection computation into transitions between state nodes, giving the adjudication process a clear execution trajectory and verifiable formal semantics.

[0081] The state set S contains three subsets with specific semantics: the initial state S0 represents the starting point of the adjudication process, indicating the initial decision-making stage before any evidence of defects has been obtained; the intermediate evidence accumulation state set S... intermediate ={S1,S2,…,S m The intermediate state represents the gradual accumulation or verification of defective evidence. Each intermediate state corresponds to a snapshot of the evidence state after the execution of a specific criterion, carrying the output information of the executed criterion and the current confidence level of the ruling; the set of termination states S terminal ={S t1 ,S t2 ,…} represents the convergence endpoint of the adjudication process. Each termination state corresponds to a specific defect identification conclusion (such as defect existence, defect type coding, cause attribution, or normal state) and is associated with complete decision path and evidence summary information.

[0082] The state transition rule δ triggers directed transitions between states based on the execution result of executed criteria. Its working mechanism is as follows: the current state is either the initial state S0 or a certain intermediate evidence accumulation state S. i ∈S intermediate At that time, receive the currently executed criterion C. i The structured output is used as input, and the target state is determined based on a preset transition condition evaluation function. If the transition condition indicates that evidence should continue to be accumulated, the process is triggered to the next intermediate evidence accumulation state S. i+1 The transition involves updating the state to include all evidentiary information from the previous state and the output of newly added criteria. If the transition condition indicates that the evidence is sufficient or the short-circuit termination criterion is met, then a transition to a specific termination state S is triggered. t ∈S terminal Upon the transfer, the adjudication process converges and outputs the associated defect identification results.

[0083] The construction of the aforementioned state set can employ an explicit enumeration method, predefining a finite set of state node identifiers and their semantic labels, and configuring a corresponding evidence storage structure and adjudication parameters for each state node; alternatively, an implicit dynamic generation method can be used, dynamically instantiating intermediate state objects based on the criteria output during adjudication execution, and maintaining the transition relationships between states through hash mapping or pointer links. The definition of state transition rules can adopt a rule base based on condition-action pairs, encoding each transition condition as a logical expression or threshold comparison formula, and associating it with the corresponding next state identifier; alternatively, a graph structure based on transition matrices or adjacency lists can be used, pre-compiling legal transition paths between states, and determining the target state during execution through table lookup or graph traversal. Regardless of the implementation method, the state machine framework ensures that the adjudication process follows a strict discrete state transition mechanism, avoiding state confusion and path uncertainty in parallel fusion.

[0084] Optionally, the above-mentioned defect identification method based on hierarchical decision-making further includes: constructing a decision tree including a root node, internal criterion nodes, and leaf nodes; wherein, the root node corresponds to the first criterion to be executed; the internal criterion nodes correspond to the next criterion to be executed; the leaf nodes correspond to preset termination conditions; and the state transition rules are used to determine the branch path from the root node or internal criterion node to the next internal criterion node based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or, when the preset termination condition is met, to enter the leaf node.

[0085] The aforementioned decision tree is a hierarchical formal modeling framework. By defining a set of nodes with hierarchical topological relationships and the branching paths between nodes, it abstracts the defect identification process into an ordered adjudication flow that starts from the root node and traverses along specific branching paths to the leaf nodes. This framework explicitly expresses the hierarchical dependencies and conditional branching relationships between criteria using a tree structure. The root node, located at the top level of the tree, represents the starting point of the adjudication flow. Internal criterion nodes are distributed in the middle levels of the tree, representing the detection nodes that need to be executed. Leaf nodes are distributed at the bottom levels of the tree, representing the termination conditions and final conclusion of the adjudication flow. Each branching path corresponds to a specific combination of evidence logic and decision rules.

[0086] The root node corresponds to the first criterion to be executed, and is responsible for initializing the adjudication process. Its activation marks the start of the hierarchical decision-making process. The criteria associated with this node are determined based on a preset adjudication priority and are used to obtain initial defect evidence and initiate subsequent branch selection. Internal criterion nodes correspond to the next criterion to be executed and are distributed on the path between the root node and leaf nodes. Each internal node is associated with the criterion execution logic under specific detection conditions, and is used to further obtain supplementary detection information or conduct detailed causal analysis under the current evidence state. The hierarchical depth of the nodes reflects the progress of evidence accumulation. Leaf nodes correspond to preset termination conditions and are the terminal nodes of the tree structure, with no further branches. Each leaf node is associated with a specific defect identification result (such as defect existence determination, type encoding, cause attribution, or normal state confirmation). When the adjudication process reaches a leaf node, it converges and outputs the associated adjudication conclusion.

[0087] In the decision tree framework, the state transition rule is manifested as a branch path selection mechanism. Its working principle is as follows: When the current adjudication process is at the root node or an internal criterion node, it receives the structured output of the currently executed criterion as the basis for path selection and evaluates the matching degree between this output and the preset conditions of each branch path. If the execution result meets the combination of conditions for continuing detection, a branch path pointing to the next internal criterion node is determined, and the process moves down along this path, activating the corresponding node to execute its associated criterion. If the execution result meets the combination of conditions for terminating the adjudication, a branch path pointing to a leaf node is determined, and the process enters the termination state along this path, outputting the associated defect identification result. This branch selection process is executed iteratively until the process reaches a leaf node and completes adjudication convergence.

[0088] The construction of the aforementioned decision tree can be achieved through a predefined method based on expert knowledge. Domain experts can manually design the tree topology according to the sequence of process knowledge and the dependencies of defect features, configuring corresponding criterion identifiers and branch condition expressions for each node. Alternatively, a data-driven automatic learning approach can be used. This involves training the decision tree model using historical defect samples, automatically extracting the optimal splitting attributes and branch thresholds between criteria, and generating a tree structure that maximizes detection accuracy. The determination of branch paths can employ a rule-matching-based conditional decision method, logically matching the output of executed criteria with the branch rules associated with the node to determine the next node identifier; or a probability-based soft decision method, calculating the posterior probability of each branch path and selecting the path with the highest probability as the target path. Regardless of the implementation method, the decision tree framework ensures that the adjudication process follows a strict hierarchical branching structure. The execution of each criterion depends on the path selection result of the preceding node, enabling single-path traversal from the root node to the leaf node or controlled multi-path exploration.

[0089] Step S150: After the preset termination condition is met, the defect identification result is determined based on the execution result of the executed criteria.

[0090] The above step S150 can employ various adjudication mechanisms to determine the defect identification result, including but not limited to: a direct mapping mechanism based on the output of a single decisive criterion, where the category identifier or status label of the criterion is directly used as the final defect identification result when the output of an executed criterion meets the preset defect judgment criteria; a fusion mechanism based on the accumulation of evidence from multiple criteria, where the comprehensive index after fusion is compared with a preset adjudication threshold by weighted summation, logical voting, or probability aggregation calculation of the outputs of multiple executed criteria to determine the defect category and confidence level; and a mapping mechanism based on the state machine termination status label, where the defect identification result is determined when the process enters a specific termination state S150. t ∈S terminal When the decision process reaches a specific leaf node, the defect identification conclusion pre-associated with the termination state is directly extracted. Based on the leaf node label reading mechanism, the defect category code stored in that node is output as the identification result when the decision process reaches that specific leaf node. Furthermore, a logical rule-based reasoning mechanism is used to input the output of the executed criteria as a pre-defined reasoning rule base, generating a composite identification conclusion containing defect category, cause attribution, and anomaly level through logical deduction or rule matching. These mechanisms can be applied independently or in combination to adapt to the varying requirements of decision accuracy and interpretability in different detection scenarios.

[0091] Optionally, the above defect identification results include at least one of the following: defect category information, confidence level information, evidence summary information, decision path information, anomaly level information, spatial region information, and consistency status information.

[0092] The aforementioned defect identification result is a structured decision state output by the hierarchical decision-making process after the preset termination conditions are met. It carries the final conclusion of the defect judgment and the complete contextual information supporting the conclusion. Its multi-dimensional content makes the decision output not only include a binary judgment of whether a defect exists, but also has the engineering characteristics of being traceable, interpretable, and auditable.

[0093] Defect category information is used to characterize the type or code of defects. By outputting predefined defect type labels (such as Oxide-related Defect, Nitride-related Defect, pin_hole, block_etch, etc.) or standardized defect codes (such as defect_code∈{101,203,407}), subsequent process processing can take targeted repair or compensation measures according to the defect category, so as to realize the classification and precise response of defect management.

[0094] The evidence summary information is used to record the set of key criteria and contributing evidence involved in the final adjudication. By outputting the contributing_criteria list or evidence_set structured data, it clearly identifies which detection nodes and under what detection conditions the criterion outputs had a decisive impact on the final conclusion, so that the adjudication process has a complete chain of evidence to support it, which facilitates subsequent quality traceability and liability determination.

[0095] Decision path information is used to characterize the sequence of criteria execution and triggering rules experienced during the adjudication process. By outputting the decision_path or triggered_rules list, it records all state transition nodes and branch paths from the first criterion to the termination state, so that the complex hierarchical decision-making process has a reproducible execution trajectory, which facilitates path analysis and anomaly localization during process debugging.

[0096] Anomaly level information is used to characterize the severity or risk level of defects. By outputting severity_level (such as Normal, Weak_Anomaly, Strong_Anomaly) or risk_level (such as Low, Medium, High, Critical) grade labels, production scheduling and quality control can allocate processing resources according to risk priority, thereby realizing graded management of defect response.

[0097] Spatial region information is used to identify the spatial location or region of defects on the inspected object. By outputting region_id, ROI coordinates or pixel-level position mapping, the abstract defect judgment is associated with the specific physical spatial location, which facilitates subsequent physical location, repair operations or failure analysis.

[0098] Consistency status information is used to characterize the consistency or conflict status among multi-source criteria. By outputting consistency_score (consistency metric) or conflict_flag (conflict flag), it indicates whether the criterion outputs from different detection conditions have reached a consensus or are contradictory, providing a status identifier for evidence reliability assessment and conflict arbitration.

[0099] The generation of the aforementioned defect identification results can be achieved using a structured object encapsulation approach, organizing the multidimensional information into a composite data structure containing multiple fields (e.g., DecisionState={result,confidence,evidence_summary,decision_path,...}), and outputting it to downstream modules through a standardized interface. Alternatively, serialized text or binary stream methods can be used to encode and transmit various types of information according to a preset protocol. A phased output approach can also be employed, where the core conclusion (defect category, confidence level) is output first upon reaching the termination state, followed by asynchronous output of auxiliary information (evidence summary, decision path) to balance response speed and information integrity. Regardless of the implementation method, the defect identification results serve as a complete carrier of the decision state, providing sufficient contextual support and traceable evidence for defect determination.

[0100] Optionally, the above-mentioned defect identification method based on hierarchical decision-making further includes: when the preset degradation constraint condition is met, all criteria are executed in parallel, and the defect identification result is determined based on the execution results of all criteria.

[0101] The aforementioned pre-defined degradation constraints are a set of logical constraints used to determine whether the hierarchical decision-making method degenerates into a traditional parallel fusion mode. When all constraints are satisfied simultaneously, the sequential decision-making mechanism is abandoned, and all criteria are executed in parallel, with the final state determination completed based on a unified logical expression. This mechanism enables the technical solution to maintain its ordered decision-making capability while possessing the flexibility to be compatible with traditional parallel decision-making, making it suitable for simple scenarios or production environments with high real-time requirements.

[0102] Preset degradation constraints may include: (1) No intermediate state constraints, which require that the state set only includes the initial state and the final state and there are no intermediate evidence accumulation states, i.e., S={S0,S... final}, evidence directly participates in the final judgment without going through a gradual accumulation process; (2) there is no execution order dependency constraint, which requires that there is no sequential dependency relationship between all criteria, that is C i C j ∈C, C does not exist i →C j(2) There is no causal dependency, all criteria can be calculated independently and input into the same decision function; (3) There is no short-circuit mechanism constraint, which requires that no criterion can trigger the early termination process, and a unified decision must be made after all criteria have been executed. C i ∈C, ShortCircuit(C i (4) No branch structure constraint, which requires that there are no multi-path decision branches, and all criterion outputs are input to a single final decision expression, i.e., Defect=F(C1,C2,…,C n ), where F is a logical expression composed of basic Boolean logic operators.

[0103] When all the above degradation constraints are met, the parallel execution mode is activated, and all detection nodes corresponding to the criteria are activated to perform independent calculations. After obtaining the structured output of each criterion, it is input into a unified logic fusion function. F The final state determination is performed, the defect identification result is output, and the process is terminated. This degenerate mode is essentially a special case of the hierarchical decision state machine or decision tree in the above scheme under specific constraints. At this time, the ordered decision structure degenerates into a one-time parallel Boolean decision, which not only maintains compatibility with existing technologies, but also restores the complete function of hierarchical decision-making when the constraints are removed.

[0104] The following are two specific application scenarios of the above-mentioned defect identification method based on hierarchical decision-making: First application scenario: an orderly decision-making scenario involving top and side views; like Figure 3 As shown, optionally, the above-mentioned different detection conditions include top-view imaging conditions and side-view imaging conditions; The aforementioned criteria include the side-view structure criterion corresponding to the side-view imaging condition and the top-view criterion corresponding to the top-view imaging condition; wherein, the side-view structure criterion is used to evaluate the consistency of the inter-layer structure, and the top-view criterion is used to assist in the judgment of the cause of defects. The above-mentioned determination and execution of the first criterion to be executed among multiple criteria based on the preset decision priority includes: determining the side view structure criterion as the first criterion to be executed and executing it; The above-mentioned determination and execution of the next pending judgment based on the execution result of the executed judgment and the state transition rules includes: when the execution result of the side view structure judgment indicates that the structure is consistent, the top view judgment is determined as the next pending judgment and executed.

[0105] The aforementioned top-view imaging and side-view imaging conditions constitute a geometrically differentiated configuration for multi-view defect detection. The top-view imaging condition acquires planar morphology information through an optical axis perpendicular to the wafer surface, focusing on the geometric features and planar distribution of the surface pattern. The side-view imaging condition acquires interlayer structure information through an optical axis tilted towards the wafer surface, focusing on interlayer contours, sidewall morphology, and structural continuity in the depth direction. The side-view structure criterion corresponds to the detection node under the side-view imaging condition. Its function is to evaluate the consistency of the interlayer structure, obtaining highly sensitive evidence of interlayer structural anomalies by analyzing the integrity of the interlayer contour, compliance of sidewall angles, and structural continuity in the depth direction. The top-view criterion corresponds to the detection node under the top-view imaging condition. Its function is to assist in the determination of defect causes, distinguishing potential sources of defects based on the matching relationship between pattern morphology and process features through planar morphology analysis of candidate regions.

[0106] In this application scenario, the preset decision priority is determined based on the order of process cognition and the difference in detection sensitivity. The side-view structure criterion is determined as the first criterion to be executed and is executed first. This priority setting stems from the high sensitivity of the side view to interlayer structural anomalies, which can obtain the most identifiable evidence of structural defects first. After the side-view structure criterion is executed, its execution result triggers subsequent state transitions: when the execution result indicates that the structure is inconsistent, it means that an interlayer structural anomaly has been detected, and the short-circuit termination mechanism is directly triggered to enter the termination state and output the defect existence determination. At this time, the top-view structure criterion is no longer executed, thereby avoiding the introduction of unnecessary planar analysis load when the structural defect has been established; when the execution result indicates that the structure is consistent, it means that the defect short-circuit decision has not been triggered, and the execution is transferred to the top-view structure criterion node, which uses the high-resolution characteristics of the planar morphology of the top-view imaging to perform auxiliary attribution analysis of the defect cause.

[0107] The aforementioned ordered adjudication process is based on the cognitive sequence of "side-view adjudication first, top-view attribution later." The side-view stage is responsible for the rapid identification and triggering of structural defects, utilizing the high sensitivity of side-view imaging to structural features in the depth direction to achieve early interception of interlayer anomalies. The top-view stage is responsible for auxiliary attribution of causes, activating only when the side-view stage has not triggered a defect short circuit. Utilizing the clear presentation of surface pattern details by top-view imaging, it assists in distinguishing the potential causes of defects through matching analysis (such as Oxide-related and Nitride-related feature patterns) between pattern morphology and a process feature library (which can be manually collected and then organized by an intelligent agent into structured information to form a process feature library that can be aligned with the image). This hierarchical structure ensures that the process can be quickly terminated and results output when structural defects are present, while providing deeper causal information through top-view analysis when the structure is normal, achieving a balance between detection efficiency and attribution accuracy.

[0108] The second application scenario: the stratification of high voltage and low voltage conditions; like Figure 4 As shown, optionally, the above-mentioned different detection conditions include low-voltage imaging conditions and high-voltage imaging conditions; The aforementioned criteria include low-voltage criteria corresponding to low-voltage imaging conditions and high-voltage criteria corresponding to high-voltage imaging conditions; among them, the low-voltage criteria are used for current station defect trigger detection, and the high-voltage criteria are used for defect classification and previous station source determination. The above-mentioned determination and execution of the first pending criterion among multiple criteria based on preset decision priority includes: determining the low voltage criterion as the first pending criterion and executing it; The above-mentioned determination and execution of the next pending criterion based on the execution results and state transition rules of the executed criteria includes: when the execution results of the low voltage criteria are consistent, the high voltage criteria are determined as the next pending criterion and executed.

[0109] The aforementioned low-voltage imaging and high-voltage imaging conditions constitute the voltage contrast differentiation configuration in electron beam defect detection. Low-voltage imaging uses a lower accelerating voltage to confine the electron beam to the surface of the current layer, preventing penetration into the underlying structure, focusing on highly sensitive detection of the current layer's surface morphology and shallow defects. High-voltage imaging uses a higher accelerating voltage to enable the electron beam to penetrate the current layer, focusing on observing the underlying structure or problems left over from previous layers. The low-voltage criterion corresponds to the detection node under low-voltage imaging conditions, its function being to trigger detection when a defect occurs at the current station, quickly identifying the actual defect at that station by evaluating the abnormal features of the current layer's surface. The high-voltage criterion corresponds to the detection node under high-voltage imaging conditions, its function being to classify defects and determine their origin from previous stations, distinguishing whether the defect is a current layer process problem or a problem left over from previous layers through penetration analysis.

[0110] In this application scenario, the preset decision priority is determined based on the physical relationship between the electron beam penetration depth and the detection target. The low voltage criterion is determined as the first criterion to be executed and is executed first. This priority setting is based on the process cognition order of low voltage imaging not penetrating to see the current station first, and high voltage penetration to judge the previous layer. This allows the high sensitivity of low voltage to the surface defects of the current layer to quickly trigger detection, avoiding the waste of computing resources caused by performing penetration analysis when there are already real defects in the current station.

[0111] After the low-voltage criterion is executed, its execution result triggers subsequent state transitions: When the execution results are inconsistent, it indicates that a real defect sensitive to low-voltage conditions has been detected at the current station. This directly triggers the short-circuit termination mechanism, entering a termination state and outputting a defect determination. At this point, the high-voltage criterion is no longer executed, thus achieving rapid determination of defects at the current station. When the execution results are consistent, it indicates that no defect determination was triggered under low-voltage conditions. The process then transitions to the high-voltage criterion node for further analysis. The high-voltage penetration capability is used to observe the preceding structure, enabling further classification and source determination of defects. For example, it can distinguish between current station defect types such as pinholes or block etches, or determine whether a defect is a legacy issue related to the prelayer. Of course, defects are not limited to pinholes and block etches; specific defect types can be adaptively set based on the application scenario.

[0112] The aforementioned application scenario is based on a decision-making logic that prioritizes the current imaging station over the preceding layers. The low-voltage phase serves as a preliminary trigger criterion, rapidly identifying the actual defect samples in the current layer and utilizing its non-penetrating characteristic to avoid interference from preceding layers. The high-voltage phase serves as a subsequent classification criterion, performing penetration analysis when the current imaging station is functioning normally, and using its penetration characteristic to trace the source of the defect. This sequential execution and state transition mechanism based on voltage comparison differs from a one-time judgment method that fuses different imaging conditions in parallel, achieving rapid response to defect triggering and accurate attribution of defect classification.

[0113] like Figure 5 As shown, based on the same inventive concept, this application also provides a defect identification system 200 based on hierarchical decision-making. This system includes: a criterion acquisition module 210, a hierarchical decision-making module 220, and a result output module 230, which are connected sequentially. The criterion acquisition module 210 is used to acquire multiple criteria from different detection conditions; The hierarchical decision module 220 is used to determine and execute the first judgment to be executed among multiple judgments based on a preset decision priority, and to determine and execute the next judgment to be executed based on the execution result and state transition rules of the executed judgments, and to repeat the determination and execution steps of the next judgment to be executed until a preset termination condition is met. The result output module 230 is used to determine the defect identification result based on the execution result of the executed criteria after the preset termination condition is met.

[0114] The defect identification system 200 based on hierarchical decision-making provided in this application embodiment can be used to execute the defect identification method based on hierarchical decision-making provided in this application embodiment. Its implementation principle and the resulting technical effects have been described in the foregoing method embodiment. For the sake of brevity, any part not mentioned in the system embodiment can be referred to the corresponding content in any of the foregoing method embodiments.

[0115] Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of this application. (Refer to...) Figure 6 The electronic device 300 includes a processor 310, a memory 320, and a communication interface 330. These components are interconnected and communicate with each other via a communication bus 340 and / or other forms of connection mechanism (not shown).

[0116] The memory 320 includes one or more (only one is shown in the figure), which may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor 310 and other possible components may access the memory 320 to read and / or write data therein.

[0117] Processor 310 includes one or more (only one is shown in the figure), which can be an integrated circuit chip with signal processing capabilities. The processor 310 described above can be a general-purpose processor, including a central processing unit (CPU), a microcontroller unit (MCU), a network processor (NP), or other conventional processors; it can also be a special-purpose processor, including a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0118] Communication interface 330 includes one or more (only one is shown in the figure) that can be used to communicate directly or indirectly with other devices to exchange data. For example, communication interface 330 can be an Ethernet interface; it can be a mobile communication network interface, such as an interface for 3G, 4G, or 5G networks; or it can be other types of interfaces with data transmission and reception capabilities.

[0119] One or more computer program instructions may be stored in the memory 320, and the processor 310 may read and run these computer program instructions to implement the hierarchical decision-based defect identification method and other desired functions provided in the embodiments of this application.

[0120] Understandable. Figure 6 The structure shown is for illustrative purposes only; the electronic device 300 may also include components that are more advanced than those shown. Figure 6 The more or fewer components shown, or having the same Figure 6 The different configurations shown. Figure 6 The components shown can be implemented using hardware, software, or a combination thereof. For example, electronic device 300 can be a single server (or other device with computing power), a combination of multiple servers, a cluster of a large number of servers, etc., and can be either a physical device or a virtual device.

[0121] This application also provides a computer-readable storage medium storing computer program instructions. These instructions are read and executed by a processor to perform the hierarchical decision-based defect identification method provided in this application. For example, the computer-readable storage medium can be implemented as follows: Figure 6 The memory 320 in the electronic device 300, or a separate storage product (such as a USB flash drive, portable hard drive, etc.).

[0122] This application also provides a computer program product comprising computer program instructions. These computer program instructions are read and executed by a processor to perform the hierarchical decision-based defect identification method provided in this application. For example, these computer program instructions may be stored in... Figure 6 The memory 320 in the electronic device 300 is located inside the memory, or it is stored in a separate storage product (such as a USB flash drive, portable hard drive, etc.).

[0123] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0124] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0125] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0126] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0127] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0128] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0129] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0130] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A defect identification method based on hierarchical decision-making, characterized in that, The method includes: Multiple criteria from different detection conditions are obtained; wherein, the different detection conditions are imaging configuration methods used in the defect identification process that can produce differentiated defect feature display effects; the criteria are structured evidence representations output by the detection node after performing defect detection calculations, and the criteria are derived from detection algorithms or models configured for different detection conditions. Based on a preset decision priority, the first decision to be executed is determined from among the multiple decisions and then executed. Based on the execution results of the executed criteria and the state transition rules, determine and execute the next criterion to be executed; Repeat the steps of determining and executing the next criterion to be executed until a preset termination condition is met; wherein, the preset termination condition includes at least one of the following: a preset short-circuit condition, a termination condition based on complete traversal, a forced convergence condition based on the maximum number of execution steps, a termination condition based on external intervention instructions, and a termination condition based on evidence accumulation saturation. After the preset termination condition is met, the defect identification result is determined based on the execution result of the executed criteria; The method further includes: constructing a state set including an initial state, an intermediate evidence accumulation state, and a termination state; wherein, the state transition rule is used to trigger a transition from the initial state or the current intermediate evidence accumulation state to the next intermediate evidence accumulation state based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or to trigger entering the termination state when the preset termination condition is met.

2. The defect identification method based on hierarchical decision-making according to claim 1, characterized in that, The method further includes: A decision tree is constructed, comprising a root node, internal criterion nodes, and leaf nodes; wherein the root node corresponds to the first criterion to be executed; the internal criterion nodes correspond to the next criterion to be executed; and the leaf nodes correspond to the preset termination condition; the state transition rule is used to determine the branch path from the root node or the internal criterion node to the next internal criterion node based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or, when the preset termination condition is met, to enter the leaf node.

3. The defect identification method based on hierarchical decision-making according to claim 1, characterized in that, The preset termination conditions include preset short-circuit conditions; After determining and executing the first criterion to be executed among the multiple criteria based on a preset decision priority, the method further includes: Determine whether the execution result of the executed criterion satisfies the preset short-circuit condition; wherein the preset short-circuit condition includes at least one of the following: key criterion short-circuit condition, threshold short-circuit condition, evidence accumulation short-circuit condition, consistency short-circuit condition, and evidence against contradiction short-circuit condition; When the execution result of the executed criterion satisfies the preset short-circuit condition, it is determined that the preset termination condition is satisfied; When the execution result of the executed criterion does not meet the preset short-circuit condition, it is determined that the preset termination condition is not met, and based on the execution result of the executed criterion and the state transition rule, the next criterion to be executed is determined and executed.

4. The defect identification method based on hierarchical decision-making according to claim 1, characterized in that, The process of determining and executing the next pending criterion based on the execution results and state transition rules of already executed criteria includes: Based on the execution result of the executed criteria, the target rule is determined in the state transition rules; wherein, the state transition rules include at least one of the following: sequence triggering rule, condition triggering rule, threshold triggering rule, evidence accumulation triggering rule, consistency triggering rule, conflict triggering rule, short-circuit triggering rule, evidence of contradiction triggering rule, and re-examination triggering rule; Based on the target rule, the next criterion to be executed is determined and executed.

5. The defect identification method based on hierarchical decision-making according to claim 4, characterized in that, The determination of the target rule in the state transition rule based on the execution result of the executed criteria includes: When the execution result of the executed criterion satisfies multiple state transition rules, the target rule is determined from the multiple state transition rules based on the preset rule priority.

6. The defect identification method based on hierarchical decision-making according to any one of claims 1 to 5, characterized in that, The execution result of the executed criterion includes at least one of the following: Boolean decision type, discrete state type, scalar metric type, vector feature type, structured object type, interval constraint type, probability confidence type, and symbolic state type.

7. The defect identification method based on hierarchical decision-making according to any one of claims 1 to 5, characterized in that, The defect identification results include at least one of the following: defect category information, confidence level information, evidence summary information, decision path information, anomaly level information, spatial region information, and consistency status information.

8. The defect identification method based on hierarchical decision-making according to any one of claims 1 to 5, characterized in that, The method further includes: When the preset degradation constraint is met, all the criteria are executed in parallel, and the defect identification result is determined based on the execution results of all the criteria.

9. The defect identification method based on hierarchical decision-making according to any one of claims 1 to 5, characterized in that, The different detection conditions include top-view imaging conditions and side-view imaging conditions; The plurality of criteria include a side-view structure criterion corresponding to the side-view imaging condition and a top-view criterion corresponding to the top-view imaging condition; wherein, the side-view structure criterion is used to evaluate the consistency of interlayer structure, and the top-view criterion is used to assist in the determination of defect causes. The step of determining and executing the first criterion to be executed among multiple criteria based on a preset decision priority includes: determining the side view structure criterion as the first criterion to be executed and executing it; The process of determining and executing the next pending criterion based on the execution result and state transition rules of the executed criterion includes: when the execution result of the side view structure criterion indicates that the structure is consistent, the top view criterion is determined as the next pending criterion and executed.

10. The defect identification method based on hierarchical decision-making according to any one of claims 1 to 5, characterized in that, The different detection conditions include low-voltage imaging conditions and high-voltage imaging conditions; The plurality of criteria include a low-voltage criterion corresponding to the low-voltage imaging condition and a high-voltage criterion corresponding to the high-voltage imaging condition; wherein, the low-voltage criterion is used for on-site defect trigger detection, and the high-voltage criterion is used for defect classification and previous station source determination. The step of determining and executing the first criterion to be executed among multiple criteria based on a preset decision priority includes: determining the low voltage criterion as the first criterion to be executed and executing it. The step of determining and executing the next pending criterion based on the execution result of the executed criterion and the state transition rule includes: when the execution result of the low voltage criterion is consistent, determining the high voltage criterion as the next pending criterion and executing it.

11. A defect identification system based on hierarchical decision-making, characterized in that, It includes a criterion acquisition module, a hierarchical decision-making module, and a result output module, which are connected in sequence, wherein: The criterion acquisition module is used to acquire multiple criteria from different detection conditions; wherein, the different detection conditions are imaging configuration methods used in the defect identification process that can produce differentiated defect feature display effects; the criteria are structured evidence representations output by the detection node after performing defect detection calculations, and the criteria are derived from detection algorithms or models configured for different detection conditions. The hierarchical decision-making module is used to determine and execute the first criterion to be executed from among multiple criteria based on a preset decision priority, and to determine and execute the next criterion to be executed based on the execution result and state transition rules of the executed criteria, and to repeat the determination and execution steps of the next criterion to be executed until a preset termination condition is met; wherein, the preset termination condition includes at least one of the following: a preset short-circuit condition, a termination condition based on complete traversal, a forced convergence condition based on the maximum number of execution steps, a termination condition based on external intervention instructions, and a termination condition based on evidence accumulation saturation; The result output module is used to determine the defect identification result based on the execution result of the executed criteria after the preset termination condition is met. The hierarchical decision-making module is further configured to construct a set of states including an initial state, an intermediate evidence accumulation state, and a termination state; wherein, the state transition rule is configured to trigger a transition from the initial state or the current intermediate evidence accumulation state to the next intermediate evidence accumulation state based on the execution result of the executed criterion, so as to determine the next criterion to be executed, or to trigger entry into the termination state when the preset termination condition is met.

12. An electronic device, characterized in that, include: A processor, a memory, and a communication bus, wherein the processor and the memory communicate with each other via the communication bus; The memory stores program instructions that can be executed by the processor, and the processor can execute the method as described in any one of claims 1 to 10 by calling the program instructions.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1 to 10.

14. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 10.

Images