Three-dimensional microscopic image processing method in stevioside crystallization process

Abstract

The invention relates to the technical field of image processing, and discloses a method for processing a three-dimensional microscopic image in a stevioside crystallization process. The systematic defects of crystal identity breakage, grain number statistics distortion and insufficient tracing precision in the prior art are overcome, and triple spanning from static tracing to dynamic spectrum, from batch tracing to event positioning and from open-loop observation to closed-loop correction is realized. An explainable, verifiable and optimizable intelligent analysis means is provided for the continuous crystallization process, so that the crystal identification accuracy and the quality tracing efficiency are substantially improved, and an event-level accurate regulation and control basis is provided for process optimization.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a three-dimensional microscopic image processing method for the crystallization process of stevia. Background Technology

[0002] Industrial process analysis technology, as a core support in the pharmaceutical and fine chemical industries, has been further developed into online tracking of crystal morphology in crystallization unit operations. Industrial crystallization process control uses three-dimensional microscopic imaging devices to capture crystal morphology in real time, assigning a unique identifier to each crystal, establishing a spatiotemporal continuous growth archive from nucleation to the final product, and achieving online monitoring of crystal size distribution and particle count calculation. This forms a specific technical implementation path in the context of steviol glycoside purification.

[0003] Current technologies are based on the assumption of rigid objects, forcibly maintaining the identity of crystals. When high shear in industrial processes causes fragmentation, agglomeration, or ripening, the breakage of the identity chain leads to systematic distortion of the particle count model. After fragmentation, the particle count is artificially inflated; agglomeration causes an abnormal decrease in the particle count; and the source of impurities cannot be traced back to specific process fluctuations. In continuous crystallizers, the shear of the agitator paddles causes the proportion of crystal fragmentation to reach a high level. Existing systems misclassify fragments as newly formed crystals, causing crystal size distribution data to lose its material traceability significance.

[0004] Therefore, we propose a three-dimensional microscopic image processing method for the stevia crystallization process to solve the problems mentioned above. Summary of the Invention

[0005] The purpose of this invention is to provide a three-dimensional microscopic image processing method for the crystallization process of stevia, in order to solve the problems mentioned in the background art, such as the systematic distortion of the particle number balance model caused by the breakage of identity chains when industrial high shear causes crushing, agglomeration or ripening, the false increase of particle number after crushing, the abnormal decrease of particle number caused by agglomeration, and the inability to trace the source of impurities to specific process fluctuation events.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional microscopic image processing method for the crystallization process of stevia, the specific steps of which are as follows: S1. Multi-physics synchronous acquisition and labeling: A three-dimensional microscopic imaging device, a concentration detection probe and a temperature sensor are set up in the crystallizer to synchronously acquire crystal morphology, mother liquor concentration and temperature data. Based on the time series, the trend of process parameter change is predicted and the image frame is labeled as steady state or transient mode. S2. Dynamic Coding and Event Capture: Extract the morphological and compositional features of crystals to construct dynamic identifiers. When breakage, agglomeration, or maturation events are detected, freeze the original crystal identifiers and generate event markers. In steady-state mode, simulate the feature evolution path through a prediction model and determine crystals that deviate from the predicted values ​​as abnormal events. S3. Spectral graph construction and conservation verification: Establish a parent-child node relationship chain for crystals where events occur, expand the crystal identifier into a spectral node containing feature information, parent node pointer, child node list, event type and process parameters, store and update in real time using a graph database, and embed particle number and mass conservation operators for verification. S4. Physical Information Embedding and Traceability: Input the changes in process parameters into the prediction model to update the growth kinetic parameters online. When a quality defect is detected, trace back from the target node along the parent node chain. Combine the process parameters and event types of each ancestor node to calculate the causal contribution and locate the key event. S5. Adaptive optics calibration and correction: Real-time monitoring of photoelectric signal fluctuations in the imaging device; when light source power drift is detected, automatic dark field correction and noise model recalibration are triggered, and photothermal compensation terms in the growth kinetics model are corrected simultaneously. S6. Quality Traceability Report Generation: After each batch is completed, the cause-and-effect chain of defect events is extracted from the spectrum diagram, and a quality traceability report containing the spectrum of the problematic crystal, key process parameter nodes, and physical mechanisms is automatically generated.

[0007] Preferably, in step S2, the specific steps of dynamic encoding and event capture are as follows: S2.1 Extract crystal morphology and composition characteristics. When crystal breakage, agglomeration or maturation events occur, record the original crystal identifier and generate event markers. S2.2 In steady-state mode, the evolution path of crystal characteristics is simulated by a prediction model, and crystals that deviate from the predicted values ​​are identified as abnormal events.

[0008] Preferably, step S2.1 specifically includes the following steps: S2.11 Extract crystal contour morphology features from three-dimensional microscopic images and crystal composition features from concentration detection data. Weight the two types of features and fuse them to form a dynamic crystal identifier with time-adaptive evolution capability. The weight coefficients are dynamically adjusted according to the changing trend of operating parameters. S2.12. Based on the dynamic identifier, the mutation feature is used to detect crystal breakage, agglomeration or maturation events. When an event is detected, the original crystal identifier is immediately frozen and an event tag containing the event type, timestamp and mass conservation check code is generated.

[0009] Preferably, step S2.2 specifically includes the following steps: S2.21. In steady-state mode, a time-series prediction model is constructed based on historical data of crystal dynamic identification to simulate the normal evolution path of crystal features; S2.22. Compare the actual observed crystal characteristic values ​​with the model prediction values. When the deviation exceeds the preset threshold, it is judged as an abnormal event.

[0010] Preferably, in step S3, the specific steps for pedigree chart construction and conservation verification are as follows: S3.1 Establish a phylogenetic node and parent-child relationship chain. For crystals that have experienced breakage, agglomeration or maturation events, establish a relationship chain between parent nodes and child nodes, and expand the crystal identifier into a phylogenetic node structure that includes feature information, parent node pointer, child node list, event type and process parameters. S3.2 Graph database storage and conservation verification: The graph database stores the phylogenetic node structure and updates the parent-child relationship chain in real time. During the storage process, a grain number conservation and quality conservation verification mechanism is embedded. When the conservation residual exceeds the set range, a data quality alarm is issued.

[0011] Preferably, step S3.1 specifically includes the following steps: S3.11 Receive the event markers and operating condition parameter change trends generated in the aforementioned steps. When crystal breakage, agglomeration, or ripening events are detected, dynamically create a phylogenetic node based on the event type and operating condition drift amplitude, and establish a chain of association between parent and child nodes. S3.12. Expand the crystal identifier into a phylogenetic node structure that includes feature information, parent node pointer, child node list, event type and process parameters. Simultaneously embed a particle number conservation and mass conservation verification mechanism during the structure construction process. When the conservation residual exceeds the set range, the node structure is corrected.

[0012] Preferably, step S3.2 specifically includes the following steps: S3.21. An event-driven adaptive graph database storage architecture is adopted to perform incremental storage and dynamic topology reconstruction of the genealogy node structure, optimize the index structure of the parent-child relationship chain in real time, and establish an event causal index. S3.22. Construct a bidirectional conservation verification mechanism with multi-physics coupling. Simultaneously embed particle number conservation and quality conservation verification operators during storage. Trigger closed-loop feedback correction through residual analysis. When the conservation residual exceeds the set range, start the node structure self-repair program and issue a data quality alarm.

[0013] Preferably, in step S4, the specific steps for embedding and tracing physical information are as follows: S4.1 Input the trend of the operating condition parameters into the prediction model, update the crystal growth kinetic parameters online, and make the model adaptively match the operating condition drift. S4.2 When a quality defect is detected, trace back from the target node in the spectrum along the parent node chain, combine the operating parameters of each node with the event type to calculate the causal contribution, and locate the key event node that caused the defect.

[0014] Preferably, in step S5, the adaptive optics calibration and correction steps are as follows: S5.1 Monitor the fluctuation of photoelectric signal of imaging device, determine the drift amplitude of light source power, and generate a correction command when the drift amplitude exceeds the set range; S5.2 Perform dark field correction and noise model recalibration according to the correction instructions, and simultaneously correct the photothermal compensation term in the growth kinetic model.

[0015] Preferably, in step S6, the specific steps for generating the quality traceability report are as follows: S6.1 Identify defect event nodes related to quality problems from the genealogy diagram and extract the complete causal chain along the parent-child relationship chain; S6.2. Transform the causal chain data into a quality traceability report that includes the problematic crystal system, key process parameter nodes, and physical mechanisms.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This method solves the systemic defects of existing technologies, such as crystal identification fracture, grain count distortion, and insufficient traceability accuracy, through event-driven dynamic identification, material spectrum construction, and embedded traceability of process parameters. It achieves a triple leap from static tracking to dynamic spectrum, from batch traceability to event localization, and from open-loop observation to closed-loop correction. It provides interpretable, verifiable, and optimizable intelligent analysis means for continuous crystallization processes, substantially improves the accuracy of crystal identification and the efficiency of quality traceability, and provides event-level precise control basis for process optimization.

[0017] 2. Through the synergistic effect of event tagging and predictive judgment, step S2 forms a closed-loop mechanism of "trigger-response-prediction-identification". Compared with current technologies that can only perform rigid tracking, this mechanism realizes dynamic and intelligent crystal identity management, instantly anchoring the source of matter when a topological event occurs and proactively identifying potential risks during steady-state processes. This dual capability upgrades crystal identification from simple morphological matching to intelligent discrimination based on event causality and evolutionary laws, significantly improving the accuracy and timeliness of crystal behavior analysis in complex crystallization processes, and providing a high-quality data foundation for subsequent spectral construction. Attached Figure Description

[0018] Figure 1 This is a diagram illustrating the method steps of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: Please refer to Figure 1 A method for processing three-dimensional microscopic images during the crystallization process of stevia, the specific steps of which are as follows: S1. Multi-physics synchronous acquisition and labeling: A three-dimensional microscopic imaging device, a concentration detection probe and a temperature sensor are set up in the crystallizer to synchronously acquire crystal morphology, mother liquor concentration and temperature data. Based on the time series, the trend of process parameter change is predicted and the image frame is labeled as steady state or transient mode. S2. Dynamic Coding and Event Capture: Extract the morphological and compositional features of crystals to construct dynamic identifiers. When breakage, agglomeration, or maturation events are detected, freeze the original crystal identifiers and generate event markers. In steady-state mode, simulate the feature evolution path through a prediction model and determine crystals that deviate from the predicted values ​​as abnormal events. S3. Spectral graph construction and conservation verification: Establish a parent-child node relationship chain for crystals where events occur, expand the crystal identifier into a spectral node containing feature information, parent node pointer, child node list, event type and process parameters, store and update in real time using a graph database, and embed particle number and mass conservation operators for verification. S4. Physical Information Embedding and Traceability: Input the changes in process parameters into the prediction model to update the growth kinetic parameters online. When a quality defect is detected, trace back from the target node along the parent node chain. Combine the process parameters and event types of each ancestor node to calculate the causal contribution and locate the key event. S5. Adaptive optics calibration and correction: Real-time monitoring of photoelectric signal fluctuations in the imaging device; when light source power drift is detected, automatic dark field correction and noise model recalibration are triggered, and photothermal compensation terms in the growth kinetics model are corrected simultaneously. S6. Quality Traceability Report Generation: After each batch is completed, the cause-and-effect chain of defect events is extracted from the spectrum diagram, and a quality traceability report containing the spectrum of the problematic crystal, key process parameter nodes, and physical mechanisms is automatically generated.

[0021] In this embodiment: Step S1 achieves spatiotemporal alignment of crystal growth environment data through multi-physics synchronous acquisition and operating condition mode marking, which solves the defect of process parameter drift and image data disconnection in traditional methods, and provides benchmark data with operating condition context for subsequent dynamic identification, so that crystal behavior analysis is based on real process fluctuations.

[0022] Step S2 uses a dynamic coding mechanism to replace the traditional static identification. Through topological event capture and steady-state evolution prediction, it effectively identifies identity breakage events such as crystal breakage, agglomeration, or maturation, avoiding the problems of crystal identity confusion and false number of crystals in the prior art, and providing accurate event anchors for lineage tracing.

[0023] Step S3 constructs a phylogenetic diagram and embeds conservation verification, upgrading discrete identity identifiers into continuous parent-child relationship chains. Through real-time verification using particle number and mass conservation operators, the non-conservation defects caused by identity breakage in the PBE model are corrected, allowing the crystal quantity statistics to return to the essence of matter conservation.

[0024] Step S4 introduces changes in process parameters into the prediction model to achieve dynamic updates. By tracing back the genealogy, the source of quality defects in the process can be accurately located. This breaks through the limitation of traditional methods that can only trace batches but not events, and enables process optimization to shift from batch trial and error to event-level precise control.

[0025] Step S5 eliminates the observation error caused by the power drift of the light source through real-time monitoring and adaptive correction of photoelectric signals, and simultaneously corrects the photothermal compensation term of the model, ensuring the continuous consistency between the observation system and the mechanism model under changing operating conditions and improving the stability of long-term operation.

[0026] Step S6 automatically generates an event-level quality traceability report, transforming the causal chain in the genealogy diagram into an auditable process record, which meets the compliance requirement that quality originates from design, and enables batch release to shift from post-event sampling to precise decision-making based on event evidence.

[0027] This method addresses the systemic shortcomings of existing technologies, such as crystal identification fragmentation, grain count distortion, and insufficient traceability accuracy, through event-driven dynamic identification, material spectrum construction, and embedded traceability of process parameters. It achieves a triple leap from static tracking to dynamic spectrum, from batch traceability to event localization, and from open-loop observation to closed-loop correction. This provides interpretable, verifiable, and optimizable intelligent analysis tools for continuous crystallization processes, substantially improving the accuracy of crystal identification and the efficiency of quality traceability, and providing event-level precise control basis for process optimization.

[0028] Example 2: Please refer to Figure 1 In step S2, the specific steps for dynamic encoding and event capture are as follows: S2.1 Extract crystal morphology and composition characteristics. When crystal breakage, agglomeration or maturation events occur, record the original crystal identifier and generate event markers. S2.2 In steady-state mode, the evolution path of crystal characteristics is simulated by a prediction model, and crystals that deviate from the predicted values ​​are identified as abnormal events.

[0029] In this embodiment: Step S2.1 achieves event-level anchoring of crystal identity by extracting crystal morphology and composition characteristics and recording the original crystal identifier when a topological event occurs. This step completes the task of converting from static identifiers to dynamic identifiers, freezing the identity snapshot at the moment the fragmentation, aggregation, or maturation event is triggered, avoiding the problem of particle count distortion caused by the continuous continuation of identity in the prior art. Its purpose is to establish a traceable initial marker for each material fragment, so that subsequent lineage construction has accurate starting point information, and solves the technical defect of traditional methods that cannot distinguish between natural growth and topological mutation.

[0030] Step S2.2 constructs a prediction model and identifies anomalous events in steady-state mode, achieving proactive identification of crystal behavior. This step completes the autonomous learning task of feature evolution laws. By comparing actual observations with the model's deduction path, deviations exceeding tolerance are identified as anomalous events. Its purpose is to proactively detect crystal anomalies induced by operating condition fluctuations, rather than passively recording results. This solves the problems of strong lag in post-event analysis and the inability to intervene in real time in existing technologies, providing an early warning basis for process control.

[0031] The two sub-steps described above, through the synergistic effect of event tagging and prediction, enable step S2 to form a closed-loop mechanism of "trigger-response-prediction-identification." Compared to current technologies that can only perform rigid tracking, this mechanism achieves dynamic and intelligent crystal identity management, instantly anchoring the source of matter when a topological event occurs and proactively identifying potential risks during steady-state processes. This dual capability upgrades crystal identification from simple morphological matching to intelligent discrimination based on event causality and evolutionary laws, significantly improving the accuracy and timeliness of crystal behavior analysis during complex crystallization processes, and providing a high-quality data foundation for subsequent spectral construction.

[0032] Example 3: Please refer to Figure 1 The specific steps of step S2.1 are as follows: S2.11 Extract crystal contour morphology features from three-dimensional microscopic images and crystal composition features from concentration detection data. Weight the two types of features and fuse them to form a dynamic crystal identifier with time-adaptive evolution capability. The weight coefficients are dynamically adjusted according to the changing trend of operating parameters. S2.12. Based on the dynamic identifier, the mutation feature is used to detect crystal breakage, agglomeration or maturation events. When an event is detected, the original crystal identifier is immediately frozen and an event tag containing the event type, timestamp and mass conservation check code is generated.

[0033] In this embodiment: S2.11 extracts crystal contour morphology features from three-dimensional microscopic images and crystal composition features from concentration detection data, and then weights and fuses these two types of features to form a dynamic crystal identifier with time-adaptive evolution capabilities, thus completing the task of organically integrating multi-dimensional crystal features. The weight coefficients are dynamically adjusted according to the changing trend of operating parameters, enabling the identifier to truly reflect the state evolution of the crystal under different process conditions. This solves the defect of traditional static identifiers being unable to adapt to operating condition drift, achieving the goal of constructing a dynamic and adaptive crystal identity benchmark, and providing a reliable feature basis for subsequent event detection.

[0034] S2.12 detects crystal breakage, aggregation, or maturation events based on dynamic identifier mutation features. Upon detection, the original crystal identifier is immediately frozen, and an event marker containing the event type, timestamp, and matter conservation check code is generated, completing the instantaneous capture and anchoring of topological events. This step solidifies the source information of matter at the moment the event is triggered, avoiding the particle count distortion problem caused by the continuous continuation of identity in existing technologies. This achieves the goal of providing an accurate causal starting point for lineage tracing, making the flow path of each material fragment traceable.

[0035] The two sub-steps described above, through the synergistic effect of multimodal feature fusion and real-time event anchoring, enable step S2 to achieve dynamic and event-driven crystal identity management. Compared to current technologies that can only perform rigid tracking, this mechanism automatically adjusts feature weights during operating condition fluctuations and instantly freezes identity snapshots during topological changes, forming a closed-loop capability of "adaptive representation - real-time anchoring." This improvement upgrades crystal recognition from simple morphological matching to intelligent discrimination based on multi-physics field features and event causality, significantly enhancing the robustness and accuracy of crystal behavior analysis and tracing during complex crystallization processes, and providing a high-quality, verifiable data foundation for subsequent genealogy construction.

[0036] Example 4: Please refer to Figure 1 The specific steps of step S2.2 are as follows: S2.21. In steady-state mode, a time-series prediction model is constructed based on historical data of crystal dynamic identification to simulate the normal evolution path of crystal features; S2.22. Compare the actual observed crystal characteristic values ​​with the model prediction values. When the deviation exceeds the preset threshold, it is judged as an abnormal event.

[0037] In this embodiment: S2.21, a time-series prediction model is constructed based on historical data of crystal dynamic identification in steady-state mode, completing the autonomous learning task of the normal evolution law of crystal features. This model simulates the standard evolution path of crystal features under stable operating conditions by analyzing the temporal correlations in historical data, establishing a reliable benchmark for crystal behavior. This task overcomes the shortcomings of traditional methods that rely solely on static threshold judgments and lack a dynamic behavior prediction mechanism, achieving the goal of proactively grasping the crystal growth trend and evolution law during stable operating periods.

[0038] S2.22 compares the actual observed crystal characteristic values ​​with the model prediction values ​​to complete the task of real-time anomaly identification. When the calculated deviation exceeds the preset threshold range, the system determines that the crystal is in an abnormal state, realizing the immediate capture and proactive warning of crystal anomalies induced by operating condition fluctuations. This task overcomes the shortcomings of existing technologies, such as strong lag in post-analysis and inability to intervene in real time, achieving the goal of proactive risk identification and advancing the timing of process control from the manifestation of abnormal results to the nascent stage of deviation.

[0039] The two sub-steps described above, through the synergistic effect of benchmark construction and real-time comparison, enable step S2 to form a closed-loop capability for prediction and detection. Compared to current technologies that can only perform post-event recording and analysis, this mechanism actively learns evolutionary patterns during the steady-state period and triggers judgments immediately when anomalies occur, achieving a fundamental shift from passive recording to proactive prediction, and from delayed analysis to real-time early warning. This improvement enhances the timeliness and accuracy of crystal anomaly detection, significantly advancing the timing of process intervention and avoiding batch quality accidents caused by delayed detection in traditional methods. Simultaneously, this mechanism provides interpretable behavioral prediction criteria for complex crystallization processes, enabling operators to understand crystal growth trends based on model deduction rather than relying solely on experience, thereby improving the scientific rigor and repeatability of process control.

[0040] Example 5: Please refer to Figure 1 In step S3, the specific steps for constructing the phylogenetic chart and verifying conservation are as follows: S3.1 Establish a phylogenetic node and parent-child relationship chain. For crystals that have experienced breakage, agglomeration or maturation events, establish a relationship chain between parent nodes and child nodes, and expand the crystal identifier into a phylogenetic node structure that includes feature information, parent node pointer, child node list, event type and process parameters. S3.2 Graph database storage and conservation verification: The graph database stores the phylogenetic node structure and updates the parent-child relationship chain in real time. During the storage process, a grain number conservation and quality conservation verification mechanism is embedded. When the conservation residual exceeds the set range, a data quality alarm is issued.

[0041] In this embodiment: S3.1 expands the crystal identifier into a spectral node structure containing feature information, parent node pointers, a list of child nodes, event types, and process parameters by establishing spectral nodes and parent-child relationship chains, thus completing the transformation of crystal identity from a static label to a dynamic spectral system. This structure immediately establishes the association chain between parent and child nodes when breakage, agglomeration, or maturation events occur, enabling a complete record of the material flow path. This solves the defect in traditional methods where the source of material cannot be traced after crystal identity fracture, achieving the goal of establishing a traceable spectral file for each material fragment, providing a complete topological foundation for subsequent reverse tracing.

[0042] S3.2 employs a graph database to store the phylogenetic node structure and updates the parent-child relationship chain in real time. During storage, a grain number conservation and mass conservation verification mechanism is embedded. When the conservation residual exceeds a set range, a data quality alarm is issued, completing the dual tasks of phylogenetic data storage and physical consistency verification. This mechanism performs bidirectional conservation verification synchronously during the data persistence phase, ensuring the integrity of grain number and mass data after topological events occur. It solves the problem of grain number statistical distortion in PBE models caused by identity breaks in existing technologies, achieving the goal of timely detection and correction of anomalies in the data storage stage, and restoring crystal quantity statistics to the essence of matter conservation.

[0043] The two sub-steps described above, through the synergistic effect of phylogenetic node construction and storage verification, enable step S3 to achieve both phylogenetic management of crystal identity and data reliability assurance. Compared to current technologies that can only perform isolated identification and simple recording, this mechanism establishes parent-child relationships instantly upon event occurrence and performs conservation verification synchronously during storage, achieving a fundamental shift from discrete identification to continuous phylogenetics and from single storage to verification storage. This improvement enhances the integrity and accuracy of crystal traceability, avoiding traceability failures caused by identity fragmentation and storage errors in traditional methods. Simultaneously, this mechanism provides verifiable material flow records for complex crystallization processes, enabling process analysis based on reliable phylogenetic data, thereby enhancing the credibility and compliance of quality traceability and meeting the stringent data integrity requirements of continuous crystallization processes.

[0044] Example 6: Please refer to Figure 1 The specific steps of step S3.1 are as follows: S3.11 Receive the event markers and operating condition parameter change trends generated in the aforementioned steps. When crystal breakage, agglomeration, or ripening events are detected, dynamically create a phylogenetic node based on the event type and operating condition drift amplitude, and establish a chain of association between parent and child nodes. S3.12. Expand the crystal identifier into a phylogenetic node structure that includes feature information, parent node pointer, child node list, event type and process parameters. Simultaneously embed a particle number conservation and mass conservation verification mechanism during the structure construction process. When the conservation residual exceeds the set range, the node structure is corrected.

[0045] In this embodiment: S3.11, by receiving event markers and the changing trends of operating parameters, dynamically creates phylogenetic nodes and establishes parent-child relationship chains based on the event type and the amplitude of operating condition drift when crystal breakage, agglomeration, or maturation events are detected, thus completing the task of transforming crystal identity from static labels to dynamic phylogenetic trees. This mechanism adaptively adjusts the node creation strategy under fluctuating operating conditions, enabling the complete recording of material flow paths. It solves the defect in traditional methods where the source of material cannot be traced after crystal identity breakage, achieving the goal of establishing a traceable phylogenetic file for each material fragment and providing a complete topological relationship foundation for subsequent reverse tracing.

[0046] S3.12 expands the crystal identifier into a phylogenetic node structure containing feature information, parent node pointers, a list of child nodes, event types, and process parameters. During structure construction, a particle count conservation and mass conservation verification mechanism is simultaneously embedded. When the conservation residual exceeds a set range, node structure correction is triggered, completing the dual tasks of phylogenetic data storage and physical consistency verification. This mechanism performs bidirectional conservation verification simultaneously during the data persistence phase, ensuring the integrity of particle count and mass data after topological events occur. It solves the problem of particle count statistical distortion in PBE models caused by identity fragmentation in existing technologies, achieving the goal of timely detection and correction of anomalies during data storage.

[0047] The two sub-steps described above, through the synergistic effect of spectral node construction and storage verification, enable step S3 to achieve both spectralization of crystal identity management and dual assurance of data reliability. Compared to current technologies that can only perform isolated identification and simple recording, this mechanism establishes parent-child relationships instantly upon event occurrence and performs conservation verification synchronously in the storage stage, achieving a fundamental shift from discrete identification to continuous spectral systems and from single storage to verified storage. This improvement enhances the integrity and accuracy of crystal traceability, avoiding traceability failures caused by identity fragmentation and storage errors in traditional methods. This mechanism provides verifiable material flow records for complex crystallization processes, enabling process analysis based on reliable spectral data, enhancing the credibility and compliance of quality traceability, and meeting the stringent data integrity requirements of continuous crystallization processes. Simultaneously, the combination of dynamic node creation and embedded verification allows the system to adaptively adjust the traceability granularity during operating condition fluctuations and automatically trigger corrections when data anomalies occur, forming a closed-loop capability of adaptive construction and real-time verification. This capability ensures that the accuracy and consistency of pedigree data are maintained during long-term continuous operation, providing a high-quality and reliable data foundation for process optimization and upgrading quality traceability from post-event verification to a reliable mechanism built into the process.

[0048] Example 7: Please refer to Figure 1 The specific steps of step S3.2 are as follows: S3.21. An event-driven adaptive graph database storage architecture is adopted to perform incremental storage and dynamic topology reconstruction of the genealogy node structure, optimize the index structure of the parent-child relationship chain in real time, and establish an event causal index. S3.22. Construct a bidirectional conservation verification mechanism with multi-physics coupling. Simultaneously embed particle number conservation and quality conservation verification operators during storage. Trigger closed-loop feedback correction through residual analysis. When the conservation residual exceeds the set range, start the node structure self-repair program and issue a data quality alarm.

[0049] In this embodiment: S3.21 adopts an event-driven adaptive graph database storage architecture, incrementally storing and dynamically reconstructing the phylogenetic node structure, optimizing the index structure of parent-child relationship chains in real time, and establishing an event causal index, thus completing the tasks of efficient storage and fast retrieval of phylogenetic data. This architecture stores node information in real time when an event is triggered and dynamically adjusts the topological relationship, solving the problems of poor timeliness and low query efficiency caused by batch storage in traditional databases. It achieves the goal of maintaining the integrity of the phylogenetic system in real time under fluctuating operating conditions, and shortens the response time for traceability queries to the millisecond level.

[0050] S3.22 constructs a bidirectional conservation verification mechanism coupled with multi-physics fields. Granularity conservation and mass conservation verification operators are simultaneously embedded during storage. Closed-loop feedback correction is triggered through residual analysis. When the conservation residual exceeds a set range, a node structure self-repair procedure is initiated and a data quality alarm is issued, completing the data integrity verification and anomaly self-repair tasks in the storage stage. This mechanism performs bidirectional conservation verification synchronously during the data persistence phase, identifies anomalies through residual analysis, and triggers automatic correction. This solves the problem of timely detection and correction of storage errors in existing technologies, achieving the goal of ensuring physical consistency in real time during data entry and avoiding traceability failures caused by storage errors.

[0051] The two sub-steps described above, through the synergy of an adaptive storage architecture and an embedded verification mechanism, enable step S3 to achieve a dual improvement in the efficiency and reliability of spectral data management. Compared to current technologies that can only perform simple recording and post-event verification, this mechanism stores and optimizes the index in real time when an event occurs, and performs conservation verification and self-repair simultaneously during the storage process, achieving a fundamental shift from batch storage to incremental storage, and from manual verification to automatic verification. This improvement enhances the timeliness, completeness, and accuracy of spectral data, avoiding the traceability failures caused by storage delays, rigid indexes, and missing verification in traditional methods. This mechanism provides a reliable record of material flow for complex crystallization processes, enabling process analysis to be based on real-time and accurate spectral data, enhancing the credibility and compliance of quality traceability. Simultaneously, the combination of event-driven storage and closed-loop verification allows the system to adaptively adjust its storage strategy when operating conditions fluctuate and automatically trigger repairs when data anomalies occur, forming a closed-loop capability of efficient storage and real-time assurance. This capability ensures that the accuracy and consistency of spectral data are maintained during long-term continuous operation, providing a high-quality, verifiable data foundation for process optimization. It also upgrades quality traceability from post-event verification to a reliable mechanism built into the process, meeting the stringent requirements of continuous crystallization processes for data timeliness and integrity.

[0052] Example 8: Please refer to Figure 1 In step S4, the specific steps for embedding and tracing physical information are as follows: S4.1 Input the trend of the operating condition parameters into the prediction model, update the crystal growth kinetic parameters online, and make the model adaptively match the operating condition drift. S4.2 When a quality defect is detected, trace back from the target node in the spectrum along the parent node chain, combine the operating parameters of each node with the event type to calculate the causal contribution, and locate the key event node that caused the defect.

[0053] In this embodiment: S4.1 By inputting the trend of changes in operating parameters into the prediction model and updating the crystal growth kinetic parameters online, the adaptive matching task between the model and the operating condition drift is completed. This mechanism enables the growth kinetic parameters to be dynamically adjusted with changes in process conditions, solving the defect that traditional model parameters are fixed and cannot reflect the fluctuations of actual operating conditions. It achieves the goal of maintaining the accuracy of model prediction under continuous operating condition drift, and provides a model basis synchronized with the actual process state for subsequent quality defect tracing.

[0054] S4.2 achieves precise localization of the root cause of quality defects by tracing back along the parent node chain from the target node in the genealogy and calculating the causal contribution by combining the operating parameters of each node with the event type. This mechanism, upon detecting a quality defect, traverses the genealogical structure backward and quantifies the contribution of each ancestor node to the defect. This overcomes the limitation of traditional methods, which can only trace back to the production batch and cannot pinpoint specific process events. It achieves the goal of improving the accuracy of quality traceability from the batch level to the event level, enabling process optimization to precisely control specific fault events.

[0055] The two sub-steps described above, through the synergistic effect of model adaptive updates and spectral reverse tracing, enable step S4 to achieve a deep coupling between the process model and the material spectrum. Compared to current technologies that can only perform batch-level tracing, this mechanism automatically corrects model parameters when operating conditions fluctuate and accurately locates event nodes when quality defects occur, achieving a fundamental shift from static to dynamic models and from batch tracing to event tracing. This improvement enhances the accuracy of quality tracing and the targeting of process control, avoiding optimization failures caused by model mismatch and coarse tracing granularity in traditional methods. This mechanism provides an interpretable path for analyzing the causes of defects in complex crystallization processes, enabling process improvements to be based on specific event data, thus enhancing the scientific rigor and effectiveness of quality optimization. Simultaneously, the combination of model adaptation and event localization allows the system to maintain tracing accuracy when operating conditions change and quickly pinpoint the root cause when defects occur, forming a closed-loop capability of model correction and precise tracing. This capability ensures that the reliability of quality traceability and the effectiveness of process optimization are maintained during long-term continuous operation, providing event-level data support for quality assurance in continuous crystallization processes, and enabling process optimization to shift from experience-based trial and error to precise decision-making based on event evidence.

[0056] Example 9: Please refer to Figure 1 In step S5, the specific steps for adaptive optics calibration and correction are as follows: S5.1 Monitor the fluctuation of photoelectric signal of imaging device, determine the drift amplitude of light source power, and generate a correction command when the drift amplitude exceeds the set range; S5.2 Perform dark field correction and noise model recalibration according to the correction instructions, and simultaneously correct the photothermal compensation term in the growth kinetic model.

[0057] In this embodiment: S5.1 monitors the photoelectric signal fluctuations of the imaging device and determines the magnitude of the light source power drift, thus completing the real-time monitoring task of the imaging system stability. When the drift exceeds the set range, a correction command is generated, realizing the active detection and response to light source fluctuations. This solves the defect in traditional methods where light source drift causes systematic deviations in the observation data that cannot be detected in time. It achieves the purpose of instant identification and triggering correction in the observation stage, avoiding the accumulation of crystal feature extraction errors caused by light source instability.

[0058] S5.2 executes dark-field correction and noise model recalibration according to the correction command, and simultaneously corrects the photothermal compensation term in the growth kinetic model, completing the collaborative correction task of the observation system and the mechanism model. This step updates the noise model and adjusts the model's photothermal parameters immediately after the light source correction, solving the defect of the disconnect between observation error and model parameters in traditional methods. It achieves the goal of maintaining the consistency between observation data and mechanism model, ensuring the accuracy of crystal identification and growth prediction in long-term operation.

[0059] The two sub-steps described above, through the linkage of photoelectric signal monitoring and model collaborative correction, enable step S5 to form a closed-loop observation-correction capability. Compared to current techniques that can only perform periodic manual calibration, this mechanism triggers correction instantly when the light source fluctuates and corrects synchronously when model parameters drift, realizing a shift from manual intervention to automatic response and from separate correction to collaborative optimization. This improvement enhances the stability of the imaging system and the accuracy of model predictions, avoiding the crystal identification failure problem caused by unstable light sources and model mismatch in traditional methods. This mechanism provides a self-maintaining observation guarantee system for continuous crystallization processes, enabling process analysis to be based on stable and reliable image data, and enhancing the data credibility of quality traceability. Simultaneously, the combination of photoelectric monitoring and model correction allows the system to adaptively adjust observation parameters when operating conditions change and automatically trigger correction when data is abnormal, forming a closed-loop capability of real-time monitoring and instant correction. This capability ensures that observation quality and model accuracy are maintained during long-term continuous operation, providing a stable and reliable data foundation for process optimization, and shifting quality traceability from reliance on manual maintenance to a built-in guarantee mechanism based on automatic correction.

[0060] Example 10: Please refer to Figure 1 In step S6, the specific steps for generating the quality traceability report are as follows: S6.1 Identify defect event nodes related to quality problems from the genealogy diagram and extract the complete causal chain along the parent-child relationship chain; S6.2. Transform the causal chain data into a quality traceability report that includes the problematic crystal system, key process parameter nodes, and physical mechanisms.

[0061] In this embodiment: S6.1 Identify defect event nodes related to quality problems from the genealogy diagram, extract the complete causal chain along the parent-child relationship chain, and complete the task of accurately locating the root cause of quality defects. This step automatically retrieves event nodes associated with quality problems in the genealogy structure and traces back to the original ancestor node, solving the defect of traditional methods that can only trace back to the production batch and cannot locate specific process events. This achieves the goal of improving the accuracy of quality traceability from the batch level to the event level, enabling process optimization to be precisely controlled for specific failure events.

[0062] S6.2 transforms causal chain data into a quality traceability report containing the problematic crystal spectrum, key process parameter nodes, and physical mechanisms, thus completing the automatic generation of the traceability report. This step automatically converts the causal chain information in the spectrum diagram into auditable process records, solving the drawbacks of incomplete, untimely, and non-compliant traditional manual records. It achieves the goal of ensuring quality originates from design requirements, enabling batch release to make accurate decisions based on event evidence, and enhancing the credibility and compliance of quality traceability.

[0063] The two sub-steps described above, through the synergistic effect of causal chain extraction and automatic report generation, enable step S6 to achieve overall automation and precision in quality traceability. Compared to current technologies that can only perform batch-level traceability and manual recording, this mechanism automatically locates the event node when a defect occurs and automatically generates a compliance report after traceability is completed, achieving a fundamental shift from manual verification to automatic traceability, and from batch records to event evidence. This improvement enhances the efficiency and accuracy of quality traceability, avoiding the optimization failure problems caused by the coarse granularity of traceability and the lag in recording in traditional methods. This mechanism provides a verifiable path for defect cause analysis in complex crystallization processes, enabling process improvements to be based on specific event data, thus enhancing the scientific rigor and effectiveness of quality optimization. Simultaneously, the combination of automatic extraction and report generation allows the system to quickly pinpoint the root cause when a quality defect occurs and immediately output audit records after traceability is completed, forming a closed-loop capability of rapid location and immediate reporting. This capability ensures that the timeliness and compliance of quality traceability are maintained during long-term continuous operation, providing event-level data support for process optimization. It enables quality assurance to shift from relying on human experience to a built-in mechanism based on automatic traceability, meeting the stringent requirements of continuous crystallization processes for the timeliness and reliability of quality traceability.

[0064] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0065] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for processing three-dimensional microscopic images during the crystallization process of stevia, characterized in that: The specific steps are as follows: S1. Multi-physics synchronous acquisition and labeling: A three-dimensional microscopic imaging device, a concentration detection probe and a temperature sensor are set up in the crystallizer to synchronously acquire crystal morphology, mother liquor concentration and temperature data. Based on the time series, the trend of process parameter change is predicted and the image frame is labeled as steady state or transient mode. S2. Dynamic Coding and Event Capture: Extract the morphological and compositional features of crystals to construct dynamic identifiers. When breakage, agglomeration, or maturation events are detected, freeze the original crystal identifiers and generate event markers. In steady-state mode, simulate the feature evolution path through a prediction model and determine crystals that deviate from the predicted values ​​as abnormal events. S3. Spectral graph construction and conservation verification: Establish a parent-child node relationship chain for crystals where events occur, expand the crystal identifier into a spectral node containing feature information, parent node pointer, child node list, event type and process parameters, store and update in real time using a graph database, and embed particle number and mass conservation operators for verification. S4. Physical Information Embedding and Traceability: Input the changes in process parameters into the prediction model to update the growth kinetic parameters online. When a quality defect is detected, trace back from the target node along the parent node chain. Combine the process parameters and event types of each ancestor node to calculate the causal contribution and locate the key event. S5. Adaptive optics calibration and correction: Real-time monitoring of photoelectric signal fluctuations in the imaging device; when light source power drift is detected, automatic dark field correction and noise model recalibration are triggered, and photothermal compensation terms in the growth kinetics model are corrected simultaneously. S6. Quality Traceability Report Generation: After each batch is completed, the cause-and-effect chain of defect events is extracted from the spectrum diagram, and a quality traceability report containing the spectrum of the problematic crystal, key process parameter nodes, and physical mechanisms is automatically generated.

2. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 1, characterized in that: In step S2, the specific steps of dynamic encoding and event capture are as follows: S2.1 Extract crystal morphology and composition characteristics. When crystal breakage, agglomeration or maturation events occur, record the original crystal identifier and generate event markers. S2.2 In steady-state mode, the evolution path of crystal characteristics is simulated by a prediction model, and crystals that deviate from the predicted values ​​are identified as abnormal events.

3. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 2, characterized in that: The specific steps of step S2.1 are as follows: S2.11 Extract crystal contour morphology features from three-dimensional microscopic images and crystal composition features from concentration detection data. Weight the two types of features and fuse them to form a dynamic crystal identifier with time-adaptive evolution capability. The weight coefficients are dynamically adjusted according to the changing trend of operating parameters. S2.

12. Based on the dynamic identifier, the mutation feature is used to detect crystal breakage, agglomeration or maturation events. When an event is detected, the original crystal identifier is immediately frozen and an event tag containing the event type, timestamp and mass conservation check code is generated.

4. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 3, characterized in that: The specific steps of step S2.2 are as follows: S2.

21. In steady-state mode, a time-series prediction model is constructed based on historical data of crystal dynamic identification to simulate the normal evolution path of crystal features; S2.

22. Compare the actual observed crystal characteristic values ​​with the model prediction values. When the deviation exceeds the preset threshold, it is judged as an abnormal event.

5. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 4, characterized in that: In step S3, the specific steps for constructing the phylogenetic diagram and verifying conservation are as follows: S3.1 Establish a phylogenetic node and parent-child relationship chain. For crystals that have experienced breakage, agglomeration or maturation events, establish a relationship chain between parent nodes and child nodes, and expand the crystal identifier into a phylogenetic node structure that includes feature information, parent node pointer, child node list, event type and process parameters. S3.2 Graph database storage and conservation verification: The graph database stores the phylogenetic node structure and updates the parent-child relationship chain in real time. During the storage process, a grain number conservation and quality conservation verification mechanism is embedded. When the conservation residual exceeds the set range, a data quality alarm is issued.

6. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 5, characterized in that: The specific steps of step S3.1 are as follows: S3.11 Receive the event markers and operating condition parameter change trends generated in the aforementioned steps. When crystal breakage, agglomeration, or ripening events are detected, dynamically create a phylogenetic node based on the event type and operating condition drift amplitude, and establish a chain of association between parent and child nodes. S3.

12. Expand the crystal identifier into a phylogenetic node structure that includes feature information, parent node pointer, child node list, event type and process parameters. Simultaneously embed a particle number conservation and mass conservation verification mechanism during the structure construction process. When the conservation residual exceeds the set range, the node structure is corrected.

7. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 6, characterized in that: The specific steps of step S3.2 are as follows: S3.

21. An event-driven adaptive graph database storage architecture is adopted to perform incremental storage and dynamic topology reconstruction of the genealogy node structure, optimize the index structure of the parent-child relationship chain in real time, and establish an event causal index. S3.

22. Construct a bidirectional conservation verification mechanism with multi-physics coupling. Simultaneously embed particle number conservation and quality conservation verification operators during storage. Trigger closed-loop feedback correction through residual analysis. When the conservation residual exceeds the set range, start the node structure self-repair program and issue a data quality alarm.

8. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 7, characterized in that: In step S4, the specific steps for embedding and tracing physical information are as follows: S4.1 Input the trend of the operating condition parameters into the prediction model, update the crystal growth kinetic parameters online, and make the model adaptively match the operating condition drift. S4.2 When a quality defect is detected, trace back from the target node in the spectrum along the parent node chain, combine the operating parameters of each node with the event type to calculate the causal contribution, and locate the key event node that caused the defect.

9. The three-dimensional microscopic image processing method for the stevia crystallization process according to claim 8, characterized in that: In step S5, the specific steps for adaptive optics calibration and correction are as follows: S5.1 Monitor the fluctuation of photoelectric signal of imaging device, determine the drift amplitude of light source power, and generate a correction command when the drift amplitude exceeds the set range; S5.2 Perform dark field correction and noise model recalibration according to the correction instructions, and simultaneously correct the photothermal compensation term in the growth kinetic model.

10. A three-dimensional microscopic image processing method for the crystallization process of stevia according to claim 9, characterized in that: In step S6, the specific steps for generating the quality traceability report are as follows: S6.1 Identify defect event nodes related to quality problems from the genealogy diagram and extract the complete causal chain along the parent-child relationship chain; S6.

2. Transform the causal chain data into a quality traceability report that includes the problematic crystal system, key process parameter nodes, and physical mechanisms.

Images