A real-time monitoring method and system for a cistanche crystallization process

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a real-time monitoring method and system for a Cistanche crystallization process, which comprises the following steps: performing segmentation model processing and image analysis on the obtained Cistanche image to obtain crystal particle geometric parameters including particle area, contour perimeter, and length and width of the minimum circumscribed rectangle; obtaining a crystal morphology isomerism index; calculating a crystal neighborhood disturbance degree; calculating a crystal signal-to-noise ratio; calculating a crystal comprehensive early warning value; performing time sequence confirmation on candidate particles with high matching degrees; performing global confirmation of abnormal states on the candidate particles confirmed by time sequence based on the global crystal particle distribution of the current process stage; and comparing the candidate particles confirmed by the global confirmation with a final early warning threshold and giving an early warning. The present application solves the technical problem of detection insensitivity of the prior art to non-normal crystal forms in the background of a large number of normal crystal samples.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a real-time monitoring method and system for the crystallization process of Cistanche. BACKGROUND

[0002] In the industrial production of active ingredients of Cistanche, such as strophanthin, purification by crystallization process is a key link to obtain high-purity products. The core goal of the crystallization process is not only to precipitate crystals, but also to stably produce normal crystal forms with specific solubility, high stability and bioavailability. However, in the crystallization kettle, abnormal crystal forms with different physical properties, such as metastable crystal forms and solvent compounds, coexist and compete with solute molecules in the solution. Any slight fluctuation in process parameters may induce preferential nucleation of abnormal crystal forms and trigger catastrophic and irreversible crystal form transformation, resulting in unqualified product quality.

[0003] In order to realize accurate control of the crystallization process, the prior art usually adopts an online image analysis method, which captures microscopic images in the crystallization kettle and uses image processing algorithms to statistically analyze the particle size and quantity of the crystals, greatly improving the control efficiency of the Cistanche crystallization process by the staff.

[0004] However, in the early or stable period of the crystallization process, the crystal nucleus of the abnormal crystal form is a rare event, and its quantity and volume are much smaller than those of the dominant normal crystal form. Traditional image analysis methods usually rely on standard classification models, which will be severely biased towards the majority class, i.e., the normal crystal form, when dealing with scenes with extremely unbalanced class numbers. This results in extremely low recognition sensitivity for the small number of abnormal crystal nuclei, making it difficult to effectively identify and warn before the abnormal crystal form triggers large-scale crystal transformation, and failing to meet the ultra-high sensitivity detection requirements of accurately identifying a single abnormal crystal nucleus in a million normal crystals. SUMMARY

[0005] To solve the technical problem of the prior art that the detection sensitivity of the abnormal crystal form is lost in the background of a large number of normal crystal samples, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a real-time monitoring method for the crystallization process of Cistanche, comprising:

[0007] The obtained Cistanche image is subjected to segmentation model processing and image analysis to obtain crystal particle geometric parameters including particle area, contour perimeter, length and width of minimum circumscribed rectangle; the crystal theoretical area is calculated according to the contour perimeter of the crystal particle, and the crystal morphology isomerism index is obtained by comparing the crystal theoretical area with the actual area; a preset circular neighborhood of a single crystal particle is taken as the center, and the crystal morphology isomerism index and the Euclidean distance of all crystal particles in the circular neighborhood except the center crystal are calculated, and the crystal neighborhood disturbance degree is calculated according to the distribution sparsity and the crystal morphology isomerism index of the single crystal particle and the neighborhood crystal particle; the crystal signal-to-noise ratio is calculated according to the difference between the crystal morphology isomerism index and the crystal neighborhood disturbance degree of the current crystal particle; the crystal comprehensive early warning value is calculated in combination with the numerical correlation and dynamic adjustment relationship of the crystal signal-to-noise ratio and the crystal morphology isomerism index; the candidate crystal particles that first exceed the average value of all crystal comprehensive early warning values are subjected to time sequence growth fingerprint matching verification, and the candidate particles with high matching degree are subjected to time sequence confirmation; the candidate particles confirmed by the time sequence are subjected to global confirmation of abnormal state based on the global crystal particle distribution of the current process stage; and the candidate particles confirmed globally are compared with the final early warning threshold and early warning is performed.

[0008] The traditional method is difficult to identify a small number of abnormal crystals from a large number of normal crystals, and is prone to the problem of insensitive detection. The present application first accurately acquires crystal data and screens out useless information, thereby laying a reliable foundation for subsequent analysis. Then, the abnormal crystals are acutely captured by comprehensively judging from the crystal morphology, the surrounding environment, the time change, the overall distribution and the like, and the misjudgment is excluded. Whether it is a slight abnormality in the overall stable environment or a prominent abnormality in the deteriorating environment, the abnormal crystals can be accurately identified, potential risks are not missed, and blind early warning is not performed. Finally, only the high-confidence abnormal crystals trigger early warning, thereby effectively solving the problem of insensitive detection of the traditional method, ensuring the stability of the crystallization process, and avoiding the unqualified product quality caused by abnormal crystals.

[0009] Preferably, the crystal particle geometric parameters including particle area, contour perimeter, length and width of minimum circumscribed rectangle are obtained, including:

[0010] A high-resolution immersion microscope imaging probe is used to continuously collect real-time image streams of the solid-liquid two-phase mixture in the crystallization kettle at a preset frame rate. Each frame of image obtained is processed by using an instance segmentation model to accurately identify and segment each independent crystal particle in the image. For each identified particle, crystal particle geometric parameters are obtained by using an image analysis algorithm, including particle area, contour perimeter, and length and width of minimum circumscribed rectangle. A screening algorithm is applied to screen and exclude crystal agglomerates that are too large in size or irregular in shape according to the crystal particle geometric parameters.

[0011] Preferably, the crystal morphology isomerism index satisfies the following expression:

[0012] ;

[0013] wherein, is a crystal morphology heterogeneity index of the i-th crystal particle, dimensionless; is a contour perimeter of the i-th crystal particle; is a contour perimeter of the i-th crystal particle; is a minimum circumscribed rectangle length of the i-th crystal particle; is a minimum circumscribed rectangle length of the i-th crystal particle; is a minimum circumscribed rectangle width of the i-th crystal particle; is a minimum circumscribed rectangle width of the i-th crystal particle; is a natural exponential function; is a natural exponential function; is a natural exponential function;

[0014] The present application can clearly distinguish different morphologies of crystal particles, i.e. normal and abnormal crystal particles, by calculating the crystal morphology heterogeneity index, so that the abnormal crystal particles can be quickly locked, and a clear basis can be provided for further determining whether the crystal particles have problems, so that the abnormal crystal particles can be found earlier.

[0015] Preferably, the crystal neighborhood disturbance degree satisfies the following expression:

[0016] ;

[0017] wherein, is a crystal neighborhood disturbance degree of the i-th crystal particle, dimensionless; is a crystal neighborhood disturbance degree of the i-th crystal particle, dimensionless; is a set of all crystal particles in a circular neighborhood of the i-th crystal particle; is a set of all crystal particles in a circular neighborhood of the i-th crystal particle; is a crystal morphology heterogeneity index of the j-th crystal particle in a circular neighborhood of the i-th crystal particle; is a crystal morphology heterogeneity index of the j-th crystal particle in a circular neighborhood of the i-th crystal particle; is an Euclidean distance between the j-th crystal particle and the k-th crystal particle in a circular neighborhood of the i-th crystal particle and the j-th crystal particle; is an Euclidean distance between the j-th crystal particle and the k-th crystal particle in a circular neighborhood of the i-th crystal particle and the j-th crystal particle; is an Euclidean distance between the j-th crystal particle and the k-th crystal particle in a circular neighborhood of the i-th crystal particle and the j-th crystal particle; is an Euclidean distance between the j-th crystal particle and the k-th crystal particle in a circular neighborhood of the i-th crystal particle and the j-th crystal particle; is an Euclidean distance between the j-th crystal particle and the k-th crystal particle in a circular neighborhood of the i-th crystal particle and the j-th crystal particle; is a very small positive number, used to prevent the denominator in the inner layer from being zero; is a very small positive number, used to prevent the denominator in the outer layer from being zero.

[0018] When calculating the perturbation degree of a crystal neighborhood, this invention considers the situation of surrounding crystal particles and their distance from the central crystal. The impact of abnormal crystals that are close to the central crystal on the stability of the surrounding environment of the central crystal will be more accurately reflected, which can more realistically reflect the state of the microenvironment in which the central crystal is located. This helps staff to fully understand the situation around the crystal and determine whether there are any local environmental anomalies.

[0019] Preferably, calculating the crystal signal-to-noise ratio includes:

[0020] The crystal morphological heterogeneity index of the current crystal particle is compared with the perturbation degree of the crystal neighborhood, and the calculation result is used as the crystal signal-to-noise ratio. The crystal signal-to-noise ratio is used to evaluate the degree of anomaly of the current crystal particle relative to its surrounding background.

[0021] Preferably, the calculated crystal comprehensive early warning value satisfies the following expression:

[0022] ;

[0023] In the formula, Indicates the first The overall early warning value of each crystal particle is dimensionless. Indicates the first Crystal morphological isomerism index of individual crystal grains; Indicates the first The signal-to-noise ratio of each crystal particle; This represents the mean crystal morphological isomerism index of all crystal particles; It is a very small dimensionless positive number used to prevent the denominator from being zero; It is a natural exponential function.

[0024] When calculating the comprehensive early warning value of a crystal, this invention combines the degree of anomaly of the crystal particle itself with its prominence in the surrounding environment. It can also dynamically adjust according to the overall crystal environment. When the overall environment is stable, even minor anomalies can be keenly detected. When the environment deteriorates, it can focus on the most prominent anomalies, making the early warning more in line with the actual situation, neither omitting potential risks nor issuing blind warnings.

[0025] Preferably, time-series verification is performed on candidate particles with high matching degree, including:

[0026] For candidate crystal particles whose comprehensive early warning value exceeds the average of all comprehensive early warning values ​​for the first time, the system initiates tracking and records its subsequent continuous values. A time series of the crystal polymorph index in the frame image is recorded as a crystal growth sequence to be measured; the crystal growth sequence to be measured is matched with a typical abnormal crystal growth sequence stored in an expert knowledge base in dynamic time warping; only when the matching degree of the crystal growth sequence to be measured with a certain typical abnormal crystal growth sequence is higher than the average matching similarity, the abnormal state of the candidate particle is time-sequentially confirmed.

[0027] Preferably, the global confirmation of the abnormal state of the candidate particle confirmed by time sequence includes:

[0028] For the candidate particle confirmed by time sequence, the system further statistically calculates the global distribution of the crystal polymorph index of all crystal particles in the current field of view in real time, and calculates the 95th percentile of all crystal particles; then, the relative significance of the candidate particle is calculated according to the absolute difference of the crystal polymorph index exceeding the 95th percentile; when the relative significance of the candidate particle exceeds a certain multiple, the abnormal state of the candidate particle is globally confirmed.

[0029] According to the overall distribution of all crystal particles, the application determines the dynamic standard for judging whether the crystal particle is abnormal, which is adjusted accordingly at different process stages, avoiding the use of a fixed standard to judge crystal particles at different stages, and ensuring that slight abnormalities can be detected in the early stages of the process, and that serious abnormalities that need to be paid attention to can be accurately identified in the later stages.

[0030] Preferably, the candidate particle confirmed globally is compared with the final warning threshold and a warning is given, including:

[0031] Only when a candidate particle After passing the time sequence growth fingerprint matching verification and the global confirmation of the abnormal state, the system can confirm it as a high-confidence warning target, and compare its final crystal comprehensive warning value with a fixed final decision threshold, and once the condition that the crystal comprehensive warning value is greater than the final decision threshold is met, the system immediately triggers the highest level of warning, and highlights the position, image and all calculated features of the abnormal crystal nucleus of the candidate particle

[0032] In the second aspect, the application provides a real-time monitoring system for a cistanche crystallization process, which comprises a processor and a memory, and the memory stores computer program instructions, which, when executed by the processor, implement the above-mentioned real-time monitoring method for a cistanche crystallization process.

[0033] By adopting the above technical solution, the above-mentioned real-time monitoring method for a cistanche crystallization process is generated into a computer program and stored in a memory to be loaded and executed by a processor, so that a terminal device is made according to the memory and the processor, and the use is facilitated. ​

[0034] The present application has the advantages that the present application can monitor the crystallization process in real time and accurately, discover abnormal conditions in time, help workers take measures at the early stage of the problem, avoid the problem from expanding to cause the whole batch of products to be out of standard, reduce production loss. Meanwhile, the whole monitoring process has high automation degree and convenient operation, reduces the error and workload of manual operation, and improves production efficiency. In addition, the stable crystallization process can ensure the consistency of product quality, improve the competitiveness of products in the market, bring better economic benefits to enterprises, and promote the development of monitoring technology in the field of production of active ingredients of Cistanche. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a flow chart schematically showing a real-time monitoring method of a Cistanche crystallization process in the present application;

[0036] Figure 2 is a schematic diagram of a Cistanche in-pot live in the present application;

[0037] Figure 3 is a Cistanche crystallization extraction diagram in the present application. DETAILED DESCRIPTION

[0038] The present application discloses a real-time monitoring method of a Cistanche crystallization process, referring to Figure 1 , comprising steps S1-S4:

[0039] S1: performing segmentation model processing and image analysis on the obtained Cistanche image to obtain crystal particle geometric parameters including particle area, contour perimeter, and length and width of the minimum circumscribed rectangle.

[0040] It should be noted that the monitoring method of the present application is based on the uninterrupted exploration of the microcosmic world inside the crystallization kettle. First, the original image information needs to be obtained from the data source and standardized to generate the basic data required for subsequent calculation, such as Figure 2 : Cistanche in-pot live diagram, Figure 3 : Cistanche crystallization extraction diagram. The semi-continuous crystallization process of Cistanche active ingredients aims to maintain the stable growth of normal needle-shaped crystals in the kettle. The traditional offline sampling and inspection method not only has a serious time lag, but also may cause crystal dissolution or secondary nucleation due to the change of temperature and pressure during the sampling process, which cannot reflect the real state in the kettle. The present application automatically and standardizedly obtains in-situ and real-time full-particle image data in the kettle, which establishes a clean and regular data basis for subsequent intelligent analysis.

[0041] Specifically, the obtained Cistanche image is subjected to segmentation model processing and image analysis to obtain crystal particle geometric parameters including particle area, contour perimeter, and length and width of the minimum circumscribed rectangle, including:

[0042] Using an immersion high-resolution microscopic imaging probe, real-time image streams of the solid-liquid two-phase mixture inside the crystallization vessel are continuously acquired at a preset frame rate. Each acquired image frame is processed using an instance segmentation model to accurately identify and segment each individual crystal particle in the image. For each identified particle, image analysis algorithms are used to obtain the crystal particle's geometric parameters, including: particle area, outline perimeter, and the length and width of the minimum bounding rectangle. A screening algorithm is then applied to identify and exclude crystal agglomerates that are too large or have irregular shapes based on the crystal particle's geometric parameters.

[0043] Thus, the geometric parameters of the crystal grains were obtained.

[0044] S2: Calculate the theoretical area of ​​the crystal based on the perimeter of the crystal grains, and compare the theoretical area with the actual area of ​​the crystal to obtain the crystal morphological isomerism index.

[0045] It should be noted that in real crystallization scenarios, an ideal crystal particle exhibits a specific needle-like or rod-like shape, with a regular and highly anisotropic morphology. In contrast, the metastable nucleus of an abnormal crystal particle often appears as a blocky or flocculent mass, with a more compact and isotropic morphology. Therefore, this invention constructs an evaluation criterion that allows the system to automatically identify and calculate the degree of morphological degradation from needle-like to blocky, thereby enabling the identification of individual abnormal particles at the starting point of the analysis.

[0046] Specifically, the theoretical area of ​​the crystal is calculated based on the perimeter of the crystal grains, and the theoretical area is compared with the actual area of ​​the crystal to obtain the crystal morphological isomerism index, including:

[0047] The crystal morphological isomerism index satisfies the following expression:

[0048] ;

[0049] In the formula, Indicates the first The crystal morphological isomerism index of each crystal grain, dimensionless; Indicates the first The perimeter of the outline of a crystal grain; Indicates the first The length of the smallest bounding rectangle of a crystal grain; Indicates the first The width of the minimum bounding rectangle of a crystal grain; Represents the natural exponential function; Pi is the mathematical constant of a circle.

[0050] In the formula, Indicates the first crystal theoretical area of a single crystal particle at a given perimeter, denominator is the crystal actual area actually occupied by a single crystal particle; The imbalance between the crystal theoretical area and the crystal actual area is calculated, and the ratio is extremely large for acicular crystals and small for blocky crystals; The nonlinear mapping is represented by a natural exponential function, which maps the extremely large ratio to a value close to 0 and maps the smaller ratio to a value significantly larger than 0, thereby achieving nonlinear amplification of the degradation of the morphology from acicular to any non-acicular morphology.

[0051] For example, assuming that there is an ideal acicular crystal particle with a geometric feature of microns, microns, and a perimeter of approximately microns, the crystal morphology isomerism index is: which is extremely close to 0; there is also a case of an abnormal crystal nucleus in the form of an approximate square with a geometric feature of microns, microns, and a perimeter of microns, and the crystal morphology isomerism index is: The above calculation results, are rounded to three decimal places, and the above calculation results are rounded to eight decimal places.

[0052] S3: Taking a single crystal particle as the center, a crystal preset circular neighborhood is calculated, the crystal morphology isomerism index and the Euclidean distance of all crystal particles in the circular neighborhood except the center crystal are calculated, and the crystal neighborhood disturbance degree is calculated according to the distribution sparsity and the crystal morphology isomerism index of the single crystal particle and the neighborhood crystal particles.

[0053] It should be noted that in a real crystallization kettle, the abnormal morphology of a single crystal particle may sometimes be caused by image noise or accidental overlap and posture of the particle, and does not necessarily represent a real abnormal crystal nucleus event. A real, newly born abnormal crystal nucleus is the result of a local physical and chemical environment mutation, and such mutation often has a spatial aggregation effect. Therefore, the present application introduces a crystal neighborhood disturbance degree index, which not only considers the morphology of the adjacent particles, but also weights the distance of the neighborhood crystal particles from the center single crystal particle, to more accurately calculate the overall stability of the microenvironment of the single crystal particle.

[0054] Specifically, a circular neighborhood is predefined with a single crystal grain as the central crystal. The crystal morphological isomerism index and Euclidean distance from the central crystal are calculated for all crystal grains within the circular neighborhood except for the central crystal. Based on the sparse distribution of the single crystal grain and the neighboring crystal grains and the crystal morphological isomerism index, the crystal neighborhood perturbation is calculated, including:

[0055] The crystal neighborhood perturbation degree satisfies the following expression:

[0056] ;

[0057] In the formula, Indicates the first The crystal neighborhood perturbation degree of each crystal grain, dimensionless; Indicates the first The set of all crystal particles in the circular neighborhood of a crystal particle, excluding the central crystal. Indicates the first In the circular neighborhood of the nth crystal grain Crystal morphological isomerism index of individual crystal grains; Indicates the first The first crystal grain and the first In the circular neighborhood of the nth crystal grain Euclidean distance between individual crystal grains; It is a very small positive number, used to prevent the inner denominator from being zero; It is a very small positive number used to prevent the outer denominator from being zero.

[0058] In the formula, Indicates the first In the circular neighborhood of the nth crystal grain The crystal morphology isomerism index of the first crystal grain, the crystal morphology isomerism index being related to the crystal morphology isomerism index of the first crystal grain. The contribution of the distance between individual crystal grains; the closer the distance, the greater the influence. The smaller the value, the higher the contribution. The larger; This represents the summation of the contribution values ​​corresponding to all neighboring crystal particles; A weighted average was calculated such that anomalous crystal particles closer to the central particle contribute significantly more to the perturbation of the crystal neighborhood than anomalous crystal particles farther away.

[0059] For example, suppose the first There are two anomalous crystal particles surrounding the first crystal particle. and Their crystal morphological isomerism indices are all ,but and Adjacent, distance micrometer; and far away, microns, using distance-weighted formula, and negligible, its crystal neighborhood disturbance degree and if using simple arithmetic average, the result is Assume the abnormality is higher, and the abnormality is lower, at this time, the distance-weighted crystal neighborhood disturbance degree is which is significantly higher than the arithmetic average accurately reflecting the strong influence of the high-risk state of the near neighbor on the microenvironment of the current crystal particle. The above calculation results to one decimal place, to three decimal places.

[0060] S4: According to the difference between the crystal morphology isomerization index of the current crystal particle and the crystal neighborhood disturbance degree, the crystal signal-to-noise ratio is calculated; combined with the numerical correlation and dynamic adjustment relationship of the crystal signal-to-noise ratio and the crystal morphology isomerization index, the crystal comprehensive early warning value is calculated; the candidate crystal particles that first exceed the average value of all crystal comprehensive early warning values are subjected to time sequence growth fingerprint matching verification, and the candidate particles with high matching degree are subjected to time sequence confirmation; based on the global crystal particle distribution of the current process stage, the candidate particles that pass the time sequence confirmation are subjected to global confirmation of abnormal state; the candidate particles that pass the global confirmation are compared with the final early warning threshold and early warning is performed.

[0061] It should be noted that the ultimate goal of the present application is to achieve ultra-high sensitivity early warning of rare abnormal crystal nuclei through a set of completely data-driven, physically process-constrained calculation framework. Thus, the present application has completed the step-by-step progressive calculation from the standardization processing of raw image data to the calculation of the crystal morphology isomerization index of a single particle and the crystal neighborhood disturbance degree of its microenvironment. In the following content, the present application proposes a peak amplification calculation mode, that is, comparing and amplifying the crystal morphology isomerization index representing individual abnormalities with the crystal neighborhood disturbance degree representing the background environment.

[0062] Specifically, according to the difference between the crystal morphology isomerization index of the current crystal particle and the crystal neighborhood disturbance degree, the crystal signal-to-noise ratio is calculated, including:

[0063] The crystal morphology isomerization index of the current crystal particle is compared with the crystal neighborhood disturbance degree, and the calculation result is taken as the crystal signal-to-noise ratio, which is used to evaluate the prominence of the abnormality degree of the current crystal particle compared with the background of its surrounding environment.

[0064] It should be noted that the ultimate goal of this invention is to achieve ultra-high sensitivity early warning for rare and abnormal crystal nuclei. To this end, this invention has obtained, through a series of calculations, a crystal morphology heterogeneity index characterizing the degree of anomalousness of a single crystal particle, and a crystal signal-to-noise ratio characterizing the clarity of the anomalous signal relative to its local background. Now, this invention faces a final core problem: how to fuse the crystal morphology heterogeneity index, representing signal intensity, and the crystal signal-to-noise ratio, representing signal clarity, into a final, decision-level early warning value. A simple linear combination or product cannot reflect the dynamic decision-making needs of complex industrial scenarios. For example, in an extremely stable system, even a clear, minute anomaly should be greatly amplified; while in a chaotic system where the system has generally deteriorated, only those truly outstanding anomalous signals deserve attention. Therefore, this invention proposes a calculation method that can dynamically adjust the amplification based on the global environment.

[0065] Specifically, by combining the numerical correlation and dynamic adjustment relationship between the crystal signal-to-noise ratio and the crystal morphological isomerism index, a comprehensive early warning value for the crystal is calculated, including:

[0066] The overall early warning value for crystals satisfies the following expression:

[0067] ;

[0068] In the formula, Indicates the first The overall early warning value of each crystal particle is dimensionless. Indicates the first Crystal morphological isomerism index of individual crystal grains; Indicates the first The signal-to-noise ratio of each crystal particle; This represents the mean crystal morphological isomerism index of all crystal particles; It is a very small dimensionless positive number used to prevent the denominator from being zero; It is a natural exponential function.

[0069] In the formula, Indicates the first The degree of deviation of the crystal morphological isomerism index of an individual crystal grain from the mean of the crystal morphological isomerism indices of all crystal grains. This indicates that when the crystallization process is stable, When I was very young, even Smaller It will rapidly approach the upper limit of 2, making the system extremely sensitive to small anomalies; when the crystallization process deteriorates, When it rises, only Only crystal grains far exceeding the average level can make Maintain a high value; The combination of abnormal signal definition and dynamic sensitivity is realized, so that the enhancement degree of the abnormal signal depends on both the definition of the signal itself and the sensitivity adjustment under the current environment. is the amplification coefficient of , through the structure of , the information of is retained, and the dynamic enhancement effect is superimposed. By multiplying the of a single crystal particle with , dynamic and adaptive early warning of the risk of crystal particles is realized, that is, in a stable environment, it is extremely sensitive to small abnormalities, and in a deteriorating environment, it focuses on the most prominent risks.

[0070] For example, when the overall crystallization environment is stable, an isolated abnormal crystal particle appears, with a high crystal morphology isomerization index of ; but all the surrounding crystal particles are normal crystal particles, resulting in a very low crystal neighborhood disturbance degree of the abnormal crystal particle , that is , so the signal-to-noise ratio . At this time, the system is stable, and the average value of the crystal morphology isomerization index of all crystal particles is very low, that is , , . At this time, the crystal comprehensive early warning value is: ; when the overall crystallization environment has deteriorated to a certain extent, there is an abnormal abnormal crystal particle , which has the same crystal morphology isomerization index as the abnormal crystal particle , that is ; but there are also other abnormal crystal particles around it, resulting in a high crystal neighborhood disturbance degree of , so the signal-to-noise ratio . At this time, the system has generally deteriorated, and the average value of the crystal morphology isomerization index of all crystal particles is high, that is , . At this time, the crystal comprehensive early warning value is: . By comparison, it can be seen that by introducing the global average value as a dynamic scale, the system automatically becomes extremely sensitive in a stable environment, and the system is automatically passivated to focus on the most prominent risks in a deteriorating environment. The above calculation results , , are retained to three decimal places.

[0071] It should be noted that the crystal comprehensive early warning value calculated above is only a momentary snapshot based on the spatial information of a single frame of image, and it is easy to misjudge the momentary artifacts caused by image noise, accidental overlap of particles, etc. as real threats. A real, process-risk abnormal crystal nucleus must follow certain physical and chemical laws in its morphological evolution, and show continuity or specific growth patterns in the time dimension. Therefore, this step aims to filter out isolated abnormal points without historical roots by checking the life history of the crystal particles, and improve the accuracy of the early warning.

[0072] Preferably, the candidate crystal particles that first exceed the mean value of all crystal comprehensive early warning values are subjected to time sequence growth fingerprint matching verification, and the candidate particles with high matching degree are subjected to time sequence confirmation, including:

[0073] For the candidate crystal particles whose crystal comprehensive early warning value first exceeds the mean value of all crystal comprehensive early warning values, the system starts tracking and records the subsequent continuous time sequence of the crystal morphological isomerization index in the frame image, denoted as the measured crystal growth sequence; the measured crystal growth sequence is matched with the typical abnormal crystal growth sequence pre-stored in the expert knowledge base; only when the matching degree of the measured crystal growth sequence with a certain typical abnormal crystal growth sequence is higher than the mean value of the matching similarity, the abnormal state of the candidate particle is confirmed in time sequence.

[0074] It should be noted that the abnormality of a crystal particle is a relative concept, and its judgment standard should be dynamically adjusted according to the overall state of the crystallization process. For example, in the early stage of crystallization, the overall system environment is relatively stable, and a slight crystal morphological abnormality of a crystal particle is worth warning, while in the late stage of crystallization, the crystal particle population itself may have a certain degree of morphological generalization, and a higher abnormal threshold is needed. This step aims to establish an adaptive reference system to ensure that the system only warns about truly significant abnormal points under the current process background.

[0075] Preferably, the global confirmation of the abnormal state of the candidate particles confirmed by time sequence is based on the global crystal particle distribution of the current process stage, including:

[0076] For the candidate particles confirmed by time sequence, the system further real-time statistics the global distribution of the crystal morphological isomerization index of all crystal particles in the current field of view, and calculates the 95th percentile of all crystal particles; then according to the absolute difference of the crystal morphological isomerization index exceeding the 95th percentile, the relative significance of the candidate particle is calculated; when the relative significance of the candidate particle exceeds a certain multiple, the abnormal state of the candidate particle is globally confirmed.

[0077] It should be noted that the 95th percentile is a statistical indicator, which specifically means that the crystal morphology index of all crystal particles in the current field is sorted from low to high, and the value at the 95th position. In the present application, it represents the statistical upper limit of the morphology of most crystal particles at the current time, and is therefore used as a dynamic and automatically following process state change abnormality judgment baseline.

[0078] It should be noted that after spatial correlation analysis, time sequence fingerprint verification and global confirmation of abnormal state, the system has accumulated multi-dimensional evaluation criteria for each candidate crystal particle. This step is the last link of decision-making, and its purpose is to integrate all evaluation criteria. Only those candidate crystal particles that pass the spatial, time and global triple test at the same time are identified as high-confidence early warning targets and trigger the final early warning action.

[0079] Specifically, the candidate particles confirmed globally are compared with the final early warning threshold and early warning, including:

[0080] Only when a candidate particle After passing the time sequence growth fingerprint matching verification and global confirmation of abnormal state, the system can confirm it as a high-confidence early warning target, and compare its final crystal comprehensive early warning value with the fixed final decision threshold. Once the condition that the crystal comprehensive early warning value is greater than the final decision threshold is met, the system immediately triggers the highest level of early warning, and highlights the position, image and all calculated characteristics of the abnormal crystal nucleus.

[0081] It should be noted that the fixed final decision threshold is an optimal risk boundary value determined by offline statistical analysis of historical successful and failed batch production data, after balancing between high detection rate and acceptable false positive rate.

[0082] By then, the real-time monitoring and early warning of the cistanche crystallization process are completed.

[0083] The embodiment of the present application also discloses a real-time monitoring system for the cistanche crystallization process, comprising a processor and a memory, the memory storing computer program instructions, when the computer program instructions are executed by the processor, realizing the real-time monitoring method for the cistanche crystallization process according to the present application.

[0084] The above system also includes a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and therefore will not be repeated here.

[0085] While the specification has illustrated and described various embodiments of the application, it will be clear to those of ordinary skill in the art that various changes, modifications, and substitutions can be made thereto without departing from the spirit and scope of the application. It is understood that in the process of practicing the application, various alternatives, modifications, and equivalents can be employed.

Claims

1. A method for real-time monitoring of a process for crystallization of Cistanche, characterized in that, The method comprises the following steps: Segmentation model processing and image analysis are performed on the obtained Cistanche images to obtain crystal particle geometric parameters including particle area, contour perimeter, and length and width of the minimum circumscribed rectangle; The crystal theoretical area is calculated according to the contour perimeter of the crystal particle, and the crystal theoretical area is compared with the actual crystal area to obtain a crystal morphology isomerism index; a circular neighborhood is preset around a single crystal particle as a center crystal, and the crystal morphology isomerism index and the Euclidean distance of all crystal particles in the circular neighborhood except the center crystal are calculated, and the crystal neighborhood disturbance degree is calculated according to the distribution sparsity and the crystal morphology isomerism index of the single crystal particle and the neighborhood crystal particles; The crystal signal-to-noise ratio is calculated according to the difference between the crystal morphology isomerism index and the crystal neighborhood disturbance degree of the current crystal particle; the crystal comprehensive early warning value is calculated in combination with the numerical correlation and dynamic adjustment relationship of the crystal signal-to-noise ratio and the crystal morphology isomerism index; the candidate crystal particles that first exceed the average value of all crystal comprehensive early warning values are subjected to time sequence growth fingerprint matching verification, and the candidate particles with high matching degree are subjected to time sequence confirmation; the candidate particles that pass the time sequence confirmation are subjected to global confirmation of abnormal state based on the global crystal particle distribution of the current process stage; and the candidate particles that pass the global confirmation are compared with the final early warning threshold and are warned.

2. A method of real time monitoring of the process of C. ferax crystallization according to claim 1, characterized in that, The crystal particle geometric parameters including the particle area, the contour perimeter, and the length and width of the minimum circumscribed rectangle are obtained by the following steps: A high-resolution immersion microscope imaging probe is used to continuously collect real-time image streams of a solid-liquid two-phase mixture in a crystallization kettle at a preset frame rate, and each frame of image collected is processed by using an instance segmentation model to accurately identify and segment each independent crystal particle in the image, and for each identified particle, the crystal particle geometric parameters including the particle area, the contour perimeter, and the length and width of the minimum circumscribed rectangle are obtained by using an image analysis algorithm; and a screening algorithm is applied to screen and exclude crystal agglomerates that are too large in size or irregular in shape according to the crystal particle geometric parameters.

3. A method of real time monitoring of the process of C. ferax crystallization according to claim 1, characterized in that, The crystal morphology isomerism index satisfies the following expression: ； wherein represents a crystal morphology heterogeneity index of the jth crystal particle, dimensionless; represents a crystal morphology heterogeneity index of the jth crystal particle, dimensionless; represents the perimeter length of the jth crystal particle; represents the perimeter length of the jth crystal particle; represents the length of the minimum circumscribed rectangle of the jth crystal particle; represents the length of the minimum circumscribed rectangle of the jth crystal particle; represents the width of the minimum circumscribed rectangle of the jth crystal particle; represents the width of the minimum circumscribed rectangle of the jth crystal particle; represents a natural exponential function; is the mathematical constant pi.

4. A method of real time monitoring of the process of C. ferax crystallization according to claim 1, characterized in that, The crystal neighborhood disturbance degree satisfies the following expression: ； In the formula, Indicates the first The crystal neighborhood perturbation degree of each crystal grain, dimensionless; Indicates the first The set of all crystal particles in the circular neighborhood of a crystal particle, excluding the central crystal. Indicates the first In the circular neighborhood of the nth crystal grain Crystal morphological isomerism index of individual crystal grains; Indicates the first The first crystal grain and the first In the circular neighborhood of the nth crystal grain Euclidean distance between individual crystal grains; It is a very small positive number, used to prevent the inner denominator from being zero; It is a very small positive number used to prevent the outer denominator from being zero.

5. A method of real time monitoring of the process of C. ferax crystallization according to claim 1, characterized in that, The crystal signal-to-noise ratio is calculated by the following steps: The crystal morphology isomerism index of the current crystal particle is compared with the crystal neighborhood disturbance degree, and the calculation result is taken as the crystal signal-to-noise ratio, which is used to evaluate the abnormality degree of the current crystal particle compared with the prominence degree of the surrounding environment background.

6. A method of real time monitoring of the process of C. ferax crystallization according to claim 1, characterized in that, The crystal comprehensive early warning value is calculated by the following expression: ； wherein represents the crystal integrated early warning value of the jth crystal particle, dimensionless; represents the crystal integrated early warning value of the jth crystal particle, dimensionless; represents the crystal integrated early warning value of the jth crystal particle, dimensionless; represents the crystal integrated early warning value of the jth crystal particle, dimensionless; represents the crystal integrated early warning value of the jth crystal particle, dimensionless; represents the crystal integrated early warning value of the jth crystal particle, dimensionless; represents the crystal integrated early warning value of the jth crystal particle, dimensionless; is a very small dimensionless normal number for preventing the denominator from being zero; is a natural exponential function.

7. A method of real time monitoring of the process of Cistanche tubulosa crystallization according to claim 1, characterized in that, The candidate particles with high matching degree are subjected to time sequence confirmation by the following steps: For the candidate crystal particles whose first crystal comprehensive early warning value exceeds the average of all crystal comprehensive early warning values, the system starts tracking and recording the subsequent continuous The time sequence of the crystal morphology isomerization index in the frame image is recorded as a crystal growth sequence to be tested; the crystal growth sequence to be tested is dynamically time warping matched with the typical abnormal crystal growth sequences pre-stored in the expert knowledge base; only when the matching degree of the crystal growth sequence to be tested with a certain typical abnormal crystal growth sequence is higher than the average of the matching similarity, the abnormal state of the candidate particle is time-sequentially confirmed.

8. A method of real time monitoring of the process of Cistanche tubulosa crystallization according to claim 1, characterized in that, The candidate particles that pass the time sequence confirmation are subjected to global confirmation of abnormal state by the following steps: For the candidate particles that pass the time sequence confirmation, the system further statistically calculates the global distribution of the crystal morphology isomerism index of all crystal particles in the current field of view in real time, and calculates the 95th percentile of all crystal particles; then, the relative significance of the candidate particle is calculated according to the absolute difference of the crystal morphology isomerism index exceeding the 95th percentile; when the relative significance of the candidate particle exceeds a specific multiple, the abnormal state of the candidate particle is globally confirmed.

9. A method of real time monitoring of the process of C. ferax crystallization according to claim 1, characterized in that, comparing the globally confirmed candidate particles to a final alert threshold and alerting, including: Only when one candidate particle After passing the timing growth fingerprint matching verification and the global confirmation of abnormal state at the same time, the system can confirm it as a high-confidence early warning target, and compare its final crystal comprehensive early warning value with the fixed final decision threshold. Once the condition of crystal comprehensive early warning value greater than the final decision threshold is met, the system immediately triggers the highest level of early warning, and highlights the final early warning target The position, image and all calculated features of the determined abnormal crystal nucleus.

10. A real-time monitoring system for a process of crystallization of Cistanche, characterized in that, comprising: a processor and a memory, said memory storing computer program instructions which, when executed by the processor, implement a method of real-time monitoring of a Cistanche crystallization process according to any one of claims 1-9.

Images