Intelligent monitoring and adjusting method and system for frozen slices

Abstract

The invention discloses a frozen section intelligent monitoring and adjusting method and system, and relates to the field of intelligent monitoring and adjusting systems, and the method comprises the steps: obtaining frozen section image data through interaction; an adaptive local threshold method is adopted to carry out binarization processing on an image, an edge detection algorithm is combined to identify defect types in slices, quantitative analysis is carried out on the defect types, correlation analysis is carried out on data of the defect types and parameter information in the slicing process, and specific influence of each parameter on defect occurrence is quantified; calculating the influence weight of each parameter on the defect; building a training data set by using the parameter associated data so as to build a prediction model of a relationship among sample types, slice parameters and defects, enabling the model to accept new slice parameter input, and predicting a corresponding defect occurrence probability and number; and taking the parameter associated data and the output of the prediction model as an objective function, and searching an optimal slice parameter combination in a given parameter space in combination with an optimization algorithm to obtain an optimal slice parameter.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and adjustment systems, and particularly to an intelligent monitoring and adjustment method and system for frozen sections. Background Art

[0002] The frozen section technique is an important tool for pathological diagnosis and plays an irreplaceable role in histological examination and rapid pathological diagnosis. By rapidly freezing tissue samples and slicing them, traditional frozen section techniques rely on the operator's experience to complete key steps, including the adjustment of temperature and humidity, the setting of section thickness, and the maintenance and operation of the blade. This technique mainly based on manual experience has certain limitations in practical applications. Due to the uneven technical levels of operators, there is subjectivity in the setting and adjustment of section parameters, and the section quality is often difficult to reach a unified standard, resulting in significant differences in aspects such as section thickness, edge integrity, and texture clarity.

[0003] In addition, variables such as blade wear, sample hardness, and water content during the frozen section process will affect the section effect, and these parameters usually cannot be monitored and adjusted precisely in real time manually. Especially in complex or high-frequency diagnostic scenarios, it is difficult for operators to detect and correct problems during the sectioning process in a timely manner, such as uneven section thickness, edge breakage, or cracks. These problems directly lead to a reduction in diagnostic efficiency and affect the reliability of the results. At the same time, when traditional frozen section techniques detect defects, although some problems in the sections can be found through manual observation, the specific location, size, and shape of the defects cannot be accurately obtained, nor can these data be systematically recorded and analyzed, further limiting the improvement of section quality and the accuracy of diagnosis.

[0004] In summary, the existing frozen section techniques have significant deficiencies in terms of quality consistency, real-time adjustment ability, and defect detection and data management, and there is an urgent need for a technology that can achieve intelligent monitoring and dynamic adjustment to solve these problems. Summary of the Invention

[0005] In order to overcome the defects of the prior art in terms of sample limitations, time cost, and data dependence, the present application provides an intelligent monitoring and adjustment method and system for frozen sections, which while overcoming the above defects also improves the efficiency and accuracy of soil fertility assessment through intelligent means. The specific solutions are as follows: In the first aspect of the present application, an intelligent monitoring and adjustment method and system for frozen sections are provided, and the method includes: . An intelligent monitoring and adjustment method for frozen sections, characterized in that the method includes: Interactively obtain frozen section image data; The image is binarized using an adaptive local threshold method, and edge detection algorithms are combined to identify the types of defects in the slices, including defects such as cracks, wrinkles, bubbles, and knife marks. Quantitative analysis is performed on the detected defect data, such as counting the number and length of cracks, the area of wrinkles, and the number and size of bubbles, as the input data for subsequent analysis; The defect data after quantitative analysis is correlated with the parameter information during the slicing process, and the specific impact of each parameter on the occurrence of defects is quantified. Furthermore, the influence weight of each parameter on the defects is calculated to ensure the detailed quantification of the detected defects and provide reliable data input for subsequent parameter correlation and model construction; A training dataset is constructed using the parameter correlation data; a regression model, random forest, or deep learning model can be selected to construct a prediction model for the relationship between the sample type, slicing parameters, and the occurrence of defects, enabling the model to accept new slicing parameter inputs and predict the corresponding defect occurrence probability and quantity; The parameter correlation data and the output of the prediction model are used as the objective function, and combined with an optimization algorithm, the optimal slicing parameter combination is searched within the given parameter space as the optimal slicing parameters.

[0006] Furthermore, the method further includes; The optimal parameters are input into the prediction model to simulate and predict the occurrence of defects under these parameters to verify whether the results meet the preset quality requirements. For example, when the defect rate is lower than 5%, the slicing quality meets the requirements; If the predicted defect rate meets the requirements, accept this parameter combination as an optimization suggestion and generate an optimization report; If the predicted defect rate is still high, adjust the search range of the optimization algorithm or update the prediction model, and then re - optimize the parameters.

[0007] Furthermore, the parameter information includes; key parameters such as sample type, slicing temperature, humidity, slicing thickness, slicing speed, and the number of times the blade is used.

[0008] Furthermore, the method of binarizing the image using an adaptive local threshold method and combining edge detection algorithms to identify the types of defects in the slices is as follows; Remove the background part of the slice image and extract the image information of the slice area of the slice image; Use the adaptive local threshold method to binarize the grayscale image to ensure accurate segmentation of the slice area even under uneven illumination; Use the Canny edge detection algorithm to extract the main contours of the slice; Set appropriate thresholds (such as a low threshold of 50 and a high threshold of 150) to ensure that the details of the slice boundary can be detected; Repair the noise and holes in the segmented image through morphological operations (such as opening and closing operations); Extract the largest connected component as the slice region according to the connected component analysis, and eliminate the non-target regions; Use the Hough line transform algorithm to detect the thin and continuous linear structures in the image; Judge whether it is a crack by the thresholds of the line segment length and width; If it is judged that there are cracks in the image, output the starting and ending coordinates of the cracks, the crack length, and the number of cracks; Use the binary result of the image and combine with the circular Hough transform algorithm to identify circular structures; Judge whether it is a bubble by the thresholds of the region area and roundness (such as area > 20 pixels, roundness > 0.8); If it is judged that there are bubbles in the image, output the center coordinates, area, diameter, and the number of each bubble; Apply local texture analysis (such as LBP features) to extract texture irregular regions; Use morphological features (such as irregularity and distribution density) to judge whether the texture irregular region is a wrinkle; If it is judged that there are wrinkles in the image, output the contour of the wrinkle region, the polygon coordinate points, and the wrinkle coverage area; Compare the edge detection result with the theoretical slice boundary to calculate the difference region of the boundary; Statistical defect area and its distribution position, and output the number, position, and area ratio of the defect regions.

[0009] Furthermore, the method for quantitative analysis of the detected defects is as follows: Statistical number of each type of defect and the defect distribution position, such as the number of cracks, the number of bubbles, the number of wrinkles, etc.; Calculate the total defect area and quantify it with the ratio to the total slice area (coverage rate); According to the preset quality scoring criteria (such as the weights of defect types, quantities, coverage rates, etc.), calculate the comprehensive quality score of the slice.

[0010] Furthermore, the method also includes: Overlay the defect detection results on the slice image, such as the contours and annotations of defect regions such as cracks and bubbles, and use different colors to represent different defect types (such as red for cracks and blue for bubbles); Generate a defect detection report, which includes the sample number and slice number, defect types, quantities, areas and coverage rates, and the comprehensive slice quality score.

[0011] Further, the method for associating the detected defect data with the parameter information during the slicing process, quantitatively analyzing the specific impact of each parameter on the occurrence of defects, and then calculating the influence weight of each parameter on the defects to obtain the specific association between each defect type and the slicing parameters to obtain the association data is as follows: Interactively obtain the detected defect data, which includes defect types (such as cracks, bubbles, wrinkles, etc.), the number of defects, location, area coverage rate, distribution density, etc.; Interactively obtain the slicing parameter information, which includes slicing temperature, humidity, slicing thickness, slicing speed, blade status; Match the defect data generated by each slicing with the corresponding slicing parameters one by one to form an analysis data set; Use the Pearson correlation coefficient to calculate the linear correlation between the defect data and the slicing parameters; Through PCA dimensionality reduction analysis, analyze the comprehensive impact of multiple parameters on the defects and extract the parameters with the greatest impact; Use regression analysis methods (such as linear regression, logistic regression) to quantitatively analyze the specific impact of parameters on the defects; Calculate the influence weight of each parameter on the defects according to the correlation results to obtain the specific association between each defect type and the slicing parameters.

[0012] Further, the method for using the parameter association data and the output of the prediction model as the objective function, and combining with an optimization algorithm to search for the optimal slicing parameter combination within the given parameter space as the optimal slicing parameters is as follows: Define the objective function as the defect rate or the comprehensive quality loss index output by the prediction model; Adopt the Bayesian optimization algorithm to search for the optimal parameter combination within the preset parameter range as the preliminary optimal solution; In this stage, use the historical data (data in S300, S400) as the basis, and estimate the objective function distribution through a surrogate model (such as Gaussian process regression); After obtaining the preliminary optimal solution through global optimization, adopt local optimization (such as simulated annealing or gradient descent) to further fine-tune the parameters to obtain a more refined optimal result as the optimal slicing parameters In the second aspect of the present application, a frozen section intelligent monitoring and adjustment system is provided, and the system includes: A data acquisition and preprocessing module, which is used to obtain the original image and slicing parameter data to ensure the data quality.

[0013] An image analysis and defect detection module, which is used to process the image data, quantify various types of defects, and provide quantitative data for association analysis.

[0014] Defect data and slicing parameter correlation analysis module, which is used to match the detected defect data with parameter data, determine the key influencing factors through correlation, PCA and regression analysis, and calculate the influence weights of each parameter on defects.

[0015] Prediction model construction module, which constructs a prediction model of the relationship between sample types, slicing parameters and defect occurrence using defect data, slicing parameters and historical data, so as to predict the defect occurrence probability and quantity under the new parameter combination.

[0016] Parameter optimization and dynamic adjustment module, which uses the parameter correlation data and the output of the prediction model as the objective function, and combines with the optimization algorithm to search for the optimal slicing parameter combination in the given parameter space as the optimal slicing parameters; Optimized parameter verification and feedback module, which verifies whether the result meets the preset quality requirements by inputting the optimal parameters into the prediction model and simulating and predicting the defect occurrence situation under these parameters.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages: 1. By introducing machine learning and image recognition technologies, this application can evaluate the slicing quality in real time, ensure the uniformity of slicing thickness and the clarity of tissue structure, and significantly reduce the defect occurrence rate.

[0018] 2. The system of this application can automatically monitor the key parameters during the slicing process and make dynamic adjustments based on historical data, reducing the dependence on the operator's experience and improving the slicing quality and efficiency.

[0019] 3. Through image analysis and historical data learning, this application enables the system to quickly detect defects such as cracks and voids in the slices and adjust the slicing parameters in a timely manner to reduce the defective product rate.

[0020] 4. With this application, while making operation records, the data can be stored in the system. The system can not only organize the data, but also analyze these data to make real-time adjustments to the parameters of the next slicing. No matter the operator's experience level, stable and high-quality slicing results can be obtained through the system. Description of the Drawings

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic flowchart of a method for intelligent monitoring and adjustment of frozen sections provided by an embodiment of this application; Figure 2 It is a schematic structural diagram of a system for intelligent monitoring and adjustment of frozen sections provided by an embodiment of this application.

[0023] Explanation of reference numerals: Data acquisition and monitoring module 11, image processing and defect detection module 12, defect data and parameter correlation analysis module 13, prediction model construction module 14, soil fertility evaluation module 15, visualization module 16. Detailed implementation manners

[0024] This application provides a method and system for intelligent monitoring and adjustment of frozen sections, aiming to solve the technical problems in the prior art that in the traditional frozen section process, relying on manual experience leads to uneven slice quality, it is difficult to ensure the thickness uniformity and texture clarity of different slices, and for defects such as cracks and voids that may occur during the sectioning process, it is difficult for traditional technologies to achieve rapid identification and feedback. A method and system for intelligent monitoring and adjustment of frozen sections are provided, enabling the system proposed in this application to not only overcome the defects of the prior art in terms of sample limitations, time cost, and data dependence, but also obtain stable and high-quality sectioning results through intelligent means. It can also be set not inside the cryostat, but in an external computer, and obtain the required information for analysis by acquiring the records of the sectioning information made by the staff to obtain the specific parameters for the next sectioning. The staff can directly section according to these parameters for the next sectioning, so that in the subsequent sectioning records, only the sectioning images need to be entered into the system, and most of the sectioning data can directly use the data generated by the system, which can also greatly reduce the workload of the staff in recording sectioning information later, and it has broad application prospects.

[0025] Next, the technical solutions in the embodiments of this application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, rather than all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0026] It should be noted that the terms "include" and "have" and any of their variants are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products, or devices.

[0027] Embodiment 1, as Figure 1 shown, this application provides an intelligent monitoring and adjustment method for frozen sections. The method includes: S100 Obtain frozen section image data through interaction; Specifically: It can provide an interaction interface through the system, allowing operators to input sample information (such as image data, sample type, sample number, processing parameters, etc.), ensuring the association between the image data and the sample background information. And the operator can view the real-time acquisition status through a touch screen or a computer terminal to confirm whether the acquisition parameters are correct; When observing the section after each sectioning is completed, obtain the image information of the section, and automatically input the image information and other information in the sample information into the system by the system or manually by the operator, and store these data together with the image data as an archive for subsequent correlation analysis and quality assessment; The interaction interface displays all the sample information of the current section, facilitating the operator to confirm the validity of the data.

[0028] S200 Perform binary processing on the image using an adaptive local threshold method, and combine an edge detection algorithm to identify the defect types in the section, including defect types such as cracks, wrinkles, bubbles, and knife marks; The specific method is as follows: S210 Separate the background from the foreground using a segmentation algorithm (such as the Otsu algorithm) according to the image histogram or color information.

[0029] Use connected component analysis to retain the region with the largest area (usually the main body of the section) and eliminate other isolated regions.

[0030] Crop the image to extract the complete section region to remove the irrelevant background part in the section image and extract the image information of the main region of the section image; Apply Gaussian filtering and median filtering to smooth the image and eliminate random noise.

[0031] Adopt histogram equalization or adaptive histogram equalization (CLAHE) to enhance the overall contrast of the image, making the section structure more obvious, so as to reduce noise, improve the image contrast, and ensure the effect of subsequent defect detection; Scale the image to a preset size and convert the color image to grayscale to reduce the computational complexity while retaining the main structural information for subsequent algorithm processing; S220 uses an adaptive local threshold method to binarize the grayscale image. The specific method to ensure accurate segmentation of the slice area under uneven illumination is as follows; Divide the preprocessed grayscale image into several local small regions (such as 32×32 pixel or 64×64 pixel blocks).

[0032] For each small region, calculate the optimal threshold using the local Otsu threshold method and binarize the region.

[0033] Eliminate the sudden change at the boundary between blocks through smoothing processing (such as Gaussian filtering) to obtain an overall uniform binary image; The specific method to extract the main contour of the slice using the Canny edge detection algorithm is as follows; Process the binarized image using the Canny edge detection algorithm.

[0034] Set appropriate low threshold (such as 50) and high threshold (such as 150) to ensure that the details of the slice boundary can be detected.

[0035] Output a binary edge map to mark the contour of the slice and possible defect edges; The method to repair the noise and holes in the segmented image through morphological operations (such as opening and closing operations) is as follows; Apply the opening operation (erosion first and then dilation) to the binary image to remove small noise; Apply the closing operation (dilation first and then erosion) to fill the small holes in the slice area; Extract the largest connected component as the final slice area through connected component analysis and eliminate non-target areas to repair the noise and holes generated during the segmentation process and enhance the continuity of the slice contour; S230 uses the edge map processed by S220 and adopts the Hough line transform algorithm to detect and extract the thin and continuous linear structures in the image; Judge whether it is a crack by preset thresholds of line segment length and width; If it is judged that there are cracks in the image, output the starting and ending coordinates of the cracks, the crack length, and the number of cracks; Use the binarization result of the image and combine with the circular Hough transform algorithm to identify circular structures; Judge whether it is a bubble by thresholds of region area and roundness (such as area > 20 pixels, roundness > 0.8); If it is determined that there are bubbles in the image, then output the center coordinates, area, diameter, and the number of bubbles of each bubble; Apply local texture analysis (such as LBP features) to extract texture irregular regions; Use morphological features (such as irregularity and distribution density) to determine whether the texture irregular region is a wrinkle; If it is determined that there are wrinkles in the image, then output the contour, polygon coordinate points, and the covered area of the wrinkle region; Compare the edge detection results with the theoretical slice boundary to calculate the difference region of the boundary.

[0036] Statistically analyze the defect area and its distribution position, and output the number, position, and area ratio of the defect regions.

[0037] Statistically analyze the number of each type of defect and the defect distribution position, such as the number of cracks, bubbles, wrinkles, etc.; Calculate the total defect area and quantify it as the ratio (coverage rate) to the total slice area; According to the preset quality scoring criteria (such as the weights of indicators such as defect types, quantities, coverage rates, etc.), calculate the comprehensive quality score of the slice; S300 performs quantitative analysis on the detected defect data, such as statistically analyzing the number and length of cracks, the area of wrinkles, and the number and size of bubbles, etc., as the input data for subsequent analysis; S310 obtains the detected defect data through the interaction of various types of defects detected in S200, which includes defect types (such as cracks, bubbles, wrinkles, etc.), defect quantities, positions, area coverage rates, distribution densities, etc.; For each defect, extract its coordinate information in the image (for example, the starting and ending coordinates of a crack, the center position of a bubble).

[0038] Establish defect position data to facilitate subsequent spatial analysis of defect distribution; S320 uses the edge detection results to calculate the length (for example, by measuring the distance from the starting point to the ending point of the crack) and width of each crack region.

[0039] Record the size information of each crack to provide a basis for subsequent comprehensive scoring; For the bubble regions detected by circular Hough transform, calculate the area and diameter of each bubble.

[0040] Convert the pixel area of the bubble into the actual physical size.

[0041] For the detected wrinkle regions (extracted from local texture irregularities), calculate their covered areas.

[0042] Record the contour and area of the wrinkled area to provide data support for quality scoring; Calculate the area and location distribution of the defect area according to the comparison between the theoretical slice boundary and the actual edge; S330 Accumulate the area of each type of defect area to obtain the total defect area.

[0043] At the same time, calculate the total area of the entire slice image; Divide the total defect area by the total area of the slice to obtain the defect coverage rate, which reflects the degree of influence of the defect on the overall slice quality.

[0044] Analyze the spatial distribution of defects in the slice to determine whether the defects are concentrated at the edge, center, or evenly distributed.

[0045] Record the defect-dense area and the sparse area to provide a spatial distribution reference for subsequent parameter optimization; S340 Set the weights of various types of defects according to the importance of the defect types; Convert the quantity, area, coverage rate, etc. of each type of defect into scoring indicators.

[0046] For example, the scoring of cracks can be based on the quantity and total length, and the scoring of air bubbles is based on the quantity and average area; According to the preset weights, perform a weighted average of each score to obtain the comprehensive quality score of the slice.

[0047] The comprehensive score can be used to determine whether the slice meets the quality standard (for example, the total score is between 0 and 100 points, and a score lower than a certain threshold indicates that the quality does not meet the standard); Overlay the defect detection results on the slice image, such as the contours and labels of defect areas such as cracks and air bubbles, and use different colors to represent different defect types (such as red for cracks and blue for air bubbles); Generate a defect detection report, which includes the sample number and slice number, defect type, quantity, area and coverage rate, and the comprehensive quality score of the slice.

[0048] S400 Perform a correlation analysis on the defect data after quantitative analysis and the parameter information during the slicing process (the parameter information includes: key parameters such as sample type, slicing temperature, humidity, slicing thickness, slicing speed, number of blade uses, etc.), quantify the specific impact of each parameter on the occurrence of defects, and then calculate the influence weight of each parameter on the defects to ensure the detailed quantification of the detected defects and provide reliable data input for subsequent parameter correlation and model construction; S410 Interactively obtain the quantified defect data in S300 (including various defect types, quantities, areas, coverage rates, distribution locations, etc.) and match it with the slice parameter data collected in S100 (such as sample types, slice temperatures, humidity, thickness, slice speed, number of blade uses, etc.) through a unique identifier (such as sample number and slice serial number) to form a complete record; Organize the matched data into a table format, with each row representing a slice operation, and its columns including information such as sample type, temperature, humidity, thickness, speed, number of blade uses, quantities of various defects, area, coverage rate, etc.; S420 Match the defect data generated for each slice with the corresponding slice parameters one by one, and normalize the slice parameters and defect indicators with different dimensions to the same scale (such as using Min - Max normalization or Z - score standardization) to form an analysis dataset so that subsequent correlation analysis and regression analysis are not affected by different dimensions; S430 Calculate the linear correlation between the defect data and the slice parameters using the Pearson correlation coefficient; That is: calculate the linear correlation between each slice parameter (independent variable) and each defect indicator (dependent variable) using the Pearson correlation coefficient.

[0049] For example, calculate the correlation coefficients between temperature and the number of cracks, and between thickness and the number of bubbles; Based on the correlation coefficient values, judge the relationship (positive or negative) between each parameter and the defect, and screen out the key parameters that have a significant impact on the defect; Record the values of the correlation coefficients and the significance levels to provide a preliminary basis for subsequent modeling; S440 Use PCA dimensionality reduction to analyze the comprehensive impact of multiple parameters on the defect and extract the parameters with the greatest impact; That is: input the normalized parameter data into the PCA algorithm to calculate each principal component and the proportion of variance it explains.

[0050] Select the first few principal components with a higher total variance explained, and analyze the loadings of each principal component to determine which original parameters have the greatest impact on the defect, so as to obtain the combined data of the main influencing factors and provide concise features for regression analysis and subsequent modeling; S450 Use regression analysis methods (such as linear regression, logistic regression) to quantify the specific impact of the parameters on the defect; Calculate the influence weights of each parameter on the defect according to the correlation results to obtain the specific association between each defect type and the slice parameters.

[0051] That is: according to the defect type (such as the number of cracks or defect coverage rate), establish a prediction relationship using a multiple linear regression or logistic regression model.

[0052] The input variables are the parameters of each slice (or the main components after PCA), and the output variable is the defect index; Use the training dataset to fit the model to obtain the regression coefficients of each parameter; Normalize the absolute values of the regression coefficients and calculate the relative influence weights of each parameter on the occurrence of defects.

[0053] Analyze the results of the regression model to confirm which parameters have a significant impact on different defects (such as cracks, wrinkles, and bubbles), and record the corresponding quantitative relationships.

[0054] S460 summarizes the results of S430 and S450, outputs the specific association data and influence weights between each defect type and the slice parameters, and outputs them in the form of a structured table or database record. These association data will serve as the input basis for subsequent S500 (prediction model construction) and S600 (parameter optimization), guiding the construction of the prediction model and the selection of optimized parameters.

[0055] S500 constructs a training dataset using the parameter association data; it can select a regression model, a random forest, or a deep learning model to construct a prediction model for the relationship between the sample type, slice parameters, and the occurrence of defects, enabling the model to accept new slice parameter inputs and predict the corresponding defect occurrence probability and quantity. The specific steps are as follows: S510 organizes the association data obtained in S400 into a structured dataset, and each record contains the input variable and the target variable: Input variables: sample type, slice temperature, humidity, slice thickness, slice speed, number of blade uses, etc.

[0056] Target variable: corresponding defect data, such as the number of cracks, defect area, defect coverage rate, and comprehensive defect score; Normalize the data in the organized dataset and divide it into a training set and a test set (for example, the training set accounts for 80% and the test set accounts for 20%) for subsequent model training and verification; S520, based on the parameter association data obtained in S400, screens out the key parameters that are most closely related to the occurrence of defects, such as temperature, humidity, and slice thickness, and performs one-hot encoding on categorical variables (such as sample type) to ensure that the model can handle them correctly; Based on experience and data analysis, construct interaction features or derived features. For example, use the ratio of temperature to humidity, the product of temperature and slice thickness, etc. as new features to help the model capture non-linear relationships; Those skilled in the art can select one or more models for trial according to the data characteristics and target requirements. Preferably, in this embodiment, a random forest model is used to construct a prediction model for the relationship between sample types, slicing parameters, and defect occurrence, enabling it to automatically capture the non-linear relationships between features through the tree model and output feature importance, making it easier to understand the impact of each parameter on defects; Train the random forest model using the training set data: Set key parameters, such as the number of trees, maximum depth, minimum sample split number, etc.; Use cross-validation (e.g., K-fold cross-validation) to evaluate the stability and generalization ability of the model.

[0057] During the training process, the random forest will reduce the risk of overfitting through the voting mechanism of multiple decision trees and automatically screen out the features that have the greatest impact on the target variable; It can also use grid search or random search to tune the hyperparameters of the random forest to achieve the best prediction performance. The parameters to be adjusted include the number of trees, maximum depth, feature subset size, etc.

[0058] Input the test set data into the trained random forest model, predict the defect occurrence of each sample, and analyze the feature importance output by the random forest to verify the influence weights of the parameters calculated in S400 on defects, ensuring the interpretability and reliability of the model, enabling the trained random forest model to receive new slicing parameter inputs and predict the corresponding defect occurrence probability, defect quantity, or comprehensive defect score; Finally, encapsulate the model as a service interface and integrate it into the entire system to achieve real-time prediction, providing a basic prediction result for subsequent parameter optimization (S600) and dynamic adjustment.

[0059] Take the parameter correlation data and the output of the prediction model as the objective function, and combine it with an optimization algorithm to search for the optimal slicing parameter combination within the given parameter space as the optimal slicing parameter.

[0060] The specific steps are as follows: Using the prediction model constructed in S500, define the objective function as the predicted defect rate or comprehensive quality loss under the input parameter combination. For example, take the defect rate output by the prediction model as the objective function value, that is, the goal is to minimize this defect rate.

[0061] Based on historical data and the results of S400 / S500, determine the reasonable range of the slicing parameters to be optimized, and set the upper and lower limits of each parameter as the search interval to ensure that all candidate solutions are within the reasonable range.

[0062] S620 generates a number of initial parameter combinations (e.g., 5 - 10 groups) in the set parameter space using random sampling or Latin hypercube sampling.

[0063] For each initial parameter combination, call the prediction model in S500 to calculate its corresponding objective function value (i.e., predicted defect rate); Adopt Gaussian process regression as a surrogate model to fit the distribution of the objective function according to the initial sampling data, and at the same time obtain the uncertainty information of each candidate solution to provide a basis for subsequent sampling; S630 uses an acquisition function (such as the expected improvement function) to balance finding potentially better solutions in unknown regions and exploiting parameters near the current known optimal solution, and selects the next optimal experimental point; That is, use the acquisition function to select the next experimental point, i.e., a new parameter combination, in the region where the uncertainty predicted by the surrogate model is high and the objective function value is low; For the newly sampled parameter combination, calculate its defect rate, i.e., the objective function value, through the prediction model of S500; Add the newly sampled point and its objective function value to the training set, and update the Gaussian process surrogate model to improve the overall prediction accuracy; Repeat the above steps and continue to iterate until the set termination condition is reached (e.g., reaching the maximum number of iterations or the change in the global optimal value is less than a predetermined threshold); After obtaining the preliminary optimal parameter combination in the global Bayesian optimization stage, in order to improve the accuracy, local optimization methods (such as simulated annealing or gradient descent) can be used to refine and adjust the parameters. It takes the global optimal parameters as the initial point, randomly samples in a smaller neighborhood, calculates the objective function value, and gradually adjusts the parameters until the objective function value tends to be stable. Generally, in the global Bayesian optimization stage, the preliminary optimal parameters are obtained, which are sufficient to solve the technical problems of this application. However, if a better and more accurate slicing effect is required, the method in step S640 can also be used to further optimize its parameters.

[0064] S700 inputs the optimal parameters into the prediction model to simulate and predict the occurrence of defects under these parameters to verify whether the results meet the preset quality requirements. For example, when the defect rate is lower than 5%, the slicing quality meets the requirements; If the predicted defect rate meets the requirements, accept this parameter combination as an optimization suggestion, generate an optimization report, and output the corresponding predicted defect rate while outputting the optimal parameter combination, and record the convergence curve and parameter change trend of the optimization process; If the predicted defect rate is still high, resulting in an unsatisfactory prediction result, adjust the search range of the Bayesian optimization, or update the surrogate model and re - iterate the optimization until a parameter combination that meets the requirements is obtained.

[0065] Embodiment 2. Based on the same inventive concept as the intelligent monitoring and adjustment method and system for frozen sections in the foregoing embodiment, as Figure 2 shown, the present application provides an intelligent monitoring and adjustment system for frozen sections. The system in the embodiment of the present application and the method embodiment are based on the same inventive concept. Among them, the system includes: A data acquisition and monitoring module 11, which is used to collect frozen section images and sectioning process parameters (such as sample type, temperature, humidity, thickness, speed, number of blade uses, etc.); An image processing and defect detection module 12, which is used to preprocess, segment, and perform edge detection on the collected images, identify and quantify defects such as cracks, wrinkles, bubbles, and knife marks, and provide quantitative data for correlation analysis; A defect data and parameter correlation analysis module 13, which is used to match the quantified defect data with the sectioning parameters, quantify the specific influence of each parameter on the occurrence of defects, and then calculate the influence weight of each parameter on the defects, and output parameter correlation data; A prediction model construction module 14, which constructs a prediction model of the relationship between sample type, sectioning parameters, and defect occurrence based on the correlation data using machine learning methods, so that the model can accept new sectioning parameter inputs and predict the corresponding defect occurrence probability and quantity; A parameter optimization and dynamic adjustment module 15; the parameter optimization and dynamic adjustment module 15 uses the parameter correlation data and the output of the prediction model as the objective function, and combines an optimization algorithm to search for the optimal sectioning parameter combination within a given parameter space as the optimal sectioning parameters; An optimized parameter verification and feedback module 16; the optimized parameter verification and feedback module 16 is used to input the optimal parameters into the prediction model for simulation verification, and detect and verify the optimization effect through actual sectioning experiments; if the predicted defect rate is lower than the set threshold, accept the parameter combination; otherwise, adjust the search range or update the model and then optimize again.

[0066] It should be noted that the above sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above describes specific embodiments of this specification. The processes depicted in the drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0067] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0068] This specification and the accompanying drawings are merely illustrative of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A method for intelligent monitoring and adjustment of frozen sections, characterized in that: The method comprises: Interactively obtain frozen section image data; Adaptive local threshold method is used to binarize the image, and edge detection algorithm is used to identify the defect type in the slice, and the detected defect data is quantitatively analyzed as input data for subsequent analysis; The defect data after quantitative analysis is correlated with the parameter information in the slicing process, and the specific influence of each parameter on the defect is quantified, and then the influence weight of each parameter on the defect is calculated; Use parameter association data to build a training data set to construct a prediction model of the relationship between sample type, slicing parameters and defect occurrence, so that the model can accept new slicing parameter inputs and predict the corresponding defect probability and quantity; The parameter association data and the output of the prediction model are used as the objective function, and combined with the optimization algorithm, the optimal slicing parameter combination is searched in the given parameter space as the optimal slicing parameter.

2. A frozen section intelligent monitoring and adjustment method according to claim 1, characterized in that: The method further comprises: Input the optimal parameters into the prediction model, simulate and predict the occurrence of defects under the parameters to verify whether the results meet the preset quality requirements; If the predicted defect rate meets the requirements, the parameter combination is accepted as the optimization suggestion and an optimization report is generated; If the predicted defect rate is still high, adjust the optimization algorithm search range or update the prediction model, and then re-optimize the parameters.

3. The method for intelligent monitoring and adjustment of frozen sections according to claim 1, characterized in that: The parameter information includes: sample type, slice temperature, humidity, slice thickness, slice speed, and number of times the blade is used.

4. The method for intelligent monitoring and adjustment of frozen sections according to claim 1, characterized in that: The method of using the adaptive local threshold method to binarize the image and combining it with the edge detection algorithm to identify the defect type in the slice is as follows; removing the background portion of the slice image and extracting image information of the slice area of ​​the slice image; Binarize the grayscale image using the adaptive local threshold method; Use the Canny edge detection algorithm to extract the main contours of the slice; Repair the noise and holes in the segmented image through morphological operations; According to the connected domain analysis, the largest connected domain is extracted as the slice area, and the non-target area is eliminated; Use Hough line transform algorithm to detect thin and continuous linear structures in the image; Whether it is a crack is determined by the threshold of line segment length and width; If it is determined that there are cracks in the image, the coordinates of the starting and ending points of the cracks, the length of the cracks, and the number of cracks are output; Use the binarization result of the image and the circular Hough transform algorithm to identify the circular structure; Whether it is a bubble is determined by the threshold of area and roundness; If it is determined that there are bubbles in the image, the center coordinates, area, diameter, and number of bubbles of each bubble are output; Apply local texture analysis to extract texture irregular areas; Use morphological features to determine whether the irregular texture area is a wrinkle; If it is determined that there are wrinkles in the image, the contour of the wrinkle area, the polygonal coordinate points, and the wrinkle coverage area are output; By comparing the edge detection results with the theoretical slice boundaries, the difference area of ​​the boundaries is calculated; Count the defect areas and their distribution locations, and output the number, location, and area ratio of the defect areas.

5. The method for intelligent monitoring and adjustment of frozen sections according to claim 1, characterized in that: The method for quantitative analysis of detected defects is as follows: Count the number of defects of each type and the distribution location of the defects; Calculate the total defect area and quantify the ratio with the total area of ​​the slice; The comprehensive quality score of the slice is calculated according to the preset quality scoring criteria.

6. A frozen section intelligent monitoring and adjustment method according to claim 5, characterized in that: The method also includes: Overlay defect detection results on slice images and use different colors to represent different defect types; Generate a defect detection report, which includes the sample number and slice number, defect type, quantity, area and coverage, and a comprehensive score of slice quality.

7. The method for intelligent monitoring and adjustment of frozen sections according to claim 1, characterized in that: The defect data after quantitative analysis is correlated with the parameter information in the slicing process, and the specific influence of each parameter on the defect is quantified, and then the influence weight of each parameter on the defect is calculated to obtain the specific correlation between each defect type and the slicing parameter to obtain the correlation data as follows: Interactively obtain the detected defect data, including defect type, defect quantity, location, area coverage, distribution density, etc.; Interactively obtain slice parameter information, including slice temperature, humidity, slice thickness, slice speed, and blade status; Match the defect data generated by each slicing with the corresponding slicing parameters one by one to form an analysis data set; The linear correlation between defect data and sectioning parameters was calculated using the Pearson correlation coefficient; Through PCA dimensionality reduction analysis, the comprehensive impact of multiple parameters on defects is analyzed and the parameters with the greatest impact are extracted; Use regression analysis to quantify the specific impact of parameters on defects; The influence weight of each parameter on the defect is calculated based on the correlation results to obtain the specific association between each defect type and the slicing parameters.

8. The method for intelligent monitoring and adjustment of frozen sections according to claim 1, characterized in that: The parameter association data and the output of the prediction model are used as the objective function, and combined with the optimization algorithm, the optimal slicing parameter combination is searched in the given parameter space. The method for the optimal slicing parameter is as follows: Define the objective function as the defect rate or comprehensive quality loss index output by the prediction model; The Bayesian optimization algorithm is used to search for the optimal parameter combination within the preset parameter range as the preliminary optimal solution; This phase uses historical data as a basis to estimate the target function distribution through a surrogate model; After obtaining the preliminary optimal solution through global optimization, local optimization is used to further fine-tune the parameters to obtain a more refined optimal result as the optimal slicing parameters.

9. A frozen section intelligent monitoring and adjustment system, the system is used to execute the method according to any one of claims 1 to 8, characterized in that: The system comprises: A data acquisition and monitoring module, which is used to acquire frozen section images and sectioning process parameters; An image processing and defect detection module, which is used to preprocess, segment and detect edges of the collected images, identify and quantify defects such as cracks, wrinkles, bubbles, and knife marks, and provide quantitative data for correlation analysis; A defect data and parameter association analysis module, which is used to match the quantized defect data with the slice parameters, and quantify the specific impact of each parameter on the occurrence of defects, and then calculate the impact weight of each parameter on the defect, and output parameter association data; A prediction model building module, which uses a machine learning method to build a prediction model of the relationship between sample type, slicing parameters and defect occurrence based on the associated data, so that the model can accept new slicing parameter inputs and predict the corresponding defect occurrence probability and quantity; Parameter optimization and dynamic adjustment module; the parameter optimization and dynamic adjustment module uses parameter association data and the output of the prediction model as the objective function, and combines the optimization algorithm to search for the optimal slicing parameter combination in a given parameter space as the optimal slicing parameter; Optimization parameter verification and feedback module; the optimization parameter verification and feedback module is used to input the optimal parameters into the prediction model for simulation verification, and verify the optimization effect through actual slicing experiment detection; if the predicted defect rate is lower than the set threshold, the parameter combination is accepted; otherwise, the search range is adjusted or the model is updated and re-optimized.

Images