Intelligent operating room environment status monitoring method and system based on IoT sensing

Abstract

The present application provides a method and system for monitoring the environmental status of an intelligent operating room based on IoT sensing, which relates to the field of data processing. The method includes: performing data encoding on target-related monitoring data of a target operating room to obtain a coding representation corresponding to the target-related monitoring data; performing high-level feature extraction on the coding representation to obtain high-level features; determining x operating modes corresponding to the target-related monitoring data, and performing environmental state prediction on the high-level features according to each operating mode to obtain the environmental state prediction confidence corresponding to each operating mode; performing contribution extraction on the coding representation to obtain the contribution corresponding to each operating mode; and determining the environmental state monitoring result of the target operating room according to the environmental state prediction confidence corresponding to each operating mode and the contribution corresponding to each operating mode. The present application can increase the accuracy of environmental state monitoring in operating rooms.

Description

Technical Field

[0001] The present application relates to the field of data processing, and more specifically, to a method and system for monitoring the environmental status of an intelligent operating room based on IoT sensing. Background Art

[0002] In the modern medical system, the operating room is a special place with extremely high environmental requirements. The success of the operation depends not only on the professional skills of medical staff and the advanced level of medical equipment, but also on the environmental status of the operating room. Environmental factors such as appropriate temperature, humidity, air quality, and the normal operation of various medical equipment are directly related to the safety of patients during surgery, the postoperative recovery effect, and the work efficiency of medical staff. Traditional operating room environmental monitoring methods are often designed based on a single operating mode. This single operating mode monitoring method only considers the environmental requirements under a preset, relatively fixed operating room usage scenario or operation process. However, in the actual use of the operating room, there are many different operating modes, such as the operation process of different types of surgery (such as cardiac surgery, orthopedic surgery, etc.), the preoperative preparation stage, and the postoperative care stage. Different operating modes have different requirements and focuses on environmental status. Traditional operating room environmental monitoring methods have obvious deficiencies in accuracy and precision. Existing operating room environmental monitoring technologies show obvious inadaptability when facing the complex and diverse operating modes of the operating room and the differences in the importance of different data in different modes. Summary of the invention

[0003] The object of the present invention is to provide a method and system for monitoring the environmental status of an intelligent operating room based on Internet of Things sensors. The present application is implemented as follows: In the first aspect, an embodiment of the present application provides a method for monitoring the environmental status of an intelligent operating room based on Internet of Things sensors, and the method includes: performing data encoding on target-related monitoring data of a target operating room to obtain a coded representation corresponding to the target-related monitoring data; performing high-level feature extraction on the coded representation to obtain high-level features corresponding to the coded representation; determining x operating modes corresponding to the target-related monitoring data, and performing environmental state prediction on the high-level features based on each operating mode to obtain the environmental state prediction confidence corresponding to each operating mode; x≥1; performing contribution extraction on the coded representation to obtain the contribution corresponding to each operating mode; determining the environmental state monitoring result of the target operating room according to the environmental state prediction confidence corresponding to each operating mode and the contribution corresponding to each operating mode, and the environmental state monitoring result includes the confidence of each target warning state.

[0004] In a second aspect, an embodiment of the present application provides a computer system, comprising: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method described above is implemented.

[0005] Beneficial effects of the present application: The present application provides an intelligent operating room environment state monitoring method and system based on IoT sensing, which performs data encoding on target-related monitoring data, obtains a coding representation, and determines x operation modes, so that in the process of determining high-level features based on the coding representation, the environmental state prediction confidence corresponding to each operation mode is determined according to the high-level features, and the contribution corresponding to each operation mode is determined according to the coding representation, so as to integrate the environmental state prediction confidence corresponding to each operation mode, and determine the final environmental state monitoring result. Based on this, the environmental state monitoring capabilities of multiple operation modes can be integrated, and the contribution of target-related monitoring data for each operation mode can be adaptively determined to determine the contribution distribution that matches the current target-related monitoring data in each operation mode, so that the environmental state monitoring result of the operation mode corresponding to the target-related monitoring data can be determined according to the distribution of the determined contribution, so that the environmental state monitoring result of the operation mode corresponding to the target-related monitoring data can be obtained based on the environmental state prediction confidence of multiple operation modes, and the environmental state monitoring result with the highest accuracy can be obtained, replacing the environmental state monitoring method of one operation mode, and the environmental state monitoring of the target operating room can be accurately completed in each operation mode, which helps to increase the accuracy of the environmental state monitoring of the target operating room. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a flow chart of a method for monitoring the environmental status of an intelligent operating room based on IoT sensing provided in an embodiment of the present application;

[0007] Figure 2 It is a schematic diagram of the composition of a computer system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0008] The execution subject of the intelligent operating room environment status monitoring method based on IoT sensing in the embodiment of the present application is a computer system, including but not limited to a server, a personal computer, a laptop, a tablet computer, a smart phone, etc. Figure 1 As shown, the method includes: Step S100: performing data encoding on target-related monitoring data of a target operating room to obtain a coding representation corresponding to the target-related monitoring data.

[0009] In the embodiment of the present application, the target-related monitoring data of the target operating room is a comprehensive data set, which covers various information related to the environmental status of the operating room. For example, it includes operating room temperature data collected by the temperature sensor, humidity data obtained by the humidity sensor, air quality related indicators (such as oxygen concentration, carbon dioxide concentration, harmful gas concentration, etc.) provided by the air quality sensor, illumination data of the illumination intensity sensor, and data related to the operating status of the surgical equipment (such as equipment power, operating time, equipment temperature, etc.), which reflect the environmental status of the operating room from different angles. Data encoding is the process of converting these original target-related monitoring data into a representation in a specific format. When encoding data, a computer system can use a variety of technical means. One technical means is to encode based on the type and value range of the data. For example, for temperature data, if its value range is between 0℃-50℃, the computer system can divide this range into several intervals. Assume that it is divided into 5 intervals, namely 0℃-10℃, 10℃-20℃, 20℃-30℃, 30℃-40℃, and 40℃-50℃. When the collected temperature value is 25℃, it can be encoded as the identifier of the corresponding interval, such as 3 (representing the interval of 20℃-30℃). For humidity data, if the value range is 0%-100%, it can also be encoded in a similar interval division method. Another encoding method can be based on the binary representation of data or a specific mapping relationship. For example, the collected decimal data is converted into binary data, and then the binary data is recombined or mapped according to certain rules to obtain a coded representation. Assuming that the collected humidity data is 60%, the corresponding decimal number converted to binary is 111100. According to the pre-set mapping rules, every two binary digits are mapped to a new coded value, such as 00-0, 01-1, 10-2, 11-3, then 111100 can be mapped to 33 (first map 11 to 3, 11 to 3, 10 to 2, 00 to 0, and the combination is 33). Through such a data encoding process, the computer system converts the originally complex and diverse target-related monitoring data in different formats into a unified coded representation that is easy to process later. This coding representation can be stored, transmitted and calculated more efficiently within the computer system, laying the foundation for subsequent high-level feature extraction, environmental state prediction and other steps. It can normalize the data without losing the key information of the original data, so that the entire intelligent operating room environmental state monitoring system can run more accurately and quickly.

[0010] Step S200: extract high-level features from the encoded representation to obtain high-level features corresponding to the encoded representation.

[0011] In the embodiments of the present application, advanced feature extraction is a process of mining more representative and abstract features from the encoded representation. The computer system can adopt various technical means to achieve advanced feature extraction. One technical means is principal component analysis (PCA). For the encoded representation, it is assumed that the encoded representation contains data in multiple dimensions (for example, encoding different types of sensor data and combining them into a high-dimensional vector). Let the encoded representation be the vector X=(x_1,x_2,…,x_n), where n is the dimension of the vector. The calculation process of PCA is first to calculate the covariance matrix , where m is the number of samples, is the sample mean vector. Then, perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues and the corresponding eigenvectors . Sort the eigenvalues from largest to smallest, select the eigenvectors corresponding to the top k eigenvalues (k < n), project the original encoded representation vector X into the subspace spanned by these k eigenvectors, and obtain the new vector , , where Y is a representation of the extracted advanced features.

[0012] Another technical means is the deep neural network (DNN). The computer system can construct a multi-layer neural network with the encoded representation as the input. Assume that the neural network has an input layer, several hidden layers, and an output layer. The number of neurons in the input layer corresponds to the dimension of the encoded representation. In the forward propagation process of the neural network, the neurons in each layer pass the information to the next layer through weighted summation and the action of an activation function (such as the ReLU function f(x)=max(0,x)). As the number of network layers increases, the network can automatically learn the complex patterns in the encoded representation, and the output of the hidden layer can be regarded as the extracted advanced features. For example, in a simple three-layer neural network, let the input layer be X (encoded representation), the output of the neurons in the first hidden layer be h1=f(W1X + b1), where W1 is the weight matrix of the first layer and b1 is the bias vector; the output of the neurons in the second hidden layer be h2=f(W2h1 + b2), and finally h2 is the extracted advanced feature.

[0013] These advanced features have more advanced semantic information compared to the original encoded representation. Taking the operating room environment as an example, the original encoded representation may be just a simple numerical mapping, but the advanced features may reflect more abstract and comprehensive information such as "comprehensive environmental state suitable for surgery" and "coordination state of equipment and environment". These advanced features can better provide an effective basis for subsequent operations such as environmental state prediction, and help the computer system more accurately judge the environmental state of the operating room, such as whether the current environment meets the best requirements for surgery and whether there are potential risk factors.

[0014] Step S300: Determine x operating modes corresponding to the target-related monitoring data, perform environmental state prediction on high-level features based on each operating mode, and obtain the environmental state prediction confidence corresponding to each operating mode; x≥1.

[0015] In the embodiment of the present application, the operating mode is a classification abstraction of different working or environmental status scenarios in the operating room. For example, the operating mode can be divided according to the different stages of the operation, such as the preoperative preparation stage, the surgical stage, the postoperative cleaning stage, etc. This is an operating mode division based on the surgical process; it can also be divided according to the different states of the operating room environment, such as normal state, emergency state (such as sudden equipment failure, sharp decline in air quality and other special circumstances), etc. After the computer system determines x operating modes, it predicts the environmental state for each operating mode. One technical means is based on a probability model, such as a naive Bayes classifier. Assume that the high-level feature vector is F=(f1,f2,…,fn), for a certain operating mode Mi (i=1,2,…,x), it is necessary to predict the environmental state E (for example, the environmental state can be divided into good, general, poor, etc.). The naive Bayes classifier is based on the Bayesian theorem. . Among them, P(E) is the prior probability of the environmental state E, which can be obtained through historical data statistics. For example, in the previous operating room environment monitoring data, the probability of the environmental state being good; P(F) is the probability of the high-level feature vector F appearing; P(F|E) is the probability of the high-level feature vector F appearing under the condition that the environmental state is E. The computer system calculates P(E|F) to obtain the probability that the environmental state is E under the operating mode Mi. This probability is the environmental state prediction confidence. For example, in the preoperative preparation stage (operating mode M1), if the temperature, humidity and other data in the high-level features are displayed within the normal range, and the equipment is in a stable operating state, the probability P(E1|F) calculated by the naive Bayes classifier that the environmental state is good (E1) is high, such as 0.8. This 0.8 is the environmental state prediction confidence that the environmental state is good under the operating mode M1.

[0016] Another technical means is the decision tree model. The decision tree predicts the environmental state by making a series of conditional judgments on high-level features. For each operating mode, the structure of the decision tree may be different. For example, in the decision tree during the surgical stage (operating mode M2), it is first determined whether the operating temperature of the equipment is too high. If so, it is further determined whether the air quality is good. If the operating temperature of the equipment is normal, it is then determined whether the humidity is within the appropriate range. Based on the results of these judgments, the environmental state is finally determined. The judgment basis for each branch node is the best split point mined from historical data. The computer system finally determines the probability that the environmental state belongs to a certain category based on the traversal path of the high-level features in the decision tree. This probability is the confidence level of the environmental state prediction.

[0017] For another example, for an emergency state (operating mode M3), if the advanced features show that the power of the equipment suddenly drops and the air quality is abnormal, the decision tree may quickly determine that the environmental state is poor (E2), and the corresponding environmental state prediction confidence may be obtained as 0.9 through the calculation mechanism inside the decision tree (such as calculating the weights of the impact of different features on the environmental state in an emergency state), indicating that in this operating mode, the environmental state is likely to be poor.

[0018] By performing such environmental state prediction for each operation mode, the computer system can obtain the environmental state prediction confidence level corresponding to each operation mode. These confidence levels can reflect the probability of the operating room environment being in different states under different operation modes.

[0019] Step S400: extract the contribution of the coded representation to obtain the contribution corresponding to each operating mode.

[0020] In the embodiment of the present application, contribution extraction is the process of determining the contribution of each part or feature in the coded representation to each operating mode (essentially a weight value). This helps to understand which information plays a more critical role in judging the environmental state in different operating modes.

[0021] Computer systems can use a variety of technical means to extract contribution. One method is an algorithm based on information gain. Information gain is an indicator that measures the contribution of a feature to a classification task. Assume that there are x operating modes, namely M1, M2, …, M x , for a feature f in the encoding representation (this feature may be the part of the encoding representation corresponding to a sensor data or data combination), calculate its information gain relative to the environmental state classification in each operating mode.

[0022] Taking the operation mode M1 as an example, suppose the environmental state classification has y categories, namely C1, C2, …, C y First, calculate the entropy H(C) without considering feature f, the formula is ,in The environmental state is The probability of a category can be obtained through historical data or existing monitoring data statistics.

[0023] Then, the data set is divided according to the value of feature f to obtain n subsets , calculate the entropy H(C|f) after partitioning, the formula is ,in Is a subset is the number of samples, |S| is the total number of samples, Is a subset The entropy of the environment state.

[0024] Then the feature f is in the running mode Information gain under By calculating the information gain of each feature in the encoding representation in each operating mode, the contribution of each feature to each operating mode can be obtained. For example, if the information gain IG(C,f1) of a feature f1 in operating mode M1 is large, it means that this feature contributes more to the classification of the environmental state in operating mode M1.

[0025] Another technical approach is to analyze the coefficients of the linear regression model. Assume that the encoding is represented by X=(x1,x2,…,x n ) as the independent variable, and a quantitative indicator Y of environmental status (such as comprehensive environmental status score) as the dependent variable to construct a linear regression model ,in is the intercept, is the regression coefficient, is the error term.

[0026] For each operation mode, the computer system can establish such a linear regression model. The absolute value of the corresponding variable The degree of influence of (a part of the coded representation) on the dependent variable Y. In the running mode Next, if The absolute value of is large, indicating that This feature is in run mode The contribution to the quantitative index Y of environmental status is relatively large.

[0027] For example, in an operating room setting, if Represents the operating power of the equipment. This means that in run mode The operating power of the equipment has a greater impact on the comprehensive environmental status score, that is, its contribution is higher. The extraction of this contribution can help the computer system better understand the importance of each part in the coding representation under different operating modes, so that it can use this information more targetedly in subsequent operations such as comprehensive environmental status evaluation. For example, based on the size of the contribution, when resources are limited, priority can be given to monitoring data corresponding to features with high contribution, or the environmental status assessment model under different operating modes can be optimized and adjusted based on the contribution.

[0028] Step S500: Determine the environmental state monitoring result of the target operating room according to the environmental state prediction confidence corresponding to each operating mode and the contribution corresponding to each operating mode, wherein the environmental state monitoring result includes the confidence of each target warning state.

[0029] In an embodiment of the present application, the computer system will comprehensively consider these two factors when determining the environmental status monitoring results of the target operating room. In essence, the contribution can be regarded as a weighted adjustment of the confidence level of the environmental status prediction. Assume that there are three operating modes M1, M2, and M3, and the corresponding environmental status prediction confidence levels are P1, P2, and P3, respectively, and their corresponding contributions are W1, W2, and W3, respectively. The confidence level of the target warning state determined by the computer system can be calculated by weighted summation. For a certain target warning state E, its final confidence level P(E) can be calculated according to the following formula: .

[0030] Computer systems can use matrix operations to perform this calculation efficiently, especially when there are many operating modes. The confidence of the environmental state prediction is formed into a column vector , the contribution forms a row vector , then the confidence level P(E) of the target warning state can be obtained by matrix multiplication Get (if normalization is required, corresponding processing can be performed).

[0031] As an implementation manner, step S100, data encoding the target-related monitoring data of the target operating room to obtain the coded representation corresponding to the target-related monitoring data, may include:

[0032] Step S110: segmenting the target-related monitoring data to obtain data segments of the target-related monitoring data;

[0033] Step S120: performing array encoding on the data segments to obtain encoded representations of the data segments.

[0034] In step S110, data fragmentation is a process in which a computer system decomposes target-related monitoring data to form smaller data units, i.e., data fragments. Its purpose is to refine complex monitoring data so that subsequent encoding operations can be performed more effectively. In operating room environment monitoring, target-related monitoring data is multi-source and complex, for example, it may include temperature values ​​collected by temperature sensors at regular intervals, humidity values ​​of humidity sensors, operating parameters of different surgical equipment, etc. The types, ranges, and acquisition frequencies of these data are different. Through data fragmentation, these different types of data can be split according to certain rules, so that subsequent processing can be performed independently for each data fragment, thereby improving the accuracy and efficiency of processing. The computer system can perform data fragmentation based on the source of the data or the characteristics of the data. For example, for environmental monitoring data in an operating room, the computer system can fragment the temperature data according to the acquisition time. Assuming that the temperature sensor collects temperature values ​​every 1 minute, the computer system can use the temperature values ​​collected every 10 minutes as a data fragment. In this way, if 60 temperature values ​​are collected within 1 hour, 6 temperature data fragments can be obtained.

[0035] For the operating parameters of surgical equipment, such as the power and current of the equipment, the computer system can segment the data according to the different functional modules of the equipment. For air quality monitoring data, including oxygen concentration and carbon dioxide concentration, the computer system can segment the data according to the type of gas.

[0036] In step S120, array encoding is the process by which the computer system converts the data fragments into a coded representation in a specific array form. Its purpose is to represent the data fragments in a form that is easier for the computer to process and store, and also facilitates subsequent operations such as high-level feature extraction. This coded representation can convert the data into a unified format while retaining the key information of the data fragments, so that different types of data fragments can interact and calculate effectively in the subsequent processing flow.

[0037] For the temperature data segments mentioned above, the computer system can use a linear mapping encoding method for array encoding. Assuming that the temperature value range in the temperature data segment is 18°C-25°C, the computer system can map this range to an array. For example, define an array with a length of 8, map 18°C ​​to the first element of the array, that is, the position with index 0, and map 25°C to the last element of the array, that is, the position with index 7. The data segments of surgical equipment power, carbon dioxide concentration, etc. are no longer given as examples.

[0038] Through sub-steps S110 and S120, the computer system converts the target-related monitoring data from the original complex form into a coded representation of data fragments with a specific format, laying the foundation for subsequent operations such as high-level feature extraction and environmental state prediction.

[0039] As an implementation manner, step S110, segmenting the target-related monitoring data to obtain data segments of the target-related monitoring data, includes:

[0040] Step S111: when the target-related monitoring data of the target operating room is obtained, a target monitoring model for monitoring the environmental status of the target operating room is obtained, where the target monitoring model includes a first mode model;

[0041] Step S112: determining the monitoring dimension corresponding to the target-related monitoring data based on the first mode model, and extracting split dimension segments of the target-related monitoring data according to the monitoring dimension;

[0042] Step S113: split the split dimension segments based on the first mode model to obtain split data corresponding to the split dimension segments;

[0043] Step S114: fragment the split data corresponding to the split dimension fragments based on the first pattern model to obtain data fragments of the target related monitoring data.

[0044] In step S111, when the target-related monitoring data of the target operating room is obtained, a target monitoring model for monitoring the environmental status of the target operating room is obtained, and this target monitoring model includes a first mode model. The target-related monitoring data covers a lot of information closely related to the environmental status of the operating room, such as temperature, humidity, air quality, light intensity, and the operating status of surgical equipment. The target monitoring model is a comprehensive, pre-built model framework, such as a transformer framework, which is used to comprehensively monitor the environmental status of the operating room. As a part of the target monitoring model, the first mode model has a specific structure and function, which is intended to assist subsequent data processing steps. For example, in an actual intelligent operating room environment monitoring system, the target monitoring model may be a complex model that integrates multiple algorithms and data processing modules, and the first mode model may be responsible for processing tasks related to preliminary analysis and feature extraction of data.

[0045] In step S112, the computer system determines the monitoring dimensions corresponding to the target-related monitoring data based on the first mode model, and then extracts the split dimension fragments of the target-related monitoring data according to the monitoring dimensions. The monitoring dimension is a way to classify and define the target-related monitoring data from the perspective of the target monitoring model. Taking the temperature data of the operating room as an example, its monitoring dimensions may include the numerical range of the temperature, the frequency of temperature change, and the distribution of the temperature in different areas of the operating room. For humidity data, the monitoring dimensions may involve the absolute value of humidity, the correlation between relative humidity and temperature, etc. Assuming that in a larger operating room, there are multiple temperature sensors distributed in different positions. The computer system will classify the temperature data collected by sensors at different positions according to these dimensions based on the temperature monitoring dimensions determined by the first mode model, thereby extracting the split dimension fragments.

[0046] In step S113, the computer system splits the split dimension segment based on the first mode model to obtain the split data corresponding to the split dimension segment. The splitting here is to further refine the split dimension segment that has been divided according to the monitoring dimension. For example, for the split dimension segment of the temperature data in the area near the operating table mentioned above, it is assumed that the first mode model has a more detailed division rule for the temperature data. If the temperature value range is set as one of the important monitoring dimensions, the computer system may split this temperature data segment again according to the temperature value range. For example, the temperature data of 18-20°C near the operating table is split into a group of split data, and the temperature data of 20-22°C is split into another group of split data. Taking the operating status data of the surgical equipment as an example, if the monitoring dimension of the equipment operating status includes the power and operating time of the equipment, for the split dimension segment of the equipment power, the computer system may split the equipment power data of different power intervals into different split data according to the pre-set power range.

[0047] Finally, in step S114, the computer system fragments (completes serialization) the split data corresponding to the split dimension fragments based on the first pattern model to obtain data fragments of the target-related monitoring data. Fragmentation is the process of converting split data into data fragments that are easier to process and store. For example, for the split data with a temperature of 18-20°C near the operating table obtained earlier, the computer system may arrange these data in the order of collection time to form a data fragment. Assuming that within this temperature range, temperature data is collected every 1 minute, and 10 temperature values ​​are collected within 10 minutes, the computer system serializes these 10 temperature values ​​in the order of collection to obtain a data fragment containing 10 elements.

[0048] Throughout the entire process, the computer system operates strictly in accordance with the rules and algorithms specified by the first mode model. The various parameters and algorithms in the first mode model are obtained through a large number of experiments and optimizations. For example, when determining the monitoring dimension, statistical analysis of environmental data of different operating rooms may be performed to identify factors that have a greater impact on the environmental status as monitoring dimensions. When splitting the fragments of the split dimension, the splitting rules are also based on the study of data distribution laws and consideration of actual operating room environmental requirements. The fragmentation operation is to meet the requirements for data format in subsequent data processing processes, to ensure that data can flow smoothly in the entire intelligent operating room environmental status monitoring system, thereby accurately monitoring the environmental status of the operating room, and providing guarantees for the smooth progress of the operation and the safety of the patient.

[0049] In addition, different types of data are interrelated and processed independently in this process. For example, although temperature data and humidity data may be related in some monitoring dimensions (such as the relationship between relative humidity and temperature), in the data fragmentation process, they are processed according to their respective monitoring dimensions, split dimension fragments, split data, and fragmentation processes. This ensures in-depth analysis of each type of data, and comprehensively considers the relationship between them in subsequent processing, fully reflecting the environmental status of the operating room.

[0050] As an implementation mode, the first mode model includes a data acquisition network; the split dimension segments include y split dimension segments, and the y split dimension segments include a target split dimension segment, and y≥1. Based on this, step S113, splitting the split dimension segments based on the first mode model to obtain split data corresponding to the split dimension segments, includes:

[0051] Step S1131: determining a projection domain corresponding to the target split dimension segment according to the monitoring dimension corresponding to the target split dimension segment; the projection domain includes a plurality of projection values, each projection value having a corresponding split dimension data range;

[0052] Step S1132: Determine the split dimension data range corresponding to the target split dimension segment based on the data acquisition network, and determine the projection value corresponding to the split dimension data range corresponding to the target split dimension segment as the split data corresponding to the target split dimension segment.

[0053] In step S1131, the computer system determines the projection domain corresponding to the target split dimension segment according to the monitoring dimension corresponding to the target split dimension segment, wherein the projection domain includes multiple projection values, each projection value has a corresponding split dimension data range. The monitoring dimension is a way to describe and classify target-related monitoring data, which defines the angles from which to view and analyze the data.

[0054] Taking the temperature data of the operating room as an example, it is assumed that in the previous steps, the temperature data of different areas of the operating room (such as the operating table area, the surrounding area, etc.) have been divided into different split dimension segments. Now take the temperature data of the operating table area as the target split dimension segment, and its corresponding monitoring dimensions may include the temperature range and the temperature change trend. The computer system determines the projection domain based on the monitoring dimension of the temperature range. Assuming that the temperature range is 18-25℃, the computer system can divide this range into several sub-ranges, such as 18-20℃, 20-22℃, and 22-25℃, and these sub-ranges correspond to different projection values. Each projection value has its corresponding split dimension data range.

[0055] In step S1132, the computer system determines the split dimension data range corresponding to the target split dimension segment based on the data acquisition network, and determines the projection value corresponding to the split dimension data range corresponding to the target split dimension segment as the split data corresponding to the target split dimension segment. The data acquisition network is a part of the first mode model, which is responsible for acquiring and processing relevant data information.

[0056] Continuing with the example of the temperature data in the operating table area, the data acquisition network will obtain the specific numerical distribution of the temperature data in the operating table area. Assuming that in a certain period of time, the temperature data in the operating table area is mainly concentrated in the range of 20-22°C, the computer system determines this split dimension data range through the data acquisition network, and then according to the projection domain determined in the previous step, the projection value corresponding to the split dimension data range of 20-22°C is 2, then the computer system determines the projection value 2 as the split data corresponding to the target split dimension segment (operating table area temperature data).

[0057] This way of processing the split dimension fragments provides a more accurate and targeted data foundation for subsequent data processing steps (such as data encoding, advanced feature extraction, etc.), which helps to improve the accuracy and effectiveness of the entire monitoring system's judgment of the operating room environment status. At the same time, with the continuous improvement of operating room environment monitoring requirements and the development of monitoring technology, this method of determining split data based on monitoring dimensions and projection domains can also be continuously optimized and expanded to adapt to more types of data and more complex environmental status monitoring needs. For example, more monitoring dimensions can be added and the division of the projection domain can be refined to improve the accuracy of data splitting and further enhance the performance of intelligent operating room environmental status monitoring.

[0058] As an implementation method, the high-level features are obtained by extracting high-level features from the encoded representation through the embedding mapping component in the first mode model; the first mode model is a target monitoring model for monitoring the environmental state of the target operating room. Based on this, the method also includes a training process of the first mode model, including:

[0059] Step S11: obtaining a first calibration template library; the first calibration template contained in the first calibration template library is a calibration template that does not carry an environmental state mark;

[0060] Step S12: obtaining s calibration targets for the first basic mode model, s≥1;

[0061] Step S13: extracting high-level features from the first calibration template based on the first basic pattern model to obtain high-level features of the template associated with each calibration target, and determining s calibration errors corresponding to the s calibration targets according to the high-level features of the template, wherein one calibration target corresponds to one calibration error;

[0062] Step S14: determining a first target calibration error according to the s calibration errors, optimizing the model parameters of the first basic mode model according to the first target calibration error, and using the optimized first basic mode model as the first mode model.

[0063] In step S11, the computer system obtains a first calibration template library. The first calibration template library contains a first calibration template that does not carry an environmental state tag. These calibration templates are a series of data samples used to train the first mode model, and the data content refers to the aforementioned related monitoring data.

[0064] In step S12, the computer system obtains s calibration targets for the first basic mode model (one calibration target here is one training task), where s≥1. These calibration targets are various training tasks set according to the needs of intelligent operating room environmental status monitoring. For example, one calibration target may be to improve the model's ability to recognize abnormal temperature fluctuations, and another calibration target may be to enhance the model's ability to understand the relationship between surgical equipment and environmental status when it is running under high load. Assuming s=3, the first calibration target is to accurately identify the potential impact of humidity changes on environmental status, the second calibration target is to accurately determine the relationship between oxygen concentration in air quality and environmental comfort, and the third calibration target is to optimize the analysis of the relationship between energy consumption and environmental stability of surgical equipment at different powers.

[0065] In step S13, the computer system extracts high-level features from the first calibration template based on the first basic pattern model, obtains high-level features of the template associated with each calibration target, and determines s calibration errors corresponding to s calibration targets based on the high-level features of the template, wherein one calibration target corresponds to one calibration error. High-level feature extraction is the process of mining more representative and abstract features from the original first calibration template data. For example, for the first calibration template containing data such as temperature and humidity, the computer system extracts high-level features through the first basic pattern model. If the temperature data in the first calibration template has a certain periodic variation pattern, the model may extract high-level features related to this periodicity, such as the frequency and amplitude of temperature fluctuations; for humidity data, it may extract features related to the humidity change rate and other environmental factors.

[0066] Taking the first calibration goal (accurately identifying the potential impact of humidity changes on environmental conditions) as an example, the computer system extracts humidity-related high-level features from the first calibration template. Assuming that there are multiple humidity data points and corresponding other environmental data (such as temperature, equipment operating status, etc.) in the first calibration template, the model may find that there is a certain correlation feature between humidity changes within a specific temperature range and the stability of equipment operation by analyzing the relationship between these data. This correlation feature is the high-level feature of the template associated with the first calibration goal. The calibration error is then determined based on this template high-level feature. If the model predicts that the impact of humidity changes on equipment operating stability deviates from the actual situation, this deviation constitutes the calibration error corresponding to the first calibration goal.

[0067] For the second calibration goal (accurately determine the relationship between oxygen concentration in air quality and environmental comfort), the computer system extracts high-level features related to oxygen concentration from the first calibration template. For example, by analyzing the work efficiency of medical staff under different oxygen concentrations (assuming that this is an indirect indicator of environmental comfort) and other relevant environmental data, the correlation characteristics between oxygen concentration and environmental comfort are obtained, such as the fact that when the oxygen concentration is within a certain range, the work efficiency of medical staff is the highest. If the model's judgment of this relationship is inconsistent with the actual statistical results, a calibration error corresponding to the second calibration goal is generated.

[0068] In step S14, the computer system determines the first target calibration error based on the s calibration errors, optimizes the model parameters of the first basic mode model based on the first target calibration error, and uses the optimized first basic mode model as the first mode model. The first target calibration error is an overall error index obtained by comprehensively considering the calibration errors corresponding to all s calibration targets. For example, the first target calibration error can be determined by a weighted average method. Assume that the calibration error corresponding to the first calibration target is e1, and the weight is w1; the calibration error corresponding to the second calibration target is e2, and the weight is w2; the calibration error corresponding to the third calibration target is e3, and the weight is w3. Then the first target calibration error E=w1*e1+ w2*e2+ w3*e3.

[0069] Once the first target calibration error is determined, the computer system can use this error to adjust the model parameters of the first basic mode model. Model parameters are key factors that determine the structure and function of the first basic mode model, such as weights, biases and other parameters in the neural network. If the first target calibration error is large, it means that the prediction results of the model deviate greatly from the actual situation, and the model parameters need to be adjusted to a greater extent; if the first target calibration error is small, the adjustment range is relatively small. Taking the neural network model as an example, if the first target calibration error is found to be large during the training process, the computer system may update the weights and biases in the neural network according to a specific optimization algorithm (such as a stochastic gradient descent algorithm). By continuously iterating this process, the first target calibration error is gradually reduced until the performance of the model reaches a satisfactory level. At this time, the first basic mode model is optimized to the first mode model, which can more effectively monitor the environmental status of the target operating room.

[0070] As embodiment 1, the s calibration targets include a data item occlusion recognition branch; based on this, step S13, based on the first basic pattern model, high-level feature extraction is performed on the first calibration template to obtain high-level features of the template associated with each calibration target, and s calibration errors corresponding to the s calibration targets are determined according to the high-level features of the template, including:

[0071] Step S131: fragmenting the first adjustment template based on the first basic mode model to obtain a first adjustment template fragment of the first adjustment template; the first adjustment template fragment includes template splitting data at multiple fragment positions;

[0072] Step S132: arbitrarily selecting a first segment position to be subjected to data item occlusion from a plurality of segment positions corresponding to the first adjustment template segment;

[0073] Step S133: performing data item occlusion processing on the template segmentation data at the first segment position in the first adjusted template segment based on the data item occlusion information to obtain first template segment data;

[0074] Step S134: performing array encoding on the first template fragment data based on the first basic pattern model to obtain a first template encoding representation of the first template fragment data, performing high-level feature extraction on the first template encoding representation to obtain a first template high-level feature corresponding to the first template encoding representation, where the first template high-level feature belongs to a template high-level feature;

[0075] Step S135: Based on the monitoring network related to the data item occlusion identification branch, the occluded data item is identified on the first template high-level features to obtain the inference split data corresponding to the data item occlusion information in the first template fragment data;

[0076] Step S136: determining the calibration error corresponding to the data item occlusion recognition branch according to the template segmentation data and the inference segmentation data at the first segment position in the first calibrated template segment.

[0077] In step S131, the computer system performs data fragmentation on the first calibration template based on the first basic mode model to obtain the first calibration template fragment of the first calibration template, and the first calibration template fragment contains template splitting data at multiple fragment positions. The first calibration template is a data sample used for training the model and does not carry environmental status labels. It contains information related to the operating room environment in many aspects, such as temperature, humidity, air quality, and the operating status of surgical equipment. Data fragmentation is to decompose this complex first calibration template into smaller data units according to certain rules, namely, first calibration template fragments.

[0078] In step S132, the computer system randomly selects the first segment position to be occluded from the multiple segment positions corresponding to the first calibration template segment. This step is to simulate the situation where data is missing or occluded, so as to test the performance of the model in this situation. For example, in the first calibration template segment containing temperature and humidity data for multiple time periods, the computer system randomly selects a segment position, assuming that this position corresponds to the temperature and humidity data at a specific moment, and takes it as the first segment position, and the data at this position will be occluded later.

[0079] In step S133, the computer system performs data item occlusion processing on the template segmentation data at the first segment position in the first adjusted template segment based on the data item occlusion information to obtain the first template segment data. The data item occlusion information defines how to occlude the selected template segmentation data, which may be occluding data according to a certain ratio, randomly occluding part of the data, or completely occluding a certain data item.

[0080] For example, if the template split data at the first fragment position contains a temperature value of 22°C and a humidity value of 50%, according to the data item occlusion information, if it is occluded in proportion, assuming 50% of the data is occluded, the computer system may occlude the temperature value to an unknown value (indicated by a special symbol, such as "?"), and retain the humidity value; if part of the data is randomly occluded, the temperature value or humidity value may be randomly selected for occlusion; if a data item is completely occluded, the temperature value may be completely occluded and only the humidity value is retained. The first template fragment data is obtained after processing in this way.

[0081] In step S134, the computer system performs array encoding on the first template fragment data based on the first basic pattern model, obtains a first template encoding representation of the first template fragment data, extracts high-level features on the first template encoding representation, obtains a first template high-level feature corresponding to the first template encoding representation, and the first template high-level feature belongs to the template high-level feature. Array encoding is to convert the first template fragment data after the data item occlusion processing into an encoding representation in a specific array form, so as to facilitate the computer system to process and extract high-level features subsequently.

[0082] For example, for the temperature (partial occlusion or full occlusion) and humidity data after occlusion processing, the computer system may map it to a predefined array structure. If the temperature value is occluded and the humidity value is 50%, the humidity value is encoded to a specific position in an array according to the range of the humidity value and the predefined encoding rules. This array is the first template encoding representation. Then, high-level feature extraction is performed on this encoding representation. For example, for this array containing partial occlusion information, the computer system may extract the relationship features between the humidity value and other unoccluded data, the distribution features of the data, etc. These features are the first template high-level features.

[0083] In step S135, the computer system identifies the occluded data item on the first template high-level features based on the monitoring network associated with the data item occlusion identification branch, and obtains the inference split data corresponding to the data item occlusion information in the first template fragment data. The monitoring network associated with the data item occlusion identification branch is a network structure specifically used to identify whether a data item is occluded and infer the content of the occluded data.

[0084] For example, based on the high-level features of the first template obtained earlier, the monitoring network may analyze based on existing data relationships and feature patterns. If the humidity value is found to have a certain correlation with the surrounding environment data in the high-level features of the first template, and according to the rules learned from historical data and the model, when the humidity value is 50%, the corresponding temperature value should be within a certain range, then the monitoring network can infer the obscured temperature value based on this relationship (if it is partially obscured, the obscured part can also be supplemented or corrected), and this inferred temperature value is the inference split data.

[0085] In step S136, the computer system determines the calibration error corresponding to the data item occlusion identification branch according to the template segmentation data and the reasoning segmentation data at the first segment position in the first calibration template segment. The calibration error reflects the prediction accuracy of the model under data item occlusion.

[0086] For example, if the temperature value in the original template split data at the first fragment position is 22°C, and the inference split data obtained through the monitoring network related to the data item occlusion recognition branch is 20°C, then the difference between the two constitutes part of the calibration error. This difference may be quantified according to the specific error calculation method (such as mean square error, absolute error, etc.), and finally the calibration error corresponding to the data item occlusion recognition branch is obtained. If this calibration error is large, it means that the model has poor prediction and recovery capabilities for data under data item occlusion, and the model parameters need to be further adjusted to improve performance; if the calibration error is small, it means that the model has a better performance in this case.

[0087] In the above process, data fragmentation enables data to be processed in a more detailed manner, providing a basis for subsequent occlusion processing and feature extraction. Selecting the first fragment position of the data item to be occluded is the starting point for simulating the actual data missing or occlusion situation. Different choices may affect the final calibration error. Data item occlusion processing directly affects the subsequent model's testing of the ability to process and recover occluded data. Array encoding and advanced feature extraction convert the processed data into a more abstract and easier to analyze form, allowing the model to mine deeper information from the data. Occluded data item recognition based on the monitoring network is the link where the model plays a role when facing occluded data, and the final calibration error is an important indicator to measure the performance of the model in the data item occlusion recognition branch. According to this calibration error, the model can be optimized in a targeted manner to improve the accuracy of the model in monitoring the operating room environment status in complex environments (such as when there is partial missing or occlusion of the data).

[0088] In summary, Example 1 comprehensively evaluates and optimizes the performance of the first basic pattern model in the data item occlusion recognition branch from data processing, feature extraction, data item recognition to calibration error determination to ensure the effectiveness of the model in actual intelligent operating room environment status monitoring.

[0089] As embodiment 2, the s calibration targets include a data item change identification branch; based on this, step S13, based on the first basic pattern model, high-level feature extraction is performed on the first calibration template to obtain high-level features of the template associated with each calibration target, and s calibration errors corresponding to the s calibration targets are determined according to the high-level features of the template, including:

[0090] Step S1301: fragmenting the first adjustment template based on the first basic pattern model to obtain a first adjustment template fragment of the first adjustment template, wherein the first adjustment template fragment includes template splitting data at a plurality of fragment positions;

[0091] Step S1302: arbitrarily selecting a second segment position to be subjected to data item change from among a plurality of segment positions corresponding to the first adjustment template segment;

[0092] Step S1303: performing data item change processing on the template segmentation data at the second segment position in the first adjusted template segment based on the arbitrary value to obtain the second template segment data, wherein the arbitrary value is different from the template segmentation data at the second segment position;

[0093] Step S1304: performing array encoding on the second template fragment data based on the first basic pattern model to obtain a second template encoding representation of the second template fragment data, performing high-level feature extraction on the second template encoding representation to obtain a second template high-level feature corresponding to the second template encoding representation, wherein the second template high-level feature belongs to the template high-level feature;

[0094] Step S1305: Based on the monitoring network related to the data item change identification branch, the second template high-level feature is subjected to change position reasoning to obtain data item change reasoning results corresponding to multiple segment positions;

[0095] Step S1306: Determine the calibration error corresponding to the data item change identification branch according to the data item change inference results corresponding to the second segment position and multiple segment positions.

[0096] In step S1301, the computer system performs data fragmentation on the first calibration template based on the first basic pattern model to obtain a first calibration template fragment of the first calibration template, wherein the first calibration template fragment contains template splitting data at multiple fragment positions. The first calibration template contains various data related to the operating room environment status, such as temperature, humidity, air quality, and the operating status of surgical equipment. Data fragmentation is to decompose these complex data according to certain rules and logic to form units that are easier to process.

[0097] Taking the temperature and humidity data of the operating room as an example, assume that the first calibration template records the temperature and humidity values ​​of the operating room at different times over a period of time. The computer system may fragment the data according to the time interval, such as taking the temperature and humidity data every 5 minutes as a small fragment. For the operating status data of the surgical equipment, it may be divided according to the data of different functional modules or operating stages of the equipment. These divided small fragments are combined to form the first calibration template fragment, and each fragment has its specific position in the whole, that is, multiple fragment positions, and each position contains the corresponding template split data. For example, the template split data at a certain fragment position may be the temperature value, humidity value and corresponding equipment power value at a certain moment.

[0098] In step S1302, the computer system randomly selects a second segment position where data item change is to be performed from among multiple segment positions corresponding to the first calibration template segment. This step is to perform targeted change operations on data at specific positions to simulate possible changes in data in actual situations, thereby testing the model's ability to identify data item changes.

[0099] For example, in the first calibration template fragment containing temperature, humidity and device power data at multiple time points, the computer system randomly selects a fragment position, which may correspond to the data at a specific moment, such as the temperature of 22°C, humidity of 50%, and device power of 300 watts at 10 am. This position is determined as the second fragment position, and the data at this position will be changed subsequently.

[0100] In step S1303, the computer system performs data item change processing on the template split data at the second segment position in the first adjustment template segment based on an arbitrary value to obtain the second template segment data, and the arbitrary value is different from the template split data at the second segment position. This step changes the original data by introducing a new value, thereby creating a data situation that is different from the original data. Continuing with the above example, if the original temperature value at the second segment position is 22°C, the computer system changes the temperature value to 25°C according to the setting (this 25°C is an arbitrary value, which is different from the original 22°C); the humidity value is originally 50%, which is changed to 55%; the equipment power is originally 300 watts, which is changed to 350 watts. After such a change, the segment containing the changed data is the second template segment data.

[0101] In step S1304, the computer system performs array encoding on the second template fragment data based on the first basic pattern model to obtain a second template encoding representation of the second template fragment data, performs high-level feature extraction on the second template encoding representation to obtain a second template high-level feature corresponding to the second template encoding representation, wherein the second template high-level feature belongs to the template high-level feature. Array encoding is to convert the changed second template fragment data into a specific array form to facilitate storage, processing and subsequent analysis operations by the computer system.

[0102] For example, for the data of the changed temperature of 25°C, humidity of 55%, and device power of 350 watts, the computer system encodes the array according to the predefined encoding rules. Assume that for temperature, its numerical range is divided into several intervals, and 25°C falls within a certain interval, which corresponds to a specific element position in the array; there are similar encoding rules for humidity and device power. The array obtained in this way is the second template encoding representation. Then, high-level features are extracted from this encoding representation. For example, the computer system may extract new relationship features between the changed temperature, humidity, and device power, such as the correlation features between temperature increase and humidity increase and device power increase. These features are the second template high-level features.

[0103] Subsequently, in step S1305, the computer system performs change position reasoning on the second template high-level features based on the monitoring network associated with the data item change identification branch, and obtains the data item change reasoning results corresponding to multiple fragment positions. The monitoring network associated with the data item change identification branch is a network structure specifically used to identify whether a data item has changed and determine the change position. Taking the previously obtained second template high-level features as an example, this monitoring network performs analysis based on the normal data relationship patterns and feature rules learned in advance. Since the temperature, humidity and equipment power data have been changed before, the monitoring network infers which positions of the data have changed based on the relationship between these data and other unchanged data and the overall data pattern. For example, the monitoring network may accurately determine that the data at the previously selected second fragment position has changed based on the correlation between temperature, humidity and equipment power and the relationship with other environmental data. This is the data item change reasoning result corresponding to multiple fragment positions.

[0104] Finally, in step S1306, the computer system determines the calibration error corresponding to the data item change identification branch according to the inference results of the data item change corresponding to the second segment position and the multiple segment positions. The calibration error reflects the accuracy of the model in identifying data item changes.

[0105] For example, if the data at the second segment position actually changed, and the monitoring network related to the data item change identification branch accurately infers that the data at this position has changed, then the calibration error may be relatively small; conversely, if the monitoring network mistakenly determines that the data at other positions has changed or fails to identify the data change at the second segment position, then the calibration error will be large. The specific calibration error calculation may be based on different evaluation indicators, such as using accuracy, recall and other indicators to quantify the model's performance in data item change identification, thereby determining the calibration error corresponding to the data item change identification branch.

[0106] As embodiment 3, the s calibration targets include a perturbation segment branch; based on this, step S13, based on the first basic pattern model, high-level feature extraction is performed on the first calibration template to obtain high-level features of the template associated with each calibration target, and s calibration errors corresponding to the s calibration targets are determined according to the high-level features of the template, including:

[0107] Step S13A: inject disturbance into the split dimension segments of the first calibration template to obtain a disturbance calibration template, and perform data segmentation on the disturbance calibration template to obtain disturbance data segments of the disturbance calibration template;

[0108] Step S13B: segmenting the first calibration template into data segments to obtain first calibration template segments of the first calibration template;

[0109] Step S13C: performing array encoding on the disturbed data segment based on the first basic pattern model to obtain a third template encoding representation of the disturbed data segment, performing high-level feature extraction on the third template encoding representation to obtain a third template high-level feature corresponding to the third template encoding representation, wherein the third template high-level feature belongs to the template high-level feature;

[0110] Step S13D: performing array encoding on the first calibrated template fragment based on the first basic mode model to obtain a fourth template encoding representation of the first calibrated template fragment, performing high-level feature extraction on the fourth template encoding representation to obtain a fourth template high-level feature corresponding to the fourth template encoding representation, wherein the fourth template high-level feature belongs to the template high-level feature;

[0111] Step S13E: Determine the calibration error corresponding to the disturbance segment branch according to the third template high-level features and the fourth template high-level features.

[0112] In step S13A, the computer system performs disturbance injection on the split dimension segments of the first calibration template to obtain a disturbance calibration template, and performs data segmentation on the disturbance calibration template to obtain disturbance data segments of the disturbance calibration template. The first calibration template contains various data information related to the operating room environment state, and its split dimension segments are data parts divided in a certain way. Disturbance injection is to introduce interference factors into these split dimension segments to simulate the instability or interference that may occur in the actual environment.

[0113] For example, for the split dimensional fragments of the operating room temperature data, if the original data is a series of temperature values ​​collected in a stable environment, the computer system can randomly add some small fluctuation values ​​to these temperature values ​​as perturbation injection. Assuming that the original temperature value sequence is [22, 23, 22.5, 23.5], the computer system follows certain rules (such as randomly adding fluctuations between -0.5 and 0.5 to each temperature value) to obtain a perturbed temperature value sequence such as [22.3, 23.2, 22.1, 23.8], which forms part of the perturbation calibration template. For the split dimensional fragments of humidity data, a similar method can be used for perturbation injection, such as adding a certain proportion of random fluctuations to the original humidity value.

[0114] In step S13B, the computer system performs data fragmentation on the first calibration template to obtain the first calibration template fragment of the first calibration template. This step is similar to the data fragmentation in step S13A, but is for the original undisturbed first calibration template. For example, for the air quality data (including oxygen concentration, carbon dioxide concentration, etc.) in the first calibration template, it is divided into different fragments according to factors such as concentration range or acquisition area, and each fragment contains air quality data within a specific range or area. These fragments are the first calibration template fragments. For the operating status data of the surgical equipment, it is data fragmented according to the different operating parameters or functional modules of the equipment to form corresponding first calibration template fragments.

[0115] In step S13C, the computer system performs array encoding on the disturbed data segment based on the first basic pattern model to obtain a third template encoding representation of the disturbed data segment, performs high-level feature extraction on the third template encoding representation, and obtains a third template high-level feature corresponding to the third template encoding representation, wherein the third template high-level feature belongs to the template high-level feature. Array encoding is an operation of converting the disturbed data segment into a specific array form to facilitate processing and storage by the computer system.

[0116] Taking the disturbed temperature data segment as an example, if the temperature value sequence in a disturbed data segment obtained through step S13A is [22.3, 23.2, 22.1, 23.8], the computer system maps these temperature values ​​to an array according to the predefined encoding rules. Assuming that the range of temperature values ​​is divided into several intervals, each interval corresponds to an element position in the array, the encoding representation of the temperature value in the array is determined according to the interval in which the temperature value is located, so that the third template encoding representation is obtained. Then, high-level feature extraction is performed on this encoding representation. For example, the computer system may extract the changing trend characteristics of the temperature value after the disturbance, such as the frequency and amplitude of the temperature value fluctuation, as well as the relationship characteristics with other data that may be associated (such as humidity data, if the encoding representation contains relevant information). These features are the third template high-level features.

[0117] In step S13D, the computer system performs array encoding on the first adjusted template fragment based on the first basic mode model to obtain a fourth template encoding representation of the first adjusted template fragment, performs high-level feature extraction on the fourth template encoding representation to obtain a fourth template high-level feature corresponding to the fourth template encoding representation, wherein the fourth template high-level feature belongs to the template high-level feature. This step is similar to step S13C, but is for the original first adjusted template fragment.

[0118] In step S13E, the computer system determines the calibration error corresponding to the disturbance segment branch according to the third template high-level feature and the fourth template high-level feature. The calibration error reflects the degree of difference between the model in processing disturbed data and non-disturbed data. For example, if the fluctuation frequency of the temperature value after disturbance in the third template high-level feature is significantly different from the fluctuation frequency of the original temperature value in the fourth template high-level feature, this difference will be reflected in the calculation of the calibration error. Assuming that the method for calculating the calibration error is to calculate the absolute value of the difference between the fluctuation frequencies of the temperature values ​​in the two high-level features, if the original temperature value fluctuation frequency fluctuates once every 10 minutes, and the temperature value fluctuation frequency after disturbance fluctuates once every 5 minutes, then this frequency difference will affect the calibration error. If there is such a difference between the disturbance and non-disturbance conditions in multiple data features (such as temperature, humidity, equipment operation status and other related high-level features), the computer system combines these differences and calculates the final calibration error according to a certain algorithm (such as weighted average, etc.). This calibration error can reflect the sensitivity of the model to disturbances and its performance in the presence of disturbances. If the calibration error is large, it means that the model may not be able to accurately process data or identify data features when facing data disturbances, and the model needs to be adjusted and optimized; if the calibration error is small, it means that the model can resist the impact of data disturbances on its performance to a certain extent. By injecting disturbances into the split dimension fragments of the first calibration template, various interference factors that may occur in the actual operating room environment can be simulated, such as minor faults of sensors, temporary fluctuations in the external environment, etc. These interference factors may affect the accuracy and stability of the data, and the model needs to be able to accurately monitor the operating room environment status under such circumstances. Operations such as data fragmentation, array encoding, and advanced feature extraction are to convert data into a form suitable for model analysis, so as to accurately compare the differences in data features with and without disturbances, thereby determining the calibration error. Optimizing the model based on this calibration error can improve the robustness and accuracy of the model in actual complex environments, ensuring that the intelligent operating room environment status monitoring system can operate reliably without being affected or being affected by these potential interference factors. At the same time, the relationships between different types of data (such as temperature, humidity, equipment operating status, etc.) also play an important role in this process, because the disturbance of one data may affect the characteristics of other data and the model's judgment of the overall environmental status through these relationships. The model needs to accurately capture these relationships to reduce calibration errors and improve performance.

[0119] As embodiment 4, the s calibration targets include a virtual marking branch; based on this, step S13, based on the first basic pattern model, high-level feature extraction is performed on the first calibration template to obtain high-level features of the template associated with each calibration target, and s calibration errors corresponding to the s calibration targets are determined according to the high-level features of the template, including:

[0120] Step S13a: segmenting the first calibration template into data segments to obtain first calibration template segments of the first calibration template;

[0121] Step S13b: performing array encoding on the first calibrated template fragment based on the first basic mode model to obtain a fourth template encoding representation of the first calibrated template fragment, performing high-level feature extraction on the fourth template encoding representation to obtain a fourth template high-level feature corresponding to the fourth template encoding representation, wherein the fourth template high-level feature belongs to the template high-level feature;

[0122] Step S13c: classifying the environment state of the fourth template high-level feature through the monitoring network related to the virtual mark branch to obtain the template environment state classification confidence corresponding to the first calibration template;

[0123] Step S13d: obtaining a historical model related to the first basic mode model, classifying the environmental state of the first calibration template according to the historical model, and obtaining a prediction confidence of the virtual marked environmental state corresponding to the first calibration template;

[0124] Step S13e: Determine the calibration error corresponding to the virtual marking branch according to the template environment state classification confidence and the virtual marking environment state prediction confidence.

[0125] In step S13a, the computer system performs data fragmentation on the first calibration template to obtain the first calibration template fragment of the first calibration template. The first calibration template contains a lot of data related to the operating room environment, such as temperature, humidity, air quality, operating status of surgical equipment, etc. Data fragmentation is to decompose these complex data into smaller and easier to process data units according to certain rules, namely, the first calibration template fragment.

[0126] In step S13b, the computer system performs array encoding on the first adjustment template fragment based on the first basic mode model to obtain a fourth template encoding representation of the first adjustment template fragment, performs high-level feature extraction on the fourth template encoding representation to obtain a fourth template high-level feature corresponding to the fourth template encoding representation, wherein the fourth template high-level feature belongs to the template high-level feature. Array encoding is to convert the first adjustment template fragment into a specific array form, which is convenient for the computer system to perform storage, processing and subsequent analysis operations.

[0127] For example, for a certain temperature data segment, if the temperature value range is 18-25°C, the computer system can map the temperature values ​​within this range to an array according to the preset encoding rules. Assume that 18-20°C is mapped to the first element position of the array, 20-22°C is mapped to the second element position, and so on. When the temperature value in this temperature data segment is 21°C, its encoding representation in the array is the element position corresponding to 20-22°C. For the data segment of the operating status of the surgical equipment, such as the equipment power data, if the power range is 0-1000 watts, it is encoded into the array according to the power interval division. In this way, the fourth template encoding representation of the first adjustment template segment is obtained. Then, the fourth template encoding representation is subjected to high-level feature extraction. For example, for the encoding representation of temperature data, the computer system may extract the distribution characteristics of the temperature value, such as the distribution frequency of the temperature value at different positions in the array within a certain time period; for the encoding representation of the equipment power data, the power change trend characteristics may be extracted, such as whether the power is gradually increasing or decreasing. These extracted features are the fourth template high-level features.

[0128] In step S13c, the computer system classifies the environmental state of the fourth template high-level features through the monitoring network associated with the virtual tag branch, and obtains the template environmental state classification confidence corresponding to the first calibration template. The monitoring network associated with the virtual tag branch is a network structure specially constructed for classifying the environmental state of data, wherein the virtual tag is a generated pseudo-label. Assume that the environmental state is divided into three states: good, general, and poor. For the fourth template high-level features extracted from data such as temperature and device power, the monitoring network classifies them according to the relationship between the pre-learned features and the environmental state. For example, if the temperature value is within the appropriate range and the device power is stable, the monitoring network may judge that the environmental state at this time is good. This judgment is based on a certain probability or confidence. If the computer system calculates that the probability that the environmental state is good is 80%, then this 80% is the template environmental state classification confidence corresponding to the first calibration template.

[0129] In step S13d, the computer system obtains a historical model related to the first basic mode model, classifies the environment state of the first calibration template according to the historical model, and obtains the prediction confidence of the virtual labeled environment state corresponding to the first calibration template. The historical model is a model that has been trained before or established on other relevant data, and it contains experience and knowledge of classifying the environment state of similar data.

[0130] For example, when the historical model processes a large amount of operating room environment data, it has learned the relationship between some data patterns and environmental states. When this historical model classifies the environmental state of the first calibration template, it will also give an environmental state classification result and the corresponding confidence level based on its own algorithm and parameters. Assume that the historical model determines that the environmental state corresponding to the first calibration template is good with a confidence level of 70%. This 70% is the prediction confidence level of the virtual labeled environmental state corresponding to the first calibration template.

[0131] Finally, in step S13e, the computer system determines the calibration error corresponding to the virtual marker branch according to the template environment state classification confidence and the virtual marker environment state prediction confidence. The calibration error reflects the degree of difference between the current model (the monitoring network associated with the virtual marker branch) and the historical model when classifying the environment state of the first calibration template.

[0132] For example, if the template environmental state classification confidence is 80% (the environmental state is considered good), and the virtual label environmental state prediction confidence is 70% (the environmental state is also considered good, but the confidence is different), then the adjustment error may be determined by calculating the difference between the two (such as 80%-70%=10%) or other more complex calculation methods (such as considering the weight of the two). If the difference is large, it means that there is a large difference between the current model and the historical model in the classification of environmental states, and the current model may need to be adjusted to make it more consistent with the actual situation or more consistent with historical experience; if the difference is small, it means that the current model is similar and stable to the historical model to a certain extent.

[0133] By determining the calibration error corresponding to the virtual marking branch, the model can be optimized to improve the accuracy and reliability of the model. Data fragmentation and array encoding provide a basis for subsequent high-level feature extraction, enabling the model to mine more effective information from the data for environmental state classification. Comparison of environmental state classification by different models (current model and historical model) helps to discover potential problems of the model, such as whether the current model is overfitting or underfitting. If the calibration error is large and persists, it may indicate that the model structure needs to be readjusted or retrained to better meet the needs of operating room environmental state monitoring. At the same time, the relationship between different types of data (such as temperature, equipment power, etc.) also plays an important role in this process, because these data jointly affect the classification results of the environmental state. The model needs to accurately capture these relationships to reduce the calibration error and improve the accuracy of classification.

[0134] As an implementation method, the high-level features are obtained by extracting high-level features from the coded representation through the embedding mapping component in the first mode model; the first mode model is a target monitoring model for monitoring the environmental status of the target operating room; the target monitoring model also includes a second mode model sequentially connected to the embedding mapping component. Based on this, in step S300, the environmental status is predicted for the high-level features based on each operating mode, and the environmental status prediction confidence corresponding to each operating mode is obtained, including:

[0135] Step S310: determining a second mode model in the target monitoring model, wherein the second mode model includes a professional adaptation model corresponding to each operation mode;

[0136] Step S320: Based on the professional adaptation model corresponding to each operating mode, the high-level features are predicted for the environmental state to obtain the environmental state prediction confidence corresponding to each operating mode, wherein a professional adaptation model is used to determine the environmental state prediction confidence corresponding to an operating mode.

[0137] In step S310, the computer system determines the second mode model in the target monitoring model, wherein the second mode model includes a professional adaptation model (an expert model) corresponding to each operating mode. The target monitoring model is a comprehensive model framework for environmental status monitoring of the target operating room, which covers multiple sub-models and different processing modules to achieve accurate monitoring of the complex environmental status of the operating room. As part of the target monitoring model, the internal structure of the second mode model includes a professional adaptation model carefully constructed for each operating mode.

[0138] The operation mode is an abstract classification of different working or environmental status scenarios in the operating room. For example, when the operation mode is divided based on the surgical process, the preoperative preparation stage is one operation mode. In this stage, the temperature and humidity of the operating room need to be adjusted to a range suitable for surgery, the equipment needs to be inspected and prepared, and the personnel need to carry out relevant preparations; the operation stage is another operation mode, at which the stability of the environment is extremely high, such as the temperature and humidity must be kept within a very precise range, the air quality must meet medical standards, and the surgical equipment must operate stably; the postoperative cleaning stage is also an operation mode, which focuses on equipment cleaning, environmental disinfection, etc. The corresponding environmental requirements are different from those of the operation stage.

[0139] For each such operating mode, there is a corresponding professional adaptation model. Taking the preoperative preparation stage as an example, its professional adaptation model may focus on factors such as the relationship between power changes during equipment preheating and the rise in ambient temperature, as well as the initial impact of personnel movement on air quality. This model is built based on a large amount of environmental data and relevant professional knowledge from the preoperative preparation stage. It contains algorithms and parameters specific to this stage. For the professional adaptation model during the surgical stage, it will focus more on precise temperature and humidity control and the impact of surgical equipment on the environment under different operations. For example, during high-energy surgical operations, the heat generated by the equipment will affect the surrounding temperature and air quality.

[0140] The computer system determines that the second mode model and its internal professional adaptation model are the basis for accurate environmental state prediction. These professional adaptation models are specially designed for the characteristics of the operating room environmental state under different operating modes. They can better capture the complex relationship between environmental data under each operating mode, thereby providing a more accurate basis for subsequent environmental state prediction.

[0141] In step S320, the computer system predicts the environment state of the high-level features based on the professional adaptation model corresponding to each operation mode, and obtains the prediction confidence of the environment state corresponding to each operation mode, wherein a professional adaptation model is used to determine the prediction confidence of the environment state corresponding to an operation mode. High-level features are more representative and abstract features extracted from the coded representation in the previous step, and these features contain comprehensive information related to the environment state of the operating room.

[0142] For example, in the preoperative preparation stage, high-level features may include comprehensive features of information such as the initial state parameters of the equipment (such as the initial power of the equipment, initial temperature, etc.), the initial temperature and humidity of the environment, and the initial indicators of air quality. Based on the professional adaptation model of the preoperative preparation stage, the computer system analyzes these high-level features to predict the state of the environment. Suppose that there is an algorithm in this professional adaptation model that determines whether the environment can reach a suitable surgical temperature range within a specified time based on the relationship between the initial power of the equipment and the rate of increase of the ambient temperature. If the model calculates that there is a 90% probability that the ambient temperature will rise to a range suitable for surgery within the next 30 minutes, then this 90% is the predicted confidence that the environmental state is "the temperature will reach a suitable range" in this operating mode during the preoperative preparation stage.

[0143] By predicting the environment state of high-level features based on the professional adaptation model corresponding to each operating mode, the prediction confidence of different environment states under each operating mode can be obtained. These confidences can reflect the probability of the operating room environment being in a certain state under a specific operating mode.

[0144] As an implementation mode, the method further includes a training process of a second mode model, comprising the following steps:

[0145] Step S21: obtaining a second calibration template library, wherein the second calibration template contained in the second calibration template library carries a mode priori tag; the mode priori tag includes x mode prior values ​​corresponding to x operating modes, one operating mode corresponds to one mode prior value, and the x mode prior values ​​are obtained according to the operating mode corresponding to the second calibration template;

[0146] Step S22: Acquire the calibrated first mode model, perform data encoding on the second calibrated template based on the first mode model, obtain a fifth template encoding representation of the second calibrated template, extract high-level features on the fifth template encoding representation, and obtain high-level features of the fifth template corresponding to the fifth template encoding representation;

[0147] Step S23: obtaining x basic professional adaptation models corresponding to the x operating modes, performing pattern recognition on the fifth template high-level features according to the x modes, and obtaining x pattern recognition confidences corresponding to the x basic professional adaptation models, wherein one operating mode corresponds to one basic professional adaptation model, and one basic professional adaptation model is used to determine one pattern recognition confidence;

[0148] Step S24: Determine the second target calibration error based on x pattern recognition confidences and x pattern prior values, optimize the model parameters of the x basic professional adaptation models based on the second target calibration error, determine the optimized x basic professional adaptation models as x professional adaptation models, and determine the second pattern model based on the x professional adaptation models.

[0149] In step S21, the computer system obtains a second calibration template library, wherein the second calibration template contained in the second calibration template library carries a mode prior mark; the mode prior mark includes x mode prior values ​​corresponding to x operating modes, one operating mode corresponds to one mode prior value, and the x mode prior values ​​are obtained according to the operating mode corresponding to the second calibration template.

[0150] The second calibration template library is a data set used to train the second mode model. These templates contain rich information related to the operating room environment status. The mode prior labeling provides each second calibration template with prior information about the operating mode, which helps the computer system better utilize this data during the training process.

[0151] In step S22, the computer system obtains the calibrated first pattern model, performs data encoding on the second calibrated template based on the first pattern model, obtains the fifth template encoding representation of the second calibrated template, performs high-level feature extraction on the fifth template encoding representation, and obtains the fifth template high-level features corresponding to the fifth template encoding representation. The first pattern model has undergone the previous training process and can effectively process the data. When the second calibrated template is encoded, the complex second calibrated template data is converted into a specific encoding representation, namely the fifth template encoding representation, just like encoding the target-related monitoring data before. This process is to facilitate subsequent high-level feature extraction operations.

[0152] Then, high-level feature extraction is performed on the fifth template encoding representation. For example, for the fifth template encoding representation containing temperature, humidity and device power encoding information, the computer system may extract the correlation features between temperature and humidity, such as the change trend of humidity when the temperature is in a certain range; it may also extract the relationship features between device power and temperature change, such as the rate of temperature increase when the device power increases, etc. These are the fifth template high-level features.

[0153] In step S23, the computer system obtains x basic professional adaptation models corresponding to x operating modes, performs pattern recognition on the fifth template high-level features respectively, and obtains x pattern recognition confidences corresponding to the x basic professional adaptation models, wherein one operating mode corresponds to one basic professional adaptation model, and one basic professional adaptation model is used to determine one pattern recognition confidence.

[0154] These basic professional adaptation models are initially constructed for each operation mode, aiming to identify the environmental status under different operation modes. Each basic professional adaptation model has its own specific structure and parameters, which is used to process the data characteristics related to a specific operation mode.

[0155] For example, for the basic professional adaptation model corresponding to the preoperative preparation mode, when the fifth template advanced features (including temperature, humidity, equipment power and other related features in the preoperative preparation mode) are input, the model will perform pattern recognition based on its internal algorithms and parameters. Assume that this model determines that the current data belongs to the preoperative preparation mode with a confidence of 80% by analyzing the features in the advanced features such as the equipment power being in the low power range and the temperature gradually rising. This 80% is the pattern recognition confidence corresponding to the preoperative preparation mode.

[0156] For the basic professional adaptation model corresponding to the surgical procedure mode, when the fifth template high-level features under the surgical procedure mode are input (such as precise temperature, humidity, high-power stable operation of equipment and other related features), if the model judges that the data belongs to the surgical procedure mode based on these features with a confidence level of 90%, this is the pattern recognition confidence level corresponding to the surgical procedure mode.

[0157] Finally, in step S24, the computer system determines the second target calibration error based on x pattern recognition confidences and x pattern prior values, optimizes the model parameters of the x basic professional adaptation models based on the second target calibration error, determines the optimized x basic professional adaptation models as x professional adaptation models, and determines the second pattern model based on the x professional adaptation models.

[0158] The second objective calibration error is an indicator that measures the difference between the basic professional adaptation model and the pattern prior value. The computer system calculates this error by comparing the pattern recognition confidence and the pattern prior value.

[0159] For example, for the preoperative preparation mode, it is assumed that the mode prior value indicates that the device power in this mode should be between 100-200 watts (this is determined based on a lot of experience and prior knowledge), and the basic professional adaptation model identifies that the device power is within this range with a confidence level of 80% based on the fifth template advanced features. If a certain error calculation method (such as mean square error) is used, the error contribution of this part can be calculated. Similar calculations are performed for other operating modes, and the error contributions in all operating modes are combined to obtain the second target calibration error.

[0160] According to this second target adjustment error, the computer system optimizes the model parameters of the basic professional adaptation model. Model parameters include weights, biases and other parameters in the model. If the second target adjustment error is large, it means that the performance of the basic professional adaptation model is poor, and these parameters need to be adjusted to a large extent. For example, for a basic professional adaptation model based on a neural network, if the error is large, it may increase or decrease the weight value of the connection between the layers of the neural network, adjust the bias term, and so on.

[0161] Through continuous optimization, when the performance of these basic professional adaptation models meets the requirements, the optimized basic professional adaptation models are determined as professional adaptation models, and these professional adaptation models together constitute the second mode model. This second mode model can more accurately identify and monitor the operating room environment status under different operating modes, providing important guarantees for the entire intelligent operating room environment status monitoring system.

[0162] As an implementation mode, the coded representation is obtained by array encoding of data segments by an encoder in a first mode model; the data segments are obtained by data segmentation of target-related monitoring data by a data acquisition network in the first mode model; the first mode model is a target monitoring model for monitoring the environmental status of a target operating room; the target monitoring model also includes a third mode model sequentially connected to the encoder; based on this, step S400, extracting the contribution of the coded representation to obtain the contribution corresponding to each operating mode, including:

[0163] Step S410: determining a third mode model in the target monitoring model;

[0164] Step S420: extract the contribution of the coded representation based on the third mode model to obtain the contribution corresponding to each operating mode.

[0165] In step S410, the computer system determines the third mode model in the target monitoring model. The target monitoring model is a comprehensive model for monitoring the environmental status of the target operating room. It contains multiple sub-models, and the third mode model is a part of it. This third mode model is specially constructed for contribution extraction. It has a specific structure and function, and cooperates with other parts in the target monitoring model to jointly realize comprehensive monitoring of the environmental status of the operating room.

[0166] For example, the target monitoring model may be a complex hierarchical structure model, including multiple layers such as data collection layer, data preprocessing layer, feature extraction layer, etc. The third mode model may be located at a certain stage of feature processing or analysis. It is connected to the previous model part (such as the part for data encoding and high-level feature extraction), receives the processed encoding representation as input, and then performs contribution extraction operation.

[0167] In step S420, the computer system extracts the contribution of the coded representation based on the third mode model to obtain the contribution corresponding to each operation mode. The coded representation is a specific representation of the monitoring data related to the target operating room obtained through a series of processing, which contains the coded content of temperature, humidity, air quality, operating status of surgical equipment and other information.

[0168] Contribution extraction is the process of determining the contribution of each part or feature in the coded representation to each operating mode. Taking the temperature data in the operating room as an example, the contribution of temperature data to the environmental state may be different in different operating modes.

[0169] In the preoperative preparation operation mode, it is assumed that the coded representation includes information such as temperature, humidity, and equipment preparation status. For this operation mode, the temperature may have a certain contribution range. If the temperature between 18-22°C is a more suitable preoperative preparation environment temperature, when the actual temperature is close to this range, the part of this temperature data in the coded representation may have a higher contribution to the environmental state in the preoperative preparation operation mode. For example, if the current temperature is 20°C, the computer system finds through the third mode model analysis that the contribution of the relevant part of this temperature data in the coded representation to the good environmental state in the preoperative preparation operation mode (such as the equipment can be preheated normally, the medical staff feel comfortable, etc.) may be evaluated as 0.6 (the 0.6 here is a hypothetical contribution value, indicating a relatively high degree of contribution).

[0170] In the operation mode of surgery, the environmental requirements are more stringent. Assume that the temperature is required to be precisely controlled at 20-21°C, the humidity is 45%-50%, and the equipment is in a stable operating state during the operation. At this time, a slight change in the temperature data in the coding representation may have a greater impact on the environmental state of the operation. If the actual temperature is 20.5°C, the computer system may conclude through the third mode model analysis that the corresponding part of this temperature data in the coding representation contributes 0.8 to the stability of the environmental state in the operation mode of surgery (such as ensuring that the performance of surgical instruments is not affected by temperature, the patient's physiological state is stable, etc.), which shows that the contribution of temperature data in this operation mode is relatively higher because it is more critical to maintaining an ideal surgical environment.

[0171] Other operating modes are not given as examples. In addition to temperature data, other data such as the operating status of surgical equipment also have different contributions in different operating modes. In the preoperative preparation operating mode, the equipment is in a low-power preheating or inspection state, and the contribution of the equipment operating status data in the coded representation to the overall environmental state may be mainly reflected in whether the equipment is ready. If the various parameters of the equipment are normal and in the expected preheating state, its contribution may be evaluated as 0.7, indicating that the equipment status is more important to the preoperative preparation environment status.

[0172] By extracting the contribution of the coding representation based on the third mode model, the computer system can clearly understand the contribution of different data parts (represented by the coding representation) to the environmental state in each operating mode. This helps to consider the impact of various factors more specifically according to the contribution under different operating modes in the subsequent comprehensive evaluation of environmental status and decision-making (such as resource allocation, exception handling, etc.). For example, if the contribution of a factor is found to suddenly decrease in the operating mode of surgery, it may indicate that there is a potential risk in the environmental state and timely measures need to be taken to adjust it. At the same time, the extraction of this contribution also provides a basis for further optimization of the model. For example, the model's attention to different data features can be adjusted according to the size of the contribution, thereby improving the accuracy and effectiveness of the model's monitoring of the operating room environmental status.

[0173] As an implementation mode, the target monitoring model also includes a second mode model for determining the prediction confidence of the environmental state corresponding to each operating mode; based on this, the method also includes:

[0174] Step S31: obtaining a third calibration template library, wherein the third calibration template contained in the third calibration template library carries an environment state tag;

[0175] Step S32: Acquire the calibrated first mode model, perform data encoding on the third calibrated template based on the first mode model, obtain a sixth template encoding representation of the third calibrated template, perform high-level feature extraction on the sixth template encoding representation, and obtain high-level features of the sixth template corresponding to the sixth template encoding representation;

[0176] Step S33: obtaining the calibrated second mode model, and performing environmental state prediction on the sixth template high-level features based on the x professional adaptation models in the second mode model to obtain the template environmental state prediction confidence corresponding to each operation mode, wherein one professional adaptation model is used to determine the template environmental state prediction confidence corresponding to one operation mode;

[0177] Step S34: extracting the contribution of the sixth coded representation through the third basic mode model to obtain the template contribution corresponding to each operation mode;

[0178] Step S35: determining the template environment state monitoring result of the template operating room related to the third calibration template according to the template environment state prediction confidence corresponding to each operation mode and the template contribution corresponding to each operation mode;

[0179] Step S36: Determine the third target calibration error according to the environmental state mark and the template environmental state monitoring result, optimize the model parameters of the third basic mode model according to the third target calibration error, use the optimized third basic mode model as the third mode model, and determine the target monitoring model according to the first mode model, the second mode model and the third mode model.

[0180] In step S31, the computer system obtains a third calibration template library, wherein the third calibration templates contained in the third calibration template library carry environmental status tags. The third calibration template library is a data set used to train and optimize the target monitoring model. These third calibration templates contain rich information related to the environmental status of the operating room, such as temperature, humidity, air quality, operating status of surgical equipment, etc., and clearly mark the environmental status.

[0181] In step S32, the computer system obtains the calibrated first pattern model, performs data encoding on the third calibrated template based on the first pattern model, obtains a sixth template encoding representation of the third calibrated template, extracts high-level features from the sixth template encoding representation, and obtains high-level features of the sixth template corresponding to the sixth template encoding representation.

[0182] The first mode model has been trained and adjusted before and has a certain data processing capability. When the third adjustment template is encoded, the various complex data therein are converted into a specific coding representation, that is, the sixth template coding representation. Taking humidity data as an example, if the humidity value range is 0-100%, the first mode model may divide this range into several intervals and encode according to the interval in which the humidity value is located. For example, 0-20% is encoded as 00, 20-40% is encoded as 01, etc. For surgical equipment operating status data, if the equipment power range is 0-1000 watts, it is also encoded according to certain rules, such as 0-200 watts is encoded as 10, 200-400 watts is encoded as 11, etc., so that the sixth template coding representation of the third adjustment template is obtained.

[0183] Then, high-level feature extraction is performed on the sixth template coding representation. Assuming that the sixth template coding representation contains coding information of temperature and humidity, the computer system may extract the correlation features between the two, such as the coordinated change trend of temperature and humidity under certain coding combinations. For the case of including the coding of device power and running time, the relationship features between the running time of the device at different power and the environmental stability may be extracted, etc. These are the high-level features of the sixth template.

[0184] In step S33, the computer system obtains the calibrated second mode model, predicts the environmental state of the sixth template high-level features based on the x professional adaptation models in the second mode model, and obtains the template environmental state prediction confidence corresponding to each operating mode, wherein a professional adaptation model is used to determine the template environmental state prediction confidence corresponding to an operating mode.

[0185] The professional adaptation model in the second mode model is specially constructed for different operation modes. For example, in the professional adaptation model corresponding to the preoperative preparation operation mode, when the advanced features of the sixth template (including temperature, humidity, equipment status and other features related to the preoperative preparation stage) are input, the model will predict the environmental state corresponding to the template based on its internal algorithm and parameters, and give a confidence level. Assume that the model predicts that the environmental state is good with a confidence level of 0.8 based on features such as the temperature being in an appropriate preoperative preparation range and the equipment being in a low-power preheating state. For the professional adaptation model of the surgical operation mode, after inputting the advanced features of the sixth template related to the surgical stage, it may predict that the environmental state is normal with a confidence level of 0.9 based on factors such as more stringent temperature, humidity and equipment stable operation requirements.

[0186] In step S34, the computer system extracts the contribution of the sixth coding representation through the third basic mode model to obtain the template contribution corresponding to each operation mode. The third basic mode model is an initial model structure for extracting the contribution.

[0187] In step S35, the computer system determines the template environment state monitoring result of the template operating room related to the third calibration template according to the template environment state prediction confidence corresponding to each operation mode and the template contribution corresponding to each operation mode. For example, in the preoperative preparation operation mode, the template environment state prediction confidence is 0.8, and the template contribution is 0.3. Assume that there are other factors such as temperature, equipment status, etc. corresponding to the prediction confidence and contribution. The computer system comprehensively considers these factors and determines the template environment state monitoring result of the template operating room in the preoperative preparation operation mode through a specific calculation method (such as weighted summation, etc.). If the calculation result shows that the overall environmental state is at a good level, this is the template environment state monitoring result under this operation mode.

[0188] Finally, in step S36, the computer system determines the third target calibration error based on the environmental state mark and the template environmental state monitoring results, optimizes the model parameters of the third basic mode model based on the third target calibration error, uses the optimized third basic mode model as the third mode model, and determines the target monitoring model based on the first mode model, the second mode model and the third mode model.

[0189] If the environmental state in the third calibration template is marked as good, but the template environmental state monitoring result is calculated as fair, there is a difference. The computer system calculates the third target calibration error based on this difference. For example, the error is quantified by calculating a certain distance metric (such as mean square error, etc.) between the two. Based on this error, the computer system adjusts the model parameters of the third basic mode model, such as changing the weights, biases and other parameters in the model. If the error is large, the adjustment may be large; if the error is small, the adjustment is small. The optimized third basic mode model becomes the third mode model.

[0190] The target monitoring model is composed of the first mode model, the second mode model and the third mode model. The first mode model is responsible for the preliminary processing such as data encoding and high-level feature extraction, the second mode model predicts the environmental status under different operating modes, and the third mode model extracts the contribution. The three models work together to enable the target monitoring model to monitor the operating room environmental status comprehensively and accurately.

[0191] As an implementation mode, the third mode model includes a feature allocation model and a weight allocation model; based on this, step S420, extracting the contribution of the coded representation based on the third mode model to obtain the contribution corresponding to each operating mode, includes:

[0192] Step S421: extracting distribution features from the coded representation based on the feature distribution model to obtain distribution features corresponding to the coded representation;

[0193] Step S422: Identify the contribution of the distribution features based on the weight distribution model to obtain the contribution corresponding to each operation mode.

[0194] In step S421, the computer system extracts the distribution features of the coded representation based on the feature distribution model in the third mode model to obtain the distribution features corresponding to the coded representation. The third mode model is responsible for the contribution extraction related operations in the target monitoring model, and the feature distribution model is an important part of the third mode model.

[0195] Taking the temperature and humidity data of the operating room as an example, suppose that the temperature and humidity data are encoded in some form in the encoding representation. The feature allocation model processes this encoding representation according to its internal algorithm and predefined rules to extract the allocation features. For example, the feature allocation model may identify specific patterns or relationships between the temperature and humidity data in the encoding representation, which reflect the potential factors that affect the environmental state under different operating modes.

[0196] In the preoperative preparation operation mode, the range of temperature and humidity may be relatively loose, but the relationship between them may have an impact on the preheating of equipment and the initial comfort of medical staff. The feature allocation model may extract the difference features of temperature and humidity, the consistency features of the change trend, etc. from the encoding representation as allocation features. For example, if the temperature is gradually rising and the humidity remains stable, this feature may imply that the environment is in a normal preheating state, which is an allocation feature.

[0197] In the operation mode of surgery, the requirements for temperature and humidity are more precise. The feature allocation model may extract the characteristics of the proximity of temperature and humidity to the optimal surgical environment setting value from the coding representation. For example, the coding representation reflects the degree to which the temperature is close to the ideal surgical temperature (such as 20-21°C) and the humidity is close to the optimal humidity range (such as 45%-50%). These features are the allocation features for the operation mode of surgery.

[0198] Other operating modes are not described in examples.

[0199] In step S422, the computer system identifies the contribution of the distribution features based on the weight distribution model in the third mode model to obtain the contribution corresponding to each operating mode. The weight distribution model is also part of the third mode model, and its function is to determine the contribution of each feature to the environmental state in different operating modes according to the distribution features.

[0200] Let's continue with the temperature and humidity example. In the preoperative preparation operation mode, assume that the distribution features obtained from the feature distribution model include the difference features of temperature and humidity and the temperature change trend features. The weight distribution model will assign weights to these distribution features according to pre-set rules or relationships learned based on historical data to determine their contribution.

[0201] For example, if historical data indicates that during the preoperative preparation stage, the temperature trend has a relatively large impact on the environmental state, while the temperature and humidity difference has a relatively small impact, then the weight allocation model may assign a higher weight, such as 0.6, to the temperature trend feature and a lower weight, such as 0.2, to the temperature and humidity difference feature. Other unmentioned allocation features may be assigned the remaining weight of 0.2 in total. Through this weight allocation, the computer system can determine the contribution of each allocation feature to the environmental state in the preoperative preparation operation mode.

[0202] In the operation mode of surgery, for the characteristics of the proximity of temperature and humidity in the distribution characteristics to the optimal surgical environment setting value, the weight distribution model will allocate weights according to the relationship between surgical success and environmental conditions. Assuming that after a large amount of statistical analysis of surgical data, it is found that the degree of temperature close to the ideal surgical temperature has an impact weight of 0.7 on surgical success, and the degree of humidity close to the optimal humidity range has an impact weight of 0.3 on surgical success, then these are their contributions to the environmental conditions in the operation mode of surgery.

[0203] Through steps S421-S422, the computer system can accurately extract the distribution features corresponding to each operating mode from the encoded representation based on the third mode model, and determine the contribution of these features. This helps to understand the degree of influence of different factors on the environmental state under various operating modes more comprehensively and meticulously in the entire intelligent operating room environmental state monitoring system. Based on these contributions, the system can make more reasonable decisions in environmental state monitoring, anomaly detection, resource allocation, etc. For example, if the contribution of a factor is found to suddenly drop during the operation, it may indicate that there is a risk in the environmental state and timely measures need to be taken to adjust it; or in terms of resource allocation, the resource requirements related to factors that have a greater impact on the environmental state can be prioritized according to the contribution of different factors.

[0204] As an implementation mode, the x operating modes include a target operating mode; then, the method may further include:

[0205] Step S41: obtaining a target monitoring model for monitoring the environmental status of a target operating room, and obtaining a basic target model for performing migration training on the target monitoring model;

[0206] Step S42: acquiring a fourth calibration template library related to the target operation mode, wherein the fourth calibration template library includes a fourth calibration template;

[0207] Step S43: processing the fourth calibration template based on the target monitoring model to obtain an environmental state monitoring result corresponding to the fourth calibration template, and determining the environmental state monitoring result corresponding to the fourth calibration template as a virtual template mark for calibrating the basic target model;

[0208] Step S44: processing the fourth calibration template based on the basic target model to obtain a migration environment state monitoring result corresponding to the fourth calibration template;

[0209] Step S45: Optimize the parameters of the basic target model according to the migration environment status monitoring results and the virtual template markings, and use the optimized basic target model as the final target model of the target monitoring model, wherein the final target model is used to perform environmental status monitoring on the target-related monitoring data under the target operation mode.

[0210] In step S41, the target monitoring model is obtained through a series of previous construction and training processes, and contains a variety of components and structures, which are intended to accurately monitor the environmental status of the operating room. At the same time, the computer system obtains a basic target model for transfer training (i.e., distillation training) in the target monitoring model. The basic target model is an initial model structure, which may have a certain degree of versatility, but may not yet achieve optimal performance in the environmental status monitoring task under the target operating mode specific to the target operating room. For example, the basic target model may be an environmental monitoring model framework obtained through general data training, and its initial parameter values ​​are set based on a wide range of environmental monitoring scenarios, but in the specific scenario of the target operating room, it is necessary to adapt to the special requirements of the target operating mode through transfer training.

[0211] The purpose of step S42 is to obtain data resources specific to the target operating mode so as to subsequently optimize the basic target model.

[0212] In step S43, the computer system processes the fourth calibration template based on the target monitoring model to obtain the environmental status monitoring result corresponding to the fourth calibration template, and determines this result as a virtual template tag (this is a pseudo tag) for calibrating the basic target model. After the previous construction and training, the target monitoring model has the ability to monitor the environmental status of the operating room. When the fourth calibration template (including the mode calibration template and the enhanced calibration template) is processed, it will obtain a monitoring result about the environmental status of the operating room according to various components and algorithms inside the model, such as data encoding, high-level feature extraction, environmental status prediction and other operations. This result may include information such as the predicted values ​​of various environmental parameters and the confidence of different target warning states. For example, the predicted value of the temperature in the operating room is 22°C, the predicted value of the humidity is 50%, and the confidence of the warning state of a certain equipment failure is 30%. These results are used as virtual template tags to provide reference standards for the optimization of the basic target model.

[0213] In step S44, the computer system processes the fourth calibration template based on the basic target model to obtain the migration environment status monitoring result corresponding to the fourth calibration template. The basic target model processes the data in the fourth calibration template according to its own structure and current parameter values. For example, the basic target model may pre-process the data according to its pre-set data processing flow, and then extract and analyze features through the internal neural network layer to finally obtain a prediction result about the operating room environment status. This result may differ from the result obtained by the target monitoring model in step S43, because the basic target model has not been optimized for the target operation mode, and its parameters and structure may not be fully suitable for the environmental status monitoring task under the target operation mode.

[0214] In step S45, the computer system optimizes the parameters of the basic target model according to the migration environment state monitoring results and the virtual template label. Specifically, the computer system compares the difference between the migration environment state monitoring results obtained by the basic target model and the virtual template label (pseudo label) provided by the target monitoring model. For example, if in terms of temperature prediction, the temperature value predicted by the basic target model is 20°C, and the temperature value in the virtual template label is 22°C, there is a difference of 2°C. The computer system will adjust the parameters of the basic target model according to this difference and a pre-set optimization algorithm, such as a back propagation algorithm.

[0215] As an implementation manner, the fourth calibration template includes a mode calibration template under a target operating mode; based on this, step S42, obtaining a fourth calibration template library related to the target operating mode, includes:

[0216] Step S421: obtaining a mode adjustment template under a target operation mode;

[0217] Step S422: determining a calibration template to be identified in a calibration template library used for template screening;

[0218] Step S423: performing commonality evaluation on the calibration template to be identified and the mode calibration template based on the template generation model to obtain a commonality measurement result between the mode calibration template and the calibration template to be identified;

[0219] Step S424: if the commonality measurement result meets the template enhancement requirement, the template to be identified and adjusted is used as an enhanced and adjusted template related to the target operation mode;

[0220] Step S425: Use the enhanced tuning template and the mode tuning template as the fourth tuning template associated with the target operating mode.

[0221] In step S421, the computer system obtains the mode calibration template under the target operating mode. The mode calibration template is a data template constructed for a specific target operating mode, and its purpose is to provide a specific data structure and information reference for operating room environmental status monitoring under the target operating mode. In the context of operating room environment monitoring, the target operating mode may be, for example, a cardiac surgery mode. In this mode, the mode calibration template contains data information closely related to the cardiac surgery environment. For example, the temperature requirements for the operating room may be very strict, and usually need to be maintained between 22-24 degrees Celsius to ensure the stability of the patient's physiological state during the operation and the normal operation of the surgical equipment. The mode calibration template will clearly record this temperature range as a key data feature.

[0222] In step S422, the computer system determines the adjustment template to be identified in the adjustment template library used for template screening. The adjustment template library is a collection of many different types of adjustment templates, which cover various possible operating room environmental status conditions and different surgical types, equipment configurations and other factors. For example, this adjustment template library may contain adjustment templates corresponding to different surgical types such as general surgery, neurosurgery, and obstetrics and gynecology. It also includes environmental status data templates corresponding to different hospital sizes and different equipment brands and models. The computer system determines the adjustment template to be identified based on certain algorithms and rules. For example, screening may be based on preliminary correlation features with the target operating mode.

[0223] In step S423, the computer system evaluates the commonality of the calibration template to be identified and the pattern calibration template based on the template generation model, and obtains the commonality measurement result between the pattern calibration template and the calibration template to be identified (for example, calculating the Euclidean distance between the two to obtain the similarity as the commonality measurement result). The template generation model is a model specifically used to analyze the similarities and differences between templates.

[0224] In step S424, if the commonality measurement result meets the template enhancement requirement, the template to be identified and adjusted is used as an enhanced and adjusted template related to the target operation mode. The template enhancement requirement is a pre-set standard used to determine whether the template to be identified and adjusted has sufficient supplementary value for environmental status monitoring under the target operation mode.

[0225] In step S425, the computer system uses the enhanced adjustment template and the mode adjustment template as the fourth adjustment template related to the target operation mode. The purpose of this step is to build a more comprehensive and rich data resource set for environmental status monitoring in the target operation mode.

[0226] For the target operation mode (such as cardiac surgery mode), the mode calibration template itself provides basic environmental status data information for this specific operation mode. The enhanced calibration template (such as the calibration template corresponding to the neurosurgery operation mentioned above) brings some additional similar but complementary data features and information. Combining the two as the fourth calibration template allows the computer system to have more reference data when optimizing the basic target model and monitoring the environmental status under the target operation mode, thereby improving the accuracy and applicability of the model.

[0227] As an implementation manner, in step S500, the environmental state monitoring result of the target operating room is determined according to the environmental state prediction confidence corresponding to each operating mode and the contribution corresponding to each operating mode, including:

[0228] Step S510: adjusting the environmental state prediction confidence corresponding to each operating mode based on the contribution corresponding to each operating mode to obtain an adjusted environmental state prediction value corresponding to each operating mode, and a weighted projection value corresponding to an operating mode is used to adjust the environmental state prediction confidence corresponding to the corresponding operating mode;

[0229] Step S520: Determine the environmental status monitoring result of the target operating room according to the adjustment environmental status prediction value corresponding to each operation mode.

[0230] In step S510, the computer system adjusts (i.e., weights) the environmental state prediction confidence corresponding to each operating mode based on the contribution corresponding to each operating mode to obtain the adjusted environmental state prediction value corresponding to each operating mode. The contribution corresponding to each operating mode here reflects the relative importance or influence of the operating mode in the overall environmental state monitoring. For example, assume that there are two operating modes, mode A and mode B. Mode A may be the operating mode during the main surgical operation, and mode B may be the operating mode during the postoperative care stage.

[0231] In the environmental status monitoring of the operating room, for some key environmental factors, the impact weights in different operating modes are different. For example, for the temperature monitoring of the operating room, in mode A, due to the high precision requirements of the surgical operation on the patient's physiological state and surgical equipment, the contribution of temperature accuracy to the overall environmental status judgment may be relatively high, assuming 0.8. In mode B, the impact of temperature fluctuations in the postoperative care stage on the overall environmental status is relatively small, and the contribution may be 0.3.

[0232] At the same time, the prediction confidence of the environmental state corresponding to each operating mode is calculated by the model before, which indicates the reliability of the predicted environmental state under the operating mode. For example, for mode A, the computer system obtains the prediction confidence of 0.9 that the current temperature is in the appropriate range (22-24 degrees Celsius) through a series of data analysis and model calculations; for mode B, the prediction confidence of the same temperature range is 0.7.

[0233] The computer system uses the contribution of each operating mode as a weight to adjust the corresponding environmental state prediction confidence. For mode A, the adjusted environmental state prediction value is 0.8×0.9=0.72; for mode B, the adjusted environmental state prediction value is 0.3×0.7=0.21.

[0234] In step S520, the computer system determines the environmental status monitoring result of the target operating room according to the predicted value of the adjusted environmental status corresponding to each operating mode. This environmental status monitoring result includes the confidence level of each target warning status. Continuing with the above-mentioned mode A and mode B as examples, in addition to the temperature factor, the environmental status of the operating room also involves multiple factors such as humidity, air quality, and equipment operating status.

[0235] Assume that for humidity factor, the contribution of mode A is 0.7, the confidence of predicting humidity in the appropriate range (40%-60%) is 0.85, and the adjusted prediction value is 0.7×0.85=0.595; the contribution of mode B is 0.4, the prediction confidence is 0.75, and the adjusted prediction value is 0.4×0.75=0.3. For oxygen concentration factor in air quality, the contribution of mode A is 0.6, the confidence of predicting oxygen concentration to be normal (20.9%-21%) is 0.92, and the adjusted prediction value is 0.6×0.92=0.552; the contribution of mode B is 0.3, the prediction confidence is 0.88, and the adjusted prediction value is 0.3×0.88=0.264.

[0236] The computer system combines all these adjusted environmental state prediction values ​​to determine the confidence of the target warning state. For example, if it is set that when the adjusted prediction value of a certain environmental factor is lower than 0.3, a low-level warning may be triggered; when it is lower than 0.1, a high-level warning is triggered. For the equipment operation status, if the equipment is a heart extracorporeal circulation machine, its adjusted prediction value for normal operation in mode A is 0.8 (hypothesis), indicating that the reliability of the normal operation of the equipment is high; if in mode B, due to equipment maintenance inspection and other operations, its adjusted prediction value for normal operation is 0.6, although it still indicates that the equipment is likely to operate normally, but the reliability is lower than that of mode A.

[0237] The present application embodiment provides a computer system, such as Figure 2As shown, the computer system 100 includes: a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, such as through a bus 102. Optionally, the computer system 100 may also include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation on the embodiments of the present application.

[0238] An embodiment of the present application provides a computer system, and the computer system in the embodiment of the present application includes: one or more processors; a memory; one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more programs are executed by the processor, the above-mentioned intelligent operating room environment status monitoring method based on Internet of Things sensing is implemented.

Claims

1. A method for monitoring the environmental status of an intelligent operating room based on IoT sensing, characterized in that: The method includes: when acquiring target-related monitoring data of a target operating room, acquiring a target monitoring model for monitoring the environmental status of the target operating room, wherein the target monitoring model includes a first pattern model; determining a monitoring dimension corresponding to the target-related monitoring data based on the first pattern model, and extracting split dimension segments of the target-related monitoring data according to the monitoring dimension; splitting the split dimension segments based on the first pattern model to obtain split data corresponding to the split dimension segments; fragmenting the split data corresponding to the split dimension segments based on the first pattern model to obtain data segments of the target-related monitoring data; and array encoding the data segments. , obtain the coded representation of the data segment; extract high-level features from the coded representation to obtain high-level features corresponding to the coded representation; determine x operating modes corresponding to the target-related monitoring data, predict the environmental state of the high-level features based on each operating mode, and obtain the environmental state prediction confidence corresponding to each operating mode; x≥1; extract the contribution of the coded representation to obtain the contribution corresponding to each operating mode; determine the environmental state monitoring result of the target operating room according to the environmental state prediction confidence corresponding to each operating mode and the contribution corresponding to each operating mode, and the environmental state monitoring result includes the confidence of each target warning state.

2. The method according to claim 1, characterized in that The first pattern model includes a data acquisition network; the split dimension segments include y split dimension segments, and the y split dimension segments include a target split dimension segment, y≥1; the splitting of the split dimension segments based on the first pattern model to obtain the split data corresponding to the split dimension segments includes: determining the projection domain corresponding to the target split dimension segment according to the monitoring dimension corresponding to the target split dimension segment; the projection domain includes multiple projection values, each projection value has a corresponding split dimension data range; determining the split dimension data range corresponding to the target split dimension segment based on the data acquisition network, and determining the projection value corresponding to the split dimension data range corresponding to the target split dimension segment as the split data corresponding to the target split dimension segment.

3. The method according to claim 1, characterized in that The high-level features are obtained by extracting high-level features from the encoded representation by an embedding mapping component in the first mode model; The first pattern model is a target monitoring model used to monitor the environmental status of the target operating room; the method also includes: obtaining a first calibration template library; the first calibration template contained in the first calibration template library is a calibration template that does not carry an environmental status mark; obtaining s calibration targets for the first basic pattern model, s≥1; performing high-level feature extraction on the first calibration template based on the first basic pattern model to obtain high-level template features associated with each calibration target, and determining s calibration errors corresponding to the s calibration targets based on the high-level template features, wherein one calibration target corresponds to one calibration error; determining a first target calibration error based on the s calibration errors, optimizing the model parameters of the first basic pattern model based on the first target calibration error, and using the optimized first basic pattern model as the first pattern model.

4. The method according to claim 3, characterized in that The s calibration targets include a data item occlusion recognition branch; the high-level feature extraction of the first calibration template based on the first basic pattern model to obtain the high-level template features associated with each calibration target, and the s calibration errors corresponding to the s calibration targets are determined according to the high-level template features, including: data fragmentation of the first calibration template based on the first basic pattern model to obtain a first calibration template fragment of the first calibration template; the first calibration template fragment contains template splitting data at multiple fragment positions; among the multiple fragment positions corresponding to the first calibration template fragment, arbitrarily select the first fragment position to be subjected to data item occlusion; and perform data item occlusion processing on the template splitting data at the first fragment position in the first calibration template fragment based on the data item occlusion information. Processing, obtain first template fragment data; based on the first basic pattern model, array encode the first template fragment data to obtain the first template encoding representation of the first template fragment data, perform high-level feature extraction on the first template encoding representation to obtain the first template high-level feature corresponding to the first template encoding representation, and the first template high-level feature belongs to the template high-level feature; based on the monitoring network related to the data item occlusion recognition branch, perform occlusion data item recognition on the first template high-level feature to obtain the inference split data corresponding to the data item occlusion information in the first template fragment data; determine the calibration error corresponding to the data item occlusion recognition branch according to the template split data at the first fragment position in the first calibration template fragment and the inference split data.

5. The method according to claim 3, characterized in that: The s calibration targets include a data item change identification branch; the high-level feature extraction of the first calibration template based on the first basic pattern model to obtain the high-level template features associated with each calibration target, and the s calibration errors corresponding to the s calibration targets are determined according to the high-level template features, including: data fragmentation of the first calibration template based on the first basic pattern model to obtain a first calibration template fragment of the first calibration template, the first calibration template fragment containing template splitting data at multiple fragment positions; arbitrarily selecting a second fragment position to be subjected to data item change from the multiple fragment positions corresponding to the first calibration template fragment; performing data item change processing on the template splitting data at the second fragment position in the first calibration template fragment based on an arbitrary numerical value to obtain a second template plate fragment data, the arbitrary numerical value is different from the template splitting data at the second fragment position; based on the first basic pattern model, the second template fragment data is array-encoded to obtain the second template encoding representation of the second template fragment data, and the second template encoding representation is extracted to obtain the second template high-level feature corresponding to the second template encoding representation, wherein the second template high-level feature belongs to the template high-level feature; based on the monitoring network related to the data item change identification branch, the second template high-level feature is inferred to be changed position, and the data item change inference results corresponding to the multiple fragment positions are obtained; according to the data item change inference results corresponding to the second fragment position and the multiple fragment positions, the calibration error corresponding to the data item change identification branch is determined.

6. The method according to claim 3, characterized in that The s calibration targets include a perturbation fragment branch; the first calibration template is extracted with advanced features based on the first basic pattern model to obtain the template advanced features associated with each calibration target, and the s calibration errors corresponding to the s calibration targets are determined according to the template advanced features, including: perturbation injection into the split dimension fragments of the first calibration template to obtain a perturbation calibration template, data segmentation of the perturbation calibration template to obtain a perturbation data fragment of the perturbation calibration template; data segmentation of the first calibration template to obtain a first calibration template fragment of the first calibration template; array encoding of the perturbation data fragment based on the first basic pattern model, Obtain a third template encoding representation of the perturbation data segment, perform high-level feature extraction on the third template encoding representation, and obtain a third template high-level feature corresponding to the third template encoding representation, wherein the third template high-level feature belongs to the template high-level feature; perform array encoding on the first adjustment template segment based on the first basic mode model, obtain a fourth template encoding representation of the first adjustment template segment, perform high-level feature extraction on the fourth template encoding representation, and obtain a fourth template high-level feature corresponding to the fourth template encoding representation, wherein the fourth template high-level feature belongs to the template high-level feature; determine according to the third template high-level feature and the fourth template high-level feature The method comprises: extracting high-level features of the first calibration template based on the first basic mode model to obtain high-level features of the template associated with each calibration target, and determining the s calibration errors corresponding to the s calibration targets according to the high-level features of the template, including: performing data fragmentation on the first calibration template to obtain a first calibration template fragment of the first calibration template; performing array encoding on the first calibration template fragment based on the first basic mode model to obtain a fourth template encoding representation of the first calibration template fragment, performing high-level feature extraction on the fourth template encoding representation to obtain the calibration errors corresponding to the s calibration targets. The fourth template encoding represents the corresponding fourth template high-level feature, wherein the fourth template high-level feature belongs to the template high-level feature; the fourth template high-level feature is classified into environmental states through the monitoring network related to the virtual tag branch, and the template environmental state classification confidence corresponding to the first calibration template is obtained; the historical model related to the first basic pattern model is obtained, and the environmental state of the first calibration template is classified according to the historical model to obtain the virtual tag environmental state prediction confidence corresponding to the first calibration template; the calibration error corresponding to the virtual tag branch is determined according to the template environmental state classification confidence and the virtual tag environmental state prediction confidence.

7. The method according to claim 1, characterized in that The high-level features are obtained by extracting high-level features from the encoded representation through the embedded mapping component in the first model model; the first model model is a target monitoring model used to monitor the environmental status of the target operating room; the target monitoring model also includes a second model model sequentially connected to the embedded mapping component; the environmental status prediction of the high-level features based on each operating mode is performed to obtain the environmental status prediction confidence corresponding to each operating mode, including: determining a second model model in the target monitoring model, wherein the second model model includes a professional adaptation model corresponding to each operating mode; based on the professional adaptation model corresponding to each operating mode, the environmental status prediction of the high-level features is performed to obtain the environmental status prediction confidence corresponding to each operating mode, wherein a professional adaptation model is used to determine the environmental status prediction confidence corresponding to an operating mode.

8. A computer system, characterized in that: include: one or more processors; Memory; One or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method according to any one of claims 1 to 7 is implemented.

Images