Intelligent analysis method for disinfection quality of equipment for department of stomatology
By combining sensors and image acquisition modules with deep learning models, an intelligent analysis system was built, which solved the problems of accuracy and efficiency in the quality detection of dental instrument disinfection, and realized comprehensive and real-time monitoring and evaluation of the disinfection process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-30
- Publication Date
- 2026-03-31
AI Technical Summary
In existing technologies, the quality testing of disinfection of dental instruments suffers from problems such as easy errors in manual testing results, incomplete testing, long testing cycles, and inability to achieve comprehensive analysis, resulting in low accuracy of disinfection quality analysis.
The system uses sensors and image acquisition modules to collect various data during the disinfection process. Through data preprocessing, entropy weighting fusion, and deep learning models, a comprehensive analysis is performed to build an intelligent analysis system that enables real-time assessment of disinfection quality.
It improves the accuracy and efficiency of disinfection quality testing, reduces the testing cycle, enables comprehensive monitoring and real-time analysis of the disinfection process, and avoids errors caused by manual testing.
Smart Images
Figure CN121765614A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing, and in particular to an intelligent analysis method for the disinfection quality of dental instruments. Background Technology
[0002] Dental instruments (such as dental drills, extraction forceps, root canal files, etc.) come into direct contact with the patient's oral mucosa and tissues, and the quality of their disinfection is a key factor in preventing cross-infection. According to the national "Technical Operation Specifications for Disinfection and Sterilization of Dental Instruments", dental instruments must meet strict disinfection standards, such as maintaining a temperature of 121-132℃ and a pressure of 103.4kPa for ≥15 minutes for high-temperature and high-pressure sterilization, with a microbial residue of ≤5CFU / cm². Currently, most methods for analyzing the quality of equipment disinfection still rely on manual sampling and testing. However, manual sampling and testing has the following technical problems in practical use: Manual testing results are prone to errors. When manually observing the cleanliness of the equipment surface, it is easily affected by factors such as the experience of the testing personnel and external lighting, leading to structural errors. Furthermore, manual sampling can only test the sampled items, failing to achieve comprehensive testing and analysis of the disinfected equipment, resulting in incomplete analysis. Currently, the detection of residual microorganisms after equipment disinfection is still done through bacterial culture, which has a long testing cycle, cannot provide feedback on disinfection quality, and is inconvenient to use. While sensors are used to monitor data such as temperature during disinfection for auxiliary analysis, the equipment can only monitor and analyze single data points, unable to perform comprehensive analysis of multiple data points. For example, if the disinfection temperature meets the standard, the pressure may not meet the disinfection requirements, making comprehensive analysis impossible. This significantly reduces the accuracy of disinfection quality analysis and results in poor effectiveness. Summary of the Invention
[0003] The present invention aims to provide an intelligent analysis method for the disinfection quality of dental instruments, so as to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: A smart analysis method for the disinfection quality of dental instruments includes the following steps: S1. Data Acquisition: Data is acquired through a detection system, which includes a sensor acquisition module, an image acquisition module, and a data processing and transmission module. The sensor module includes a temperature sensor, a humidity sensor, a pressure sensor, a concentration sensor, and a microbial sensor. The sensor acquisition module acquires temperature data, humidity data, pressure data, disinfectant concentration values, and microbial residue data of the equipment during the disinfection process. The image acquisition module includes a high-definition industrial camera and a light source. The image acquisition module acquires surface images of the equipment. The data processing and transmission module processes and transmits the acquired data. S2. Data preprocessing; The data collected in step S1 is preprocessed, including the following steps: S21. Remove outliers; Remove sensor outliers using the 3σ criterion. First, calculate the mean μ and standard deviation σ of each sensor data. Then, identify data that exceed the range of [μ-3σ, μ+3σ] as outliers and remove them. For image data, remove salt-and-pepper noise from the image using median filtering. Set the median filtering window size to 3×3 to remove salt-and-pepper noise from the image while preserving effective detail information. S22. Standardized Data: Sensor data is converted into a common standard format using the Z-score method. The standardization conversion is completed according to the following formula: in For standardized data, The original data, The average value of the data. The standard deviation of the data; For image data, the grayscale values of the image are standardized to 0-1. The original grayscale values of the image are in the range of 0-255. The standardization transformation is performed according to the following formula: in The standardized grayscale value. The original grayscale value; S23. Data fusion: A weighted fusion algorithm based on entropy weighting is used to fuse data from multiple sensors to obtain fused data. S3. Construct the analysis model; the analysis model includes a CNN submodule, an LSTM submodule, a Bahdanau attention mechanism layer, and a fully connected layer; The CNN submodule includes three convolutional layers, two pooling layers, and one flattened layer. The convolutional layers have sizes of 3×3, 3×3, and 5×5, and their numbers are 32, 64, and 128, respectively. The pooling kernel size of the pooling layers is 2×2. The LSTM submodule includes two hidden layers and one fully connected LSTM layer. The number of hidden units in the two hidden layers are 128 and 64, respectively. The fully connected layer consists of two groups, which are located after the Bahdanau attention mechanism layer. S4. Analyze the disinfection quality; input the fused data obtained in step S2 into the analysis model, train and output the disinfection quality assessment results, which include three situations: qualified, pending re-inspection, and unqualified. S5. Generate an assessment report; Based on the results output in step S4, generate an assessment report and issue corresponding alarms. An alarm is issued when the results are unsatisfactory, and a re-examination signal is issued when a re-examination is required. Then, the assessment report and soaking signal are transmitted in real time to the hospital management system and the mobile terminals of the management personnel. S6. Store data; Store disinfection data for later viewing and for easy data retrieval to optimize the analysis model.
[0005] Preferably, after the analysis model is constructed, the method further includes analysis model training and validation, which includes the following steps: S31. Construct a dataset; Select 10 types of commonly used dental instruments, and conduct disinfection experiments on 500 instruments of each type under different disinfection conditions, using multiple sets of multi-dimensional data; Have multiple senior experts label the disinfection results, labeling the disinfection results as qualified, pending re-inspection, and unqualified, and then divide the above-collected multi-dimensional data into training set, validation set, and test set in a ratio of 7:2:1. S32, Model Training: Input the data allocated to the training set in step S31 into the model. Use 64 samples per training iteration, and perform 50 training iterations. Calculate the loss between the model output and the true label using the cross-entropy loss function. The cross-entropy loss function is as follows: in, This represents the cross-entropy loss value. This represents the probability that the model predicts a sample belongs to category C, where c represents the category. In this technical solution, c = 1, 2, and 3, corresponding to the three results of qualified, pending review, and unqualified, respectively. This represents the one-hot encoded value of the true label when the sample belongs to class c. = 1, otherwise = 0; The gradient is calculated using the backpropagation algorithm. Based on the evaluation results of the validation set, the model parameters are adjusted using the Adam optimizer. After each training iteration, the model is evaluated using the validation set samples, and the loss value L is calculated based on the validation set data. y If L y If the model parameters decrease by ≥5% compared to the previous training iteration, then the current model parameters are saved as the optimal model. y If the learning rate does not decrease after 5 consecutive training iterations, it is decayed to 50% of its current value, where the minimum learning rate is 0.00001. After the learning rate decays for three consecutive iterations, L... y If there is still no decrease, stop training and output the optimal model; when L yWhen changes occur, rework and save the optimal model according to the above procedure and continue training. When the learning rate decays to the minimum value of 0.00001, regardless of L... y The system outputs the optimal model even when changes are occurring, preventing overfitting. During training, classification accuracy and impairment values on the validation set can be calculated. The model's state can be assessed based on the validation set data, and hyperparameters can be adjusted to further avoid overfitting and improve generalization ability. S33. Model Validation: Input the data allocated to the test set into the trained model, calculate the accuracy, precision, and recall of the model on the test set, evaluate the model performance by comparing the accuracy, precision, and recall with the standard values of existing methods, and adjust the model parameters to further optimize it if the calculated values of accuracy, precision, and recall do not meet the design requirements. The formula for calculating accuracy is as follows: The formula for calculating accuracy is as follows: The formula for calculating recall rate is as follows: Where c is the class, TP is the number of samples correctly predicted as positive, TN is the number of samples correctly predicted as other classes, FP is the number of samples incorrectly predicted as positive, and FN is the number of samples incorrectly predicted as negative.
[0006] Preferably, the temperature sensor is a PT100 platinum resistance temperature sensor, the humidity sensor is an SHT30 digital humidity sensor, the pressure sensor is an MPX5700 series pressure sensor, and the microbial detection sensor is an ATP bioluminescence-based microbial sensor. Detection is performed by direct contact of the probe with the equipment surface. The probe uses a replaceable sterile sampling head to avoid cross-contamination. The image acquisition module includes a high-definition industrial camera and a light source. The image acquisition module acquires surface images of the equipment. The high-definition industrial camera is a Basler acA2500-14gm high-definition industrial camera with a resolution of 2592×1944. The high-definition industrial camera and the light source are installed inside the disinfection equipment via a robotic arm.
[0007] Preferably, when calculating the fused data in step S23, First, the sensor data is normalized to obtain normalized values. The normalized values are then calculated using the following formula: Then calculate the information entropy using the following formula: in, For the sensor's label number, =1, 2, ..., n, where n is the number of sensors and m is the amount of data per sensor. This refers to the label number of which group of data within the sensor data set. Then determine the weights, calculate the weights of each sensor data based on the information entropy, and complete the weight calculation according to the following formula: If this embodiment only sets up sensors for temperature, humidity, pressure, disinfectant concentration, and microbial quantity, then n=5; After obtaining the weights of each sensor's data, the standardized fused data is calculated using the following formula: in For standard data fusion, This represents the standardized data for the i-th sensor.
[0008] Preferably, in step S4, after the data enters the analysis model, the spatial features of the equipment surface image are extracted through the CNN submodule, the time series features of the multi-dimensional sensor data are analyzed through the LSTM submodule, and the weight ratio of key features is strengthened through the Bahdanau attention mechanism layer. After the image data enters the analysis model, features are extracted through convolutional layers, then the dimensionality is reduced through pooling layers, then it is converted into a one-dimensional feature vector through a flattening layer, and then it is input into the Bahdanau attention mechanism layer. After the fused data enters the analysis model, the time series dependencies are captured through the hidden layer, and then the feature dimensions are compressed through the LSTM fully connected layer. Then, it is also input into the Bahdanau attention mechanism layer. The similarity between two sets of feature vectors is calculated using the Bahdanau attention mechanism layer, and the calculation is performed using the following formula: in The feature vectors processed by the CNN submodule The feature vectors processed by the LSTM submodule, v, , For learnable parameters, Attention weights α are generated using the softmax function, and then the fused features are calculated using the following formula: The fused features are then fed into a fully connected layer. The first fully connected layer compresses the fused features and further extracts key features. The compressed fused features are then fed into a second fully connected layer, which outputs three class probability values via a softmax function. This leads to the analytical results.
[0009] Preferably, in step S5, when the result is unqualified, an alarm is issued directly, and the unqualified equipment number, the reason for the unqualified result, and the disinfection parameter adjustment suggestions are output to the hospital management platform in the form of an analysis report. After the staff confirms, the disinfection parameters are automatically adjusted to ensure the quality of subsequent disinfection. At the same time, the unqualified equipment is taken out and disinfected again. When the result is "requiring re-examination", the sampling frequency of each sensor in the detection system is automatically increased. At this time, the sampling frequency of each sensor is adjusted to 2 times / second, and then a second test is performed. If the second test is qualified, the detection system is restored to the original state and the test continues. If the second test is still "requiring re-examination", an alarm is issued and the equipment number and reason for re-examination are output to the hospital management platform. The management personnel take out the equipment to be re-examined for manual testing based on the alarm effect. When the result is satisfactory, a satisfactory report is automatically generated for subsequent management personnel to review.
[0010] Preferably, the detection system collects data at a frequency of 180 seconds per time.
[0011] Preferably, in step S32, the initial parameters of the Adam optimizer are set to a learning rate α = 0.001, momentum parameters β1 = 0.9, and β2 = 0.999.
[0012] The beneficial effects of this technical solution compared to existing technologies are as follows: (1) This technical solution sets up a detection system, which collects different data in the disinfection process through multiple sensors, and realizes comprehensive detection of environmental parameters, chemical parameters, biological parameters and appearance cleanliness in the disinfection process of equipment, thereby improving the quality of monitoring and analysis and avoiding errors caused by manual detection. This technical solution uses the entropy weight method to perform weighted fusion of multi-source data, dynamically allocates weights according to the information content of each parameter, and captures the correlation between parameters through the constructed analysis model, thereby realizing comprehensive analysis of disinfection quality, avoiding misjudgment caused by single data analysis, greatly improving the accuracy of quality analysis and improving the use effect. This technical solution can output the quality analysis effect by directly analyzing and transmitting the data, realizing real-time analysis of disinfection quality, reducing the detection cycle and improving the efficiency of analysis and detection. Attached Figure Description
[0013] Figure 1 This is a schematic diagram of the process of the present invention; Detailed Implementation The present invention will now be described in further detail with reference to the accompanying drawings and embodiments: A smart analysis method for the disinfection quality of dental instruments includes the following steps: S1. Data Acquisition: Data is acquired through a detection system, which includes a sensor acquisition module, an image acquisition module, and a data processing and transmission module. The sensor module includes a temperature sensor, a humidity sensor, a pressure sensor, a concentration sensor, and a microbial sensor. The sensor acquisition module collects temperature, humidity, pressure, disinfectant concentration, and microbial residue data of the equipment during the disinfection process. Specifically, a PT100 platinum resistance temperature sensor is used, an SHT30 digital humidity sensor is used, an MPX5700 series pressure sensor is used, and a microbial detection sensor based on ATP bioluminescence is used. Detection is performed by direct contact of the probe with the equipment surface. The probe uses a replaceable sterile sampling head to avoid cross-contamination. The image acquisition module includes a high-definition industrial camera and a light source. It acquires surface images of the equipment. The high-definition industrial camera is a Basler acA2500-14gm with a resolution of 2592×1944. The high-definition industrial camera and light source are mounted inside the observation window of the disinfection equipment via a robotic arm, enabling comprehensive imaging of the disinfection equipment. The data processing and transmission module completes the transmission and processing of the acquired data. S2. Data preprocessing; The data collected in step S1 is preprocessed, including the following steps: S21. Remove outliers; Remove sensor outliers using the 3σ criterion. First, calculate the mean μ and standard deviation σ of each sensor data. Then, identify data that exceed the range of [μ-3σ, μ+3σ] as outliers and remove them. For image data, remove salt-and-pepper noise from the image using median filtering. Set the median filtering window size to 3×3 to remove salt-and-pepper noise from the image while preserving effective detail information. S22. Standardized Data: Sensor data is converted into a common standard format using the Z-score method. The standardization conversion is completed according to the following formula: in For standardized data, The original data, The average value of the data. The standard deviation of the data; For image data, the grayscale values of the image are standardized to 0-1. The original grayscale values of the image are in the range of 0-255. The standardization transformation is performed according to the following formula: in The standardized grayscale value. The original grayscale value; S23. Data fusion: A weighted fusion algorithm based on entropy weighting is used to fuse data from multiple sensors to obtain fused data. First, calculate the information entropy of the data from each sensor. Information entropy reflects the uncertainty of the data; the smaller the entropy value, the greater the amount of information in the data, and the greater the weight should be. First, the sensor data is normalized to obtain normalized values. The normalized values are then calculated using the following formula: Then calculate the information entropy using the following formula: in, For the sensor's label number, =1, 2, ..., n, where n is the number of sensors and m is the amount of data per sensor. This refers to the label number of which group of data within the sensor data set. For example, if the 5 sets of data from the i-th temperature sensor are [121, 122, 120, 123, 124], then m is 5. The normalized value after normalizing the five sets of data according to the above formula is: =0.198, =0.200, =0.197, =0.202, =0.203, This represents the normalized value of the first data set out of the 5 data sets from the i-th temperature sensor. Then, the information entropy is calculated according to the formula above. If the value is 0.998, it indicates that the sensor has high data stability, a large amount of information, and should therefore have a higher weight. Then determine the weights, calculate the weights of each sensor data based on the information entropy, and complete the weight calculation according to the following formula: If this embodiment only sets up sensors for temperature, humidity, pressure, disinfectant concentration, and microbial quantity, then n=5; After obtaining the weights of each sensor's data, the standardized fused data is calculated using the following formula: in For standard data fusion, Standardized data for the i-th sensor; by calculating standardized fused data, unified data is provided for subsequent analysis and processing by the analysis model; S3. Construct the analysis model; the analysis model includes a CNN submodule, an LSTM submodule, a Bahdanau attention mechanism layer, and a fully connected layer. The CNN submodule consists of 3 convolutional layers, 2 pooling layers, and 1 flattening layer. The convolutional layers have sizes of 3×3, 3×3, and 5×5, and the number of layers is 32, 64, and 128 respectively. The pooling kernel size of the pooling layer is 2×2. The LSTM submodule consists of two hidden layers and one fully connected LSTM layer. The number of hidden units in the two hidden layers are 128 and 64, respectively. There are two sets of fully connected layers, which are located after the Bahdanau attention mechanism layer. S4. Analyze the disinfection quality; Input the fused data and image data obtained in step S2 into the analysis model, train and output the disinfection quality assessment results; After the data enters the analysis model, the spatial features of the equipment surface image are extracted through the CNN submodule, the time series features of multi-dimensional sensor data are analyzed through the LSTM submodule, and the weight ratio of key features is strengthened through the Bahdanau attention mechanism layer. After the image data enters the analysis model, features are extracted through convolutional layers, then the dimensionality is reduced through pooling layers, then it is converted into a one-dimensional feature vector through a flattening layer, and then it is input into the Bahdanau attention mechanism layer. After the fused data enters the analysis model, the time series dependencies are captured through the hidden layer, and then the feature dimensions are compressed through the LSTM fully connected layer. Then, it is also input into the Bahdanau attention mechanism layer. The similarity between two sets of feature vectors is calculated using the Bahdanau attention mechanism layer, and the calculation is performed using the following formula: in The feature vectors processed by the CNN submodule The feature vectors processed by the LSTM submodule, v, , For learnable parameters, Attention weights α are generated using the softmax function, and then the fused features are calculated using the following formula: Then, the fused features are input into the fully connected layer. The fused features are input into the first fully connected layer, which compresses the fused features to further extract key features. Then, the compressed fused features are input into the second fully connected layer, and three types of probability values are output through the softmax function , and then the analysis result is obtained. For example , when the identification qualification rate is 96%, when ≥0.9, it is determined as qualified. When 0.7 ≤ <0.9 or 0.3 ≤ , it is determined as pending recheck. When ≥0.8, it is determined as unqualified; This determination standard can be dynamically optimized according to the training situation of the analysis model
[0014] S5. Analyze the disinfection quality; Establish the disinfection level according to the result output in step S4. The disinfection level is divided into three levels: excellent, pending recheck, and unqualified. An alarm is issued when it is unqualified, and a recheck signal is issued when it is pending recheck; Then, the grading result is transmitted to the hospital management system and the mobile terminal of the management staff in real time The frequency of the detection system for collecting data is 180 seconds / time. The detection system processes the collected data according to the above steps and outputs the result after being processed by the analysis model. The result includes three situations: qualified, pending recheck, and unqualified. Then, the system automatically retrieves the preset disinfection quality standard parameters, and compares the evaluation result with the disinfection quality standard parameters to generate an evaluation report When the result is unqualified, an alarm is directly issued, and the equipment number, unqualified reason, and disinfection parameter adjustment suggestion of the unqualified equipment are output to the hospital management platform in the form of an analysis report. After the staff confirms, the disinfection parameters are automatically adjusted to ensure the subsequent disinfection quality. At the same time, the unqualified disinfected equipment is taken out for re-disinfection When the result is pending recheck, the sampling frequency of each sensor in the detection system is automatically increased. At this time, the sampling frequency of each sensor is adjusted to 2 times / second, and then secondary detection is carried out. If the secondary detection is qualified, the detection system is restored to its original state and continues to detect. If the secondary detection is still pending recheck, an alarm is issued, and the equipment number and pending recheck reason of the pending recheck equipment are output to the hospital management platform. The management staff takes out the pending recheck equipment for manual detection according to the alarm effect When the result is qualified, a qualified report is automatically generated for subsequent management staff to query
[0015] S6. Store data; Store the disinfection data for subsequent viewing, and at the same time facilitate data extraction to optimize the analysis model
[0016] The construction of the analysis model also includes the training and verification of the analysis model. The training and verification of the analysis model include the following steps S31. Construct a dataset; Select 10 types of commonly used dental instruments, and conduct disinfection experiments on 500 instruments of each type under different disinfection conditions, using multiple sets of multi-dimensional data; Have multiple senior experts label the disinfection results, labeling the disinfection results as qualified, pending re-inspection, and unqualified, and then divide the above-collected multi-dimensional data into training set, validation set, and test set in a ratio of 7:2:1. S32, Model Training: Input the data allocated to the training set in step S31 into the model. Use 64 samples per training iteration, and perform 50 training iterations. Calculate the loss between the model output and the true label using the cross-entropy loss function. The cross-entropy loss function is as follows: in, This represents the cross-entropy loss value. This represents the probability that the model predicts a sample belongs to category C, where c represents the category. In this technical solution, c = 1, 2, and 3, corresponding to the three results of qualified, pending review, and unqualified, respectively. This represents the one-hot encoded value of the true label when the sample belongs to class c. = 1, otherwise = 0; The gradient is calculated using the backpropagation algorithm. Based on the evaluation results of the validation set, the model parameters are adjusted using the Adam optimizer. The Adam optimizer parameters are set to a learning rate of α=0.001, momentum parameters β1=0.9, β2=0.999, and ε=10⁻⁸. After each training iteration, the model is evaluated using validation set samples, and the loss value L on the validation set data is calculated. y If L y If the model parameters decrease by ≥5% compared to the previous training iteration, then the current model parameters are saved as the optimal model. y If the learning rate does not decrease after 5 consecutive training iterations, it is decayed to 50% of its current value, where the minimum learning rate is 0.00001. After the learning rate decays for three consecutive iterations, L... y If there is still no decrease, stop training and output the optimal model; when L y When changes occur, rework and save the optimal model according to the above procedure and continue training. When the learning rate decays to the minimum value of 0.00001, regardless of L... y The system outputs the optimal model even when changes are occurring, preventing overfitting. During training, classification accuracy and impairment values on the validation set can be calculated. The model's state can be assessed based on the validation set data, and hyperparameters can be adjusted to further avoid overfitting and improve generalization ability. S33. Model Validation: Input the data allocated to the test set into the trained model, calculate the accuracy, precision, and recall of the model on the test set, evaluate the model performance by comparing the accuracy, recall, and F1 score with the standard values of existing methods, and adjust the model parameters to further optimize it if the calculated values of accuracy, precision, and recall do not meet the design requirements. The formula for calculating accuracy is as follows: The formula for calculating accuracy is as follows: The formula for calculating recall rate is as follows: Where c is the class, TP is the number of samples correctly predicted as positive, TN is the number of samples correctly predicted as other classes, FP is the number of samples incorrectly predicted as positive, and FN is the number of samples incorrectly predicted as negative.
[0017] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.
Claims
1. An intelligent analysis method for the disinfection quality of dental equipment, characterized in that, Comprise the following steps: S1, collect data; collect data through a detection system, the detection system comprising a sensor acquisition module, an image acquisition module and a data processing and transmission module, the sensor module comprising a temperature sensor, a humidity sensor, a pressure sensor, a concentration sensor and a microorganism sensor, collecting the temperature data, humidity data, pressure data, disinfectant concentration value and microorganism residual data of the equipment during the disinfection process through the sensor acquisition module; the image acquisition module comprises a high-definition industrial camera and a light source, and the surface image of the equipment is collected through the image acquisition module; the data processing and transmission module is used to complete the processing and transmission of the collected data; S2, data preprocessing; the collected data in step S1 is preprocessed, comprising the following steps: S21, remove outliers; remove sensor outliers by 3σ criterion, first calculate the mean μ and standard deviation σ of each sensor data, then determine the data outside the range [μ-3σ, μ+3σ] as outliers, then remove the outliers, and for image data, remove the salt and pepper noise in the image by median filtering, and set the median filtering window size to 3x3, so that the effective detail information is retained while the salt and pepper noise in the image is removed; S22, standardize the data; convert the sensor data to the same standard format by Z-score method, and complete the standardization conversion according to the following formula: wherein is the standardized data, is the raw data, is the mean of the data, is the standard deviation of the data; For image data, the gray value of the image is standardized to 0-1, and the original gray value of the image is in the range of 0-255, and the standardization conversion is completed according to the following formula: wherein is the normalized gray value, is the original gray value; S23, data fusion; fuse multiple sensor data based on the weighted fusion algorithm of entropy weight method to obtain fused data; S3, construct an analysis model; the analysis model comprises a CNN sub-module, an LSTM sub-module, a Bahdanau attention mechanism layer and a fully connected layer; The CNN sub-module comprises three convolution layers, two pooling layers and one flattening layer, the convolution layer sizes are 3x3, 3x3 and 5x5 respectively, and the numbers are 32, 64 and 128 respectively, the pooling kernel size of the pooling layer is 2x2, The LSTM sub-module comprises two hidden layers and an LSTM fully connected layer, the number of hidden units of the two hidden layers is 128 and 64 respectively, The fully connected layer is provided with two groups, and the two groups of fully connected layers are arranged after the Bahdanau attention mechanism layer; S4, analyze the disinfection quality; input the fused data obtained in step S2 into the analysis model, train the output disinfection quality evaluation results, and the evaluation results include three cases of qualified, to be reviewed and unqualified; S5, generate an evaluation report; generate an evaluation report according to the output result of step S4 and issue a corresponding alarm, wherein an alarm is issued when it is unqualified, and a review signal is issued when it is to be reviewed; then the evaluation report and the soaking signal are transmitted to the hospital management system and the mobile terminal of the management personnel in real time; S6, store data; store the disinfection data for subsequent viewing and data extraction to optimize the analysis model.
2. The intelligent analysis method for the sterilization quality of dental equipment according to claim 1, characterized in that, After the analysis model is constructed, the analysis model training and verification are also included, and the analysis model training and verification comprise the following steps: S31, construct a data set; select 10 types of equipment commonly used in the dental department, conduct 500 different disinfection experiments on each type of equipment, use multiple groups of multidimensional data; label the disinfection results by multiple experienced experts, and the disinfection results are labeled as qualified, to be reviewed and unqualified; then divide the above collected multiple groups of multidimensional data into a training set, a validation set and a test set according to a ratio of 7:2:1; S32, model training; input the data divided into the training set in step S31 into the model, use 64 samples for each training iteration, and the number of training iterations is 50 times; calculate the loss between the model output and the true label by using a cross-entropy loss function; the cross-entropy loss function is specifically: wherein, is a cross-entropy loss value, represents a probability that the model predicts the sample belongs to the category C, c represents a category of classification, in the technical solution, c = 1, 2, 3, respectively corresponding to three results of qualified, to be reviewed and unqualified, represents a one-hot encoding value of a true label, when the sample belongs to the category c, = 1, otherwise = 0; The gradient is calculated by the back propagation algorithm, the model parameters are adjusted by the Adam optimizer according to the evaluation results of the validation set, and the model is evaluated by the validation set sample after each training of the training set sample to calculate the loss value L of the validation set data y If L y If the last training decreases by more than 5%, the current model parameters are saved as the optimal model, and when L y When the learning rate is continuously attenuated three times, L y If it still does not decrease, stop training and output the optimal model; when L y If there is a change, retrain and save the optimal model according to the above process, and when the learning rate is attenuated to the minimum value 0.00001, whether L y Is still changing, the optimal model is also output to prevent overfitting; during the training process, the classification accuracy and loss value on the validation set can be calculated, the model state is judged according to the validation set data, and the model hyperparameters are adjusted to further avoid overfitting and improve the model generalization ability, S33, model verification; input the data divided into the test set into the trained model, calculate the accuracy, precision and recall of the model on the test set, and compare the accuracy, precision and recall with the standard values of the existing method to evaluate the performance of the model; when the calculated values of the accuracy, precision and recall do not meet the design requirements, adjust the model parameters to further optimize them. The calculation formula of the accuracy is: The calculation formula of the precision is: The calculation formula of the recall is: Wherein c is the class, TP is the number of samples correctly predicted as positive class by the model, TN is the number of samples correctly predicted as other class by the model, FP is the number of samples incorrectly predicted as positive class, and FN is the number of samples incorrectly predicted as negative class.
3. The intelligent analysis method for the sterilization quality of dental equipment according to claim 1, characterized in that, The temperature sensor selects a PT100 platinum resistance temperature sensor, the humidity sensor selects a SHT30 digital humidity sensor, the pressure sensor selects an MPX5700 series pressure sensor, and the microbial detection sensor selects a microbial sensor based on the ATP bioluminescence method. The probe directly contacts the surface of the equipment for detection, the probe uses a replaceable sterile sampling head to avoid cross contamination; the image acquisition module includes a high-definition industrial camera and a light source, and the surface image of the equipment is acquired by the image acquisition module. The high-definition industrial camera selects a Basler acA2500-14gm high-definition industrial camera, and the resolution is set to 2592x1944. The high-definition industrial camera and the light source are installed in the disinfection equipment through a mechanical arm.
4. The intelligent analysis method for the sterilization quality of dental equipment according to claim 1, characterized in that, When the step S23 calculates the fusion data, First, normalize the data of the sensor to obtain the normalized value of the data, and calculate the normalized value according to the following formula: Then calculate the information entropy according to the following formula: wherein, is a serial number of the sensor, = 1, 2, …, n, n is the number of sensors, and m is the amount of data of the sensor, is a serial number of the mth group of data in the amount of data of the sensor, Then determine the weight, calculate the weight of each sensor data according to the information entropy, and complete the calculation of the weight according to the following formula: If only temperature, humidity, pressure, disinfectant concentration and microbial quantity sensors are provided in this embodiment, then n=5; After obtaining the weight of each sensor data, the standardized fusion data is calculated, and the standardized fusion data is calculated according to the following formula: wherein is the standard fusion data, is the normalized data of the i-th sensor.
5. The intelligent analysis method for the sterilization quality of dental equipment according to claim 1, characterized in that, In the step S4, after the data enters the analysis model, the spatial features of the equipment surface image are extracted by the CNN submodule, the time sequence features of the multi-dimensional sensor data are analyzed by the LSTM submodule, and the weight proportion of the key features is strengthened by the Bahdanau attention mechanism layer; After the image data enters the analysis model, the features are extracted by the convolution layer, then the dimension is reduced by the pooling layer, then the features are converted into a one-dimensional feature vector by the flattening layer, and then the one-dimensional feature vector is input into the Bahdanau attention mechanism layer; After the fusion data enters the analysis model, the time sequence dependence is captured by the hidden layer, then the feature dimension is compressed by the LSTM full connection layer, and then the fusion data is also input into the Bahdanau attention mechanism layer; The similarity of the two groups of feature vectors is calculated by the Bahdanau attention mechanism layer, and the calculation is completed by the following formula: wherein is a feature vector processed by a CNN sub-module, is a feature vector processed by an LSTM sub-module, v, , is a learnable parameter, The attention weight α is generated by the softmax function, and then the fusion feature is calculated, and the fusion feature is calculated according to the following formula: Then the fusion features are input into a full connection layer, the fusion features are input into a first full connection layer, the fusion features are compressed in the first full connection layer to further extract key features, and then the compressed fusion features are input into a second full connection layer, and three class probability values are output through a softmax function , and then an analysis result is obtained.
6. The intelligent analysis method for the sterilization quality of dental equipment according to claim 1, characterized in that, In the step S5, when the result is unqualified, an alarm is directly sent out, and the unqualified equipment number, unqualified reason and disinfection parameter adjustment suggestion are output to the hospital management platform in the form of an analysis report, and the staff determines to automatically adjust the disinfection parameters to ensure the subsequent disinfection quality, and the disinfection unqualified equipment is taken out for re-disinfection; When the result is to be reviewed, the sampling frequency of each sensor in the detection system is automatically increased, at this time the sampling frequency of each sensor is adjusted to 2 times per second, then secondary detection is performed, if the secondary detection is qualified, the detection system returns to the original state and continues to detect, if the secondary detection is still to be reviewed, an alarm is sent out, and the equipment number to be reviewed and the reason to be reviewed are output to the hospital management platform, and the management personnel take out the equipment to be reviewed for manual detection according to the alarm effect; When the result is qualified, a qualified report is automatically generated, which is convenient for subsequent management personnel to query.
7. The intelligent analysis method for the sterilization quality of dental equipment according to claim 1, characterized in that, The frequency of the detection system collecting data is 180 seconds / time.
8. The intelligent analysis method for the sterilization quality of dental equipment according to claim 2, wherein, In the S32, the initial parameters of the Adam optimizer are set as a learning rate α=0.001, a momentum parameter β1=0.9, and β2=0.999.