Nuclear radiation detection system based on artificial intelligence

Abstract

The invention discloses a nuclear radiation detection system based on artificial intelligence, and relates to the technical field of nuclear radiation detection. The method comprises the following steps: acquiring radiation intensity data and corresponding interference factor data influencing the radiation intensity data, performing data marking, constructing a first output sample set, a first input sample set, a second output sample set and a second input sample set according to a marking result, and further constructing a first output sample set, a second input sample set and a third input sample set which take the first input sample set as input and take the second input sample set as output; using the first output sample set as an output neural network model, using the trained neural network model to obtain a prediction sample set corresponding to the second input sample set, and determining a correction coefficient under the currently collected real-time interference factor data level according to the second output sample set and the prediction sample set, and correcting the currently collected real-time radiation intensity data by using the correction coefficient to obtain the corrected radiation intensity data, and realizing high-precision and self-adaptive radiation data correction through fusion and weighting processing of various neural networks.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear radiation detection, and particularly relates to a nuclear radiation detection system based on artificial intelligence. Background Art

[0002] Nuclear radiation detection is of extremely important significance in multiple fields. In the nuclear energy field, the safe operation of nuclear facilities depends on the real-time monitoring of radiation levels to prevent radioactive leakage and nuclear accidents. In environmental protection, monitoring the radiation levels in the environment helps to evaluate the impact of nuclear accidents on the ecosystem and take corresponding protective measures. In addition, in the fields of medicine, industry, customs, etc., nuclear radiation detection is also a key means to ensure the health and safety of personnel.

[0003] Traditional nuclear radiation detection mainly relies on nuclear radiation detectors, such as gamma-ray detectors, neutron detectors, etc. These devices can collect radiation data in real time, but there are the following limitations in practical applications: Data distortion problem: Radiation data is affected by multiple factors, including natural background radiation (such as cosmic rays, natural radioactive substances), environmental factors (such as temperature, humidity, air pressure), the relative position of the detector and the radiation source, the shielding effect of surrounding substances, etc. These factors may cause a deviation between the detected radiation data and the actual radiation level, thus affecting the accuracy of the detection result; Limitations of a single model: Traditional methods usually use a single mathematical model or empirical formula to correct data, but these models are often difficult to comprehensively consider all influencing factors, especially the data correction effect in complex environments is not good.

[0004] Lack of adaptability: Traditional detection systems are difficult to automatically adjust the correction strategy when facing environmental changes, instrument aging or new interference sources, resulting in a decrease in detection accuracy.

[0005] In recent years, artificial intelligence technology, especially machine learning and deep learning, has made remarkable progress and demonstrated powerful data processing and pattern recognition capabilities in multiple fields. Artificial intelligence technology can automatically learn complex patterns and relationships in data, providing new ideas for solving the data distortion problem in nuclear radiation detection. For this reason, a nuclear radiation detection system based on artificial intelligence is proposed. Summary of the Invention

[0006] The main purpose of the present invention is to provide a nuclear radiation detection system based on artificial intelligence, which realizes high-precision and adaptive radiation data correction through the fusion and weighted processing of multiple neural networks. It can effectively solve the problems in the background art.

[0007] To achieve the above object, the technical solution adopted by the present invention is as follows. A nuclear radiation detection system based on artificial intelligence, the detection process of the system includes the following steps: Collecting radiation intensity data and corresponding interference factor data, wherein the interference factor data is data that affects the radiation intensity data; Marking accurate data points in the radiation intensity data, and constructing a first output sample set with the marked accurate data points , constructing a first input sample set with the interference factor data corresponding to the accurate data point , constructing a second output sample set with the unlabeled radiation intensity data , constructing a second input sample set with the interference factor data corresponding to the unlabeled radiation intensity data ; Taking the first input sample set As input, take the first output sample set For output, a neural network model is constructed, and the neural network model is trained until its accuracy reaches a set expected value; Take the second input sample set As input, using the trained neural network model, obtain the second input sample set The corresponding prediction sample set ; According to the second output sample set and the prediction sample set Determine the correction factor for the current level of real-time interference factor data collected , using the correction factor Correct the currently collected real-time radiation intensity data to obtain corrected radiation intensity data.

[0008] Furthermore, the system comprises: A data acquisition module for collecting radiation intensity data and corresponding interference factor data; The data acquisition module comprises: A first data acquisition submodule for acquiring real-time radiation intensity data; A second data acquisition submodule for acquiring historical radiation intensity data; A third data acquisition submodule for acquiring real-time interference factor data corresponding to the real-time radiation intensity data; A fourth data acquisition submodule for acquiring historical interference factor data corresponding to the historical radiation intensity data; The marked radiation intensity data is the historical radiation intensity data.

[0009] Furthermore, the system also includes: A data marking module for marking accurate data points in the radiation intensity data; A sample set construction module for constructing a data sample set, including: A first output sample set construction module for constructing a first output sample set with the marked accurate data points ; A first input sample set construction module for constructing a first input sample set with the interference factor data corresponding to the first output sample set ; A second output sample set construction module for constructing a second output sample set with the unmarked radiation intensity data ; A second input sample set construction module for constructing a second input sample set with the interference factor data corresponding to the second output sample set 。

[0010] Furthermore, the system further includes: A neural network module for constructing a neural network model with the first input sample set as the input and the first output sample set as the output; wherein, the neural network module includes a first neural network module, a second neural network module, and a third neural network module; The neural network model includes: A first neural network model constructed by using the first neural network module; A second neural network model constructed by using the second neural network module; And, a third neural network model constructed by using the third neural network module.

[0011] Furthermore, the system further includes: A prediction sample set acquisition module for using the trained neural network model with the second input sample set as the input to obtain a prediction sample set corresponding to the second input sample set ; ; wherein, the prediction sample set acquisition module includes a first prediction sample set acquisition module, a second prediction sample set acquisition module, and a third prediction sample set acquisition module; The prediction sample set includes a first prediction sample set , a second prediction sample set and a third prediction sample set ; The first prediction sample set is based on the second input sample set is used as the input, and the output of the trained first neural network model is utilized; The second prediction sample set uses the second input sample set as the input, and the output of the trained second neural network model is utilized; The third prediction sample set uses the second input sample set as the input, and the output of the trained third neural network model is utilized.

[0012] Furthermore, the system further includes: A correction coefficient acquisition module, configured to determine a correction coefficient at the current level of real-time interference factor data collected, according to the second output sample set and the prediction sample set ; ; The correction coefficient acquisition module includes a first correction coefficient acquisition module, a second correction coefficient acquisition module, and a third correction coefficient acquisition module; The correction coefficient is a weighted value of the first correction coefficient , the second correction coefficient , and the third correction coefficient ; The first correction coefficient is determined by the first correction coefficient acquisition module according to the second output sample set and the first prediction sample set ; The second correction coefficient is determined by the second correction coefficient acquisition module according to the second output sample set and the second prediction sample set ; The third correction coefficient is determined by the third correction coefficient acquisition module according to the second output sample set and the third prediction sample set ;

[0013] Furthermore, the system further includes: A data correction module, configured to correct the currently collected real-time radiation intensity data by using the correction coefficient to obtain the corrected radiation intensity data.

[0014] Furthermore, the acquisition process of the correction coefficient includes the following steps: Obtain the real-time interference factor data collected currently , where is expressed as the th real-time interference factor data, = 1, 2,..., ; is the total number of interference factor types; Define that when the Euclidean distance between data is within the threshold , two groups of interference factor data are in the same level state. According to the defined content, screen out all the interference factor data in the second input sample set that are in the same level state as the real-time interference factor data , where represents the th data of the th interference factor in the second input sample set that is in the same level state as the real-time interference factor data; Extract the unlabeled radiation intensity data in the second output sample set that corresponds to the interference factor data Extract the first predicted radiation intensity data in the first prediction sample set with the interference factor data as the input and output by the trained first neural network model. Use the extracted data to calculate the first correction coefficient , and the calculation formula is: = ; In the formula, is the total amount of extracted data; Extract the second predicted radiation intensity data in the second prediction sample set with the interference factor data as the input and output by the trained second neural network model. Use the extracted data to calculate the second correction coefficient , and the calculation formula is: = ; Extract the third predicted radiation intensity data in the third prediction sample set with the interference factor data as the input and output by the trained third neural network model. Use the extracted data to calculate the third correction coefficient , and the calculation formula is: = ; Perform linear weighted processing on the obtained first correction coefficient , the second correction coefficient and the third correction coefficient , and use the weighted value as the correction coefficient , where the calculation formula is: = × + × + × ; In the formula, is the weighted coefficient of the first correction coefficient ; is the weighted coefficient of the second correction coefficient ; is the weighted coefficient of the third correction coefficient ; , and are all within the interval (0, 1), and + + = 1.

[0015] Furthermore, the corrected radiation intensity data = real-time radiation intensity data × correction coefficient .

[0016] Furthermore, the system further includes a memory, a processor, and a computer program stored on the memory and executable on the processor.

[0017] The present invention has the following beneficial effects Compared with the prior art, this solution automatically learns the complex relationship between influencing factors and radiation data by using machine learning models such as artificial neural networks, thereby more accurately correcting distorted radiation data; Compared with the prior art, this solution combines multiple neural network architectures, integrates the advantages of different models, and improves the detection accuracy and robustness; Compared with the prior art, this solution can optimize the model structure and calculation process to ensure the efficient operation of the system in real-time monitoring scenarios; Compared with the prior art, this solution is applicable to a variety of application scenarios, including nuclear facility monitoring, environmental monitoring, emergency response, etc., and has broad promotion value.

[0018] Compared with the prior art, this solution uses artificial intelligence technology to solve the problem of data distortion in traditional nuclear radiation detection, improves the detection accuracy and the robustness of the system, and provides more reliable guarantees for nuclear safety, environmental protection, and personnel health. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the detection process of the detection system of the present invention; Figure 2 It is a block diagram of the structure of the detection system of the present invention; Figure 3 It is a schematic diagram of the structure of the first neural network model constructed by the solution of the present invention; Figure 4 It is a schematic diagram of the structure of the second neural network model constructed by the solution of the present invention; Figure 5 It is a schematic diagram of the structure of the third neural network model constructed by the solution of the present invention. Specific embodiments

[0020] The present invention will be further described below in conjunction with specific embodiments. Among them, the attached drawings are only for illustrative purposes, showing only schematic diagrams, not physical diagrams, and should not be construed as a limitation of the present invention. In order to better illustrate the specific embodiments of the present invention, some components in the attached drawings will be omitted, enlarged or reduced, and do not represent the dimensions of the actual product.

[0021] See Figure 1 The schematic diagram of the detection process of the present solution shown. The specific implementation process of the technical solution of the present invention includes the following steps: Step 1: Collect radiation intensity data and corresponding interference factor data.

[0022] Among them, the radiation intensity data includes real-time radiation intensity data and historical radiation intensity data; The interference factor data includes real-time interference factor data corresponding to the real-time radiation intensity data and historical interference factor data corresponding to the historical radiation intensity data; The interference factor data is the data that affects the radiation intensity data; specifically, the interference factors can be the following types: Physical environment factors Natural radiation background: Cosmic rays: High-energy particles from outer space will penetrate the atmosphere and generate secondary particles, which will produce background signals for radiation detectors on the ground. The intensity of cosmic rays is affected by geographical location (such as altitude) and weather conditions (such as atmospheric pressure, temperature).

[0023] Natural radioactive substances: The earth itself contains natural radioactive elements (such as uranium, thorium, potassium, etc.), and these elements will release α, β, γ rays to form a natural radiation background. The content of radioactive substances in environmental media such as soil, rock, and water will affect the readings of the detector.

[0024] Environmental media: Air density and composition: The density and composition of air can affect the penetration ability of rays. For example, air with high humidity may have a certain attenuation effect on the propagation of gamma rays.

[0025] Material shielding: If there are substances such as buildings, rocks, and metals near the detection point, these substances will shield the rays emitted by the radiation source, resulting in a weakening of the signal received by the detector.

[0026] Geographical location and terrain: Altitude: The higher the altitude, the greater the intensity of cosmic rays and the higher the background radiation level.

[0027] Terrain undulation: Complex terrain (such as valleys and hills) will affect the propagation path and intensity distribution of rays.

[0028] Characteristics of the detection instrument itself Detector type: Gamma-ray detector: Usually, scintillators (such as NaI(Tl)) or semiconductors (such as HPGe) are used as detection materials. Different materials have different sensitivities to gamma rays of different energies.

[0029] Neutron detector: Neutron detectors usually require special materials (such as He-3 tubes, lithium glass) to detect neutrons. The efficiency and sensitivity of neutron detectors are also affected by materials and designs.

[0030] Instrument performance parameters: Energy resolution: The ability of the detector to distinguish rays of different energies. Detectors with high energy resolution can identify radioactive nuclides more accurately.

[0031] Detection efficiency: The proportion of rays that the detector can detect. The detection efficiency is affected by the detector material, geometry, and working conditions.

[0032] Counting rate: The number of signals that the detector can process per unit time. Detectors with high counting rates are more suitable for detecting high radiation levels.

[0033] Instrument aging and calibration: Aging effect: During long-term use, the performance of the detector may decline, such as a decrease in detection efficiency and an increase in noise.

[0034] Calibration status: Regular calibration is the key to ensuring the accuracy of the detector. If the calibration is inaccurate, it will lead to measurement errors.

[0035] Operating conditions Detection distance: The farther the distance between the detector and the radiation source, the weaker the received radiation intensity. Changes in distance will cause significant differences in signal intensity.

[0036] Detection Angle: The relative angle between the detector and the radiation source also affects the signal intensity. Some detectors have specific response characteristics to the incident angle of the radiation.

[0037] Detection Time: The length of the detection time affects the statistical accuracy of the signal. A shorter detection time may lead to larger random errors.

[0038] Operator Skills: The experience and skill level of the operator also affect the accuracy of the data. For example, incorrect operation may lead to misreading or misjudgment of the detector.

[0039] External Interference Electromagnetic Interference: Strong electromagnetic fields (such as high-voltage power lines, electronic devices) may interfere with the electronic components of the detector, resulting in abnormal signals.

[0040] Chemical Interference: Certain chemical substances (such as strong acids, strong alkalis) may corrode the surface or internal materials of the detector, affecting its performance.

[0041] Artificial Radiation Source Interference: If there are other artificial radiation sources (such as medical equipment, industrial radiation sources) in the detection area, it may interfere with the detection of the target radiation source.

[0042] Sample Characteristics Types of Radionuclides: Different radionuclides emit different types and energies of radiation. For example, uranium-238 mainly emits alpha rays, while cobalt-60 mainly emits gamma rays.

[0043] Radiation Source Intensity: The activity of the radiation source (i.e., the number of decays per unit time) directly affects the signal intensity received by the detector.

[0044] It should be noted that for the above interference factor data, it includes numerical data, type data, and status data. For type data and status data, they can be quantified in the form of coding for subsequent data analysis.

[0045] Step 2: Mark the accurate data points in the radiation intensity data, and construct the first output sample set with the marked accurate data points , and construct the first input sample set with the corresponding historical interference factor data , and construct the second output sample set with the unmarked radiation intensity data , and construct the second input sample set with its corresponding historical interference factor data ; Among them, the marked radiation intensity data is historical radiation intensity data.

[0046] It should be noted that for the marking of accurate data points In a possible embodiment, a calibration source can be used to determine accurate data points in the historical radiation intensity data, specifically: Standard radiation source Method: Measure using a standard radiation source with known activity and energy. The radiation levels of these radiation sources are precisely calibrated and can be used as a reference for accurate radiation levels.

[0047] Steps: In a controlled environment (such as a laboratory), place the standard radiation source at a fixed position of the detector.

[0048] Measure the radiation level of the standard radiation source using a nuclear radiation detector.

[0049] Record the measurement result and label it as accurate radiation level data.

[0050] In a possible embodiment, accurate data points in the historical radiation intensity data can also be determined by the multiple repeated measurement method, specifically: Multiple repeated measurement method Method: Repeatedly measure the same radiation source under the same conditions, and identify accurate data through statistical analysis.

[0051] Steps: Under the same environmental conditions, measure the same radiation source multiple times (such as 10 times or more).

[0052] Calculate the average value and standard deviation of the multiple measurements.

[0053] Use the average value as an estimate of the accurate radiation level.

[0054] Mark the measurement values close to the average value or less than or exceeding a certain threshold (such as 2 times the standard deviation) as accurate data.

[0055] In a possible embodiment, accurate data points in the historical radiation intensity data can also be determined by comparing the measurement results of different detectors, specifically: Multiple detector comparison method Method: Use multiple different types of detectors to measure the same radiation source simultaneously, and identify data by comparing the results.

[0056] Steps: Under the same environmental conditions, use multiple detectors (such as gamma-ray detectors, neutron detectors) to measure the same radiation source simultaneously.

[0057] Record the measurement results of each detector.

[0058] Calculate the average value of the measurement results of all detectors.

[0059] Use the average value as an estimate of the accurate radiation level.

[0060] Mark the measurement values with large deviations from the average value as distorted data, and similarly, accurate data can be obtained.

[0061] In a possible embodiment, the model simulation method or the expert annotation method can also be used for data marking. Specifically: Use a physical model or numerical simulation to generate accurate radiation level data as the training target.

[0062] Steps: According to the physical characteristics of the radionuclide (such as decay constant, energy spectrum) and the response characteristics of the detector, establish a radiation transport model.

[0063] Use numerical simulation methods (such as Monte Carlo simulation) to generate accurate radiation level data.

[0064] Mark the simulation data as accurate radiation level data.

[0065] Compare the actual measurement data with the simulation data to identify the data.

[0066] Expert annotation method Method: Invite experts in the field of nuclear radiation detection to annotate the data and identify accurate and distorted radiation level data.

[0067] Step 3: Use the first input sample set as the input and the first output sample set as the output to construct a neural network model and train the neural network model until its accuracy reaches the set expected value; Among them, the neural network model includes: The first neural network model constructed using the first neural network module; The second neural network model constructed using the second neural network module; And the third neural network model constructed using the third neural network module.

[0068] In a possible embodiment, a convolutional neural network (CNN) can be used as the first neural network model. The convolutional neural network (CNN) is used to process spatial features, such as the relative position of the detector and the radiation source, the shielding effect of surrounding substances, etc., and is suitable for capturing local features and spatial correlations. See Figure 3 the structural diagram of the first neural network model shown. The specific construction process of the model includes the following steps: Data preprocessing Normalization: Normalize the input data and output data to the same range (such as 0 to 1) so that the neural network can process them better. The input data is the first input sample set The data of various interference factors therein; the output data is the first output sample set The radiation intensity data therein.

[0069] Data augmentation: If the amount of data is insufficient, more training samples can be generated by adding noise, random offsets, etc.

[0070] Data partitioning: Partition the data into a training set, a validation set, and a test set.

[0071] Construct a Convolutional Neural Network (CNN) model Input layer The dimension of the input layer should be consistent with the dimension of the input data.

[0072] Convolutional layer The convolutional layer is the core part of the CNN, used to extract local features of the input data. Multiple convolutional layers can be set, and each convolutional layer contains multiple convolutional kernels (filters).

[0073] Each convolutional kernel slides on the input data, calculates the convolution operation, and generates feature data (feature map).

[0074] Pooling layer The pooling layer is used to reduce the dimension of the feature data, reduce the amount of computation, and retain important features. Commonly used pooling operations include MaxPooling and AveragePooling.

[0075] Activation layer The activation layer is used to introduce non-linearity so that the model can learn complex feature relationships. Commonly used activation functions include ReLU (Rectified Linear Unit), Sigmoid, and Tanh, etc.

[0076] Fully connected layer The fully connected layer is used to integrate the features extracted by the convolutional layer and the pooling layer and output the final prediction result. The number of neurons in the fully connected layer can be set according to needs.

[0077] Output layer The number of neurons in the output layer should be consistent with the dimension of the output data. For the radiation intensity data, the output layer usually has only one neuron, representing the predicted radiation intensity value.

[0078] The activation function of the output layer usually selects a linear activation function (linear) because the radiation intensity is a continuous numerical value.

[0079] Model training Loss function Since the radiation intensity is a regression problem, the loss function usually chooses the mean squared error (MSE) or the mean absolute error (MAE).

[0080] Optimization algorithm The optimization algorithm usually chooses variants of the gradient descent method, such as the Adam optimizer.

[0081] Training process Forward propagation: The input data passes through the convolutional layer, pooling layer, activation layer, and fully connected layer, and finally reaches the output layer to calculate the predicted radiation intensity.

[0082] Calculate the loss: Use the loss function to calculate the difference between the predicted value and the true value.

[0083] Backward propagation: Calculate the gradient of the loss function with respect to each parameter through the backward propagation algorithm, and use the optimization algorithm to update the weights and biases of the network.

[0084] Iterative training: Repeat the above steps until the loss function converges to a smaller value.

[0085] Model evaluation Validation set evaluation Use the validation set to evaluate the performance of the model and avoid overfitting. When the performance of the model does not reach the set expected value, it is necessary to adjust the model parameters and re-train and optimize the model.

[0086] Test set evaluation Use the test set to evaluate the final performance of the model to ensure that the model has good generalization ability on unseen data.

[0087] In a possible embodiment, a long short-term memory network (LSTM) can be used as the second neural network model. The long short-term memory network (LSTM) is used to process time series features, such as continuously monitored radiation data, and is suitable for capturing dynamic changes and long-term dependencies over time. See Figure 4 The structural diagram of the second neural network model shown. The specific construction process of the model includes the following steps: Data preprocessing Normalization: Normalize the input data and output data to the same range (such as 0 to 1) so that the neural network can process it better. The input data is the interference factor data in the first input sample set ; the output data is the radiation intensity data in the first output sample set .

[0088] Time series format: Convert the data into a time series format suitable for LSTM processing, usually a three-dimensional tensor (number of samples, number of time steps, number of features).

[0089] Data Partitioning: Divide the data into a training set, a validation set, and a test set.

[0090] Build a Long Short-Term Memory (LSTM) model Input Layer The dimension of the input layer should be consistent with the dimension of the input data. For example, if the input data is a time series, the input layer should be a three-dimensional input (number of samples, number of time steps, number of features).

[0091] LSTM Layer The LSTM layer is the core part of the model, used to capture the dynamic changes and long-term dependencies in time series data. Multiple LSTM layers can be set, and each LSTM layer contains multiple units.

[0092] Each LSTM unit controls the flow of information through a gating mechanism (input gate, forget gate, output gate), thus effectively handling the long-term dependence problem.

[0093] Fully Connected Layer The fully connected layer is used to integrate the features extracted by the LSTM layer and output the final prediction result. The number of neurons in the fully connected layer can be set as needed.

[0094] Output Layer The number of neurons in the output layer should be consistent with the dimension of the output data. For radiation intensity data, the output layer usually has only one neuron, representing the predicted radiation intensity value.

[0095] The activation function of the output layer usually selects the linear activation function (linear) because the radiation intensity is a continuous numerical value.

[0096] Model Training Loss Function Since the radiation intensity is a regression problem, the loss function usually selects the mean squared error (MSE) or the mean absolute error (MAE).

[0097] Optimization Algorithm The optimization algorithm usually selects a variant of the gradient descent method, such as the Adam optimizer.

[0098] Training Process Forward Propagation: The input data passes through the LSTM layer and the fully connected layer, and finally reaches the output layer to calculate the predicted radiation intensity.

[0099] Calculate Loss: Use the loss function to calculate the difference between the predicted value and the true value.

[0100] Backward Propagation: Calculate the gradient of the loss function with respect to each parameter through the backward propagation algorithm, and use the optimization algorithm to update the weights and biases of the network.

[0101] Iterative training: Repeat the above steps until the loss function converges to a smaller value.

[0102] Model evaluation Validation set evaluation Use the validation set to evaluate the performance of the model and avoid overfitting. When the performance of the model does not reach the set expected value, the model parameters need to be adjusted and the model training and optimization need to be carried out again.

[0103] Test set evaluation Use the test set to evaluate the final performance of the model and ensure that the model has good generalization ability on unseen data.

[0104] In a possible embodiment, a fully connected neural network (FCNN) can be used as the third neural network model. The fully connected neural network (FCNN) is used to process general features, such as environmental factors like temperature, humidity, air pressure, etc., and is suitable for capturing global features and simple linear relationships. See Figure 5 The structure diagram of the third neural network model shown. The specific construction process of the model includes the following steps: Data preprocessing Normalization: Normalize the input data and output data to the same range (such as 0 to 1) so that the neural network can process it better. The input data is the interference factor data in the first input sample set ; the output data is the radiation intensity data in the first output sample set in.

[0105] Data partitioning: Divide the data into a training set, a validation set, and a test set.

[0106] Construct a fully connected neural network (FCNN) model Input layer The number of neurons in the input layer should be the same as the number of features of the input data.

[0107] Hidden layer The hidden layer is used to extract the features of the input data. Multiple hidden layers can be set, and each hidden layer contains multiple neurons.

[0108] Each hidden layer usually uses a non-linear activation function (such as ReLU) to introduce non-linearity so that the model can learn complex feature relationships.

[0109] Output layer The number of neurons in the output layer should be the same as the dimension of the output data. For the radiation intensity data, the output layer usually has only one neuron, representing the predicted radiation intensity value.

[0110] The activation function of the output layer is usually chosen as the linear activation function (linear) because the radiation intensity is a continuous value.

[0111] Model training Loss function Since the radiation intensity is a regression problem, the loss function is usually chosen as the mean squared error (MSE) or the mean absolute error (MAE).

[0112] Optimization algorithm The optimization algorithm usually chooses a variant of the gradient descent method, such as the Adam optimizer.

[0113] Training process Forward propagation: The input data passes through the hidden layer and finally reaches the output layer to calculate the predicted radiation intensity.

[0114] Calculate the loss: Use the loss function to calculate the difference between the predicted value and the true value.

[0115] Backward propagation: Calculate the gradient of the loss function with respect to each parameter through the backward propagation algorithm, and use the optimization algorithm to update the weights and biases of the network.

[0116] Iterative training: Repeat the above steps until the loss function converges to a smaller value.

[0117] Model evaluation Validation set evaluation Evaluate the performance of the model using the validation set to avoid overfitting. When the performance of the model does not meet the set expectations, the model parameters need to be adjusted and the model training and optimization need to be carried out again.

[0118] Test set evaluation Evaluate the final performance of the model using the test set to ensure that the model has good generalization ability on unseen data.

[0119] Step 4: Using the second input sample set as the input, utilize the trained neural network model to obtain the prediction sample set corresponding to the second input sample set .

[0120] Among them, the prediction sample set includes the first prediction sample set , the second prediction sample set and the third prediction sample set ; The first prediction sample set is the output using the second input sample set as the input and the trained first neural network model; The second prediction sample set is the output using the second input sample set​ is the input and uses the output of the second neural network model that has been trained; The third prediction sample set uses the second input sample set as the input and uses the output of the third neural network model that has been trained.

[0121] Step 5: Determine the correction coefficient at the level of the real-time interference factor data collected currently according to the second output sample set and the prediction sample set . .

[0122] Among them, the correction coefficient is the weighted value of the first correction coefficient , the second correction coefficient and the third correction coefficient ; the weighted calculation formula is: = × + × + × ; in the formula, is the weighting coefficient of the first correction coefficient ; is the weighting coefficient of the second correction coefficient ; is the weighting coefficient of the third correction coefficient ; , and are all within the interval (0, 1), and + + = 1.

[0123] The first correction coefficient is determined by the first correction coefficient acquisition module according to the second output sample set and the first prediction sample set . The second correction coefficient is determined by the second correction coefficient acquisition module according to the second output sample set and the second prediction sample set . The third correction coefficient is determined by the third correction coefficient acquisition module according to the second output sample set and the third prediction sample set .

[0124] Specifically, the acquisition process of the correction coefficient includes the following steps: Obtain the real-time interference factor data collected currently , where is expressed as the th item of real-time interference factor data, = 1, 2,..., ; is the total number of interference factor types; Define that when the Euclidean distance between data is within the threshold , two groups of interference factor data are in the same level state. According to the defined content, filter out all the interference factor data in the second input sample set that is in the same level state as the real-time interference factor data , where is expressed as the th data of all the interference factors in the second input sample set that is in the same level state as the real-time interference factor data; = ; ; Extract the unlabeled radiation intensity data corresponding to the interference factor data in the second output sample set ; ; Extract the first predicted radiation intensity data output by using the trained first neural network model with the interference factor data as the input in the first prediction sample set , and calculate the first correction coefficient by using the extracted data. The calculation formula is: = ; In the formula, is the total amount of the extracted data; Extract the second predicted radiation intensity data output by using the trained second neural network model with the interference factor data as the input in the second prediction sample set , and calculate the second correction coefficient by using the extracted data. The calculation formula is: = ; Extract the third predicted radiation intensity data output by using the trained third neural network model with the interference factor data as the input in the third prediction sample set , and calculate the third correction coefficient by using the extracted data. The calculation formula is: = ; Perform linear weighted processing on the obtained first correction coefficient , second correction coefficient and third correction coefficient , and use the weighted value as the correction coefficient .

[0125] Step 6: Use the correction coefficient to correct the currently collected real-time radiation intensity data and obtain the corrected radiation intensity data.

[0126] Among them, the corrected radiation intensity data = real-time radiation intensity data × correction coefficient .

[0127] The above details the method of the embodiments of the present invention. The following provides the system structure of the embodiments of the present invention, and the specific structure block diagram is as shown in Figure 2 shown.

[0128] The system includes: A data acquisition module for acquiring radiation intensity data and corresponding interference factor data; The data acquisition module includes: A first data acquisition sub-module for acquiring real-time radiation intensity data; A second data acquisition sub-module for obtaining historical radiation intensity data; A third data acquisition sub-module for obtaining real-time interference factor data corresponding to the real-time radiation intensity data; A fourth data acquisition sub-module for obtaining historical interference factor data corresponding to the historical radiation intensity data; A data marking module for marking accurate data points in the radiation intensity data; A sample set construction module for constructing a data sample set, including: A first output sample set construction module for constructing a first output sample set with the marked accurate data points ; A first input sample set construction module for constructing a first input sample set with the interference factor data corresponding to the first output sample set ; A second output sample set construction module for constructing a second output sample set with the unmarked radiation intensity data ; A second input sample set construction module for constructing a second input sample set with the interference factor data corresponding to the second output sample set .

[0129] For constructing with the first input sample set is the input, and the first output sample set is the neural network module of the neural network model for output; The neural network module includes a first neural network module, a second neural network module, and a third neural network module; The predicted sample set acquisition module is used to use the second input sample set as the input, and use the trained neural network model to obtain the predicted sample set corresponding to the second input sample set ; ; Among them, the predicted sample set acquisition module includes a first predicted sample set acquisition module, a second predicted sample set acquisition module, and a third predicted sample set acquisition module; The correction coefficient acquisition module is used to determine the correction coefficient at the current level of real-time interference factor data collected according to the second output sample set and the predicted sample set ; ; The correction coefficient acquisition module includes a first correction coefficient acquisition module, a second correction coefficient acquisition module, and a third correction coefficient acquisition module; The correction coefficient is the weighted value of the first correction coefficient , the second correction coefficient and the third correction coefficient ; The first correction coefficient is determined by the first correction coefficient acquisition module according to the second output sample set and the first predicted sample set ; The second correction coefficient is determined by the second correction coefficient acquisition module according to the second output sample set and the second predicted sample set ; The third correction coefficient is determined by the third correction coefficient acquisition module according to the second output sample set and the third predicted sample set ;

[0130] The data correction module is used to correct the currently collected real-time radiation intensity data by using the correction coefficient to obtain the corrected radiation intensity data.

[0131] The system further includes a memory, a processor, and a computer program stored on the memory and executable on the processor.

[0132] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and all these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based nuclear radiation detection system, characterized in that, The detection process of the system includes the following steps: Collect radiation intensity data and corresponding interference factor data, where the interference factor data is the data that affects the radiation intensity data; Mark the accurate data points in the radiation intensity data, and construct a first output sample set with the marked accurate data points , and construct a first input sample set with the interference factor data corresponding to the accurate data points , and construct a second output sample set with the unmarked radiation intensity data , and construct a second input sample set with the interference factor data corresponding to the unmarked radiation intensity data ; Using the first input sample set as the input and the first output sample set as the output, a neural network model is constructed and the neural network model is trained until its accuracy rate reaches the set expected value; Using the second input sample set as the input, and by means of the trained neural network model, obtain a predicted sample set corresponding to the second input sample set ; Based on the second output sample set and the predicted sample set determine the correction coefficient at the level of real-time interference factor data currently collected , and use the correction coefficient to correct the radiation intensity data currently collected, and obtain the corrected radiation intensity data.

2. The nuclear radiation detection system based on artificial intelligence according to claim 1, characterized in that The system includes: A data acquisition module for collecting radiation intensity data and corresponding interference factor data; The data acquisition module includes: A first data acquisition sub-module for collecting real-time radiation intensity data; A second data acquisition sub-module for obtaining historical radiation intensity data; A third data acquisition sub-module for obtaining real-time interference factor data corresponding to the real-time radiation intensity data; A fourth data acquisition sub-module for obtaining historical interference factor data corresponding to the historical radiation intensity data; Among them, the marked radiation intensity data is the historical radiation intensity data.

3. The nuclear radiation detection system based on artificial intelligence according to claim 2, wherein, The system further includes: A data marking module for marking accurate data points in the radiation intensity data; A sample set construction module for constructing a data sample set, including: The first output sample set construction module is used to construct a first output sample set with the marked accurate data points ; The first input sample set construction module is configured to construct a first input sample set by using the interference factor data corresponding to the first output sample set ; The second output sample set construction module is used to construct a second output sample set with the unlabeled radiation intensity data ; The second input sample set construction module is used to construct a second input sample set with the interference factor data corresponding to the second output sample set .

4. An artificial intelligence-based nuclear radiation detection system according to claim 3, characterized in that, The system further includes: A neural network module for constructing a neural network model that uses the first input sample set as input and the first output sample set as output; Among them, the neural network module includes a first neural network module, a second neural network module, and a third neural network module; The neural network model includes: A first neural network model constructed using the first neural network module; A second neural network model constructed using the second neural network module; And a third neural network model constructed using the third neural network module.

5. An artificial intelligence-based nuclear radiation detection system according to claim 4, characterized in that, The system further includes: A prediction sample set acquisition module, which uses the second input sample set as input, and utilizes the trained neural network model to obtain a prediction sample set corresponding to the second input sample set ; ; Among them, the prediction sample set acquisition module includes a first prediction sample set acquisition module, a second prediction sample set acquisition module, and a third prediction sample set acquisition module; The predicted sample set includes a first predicted sample set , a second predicted sample set and a third predicted sample set ; The first prediction sample set uses the second input sample set as input and uses the output of the trained first neural network model; The second predicted sample set uses the second input sample set as input and is the output of the trained second neural network model; The third prediction sample set uses the second input sample set as input and is the output of the trained third neural network model.

6. An artificial intelligence-based nuclear radiation detection system according to claim 5, characterized in that, The system further includes: A correction coefficient acquisition module, configured to determine a correction coefficient at the level of real-time interference factor data currently collected according to the second output sample set and the prediction sample set ; ; The correction coefficient acquisition module includes a first correction coefficient acquisition module, a second correction coefficient acquisition module, and a third correction coefficient acquisition module; The correction coefficient is the weighted value of the first correction coefficient , the second correction coefficient and the third correction coefficient ; The first correction coefficient is determined by the first correction coefficient obtaining module according to the second output sample set and the first prediction sample set ; The second correction coefficient is determined by the second correction coefficient acquisition module according to the second output sample set and the second prediction sample set ; The third correction coefficient is determined by the third correction coefficient obtaining module according to the second output sample set and the third prediction sample set for determination.

7. An artificial intelligence-based nuclear radiation detection system according to claim 6, characterized in that The system further includes: A data correction module for correcting the currently collected real-time radiation intensity data by using the correction coefficient to obtain the corrected radiation intensity data.

8. An artificial intelligence-based nuclear radiation detection system according to claim 6, characterized in that, The correction coefficient is obtained through the following steps: Obtain the real-time interference factor data collected currently , where is expressed as the th item of real-time interference factor data, = 1, 2,..., ; is the total number of interference factor types; Define the Euclidean distance between data When within the threshold range, the two sets of interference factor data are in the same level state. According to the defined content, filter out all the interference factor data in the second input sample set that are in the same level state as the real-time interference factor data , where is expressed as the th interference factor in the second input sample set that is in the same level state as the real-time interference factor data and the th data; Extract the second output sample set among which, the unlabeled radiation intensity data corresponding to the interference factor data ; Extract the first prediction sample set Among them, using the interference factor data as the input, the first predicted radiation intensity data output by the trained first neural network model , calculate the first correction coefficient using the extracted data , and the calculation formula is: = ; In the formula, is the total amount of extracted data; Extract the second prediction sample set Among them, using the interference factor data as the input, the second predicted radiation intensity data output by the trained second neural network model , and calculate the second correction coefficient using the extracted data , and the calculation formula is: = ; Extract the third prediction sample set Among them, using the interference factor data as the input, the third predicted radiation intensity data output by the trained third neural network model , and calculate the third correction coefficient using the extracted data , and the calculation formula is: = ; For the obtained first correction coefficient 、the second correction coefficient and the third correction coefficient perform linear weighting processing, and use the weighted value as the correction coefficient , where the calculation formula is: = × + × + × ; in the formula, is the weighting coefficient of the first correction coefficient ; is the weighting coefficient of the second correction coefficient ; is the weighting coefficient of the third correction coefficient ; , and are all within the interval (0, 1), and + + = 1.

9. An artificial intelligence-based nuclear radiation detection system according to claim 7, characterized in that, Corrected radiation intensity data = Real-time radiation intensity data × Correction factor .

10. An artificial intelligence-based nuclear radiation detection system according to any one of claims 1-9, characterized in that, The system further includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it can implement the steps of the detection process described in claim 1.

Images