Method for identifying risk of false triggering of detection of radioactive substances based on image recognition

Abstract

The radioactive substance detection false trigger risk identification method based on image recognition provided in the application relates to the technical field of intelligent detection and safety protection. First, based on temperature and humidity, illumination and other sensor data, the image acquisition parameters are dynamically adjusted to solve the image quality fluctuation problem in complex environments. Second, the hardware level synchronous trigger and time stamp calibration technology is used to realize the spatio-temporal alignment of multi-modal images, and the metal reflector and the thermal anomaly area are accurately associated through feature fusion. At the same time, the theoretical temperature is dynamically calibrated through the thermodynamic compensation model, which greatly improves the false trigger identification accuracy. Finally, the comprehensive false trigger risk probability is determined from multiple dimensions and the corresponding countermeasures are taken. Through dynamic environment perception and visible light / infrared image collaborative processing technology, the application constructs a full-link closed-loop system from parameter optimization to risk control, which significantly improves the accuracy and safety of foreign matter detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent detection and safety protection, and particularly relates to a radioactive substance detection false trigger risk identification method based on image recognition. BACKGROUND

[0002] In the fields of industrial detection, safety protection, etc., X-ray detection equipment is widely used for foreign object identification and risk warning. However, the traditional detection system is prone to false triggering under complex environmental interference, especially factors such as metal reflectors, temperature and humidity changes, and environmental thermal noise, which seriously affect the detection stability and accuracy. In the prior art, most systems rely on single modal images (such as visible light or infrared) for judgment, which is difficult to fully reflect the optical and thermodynamic characteristics of the target, resulting in a high false judgment rate. In addition, the image acquisition parameters are usually fixedly set, lacking dynamic adaptability to the real-time environment, further exacerbating the false trigger risk. In order to improve the detection accuracy, some schemes introduce multi-modal image fusion technology, but generally do not fully consider the spatio-temporal alignment error, environmental temperature compensation mechanism, and adaptability of feature fusion strategy, which limits its effect in practical application. SUMMARY

[0003] The purpose of the present application is to provide a radioactive substance detection false trigger risk identification method and system based on image recognition, to realize high-precision identification and rapid response of foreign object false trigger risk, and to improve the intelligent level and operation reliability of the detection system.

[0004] The present application provides a radioactive substance detection false trigger risk identification method based on image recognition, which comprises:

[0005] S1: acquiring first image acquisition data according to first environmental information, performing spatio-temporal alignment processing and first image preprocessing on the first image acquisition data to determine target multi-modal image data;

[0006] S2: performing interference source detection processing and thermodynamic verification processing on the target multi-modal image data to determine first fused foreign object features, first object categories and first deviation temperatures;

[0007] S3: determining a comprehensive false trigger risk probability according to the first fused foreign object features, the first object categories and the first deviation temperatures;

[0008] S4: determining a first response strategy according to the comprehensive false trigger risk probability.

[0009] Preferably, the S1 comprises:

[0010] S11: determining a first set of setting parameters based on at least one first environmental sensor;

[0011] S12: acquire first image acquisition data based on the first set of setting parameters, and perform spatio-temporal alignment processing on the first image acquisition data to determine first multi-modal image data;

[0012] S13: perform first image preprocessing on the first multi-modal image data to obtain target multi-modal image data.

[0013] Preferably, the S11 comprises:

[0014] S111: acquire a first set of environmental parameters according to at least one first environmental sensor;

[0015] S112: input the first set of environmental parameters into a first setting information determination model to determine a first set of setting parameters.

[0016] Preferably, the S12 comprises:

[0017] S121: perform a first setting operation on a first visible light acquisition device and a first infrared light acquisition device based on the first set of setting parameters;

[0018] S122: perform a first synchronous triggering operation on the first visible light acquisition device and the first infrared light acquisition device to obtain first visible light data and first infrared light data;

[0019] S123: perform spatio-temporal alignment processing on the first visible light data and the first infrared light data according to timestamp data to determine first multi-modal image data.

[0020] Preferably, the first image preprocessing comprises visible light image enhancement and infrared thermal image calibration.

[0021] Preferably, the S2 comprises:

[0022] S21: acquire first visible light data from the target multi-modal image data, perform interference source detection processing on the first visible light data, and obtain a first false triggering probability;

[0023] S22: when the first false triggering probability is greater than a first preset value, perform thermodynamic verification processing on the target multi-modal image data to obtain a second false triggering probability;

[0024] S23: when the second false triggering probability is greater than a third preset value, perform adaptive feature fusion processing on the target multi-modal image data to determine first fused foreign object features.

[0025] Preferably, the S21 comprises the following sub-steps:

[0026] S211: extract first visible light data from the target multi-modal image data, obtain a first object category based on the first visible light data;

[0027] S212: calculate a first reflection light intensity based on the first visible light data;

[0028] S213: input the first object category and the first reflection light intensity into a first false trigger probability determination model to obtain a first false trigger probability.

[0029] Preferably, the S22 comprises the following sub-steps:

[0030] S221: in response to the first false trigger probability being greater than a first preset value, extract an infrared thermal image sequence from the target multi-modal image data;

[0031] S222: extract a first foreign matter area based on the infrared thermal image sequence and the first visible light data;

[0032] S223: perform temperature anomaly detection processing on the infrared thermal image sequence to obtain a first temperature anomaly area;

[0033] S224: calculate a first coincidence degree of the first foreign matter area and the first temperature anomaly area, and when the first coincidence degree is greater than a second preset value, determine a second false trigger probability according to a first bias temperature.

[0034] Preferably, the S4 comprises the following sub-steps:

[0035] S41: when the comprehensive false trigger risk probability is greater than a fourth preset value, immediately cut off the power supply of the X-ray tube;

[0036] S42: when the comprehensive false trigger risk probability is greater than a fifth preset value and less than or equal to the fourth preset value, immediately cut off the power supply of the X-ray tube.

[0037] Preferably, the fusion formula in the adaptive feature fusion processing comprises:

[0038] F fusion = α × F vis + (1-α) × F IR

[0039] Wherein, F fusion is the first fused foreign matter feature, α is a first weight, F vis is a visible light feature, and F IR is an infrared feature.

[0040] The radioactive material detection false trigger risk identification method based on image recognition provided in the application relates to the technical field of intelligent detection and safety protection. First, based on temperature and humidity, illumination and other sensor data, the image acquisition parameters are dynamically adjusted to solve the image quality fluctuation problem in complex environments. Second, the hardware level synchronous trigger and timestamp calibration technology is used to realize the spatio-temporal alignment of multi-modal images, and the metal reflector and thermal anomaly area are accurately associated through feature fusion. At the same time, the theoretical temperature is dynamically calibrated through the thermodynamic compensation model, which greatly improves the false trigger identification accuracy. Finally, the comprehensive false trigger risk probability is determined from multiple dimensions and the corresponding countermeasures are taken. Through dynamic environment perception and visible light / infrared image collaborative processing technology, the application constructs a full-link closed-loop system from parameter optimization to risk control, which significantly improves the accuracy and safety of foreign matter detection. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only exemplary, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0042] Figure 1 is the execution flowchart of the radioactive material detection false trigger risk identification method based on image recognition in the application.

[0043] Figure 2 is the first fusion foreign matter feature determination flowchart in step S2 in the technical solution of the application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely in the embodiments of the present application in combination with the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0045] The technical solutions in the embodiments of the present application will be described clearly and completely in the embodiments of the present application in combination with the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0046] The radioactive material detection false trigger risk identification method based on image recognition in the present application will be described in detail.

[0047] The radioactive material detection false trigger risk identification method based on image recognition provided in the present embodiment is shown in the specific flowchart as Figure 1 ​

[0048] S1: obtaining first image acquisition data according to first environment information, performing spatio-temporal alignment processing and first image preprocessing on the first image acquisition data to determine target multi-modal image data.

[0049] In this step, visible light data and infrared light data are required for multi-modal feature analysis in step S2. In this step, the image acquisition parameters are optimized through environment data analysis to improve image quality and reduce false trigger risk.

[0050] The S1 specifically includes the following sub-steps:

[0051] S11: determining a first set of setting parameters based on at least one first environment sensor.

[0052] Since the best acquisition parameters for the image acquisition device are not the same under different environmental conditions, such as sunny and rainy days, in this step, the image acquisition parameters need to be dynamically adjusted according to real-time environment data (such as light intensity, temperature, humidity) to ensure the quality of image data acquisition.

[0053] The S11 includes the following sub-steps:

[0054] S111: obtaining a first set of environment parameters based on at least one first environment sensor.

[0055] Among them, the at least one first environment sensor includes one or more of a temperature sensor, a humidity sensor, and a light sensor. For example, it can include a temperature and humidity sensor (DS18B20), a light sensor (BH1750).

[0056] The first set of environment parameters includes one or more of temperature, humidity, and light parameters of the environment in which the image acquisition device is currently located.

[0057] S112: inputting the first set of environment parameters into a first setting information determination model to determine a first set of setting parameters.

[0058] The first setting information determination model can be obtained by training a machine learning model. In the training sample, the first set of environment parameters composed of one or more of temperature, humidity, and light parameters can be used as input information, and the exposure parameters of the light camera and the sampling frequency of the infrared detector can be used as output information, thereby completing the above training process.

[0059] Preferably, the first setting information determination model is a convolutional neural network model.

[0060] Preferably, the exposure parameters of the visible light camera specifically include ISO value and shutter speed.

[0061] In the implementation process, the FPGA can run the parameter optimization model in real time, and the acquisition parameters are updated every 10 seconds. For example, the ISO value is 1600 and the shutter speed is 1 / 500s in the rainy environment model, so that the visible light image is still clear in low light. When the environmental humidity is greater than 70%, the model automatically reduces the sampling frequency of the infrared detector to 500Hz to reduce the water vapor interference.

[0062] S12: Obtain first image acquisition data based on the first set of setting parameters, and perform spatio-temporal alignment processing on the first image acquisition data to determine first multi-modal image data.

[0063] In this step, the visible light and infrared light acquisition devices need to be set according to the image acquisition device setting parameters determined in S11, and the corresponding data acquisition is completed. The time alignment processing is performed on the acquired data to solve the time misalignment problem of visible light and infrared images, to provide high-quality input for subsequent feature fusion, so as to obtain the first multi-modal image data.

[0064] The S12 specifically can include the following sub-steps:

[0065] S121: Perform first setting operation on the first visible light acquisition device and the first infrared light acquisition device based on the first set of setting parameters.

[0066] In this step, the visible light acquisition device and the infrared light acquisition device need to be set according to the information contained in the first set of setting parameters.

[0067] S122: Perform first synchronous triggering operation on the first visible light acquisition device and the first infrared light acquisition device to obtain first visible light data and first infrared light data.

[0068] The first synchronous triggering operation includes hardware level synchronous triggering, which specifically can use the shared power trigger signal of the infrared detector and the visible light camera to realize synchronization through the GPIO interrupt pin. When the infrared detector detects an object, the rising edge signal directly drives the visible light camera to expose, and the time deviation is controlled within a preset time range.

[0069] Preferably, during the process of image data acquisition, a microsecond-level timestamp (format: YYYY-MM-DD HH:MM:SS.Microseconds) is added to each frame of data by the built-in ARM Cortex-A9 processor real-time clock of the device. The device clock is calibrated regularly (once an hour) through the NTP protocol to ensure that the time deviation between multiple devices is <500μs.

[0070] S123: Perform spatio-temporal alignment processing on the first visible light data and the first infrared light data according to the timestamp data to determine the first multi-modal image data.

[0071] In the S122, the light data and infrared light data are synchronized by hardware triggers, which may randomly produce certain time errors due to different execution environments. Since the timestamp data is embedded in each frame of image data collected in the S122, the spatio-temporal alignment processing can be performed according to the timestamp data in this step.

[0072] In the specific implementation process, the system calibrates the collection sequence of visible light and infrared images through timestamp differences. For example, if the infrared image timestamp is 14:56:20.123 and the visible light image timestamp is 14:56:20.125, the time difference between the two is considered to be 2ms, and the visible light image can be fine-tuned through an interpolation algorithm to match the time reference of the infrared image.

[0073] After performing spatio-temporal alignment processing on the first visible light data and the first infrared light data, the first multi-modal image data is obtained.

[0074] S13: Perform first image preprocessing on the first multi-modal image data to obtain target multi-modal image data.

[0075] Since the directly collected visible light and infrared light data may have certain interference and noise, in this step, image preprocessing operations are needed to improve the signal-to-noise ratio of visible light and infrared images, eliminate environmental interference, and provide clear input for subsequent feature extraction.

[0076] The first image preprocessing specifically can include:

[0077] 1) Visible light image enhancement: Since metal objects are the most likely to cause false triggering, histogram equalization and adaptive contrast enhancement algorithm (CLAHE) can be used to improve the edge sharpness of metal reflectors. For example, the outline of a metal tweezers is sharpened to make it more easily identifiable in the visible light image.

[0078] 2) Infrared thermal image calibration: the response of the infrared detector presents a certain non-uniformity at different ambient temperatures. Therefore, the non-uniform response of the infrared detector is corrected by a temperature calibration table to eliminate the interference of the ambient temperature. For example, when the ambient temperature is 25°C, the system automatically adjusts the baseline value of the infrared thermal image to ensure that ΔT = 0°C corresponds to the true ambient temperature.

[0079] Preferably, after the S13, further comprising:

[0080] A lightweight CNN model (MobileNetV2) is constructed to evaluate the image quality and output a definition score (0-1). If the score is lower than a preset threshold, a resampling mechanism is triggered. Preferably, the preset threshold can be set to 0.7.

[0081] Preferably, in the target multi-modal image data, the visible light image and the infrared thermal image are both stored as 16-bit grayscale images (size 512x512 pixels), and the original timestamp is retained for subsequent time alignment.

[0082] S2: performing interference source detection processing and thermodynamic verification processing on the target multi-modal image data to determine first foreign object features, first object categories, and first deviation temperatures.

[0083] Since using visible light or infrared images alone for foreign object detection will have the problem of low detection accuracy, in this step, the features of the visible light and infrared images in the target multi-modal image data are fused to obtain first foreign object features for verifying whether the foreign object is likely to cause the detector to misread.

[0084] The S2 includes the following sub-steps, as shown in Figure 2

[0085] S21: obtaining first visible light data from the target multi-modal image data, performing interference source detection processing on the first visible light data, and obtaining a first false trigger probability.

[0086] In this step, potential interference sources (such as metal reflectors) need to be identified and their reflectivity intensity needs to be quantified to provide a basis for subsequent thermodynamic verification.

[0087] The S21 includes the following sub-steps:

[0088] S211: extracting first visible light data from the target multi-modal image data, and obtaining a first object category based on the first visible light data.

[0089] In this step, the first visible light data needs to be subjected to preliminary object identification processing to obtain a first object category, such as metal.

[0090] ​The specific process of object recognition processing is as follows:

[0091] The input cropped 224x224 pixel visible light image region is extracted by a lightweight MobileNetV2 model to extract a multi-dimensional feature vector, and the object type (such as "metal tool" "plastic product") is recognized. An attention mechanism is introduced to weight the feature vector and highlight key material features such as metal / plastic (the weight coefficient of the above key material features can be set to >0.6).

[0092] S212: Calculate the first reflectivity intensity based on the first visible light data.

[0093] The calculation formula of the first reflectivity intensity (R vis ) is as follows:

[0094] R vis = maximum brightness / average brightness.

[0095] S213: Input the first object category and the first reflectivity intensity into the first false trigger probability determination model to obtain the first false trigger probability.

[0096] The first false trigger probability is obtained by training historical data, and an XGBoost classification model can be used. In the sample data, the object category and reflectivity intensity in the historical data are used as input data, and the historical actual false trigger probability is used as output to train the first false trigger probability determination model.

[0097] The first false trigger probability will be used as a trigger condition for the subsequent infrared light verification step.

[0098] S22: When the first false trigger probability is greater than a first preset value, perform thermodynamic verification processing on the target multi-modal image data to obtain a second false trigger probability.

[0099] In this step, the thermal anomaly pattern of the foreign object needs to be detected to verify whether it is caused by absorption of X-rays or environmental factors, thereby determining the second false trigger probability.

[0100] The S22 includes the following sub-steps:

[0101] S221: In response to the first false trigger probability being greater than a first preset value, extract an infrared thermal image sequence from the target multi-modal image data.

[0102] The infrared thermal image sequence includes multiple infrared images, and the first preset value can be set according to specific needs, for example, it can be set to 0.8.

[0103] S222: Extract a first foreign matter region based on the infrared thermal image sequence and the first visible light data.

[0104] In this step, one or more images in the infrared thermal image sequence need to be profiled with the foreign matter identified in the first visible light data, so as to determine the region with foreign matter in the infrared thermal image sequence according to the region with foreign matter in the first visible light data, and determine it as the first foreign matter region.

[0105] S223: Perform temperature anomaly detection processing on the infrared thermal image sequence to obtain a first temperature anomaly region.

[0106] In this step, the temperature anomaly detection processing specifically includes detecting the region with local overheating in the infrared thermal image sequence. Specifically, it can include:

[0107] Input a continuous 10-frame 50x50-pixel infrared thermal image sequence, extract a 128-dimensional feature vector through a 3-layer 3D-CNN (input channel = 10, convolution kernel size 3x3x3), calculate the thermal gradient direction (gradient amplitude > 1.5°C / px is considered abnormal) through the feature vector, and determine the local overheating region.

[0108] S224: Calculate a first coincidence degree of the first foreign matter region and the first temperature anomaly region, and when the first coincidence degree is greater than a second preset value, determine a second false trigger probability according to a first deviation temperature.

[0109] When the coincidence degree of the first foreign matter region and the first temperature anomaly region is greater than the second preset value, it means that the first foreign matter region containing metal is also in a local overheating condition, and therefore there is a high risk of false triggering.

[0110] At this time, the first deviation temperature of the first temperature anomaly region from the theoretical temperature needs to be calculated. The theoretical temperature is predicted based on finite element analysis (FEA) simulation of the heat conduction path of the metal tool under X-ray irradiation. If the temperature of the first temperature anomaly region deviates from the theoretical temperature by >10%, it is marked as "thermodynamic anomaly", and the second false trigger probability is generated. The second false trigger probability is positively correlated with the first deviation temperature.

[0111] In a preferred embodiment, the theoretical heat value of the infrared thermal image needs to be dynamically calibrated by a thermodynamic compensation model to eliminate the influence of the ambient temperature on the determination of the first deviation temperature.

[0112] Thermodynamic compensation model: dynamically calibrate the theoretical heat value Q based on the T-shaped heat dissipation groove temperature rise curve of the device theory :

[0113] Qtheory = η(T env ) x ∫Ddt (η is the temperature-dependent efficiency coefficient)

[0114] Wherein, η(T env ) is the heat dissipation efficiency when the ambient temperature is T env , and D is the radiation at a certain time point t.

[0115] In addition, the measurement error (noise suppression > 90%) of the first deviation temperature can be corrected in real time by Kalman filtering.

[0116] In addition, in a high temperature environment (for example, 45℃), the heat dissipation efficiency η drops to 0.7, and the risk of false triggering caused by temperature deviation increases significantly, so the system automatically reduces the determination threshold of the first deviation temperature (for example, from 5℃ to 3.5℃), to avoid false triggering due to environmental interference.

[0117] S23: When the second false triggering probability is greater than a third preset value, performing adaptive feature fusion processing on the target multi-modal image data to determine a first fused foreign matter feature.

[0118] When the second false triggering probability is large, it indicates that there is a large risk of false triggering, which needs to be processed subsequently. In the subsequent processing process, the visible light feature and the infrared feature in the target multi-modal image data need to be fused to determine the first fused foreign matter feature. The first fused foreign matter feature refers to the foreign matter feature corresponding to the first temperature anomaly region, which contains features of visible light and infrared light.

[0119] The adaptive feature fusion processing includes:

[0120] A cross-attention module (Cross-Attention) based on Transformer is used to dynamically allocate the weights of visible light and infrared features. The formula of the fused feature is as follows:

[0121] F fusion = α x F vis + (1-α) x F IR

[0122] Wherein, F fusion is the first fused foreign matter feature, α is the first weight, mainly affected by meteorological factors, F vis is the visible light feature, and F IR is the infrared feature.

[0123] For example, in rainy weather, the visible light is blurred (α can be set to 0.2), the weight of the infrared feature accounts for 80%, and the foreign matter is determined to be a wet wood piece (non-risk), to avoid false triggering of the radiation to be turned off.

[0124] S3: determining a comprehensive false trigger risk probability according to the first fusion foreign matter feature, the first object category and the first deviation temperature.

[0125] In this step, the first fusion foreign matter feature, the first object category and the first deviation temperature can be input into a second false trigger probability determination model to output the comprehensive false trigger risk probability.

[0126] The second false trigger probability determination model is preferably obtained by training a convolutional neural network model, in which multiple different layers need to be set for multiple categories of features respectively. In sample data, the fusion foreign matter feature, the object category recognized by visible light recognition and the deviation temperature detected by infrared feature detection in historical data are taken as input, and the comprehensive false trigger risk probability is taken as output to complete the training of the model.

[0127] S4: determining a first response strategy according to the comprehensive false trigger risk probability.

[0128] In this step, the comprehensive false trigger risk probability needs to be converted into a device control instruction to achieve rapid response.

[0129] The S4 specifically includes the following sub-steps:

[0130] S41: immediately cutting off the power supply of the X-ray tube when the comprehensive false trigger risk probability is greater than a fourth preset value.

[0131] When the comprehensive false trigger risk probability indicates that there is currently a high risk, the power supply of the X-ray tube needs to be immediately cut off to prevent high-risk foreign matter from causing excessive radiation.

[0132] Specific means include:

[0133] Hardware-level cutting off: sending a signal to the X-ray tube driving circuit through FPGA to cut off the power supply at the hardware level (response delay <10 ms), triggering the circuit protection relay (model: Omron G6K-2P-S-Y-DC12).

[0134] Data recording: when an event is triggered, the system automatically saves the current visible light / infrared image, environmental parameters and risk index to the SQLite database.

[0135] In one example, when the comprehensive false trigger risk probability = 0.92, the FPGA immediately cuts off the power supply of the X-ray tube (delay 8 ms) to avoid excessive radiation (theoretical maximum dose 0.15 μSv) within 0.5 seconds.

[0136] Preferably, the fourth preset value is 0.7.

[0137] S42: When the comprehensive false triggering risk probability is greater than a fifth preset value and less than or equal to the fourth preset value, immediately cut off the X-ray tube power supply.

[0138] When the comprehensive false triggering risk probability indicates that there is a moderate risk at present, the operator is prompted by warning, and the event is recorded for subsequent analysis.

[0139] Specific means include:

[0140] Multi-modal warning: activate the buzzer and LED warning light, the buzzer frequency is set to 2 kHz (for 3 seconds), and the LED uses red pulse flashing (period 0.5 s). The screen displays an infrared-visible light fusion thermal map, and labels the risk area (red box + temperature value).

[0141] Operation guidance: based on the historical case database, recommend specific measures for the operator to take (such as "remove metal tools and restart the equipment").

[0142] In one example, for example, a medium-risk foreign object triggers the buzzer and LED warning, the screen displays the thermal map and labels the area with ΔT = 5°C, and the operator can manually check and remove the interference source.

[0143] Preferably, the fifth preset value is 0.4.

[0144] The radiation substance detection false triggering risk identification method based on image recognition provided in the application relates to the technical field of intelligent detection and safety protection. First, based on temperature and humidity, light and other sensor data, the image acquisition parameters are dynamically adjusted to solve the image quality fluctuation problem in complex environments. Second, hardware-level synchronous triggering and timestamp calibration technology is used to realize multi-modal image space-time alignment, and metal reflectors and thermal anomaly areas are accurately associated through feature fusion. At the same time, the theoretical temperature is dynamically calibrated through a thermodynamic compensation model, which greatly improves the false triggering identification accuracy. Finally, the comprehensive false triggering risk probability is determined from multiple dimensions, and the corresponding countermeasures are taken. Through dynamic environment perception and visible light / infrared image collaborative processing technology, the application constructs a full-link closed-loop system from parameter optimization to risk control, which significantly improves the accuracy and safety of foreign object detection.

[0145] The above is only a preferred embodiment of the application, so equivalent changes or modifications made in accordance with the structure, features and principles described in the scope of the application are included in the scope of the application.

Claims

1. A method for identifying the risk of false triggering in radioactive material detection based on image recognition, characterized in that, The method includes: S1: Obtain first image acquisition data based on first environmental information, perform spatiotemporal alignment processing and first image preprocessing on the first image acquisition data, and determine the target multimodal image data; S2: Perform interference source detection processing and thermodynamic verification processing on the target multimodal image data to determine the first fused foreign object feature, the first object category, and the first deviation temperature; S3: Determine the overall false trigger risk probability based on the first fused foreign object characteristics, the first object category, and the first deviation temperature; S4: Determine the first response strategy based on the comprehensive false trigger risk probability; S2 includes: S21: Obtain first visible light data from the target multimodal image data, perform interference source detection processing on the first visible light data, and obtain a first false trigger probability; S22: When the first false trigger probability is greater than the first preset value, perform thermodynamic verification processing on the target multimodal image data to obtain the second false trigger probability; S23: When the second false trigger probability is greater than the third preset value, perform adaptive feature fusion processing on the target multimodal image data to determine the first fused foreign object feature; S21 includes the following sub-steps: S211: Extract first visible light data from the target multimodal image data, and obtain a first object category based on the first visible light data; S212: Calculate the first reflective intensity based on the first visible light data; S213: Input the first object category and the first reflective intensity into the first false trigger probability determination model to obtain the first false trigger probability; S22 includes the following sub-steps: S221: In response to the first false trigger probability being greater than a first preset value, extract an infrared thermal image sequence from the target multimodal image data; S222: Extract the first foreign object region based on the infrared thermal image sequence and the first visible light data; S223: Perform temperature anomaly detection processing on the infrared thermal image sequence to obtain a first temperature anomaly region; S224: Calculate the first overlap degree between the first foreign object region and the first temperature anomaly region. When the first overlap degree is greater than the second preset value, determine the second false trigger probability based on the first deviation temperature.

2. The method for identifying the risk of false triggering of radioactive material detection based on image recognition according to claim 1, characterized in that, S1 includes: S11: Determine a first set of setting parameters based on at least one first environmental sensor; S12: Obtain first image acquisition data based on the first set of setting parameters, and perform spatiotemporal alignment processing on the first image acquisition data to determine the first multimodal image data; S13: Perform first image preprocessing on the first multimodal image data to obtain target multimodal image data.

3. The method for identifying the risk of false triggering of radioactive material detection based on image recognition according to claim 2, characterized in that, S11 includes: S111: Obtain a first set of environmental parameters based on at least one first environmental sensor; S112: Input the first set of environmental parameters into the first setting information determination model to determine the first set of setting parameters.

4. The method for identifying the risk of false triggering of radioactive material detection based on image recognition according to claim 3, characterized in that, S12 includes: S121: Perform a first setting operation on the first visible light acquisition device and the first infrared light acquisition device based on the first set of setting parameters; S122: Perform a first synchronization trigger operation on the first visible light acquisition device and the first infrared light acquisition device to obtain first visible light data and first infrared light data; S123: Perform spatiotemporal alignment processing on the first visible light data and the first infrared light data based on the timestamp data to determine the first multimodal image data.

5. The method for identifying the risk of false triggering of radioactive material detection based on image recognition according to claim 4, characterized in that, The first image preprocessing includes visible light image enhancement and infrared thermal image calibration.

6. The method for identifying the risk of false triggering of radioactive material detection based on image recognition according to claim 1, characterized in that, S4 includes the following sub-steps: S41: When the overall false trigger risk probability is greater than the fourth preset value, immediately cut off the power supply to the X-ray tube; S42: When the overall false trigger risk probability is greater than the fifth preset value and less than or equal to the fourth preset value, immediately cut off the power supply to the X-ray tube.

7. The method for identifying the risk of false triggering of radioactive material detection based on image recognition according to claim 6, characterized in that, The fusion formula in the adaptive feature fusion process includes: F fusion =α×F vis +(1 a)×F IR Among them, F fusion The first fused foreign object feature is α, where α is the first weight and F is the first weight. vis For visible light characteristics, F IR It has infrared characteristics.

Images