Non-contact multi-mode thermal imager temperature measurement method

Abstract

The invention provides a non-contact multi-mode thermal imager temperature measurement method, and relates to the technical field of non-contact temperature measurement, and the method comprises the steps: collecting the three-dimensional structure information of a measured object, obtaining infrared thermal radiation data, and collecting meteorological parameters; identifying the material and the surface state of the object, and adjusting an infrared emissivity parameter by using a deep learning model; performing cross-modal feature alignment on the related data; and inputting the meteorological parameters into the environment interference model to correct the infrared signal, determining whether the temperature measurement data is qualified, and if yes, repeating until the temperature measurement of the measured area is completed. According to the invention, by fusing the multi-modal data and utilizing the deep learning model, the space-time attention mechanism, the environment interference model and the like, the dynamic calibration of the infrared emissivity, the feature alignment of the multi-modal data and the effective compensation of the environment interference are realized; the problems of large environmental interference, difficult emissivity calibration, insufficient complex scene adaptability and the like of a traditional method are solved, and the temperature measurement precision and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of non-contact temperature measurement, and in particular to a non-contact multi-modal thermal imager temperature measurement method. Background Art

[0002] With the rapid development of science and technology, the demand for precise temperature measurement is growing in numerous fields, including industrial production, healthcare, and security monitoring. In industrial production, monitoring equipment operating status requires high-precision temperature data to ensure production safety and product quality. In the medical field, rapid and accurate temperature screening is a key component of disease prevention and control. In security monitoring, precise temperature detection helps to promptly detect fire hazards. However, traditional measurement methods struggle to meet these requirements in complex environments, prompting multimodal fusion temperature measurement technology to become a research hotspot.

[0003] Traditional temperature measurement methods, such as single infrared temperature measurement, are easily affected by environmental factors. For example, when the ambient temperature, humidity, and light change, the measurement results will deviate greatly. In high temperature, high humidity, or strong light environments, it is almost impossible to provide accurate data. In addition, traditional methods have difficulty dealing with objects of complex shapes and materials, and cannot accurately identify the material and surface state of the object, resulting in inaccurate emissivity calibration, which greatly reduces the temperature measurement accuracy and cannot meet the high-precision requirements of modern industry and scientific research.

[0004] Therefore, it is necessary to design a non-contact multimodal thermal imager temperature measurement method to solve the problems of existing technology, such as the temperature measurement accuracy is greatly affected by the environment, it is difficult to identify the material and surface state of complex objects, and the emissivity calibration is inaccurate. Summary of the Invention

[0005] In view of this, the present invention proposes a non-contact multi-modal thermal imager temperature measurement method, which aims to solve the problems of traditional non-contact temperature measurement being greatly affected by environmental interference, difficult emissivity calibration, and insufficient adaptability to complex scenes by fusing three-dimensional structural information, visible light image texture features, infrared thermal radiation data and meteorological parameters.

[0006] In one aspect, the present invention provides a non-contact multi-modal thermal imager temperature measurement method, comprising:

[0007] S1, collects the three-dimensional structural information and infrared thermal radiation data of the object under test, collects the meteorological parameters of the on-site environment, and collects the texture features of the object under test in the visible light image;

[0008] S2, identifying the material and surface state of the object under test based on the three-dimensional structure information and texture features of the object under test;

[0009] S3, converting the three-dimensional structure information into point cloud form, and performing cross-modal feature alignment on the infrared thermal radiation data, point cloud spatial coordinates, and visible light image semantic information of the measured object;

[0010] S4, inputting meteorological parameters into the environmental interference model to correct the infrared thermal radiation data, and judging whether the temperature measurement data is qualified. If the temperature measurement data is qualified, execute step S5; if the temperature measurement data is abnormal, execute steps S2-S4;

[0011] S5: When the temperature measurement data is qualified, repeat steps S1-S4 until the temperature measurement of the measured area is completed.

[0012] Furthermore, the method of identifying the material and surface state of the object under test based on the three-dimensional structural information and surface texture features of the object under test includes:

[0013] Perform regional gradient analysis on texture features in visible light images;

[0014] The three-dimensional structural features and regional texture features are matched with the three-dimensional structural features and regional texture features of each sample in the preset material feature library, and the material type and surface roughness are obtained.

[0015] Furthermore, when the highest matching similarity is higher than or equal to a preset matching similarity threshold, the preset emissivity parameter mapping table is queried according to the material type and surface roughness of the sample corresponding to the highest similarity matching, to obtain the corresponding infrared emissivity parameter;

[0016] When the highest matching similarity is lower than the preset matching similarity threshold, the parameters of the deep learning model are updated. According to the material type and surface roughness output by the updated deep learning model, the preset emissivity parameter mapping table is queried to obtain the corresponding infrared emissivity parameters.

[0017] Furthermore, the updating of parameters of the deep learning model includes:

[0018] Construct a mixed training dataset containing historical material samples and currently added samples, and perform data enhancement processing on the new samples;

[0019] When the sample size of the mixed training dataset reaches the preset model update threshold, the underlying feature extraction layer of the deep learning model is frozen, and only the parameters of the top classification and regression layer are adjusted;

[0020] When the sample size of the mixed training data set reaches the preset model update threshold, the sample size of the mixed training data set continues to be expanded until it reaches the preset model update threshold;

[0021] Reduce the learning rate according to the trend of the training loss value;

[0022] When the loss value of multiple consecutive training batches decreases below the preset decrease threshold, the incremental learning process is stopped and the updated deep learning model parameters are saved; otherwise, reinforcement learning continues.

[0023] Furthermore, the cross-modal feature alignment of the infrared thermal radiation data, point cloud spatial coordinates, and visible light image semantic information of the object under test includes:

[0024] Establish a three-dimensional space coordinate system and convert the point cloud space coordinates into a coordinate matrix under a unified coordinate system;

[0025] The infrared thermal radiation data is gridded and the spatial position index corresponding to the coordinate matrix is ​​generated.

[0026] Furthermore, when there are occluded area annotations in the semantic information of the visible light image, the infrared thermal radiation data of the occluded area is compensated, and the compensated infrared thermal radiation data features, coordinate matrix features and visible light image semantic features are input into the multi-layer Transformer encoder, and cross-modal feature context association modeling is performed;

[0027] When there is no occlusion area annotation in the semantic information of the visible light image, the cross-modal feature alignment operation is performed directly.

[0028] Furthermore, the step of inputting meteorological parameters into the environmental interference model to correct infrared thermal radiation data includes:

[0029] When the rate of change of the ambient temperature in the meteorological parameters is less than or equal to the unit time threshold, the environmental interference model is trained online;

[0030] When the fluctuation range of the correction coefficient output by the trained model exceeds the preset range, the output results of the environmental interference model of multiple different training cycles are weighted averaged to obtain the final infrared thermal radiation correction parameter. The environmental noise component in the infrared thermal radiation data is separated according to the correction parameter to obtain the corrected infrared thermal radiation data;

[0031] When the rate of change of the ambient temperature in the meteorological parameters is greater than the unit time threshold, an infrared thermal radiation data correction operation is performed on the historical meteorological parameters in a preset time period before and after the current moment.

[0032] Furthermore, the determination of whether the temperature measurement data is qualified includes:

[0033] Establish a temperature measurement data quality assessment indicator system;

[0034] When any evaluation indicator exceeds the preset normal range, the temperature measurement data is judged to be abnormal, otherwise the data is judged to be qualified.

[0035] Furthermore, the evaluation indicators include a temperature gradient continuity index, a spatial consistency index of different modal data, and a material emissivity parameter rationality index.

[0036] Furthermore, when the temperature measurement data is qualified, steps S1-S4 are repeated until the temperature measurement of the measured area is completed, including:

[0037] Divide the measured area into multiple sub-areas and adjust the temperature measurement order according to the three-dimensional structure complexity and material distribution uniformity of each sub-area;

[0038] After completing the temperature measurement of one sub-area, overlapping data collection is performed on the boundary areas of adjacent sub-areas;

[0039] Perform boundary fusion processing on the temperature distribution maps of adjacent sub-regions;

[0040] When all sub-areas have completed temperature measurement and there are no abnormal data points in the temperature distribution map after boundary fusion, the temperature measurement cycle is terminated.

[0041] Compared with the prior art, the beneficial effect of the present invention lies in that a non-contact multimodal thermal imager temperature measurement method of the present invention integrates multimodal data, namely three-dimensional structural information, visible light image texture features, infrared thermal radiation data and meteorological parameters, uses a deep learning model to dynamically adjust infrared emissivity parameters, combines the spatiotemporal attention mechanism for cross-modal feature alignment, and uses the environmental interference model to compensate for the influence of environmental factors. It can effectively overcome the problems of traditional non-contact temperature measurement being subject to large environmental interference, difficult emissivity calibration, and insufficient adaptability to complex scenes, significantly improves the temperature measurement accuracy, and can be widely used in industrial equipment fault diagnosis, medical temperature screening, security fire warning and other fields, to improve the detection accuracy and safety of related scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0043] Figure 1 This is a flow chart of a non-contact multi-modal thermal imager temperature measurement method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the implementation regulations.

[0045] Reference Figure 1 As shown, in some embodiments of the present application, a non-contact multi-modal thermal imager temperature measurement method includes:

[0046] S1, collects the three-dimensional structural information and infrared thermal radiation data of the object under test, collects the meteorological parameters of the on-site environment, and collects the texture features of the object under test in the visible light image;

[0047] S2, identifying the material and surface state of the object under test based on the three-dimensional structure information and texture features of the object under test;

[0048] S3, converting the three-dimensional structure information into point cloud form, and performing cross-modal feature alignment on the infrared thermal radiation data, point cloud spatial coordinates, and visible light image semantic information of the measured object;

[0049] S4, inputting meteorological parameters into the environmental interference model to correct the infrared thermal radiation data, and judging whether the temperature measurement data is qualified. If the temperature measurement data is qualified, execute step S5; if the temperature measurement data is abnormal, execute steps S2-S4;

[0050] S5: When the temperature measurement data is qualified, repeat steps S1-S4 until the temperature measurement of the measured area is completed.

[0051] Specifically, LiDAR is used to collect the three-dimensional structural information of the object being measured.

[0052] Specifically, after completing the temperature measurement of the measured area, a temperature distribution map is generated and output according to the cross-modal feature alignment results. The process includes: establishing a temperature-color mapping table to map different temperature ranges to a preset color space; when there are pixels with missing temperature values ​​in the cross-modal feature alignment results, Kriging interpolation is performed based on the temperature values ​​and spatial coordinates of the neighborhood around the pixel to estimate the missing temperature values; edge detection is performed on the generated temperature distribution map to identify the contour boundaries of the measured object; the contour boundary information is superimposed on the temperature distribution data to generate a temperature distribution map with a structural identification; the map data is packaged according to a preset data format and output to an external device through a data interface, wherein the preset color space is a color coding system determined by referring to industrial temperature measurement standards and human visual characteristics; the preset data format is a file encoding rule determined by being compatible with the input requirements of mainstream data processing platforms.

[0053] It is understandable that building a complete set of multimodal data acquisition, processing and cyclic temperature measurement processes provides a systematic framework for achieving high-precision non-contact temperature measurement.

[0054] In some embodiments of the present application, the method of identifying the material and surface state of the object under test based on the three-dimensional structural information and surface texture features of the object under test includes:

[0055] The texture features in the visible light image are subjected to regional gradient analysis. The three-dimensional structural features and regional texture features are matched with the three-dimensional structural features and regional texture features of each sample in the preset material feature library to obtain the material type and surface roughness. When the highest match similarity is higher than or equal to the preset match similarity threshold, the preset emissivity parameter mapping table is queried based on the material type and surface roughness of the sample with the highest similarity match to obtain the corresponding infrared emissivity parameters. When the highest match similarity is lower than the preset match similarity threshold, the parameters of the deep learning model are updated. Based on the material type and surface roughness output by the updated deep learning model, the preset emissivity parameter mapping table is queried to obtain the corresponding infrared emissivity parameters. The process of performing regional gradient analysis on texture features of visible light images includes: dividing the visible light image into multiple regular grid areas, calculating the gradient amplitude in the x-direction and the y-direction for each grid area; establishing a gradient amplitude histogram, and counting the distribution ratio of pixels in different gradient intervals; when the proportion of pixels in the high gradient interval in the gradient amplitude histogram of a certain grid area exceeds a preset pixel ratio threshold, determining that the area is a complex texture area, performing local amplification processing on the area, and extracting sub-pixel texture features; cascading the gradient features of each grid area and the sub-pixel texture features to generate a global texture feature vector, wherein the preset pixel ratio threshold is a critical value of the high gradient pixel ratio determined by analyzing the texture gradient distribution characteristics of different material surfaces. The parameter updating of the deep learning model includes: constructing a mixed training data set including historical material samples and currently added samples, and performing data enhancement processing on the newly added samples; when the sample size of the mixed training data set reaches a preset model update threshold, freezing the bottom feature extraction layer of the deep learning model, and only adjusting the parameters of the top classification and regression layer; when the sample size of the mixed training data set reaches a preset model update threshold, continuing to expand the number of samples of the mixed training data set until the preset model update threshold is reached; reducing the learning rate according to the changing trend of the training loss value; when the loss value of multiple consecutive training batches decreases by less than the preset decrease threshold, stopping the incremental learning process and saving the updated deep learning model parameters; otherwise, continuing the enhancement learning.

[0056] Specifically, the preset matching similarity threshold is a specific similarity value determined based on different application scenarios and requirements; the preset material feature library is a feature set constructed by collecting three-dimensional structural data and texture image data of multiple known materials and performing feature annotation; the preset model update threshold is a sample quantity critical value determined by analyzing the overfitting risk and training efficiency of the deep learning model; the preset decline amplitude threshold is a loss decline amplitude critical value determined by statistically analyzing the loss convergence speed during the deep learning model training process.

[0057] Specifically, when there are surface features or irregular contours in the LiDAR three-dimensional structural information, the texture features of the visible light image are subjected to regional gradient analysis to extract the gray-level co-occurrence matrix features and local binary pattern features of different regions; an incremental learning mechanism is used to update the parameters of the deep learning model.

[0058] It is understandable that by analyzing structural texture features to identify materials and adjust emissivity, the temperature measurement error caused by different materials can be effectively reduced, the temperature measurement accuracy can be improved, and the environmental interference model training can be optimized so that it can better capture the impact of environmental parameter changes on infrared signals and ensure temperature measurement accuracy.

[0059] In some embodiments of the present application, the cross-modal feature alignment of infrared thermal radiation data, point cloud spatial coordinates, and visible light image semantic information of the object under test includes:

[0060] Establish a three-dimensional space coordinate system and convert the point cloud space coordinates into a coordinate matrix under a unified coordinate system;

[0061] The infrared thermal radiation data is gridded and the spatial position index corresponding to the coordinate matrix is ​​generated.

[0062] When there are occluded area annotations in the semantic information of the visible light image, the infrared thermal radiation data of the occluded area is compensated, and the compensated infrared thermal radiation data features, coordinate matrix features and visible light image semantic features are input into the multi-layer Transformer encoder, and cross-modal feature context association modeling is performed;

[0063] When there is no occlusion area annotation in the semantic information of the visible light image, the cross-modal feature alignment operation is performed directly.

[0064] Specifically, the contextual correlation modeling of cross-modal features is achieved through self-attention mechanism and cross-attention mechanism.

[0065] It is understandable that the use of cross-modal feature alignment technology can integrate multi-source data information, enhance the correlation between data, and thus improve the accuracy of the temperature distribution map.

[0066] In some embodiments of the present application, inputting meteorological parameters into the environmental interference model to correct infrared thermal radiation data includes:

[0067] When the rate of change of the ambient temperature in the meteorological parameters is less than or equal to the unit time threshold, the environmental interference model is trained online;

[0068] When the fluctuation range of the correction coefficient output by the trained model exceeds the preset range, the output results of the environmental interference model of multiple different training cycles are weighted averaged to obtain the final infrared thermal radiation correction parameter. The environmental noise component in the infrared thermal radiation data is separated according to the correction parameter to obtain the corrected infrared thermal radiation data;

[0069] When the rate of change of the ambient temperature in the meteorological parameters is greater than the unit time threshold, an infrared thermal radiation data correction operation is performed on the historical meteorological parameters in a preset time period before and after the current moment.

[0070] Specifically, the unit time threshold is a critical value of the temperature change rate determined by analyzing the stable period of the infrared signal under different environmental conditions; the preset interval is a reasonable fluctuation range determined by the distribution range of the correction coefficient output by the statistical historical environmental interference model.

[0071] Specifically, the process of online training of the environmental interference model includes:

[0072] Establishing a feature input vector of the environmental interference model, wherein the feature input vector includes real-time meteorological parameters, a meteorological parameter change sequence in a historical period, and a corresponding infrared signal correction residual;

[0073] When the time span of online training data exceeds the preset period, the historical training data is filtered and the valid data in the most recent period is retained;

[0074] Input the feature input vector into the LSTM network in time series;

[0075] Capturing the long-term dependence of meteorological parameter changes on infrared signals;

[0076] During the model training process, the root mean square error of the corrected residual is used as the loss function, and the weight parameters of the model are updated, where

[0077] The preset period is a time length determined by analyzing the frequency of changes in environmental parameters and the efficiency of model updating.

[0078] It can be understood that correcting the infrared signal through the environmental interference model can effectively eliminate the interference of environmental factors on temperature measurement and make the measurement results more reliable.

[0079] In some embodiments of the present application, determining whether the temperature measurement data is qualified includes:

[0080] Establish a temperature measurement data quality assessment indicator system;

[0081] When any evaluation indicator exceeds the preset normal range, the temperature measurement data is judged to be abnormal, otherwise the data is judged to be qualified.

[0082] The evaluation indicators include temperature gradient continuity index, spatial consistency index of different modal data and rationality index of material emissivity parameters.

[0083] Specifically, the temperature gradient continuity index is a continuity evaluation parameter determined by calculating the degree of deviation between the temperature difference of adjacent pixels and a preset temperature change threshold, and is used to judge the rationality of the surface temperature distribution of the measured object;

[0084] The preset temperature change threshold is a temperature difference critical value determined by measuring the temperature distribution of the surface of various standard materials and calculating the temperature change range of adjacent pixels;

[0085] The spatial consistency index of different modal data is an inter-modal data alignment parameter determined by comparing the spatial coordinates of infrared thermal radiation data with the spatial coordinates of LiDAR point cloud and the position matching degree of visible light image pixel coordinates, and is used to detect the consistency of multi-modal data in spatial positioning;

[0086] The material emissivity parameter rationality index is a parameter validity index determined by querying a preset material-emissivity correspondence library to verify whether the emissivity parameter of the currently identified material is within the theoretical emissivity range of the material, and is used to ensure the accuracy of the emissivity adjustment;

[0087] The preset normal range is the index fluctuation range determined by statistical analysis of multi-modal temperature measurement data of standard material samples;

[0088] The preset difference value is a critical value of structural difference that can be determined by analyzing the measurement error of the LiDAR device and the surface deformation of the object.

[0089] It is understandable that establishing an evaluation system to judge data quality and recollecting and processing data in case of anomalies ensures the reliability and accuracy of temperature measurement data.

[0090] In some embodiments of the present application, when the temperature measurement data is qualified, repeating steps S1-S4 until the temperature measurement of the measured area is completed includes:

[0091] Divide the measured area into multiple sub-areas and adjust the temperature measurement order according to the three-dimensional structure complexity and material distribution uniformity of each sub-area;

[0092] After completing the temperature measurement of one sub-area, overlapping data collection is performed on the boundary areas of adjacent sub-areas;

[0093] Perform boundary fusion processing on the temperature distribution maps of adjacent sub-regions;

[0094] When all sub-areas have completed temperature measurement and there are no abnormal data points in the temperature distribution map after boundary fusion, the temperature measurement cycle is terminated.

[0095] Specifically, the preset period is a time length determined by analyzing the frequency of changes in environmental parameters and the efficiency of model updating.

[0096] It can be understood that optimizing the division, acquisition and fusion of the measured area can reduce the temperature measurement error between sub-areas and make the overall temperature measurement results more accurate.

[0097] It should be noted that:

[0098] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this description.

[0099] Furthermore, those skilled in the art will appreciate that although some embodiments described herein include certain features and not other features included in other embodiments, the combination of features from different embodiments is meant to be within the scope of this application and to form different embodiments.

[0100] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A non-contact multi-modal thermal imager temperature measurement method, characterized in that: include: S1, collects the three-dimensional structural information and infrared thermal radiation data of the object under test, collects the meteorological parameters of the on-site environment, and collects the texture features of the object under test in the visible light image; S2, identifying the material and surface state of the object under test based on the three-dimensional structure information and texture features of the object under test; S3, converting the three-dimensional structure information into point cloud form, and performing cross-modal feature alignment on the infrared thermal radiation data, point cloud spatial coordinates, and visible light image semantic information of the measured object; S4, inputting meteorological parameters into the environmental interference model to correct the infrared thermal radiation data, and judging whether the temperature measurement data is qualified. If the temperature measurement data is qualified, execute step S5; if the temperature measurement data is abnormal, execute steps S2-S4; S5: When the temperature measurement data is qualified, repeat steps S1-S4 until the temperature measurement of the measured area is completed.

2. The non-contact multi-modal thermal imager temperature measurement method according to claim 1, characterized in that: The method of identifying the material and surface state of the object to be measured based on the three-dimensional structural information and surface texture characteristics of the object to be measured includes: Perform regional gradient analysis on texture features in visible light images; The three-dimensional structural features and regional texture features are matched with the three-dimensional structural features and regional texture features of each sample in the preset material feature library, and the material type and surface roughness are obtained.

3. The non-contact multi-modal thermal imager temperature measurement method according to claim 2, characterized in that: When the highest matching similarity is higher than or equal to the preset matching similarity threshold, the preset emissivity parameter mapping table is queried according to the material type and surface roughness of the sample corresponding to the highest similarity matching degree to obtain the corresponding infrared emissivity parameter; When the highest matching similarity is lower than the preset matching similarity threshold, the parameters of the deep learning model are updated. According to the material type and surface roughness output by the updated deep learning model, the preset emissivity parameter mapping table is queried to obtain the corresponding infrared emissivity parameters.

4. The non-contact multi-modal thermal imager temperature measurement method according to claim 3, characterized in that: The parameter updating of the deep learning model includes: Construct a mixed training dataset containing historical material samples and currently added samples, and perform data enhancement processing on the new samples; When the sample size of the mixed training dataset reaches the preset model update threshold, the underlying feature extraction layer of the deep learning model is frozen, and only the parameters of the top classification and regression layer are adjusted; When the sample size of the mixed training data set does not reach the preset model update threshold, the sample size of the mixed training data set continues to be expanded until it reaches the preset model update threshold; Reduce the learning rate according to the trend of the training loss value; When the loss value of multiple consecutive training batches decreases below the preset decrease threshold, the incremental learning process is stopped and the updated deep learning model parameters are saved; otherwise, reinforcement learning continues.

5. The non-contact multi-modal thermal imager temperature measurement method according to claim 4, characterized in that: The cross-modal feature alignment of the infrared thermal radiation data, point cloud spatial coordinates, and visible light image semantic information of the measured object includes: Establish a three-dimensional space coordinate system and convert the point cloud space coordinates into a coordinate matrix under a unified coordinate system; The infrared thermal radiation data is gridded and the spatial position index corresponding to the coordinate matrix is ​​generated.

6. The non-contact multi-modal thermal imager temperature measurement method according to claim 5, characterized in that: When there are occluded area annotations in the semantic information of the visible light image, the infrared thermal radiation data of the occluded area is compensated, and the compensated infrared thermal radiation data features, coordinate matrix features and visible light image semantic features are input into the multi-layer Transformer encoder, and cross-modal feature context association modeling is performed; When there is no occlusion area annotation in the semantic information of the visible light image, the cross-modal feature alignment operation is performed directly.

7. The non-contact multi-modal thermal imager temperature measurement method according to claim 6, characterized in that: The step of inputting meteorological parameters into the environmental interference model to correct infrared thermal radiation data includes: When the rate of change of the ambient temperature in the meteorological parameters is less than or equal to the unit time threshold, the environmental interference model is trained online; When the fluctuation range of the correction coefficient output by the trained model exceeds the preset range, the output results of the environmental interference model of multiple different training cycles are weighted averaged to obtain the final infrared thermal radiation correction parameter. The environmental noise component in the infrared thermal radiation data is separated according to the correction parameter to obtain the corrected infrared thermal radiation data; When the rate of change of the ambient temperature in the meteorological parameters is greater than the unit time threshold, an infrared thermal radiation data correction operation is performed on the historical meteorological parameters in a preset time period before and after the current moment.

8. The non-contact multi-modal thermal imager temperature measurement method according to claim 7, characterized in that: The determination of whether the temperature measurement data is qualified includes: Establish a temperature measurement data quality assessment indicator system; When any evaluation indicator exceeds the preset normal range, the temperature measurement data is judged to be abnormal, otherwise the data is judged to be qualified.

9. The non-contact multi-modal thermal imager temperature measurement method according to claim 8, characterized in that: The evaluation indicators include temperature gradient continuity index, spatial consistency index of different modal data and rationality index of material emissivity parameters.

10. The non-contact multi-modal thermal imager temperature measurement method according to claim 9, characterized in that: When the temperature measurement data is qualified, steps S1-S4 are repeated until the temperature measurement of the measured area is completed, including: Divide the measured area into multiple sub-areas and adjust the temperature measurement order according to the three-dimensional structure complexity and material distribution uniformity of each sub-area; After completing the temperature measurement of one sub-area, overlapping data collection is performed on the boundary areas of adjacent sub-areas; Perform boundary fusion processing on the temperature distribution maps of adjacent sub-regions; When all sub-areas have completed temperature measurement and there are no abnormal data points in the temperature distribution map after boundary fusion, the temperature measurement cycle is terminated.

Images