Method and system for identifying heat radiation abnormity based on fusion of multi-source heterogeneous data features

Abstract

The invention belongs to the field of heat radiation anomaly identification, and provides a heat radiation anomaly identification method and system based on fusion of multi-source heterogeneous data features, and the method comprises the steps: obtaining multi-source heterogeneous data; extracting temperature distribution characteristics from the infrared image, and extracting environment temperature, humidity and wind speed characteristics from the meteorological data; recursive feature elimination is carried out on the extracted features; calculating an intra-class scatter matrix and an inter-class scatter matrix of the first feature subset; determining an optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and projecting the first feature subset to the optimal projection matrix to obtain a dimension reduction feature subset; performing feature interaction operation on the dimension reduction feature subset to generate an interaction feature subset; screening the interactive feature subsets by adopting a packaging type feature selection algorithm to obtain a final comprehensive feature set; and based on the final comprehensive feature set, a thermal radiation anomaly detection model is constructed, and the thermal radiation anomaly detection effect and the model robustness are improved.

Description

Technical Field

[0001] The present invention belongs to the field of thermal radiation anomaly recognition, and specifically relates to a thermal radiation anomaly recognition method and system based on the fusion of multi-source heterogeneous data features. Background Art

[0002] Thermal radiation anomaly detection has important applications in many fields, including fire warning, industrial equipment monitoring, environmental anomaly identification, etc. Traditional thermal radiation anomaly detection mainly relies on the analysis of a single data source (such as infrared images or meteorological data). The detection accuracy and adaptability of this method are often limited by the coverage and accuracy of the data source. However, in practical applications, a single data source is easily affected by factors such as environmental changes, equipment interference, and data noise, resulting in high false alarm and missed alarm rates.

[0003] In recent years, with the rapid development of sensing technology and remote sensing technology, the application of multi-source heterogeneous data (such as satellite remote sensing, drone infrared imaging, ground sensor networks and data collected by meteorological stations) in thermal radiation anomaly detection has gradually attracted attention. By fusing the characteristics of multi-source heterogeneous data, the temperature distribution, meteorological conditions and environmental changes in the target area can be more comprehensively reflected. However, the fusion and analysis of multi-source data faces many technical challenges, which are mainly reflected in the following aspects: Multi-source data comes from different acquisition devices and sensors with different formats, resolutions and sampling frequencies. In addition, the inconsistency of infrared images and meteorological data in spatial and temporal scales complicates data alignment and feature fusion. Existing multi-source data fusion methods often suffer from information loss in the extraction and expression of heterogeneous features, and cannot fully reveal the potential correlation between multi-source features, resulting in poor anomaly detection results.

[0004] In the process of multi-source data fusion, the existence of redundant features and noise features will reduce the detection accuracy of the model, while increasing the computational burden and model complexity. Current methods usually use simple feature selection or dimensionality reduction techniques, which makes it difficult to ensure that the extracted features can maintain high discriminability and adaptability in a multi-source environment, resulting in insufficient accuracy and robustness of the detection model.

[0005] There are complex nonlinear relationships between infrared image features and meteorological features in multi-source data, and traditional feature combination methods are difficult to capture these potential nonlinear associations. Existing anomaly detection models mostly use linear classification or simple interaction strategies, which are difficult to fully utilize the nonlinear information between features, thus limiting the sensitivity of the detection model to thermal radiation anomalies.

[0006] In the environment of multi-source heterogeneous data, the changing characteristics and collection errors of different data sources may lead to insufficient generalization ability of the detection model in practical applications, especially in different scenarios, climate conditions or environmental factors. Traditional models are prone to overfitting when facing multi-source heterogeneous feature combinations, and lack long-term adaptability and stability to real scenarios. Summary of the invention

[0007] In order to solve the problems in the prior art, the present invention provides a method for identifying thermal radiation anomalies based on fusing multi-source heterogeneous data features, the method comprising the following steps: Acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes infrared image data and meteorological data; Extracting temperature distribution features from the infrared image, and extracting ambient temperature, humidity and wind speed features from the meteorological data; The extracted features are subjected to recursive feature elimination using the K nearest neighbor classifier as an evaluation model for feature selection to obtain the first feature subset; Calculating an intra-class scatter matrix and an inter-class scatter matrix of the first feature subset; determining an optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and projecting the first feature subset onto the optimal projection matrix to obtain a reduced-dimensional feature subset; Performing feature interaction operations on the dimension reduction feature subset to generate an interactive feature subset; Using a packaged feature selection algorithm to screen the interactive feature subset to obtain a final comprehensive feature set; Based on the final comprehensive feature set, a thermal radiation anomaly detection model is constructed, and the thermal radiation anomaly detection model is used to identify thermal radiation anomalies.

[0008] On the other hand, the present invention also provides a thermal radiation anomaly recognition system based on the fusion of multi-source heterogeneous data features, the system comprising the following modules: An acquisition module, used for acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes infrared image data and meteorological data; A first feature extraction module is used to extract temperature distribution features from the infrared image and to extract ambient temperature, humidity and wind speed features from the meteorological data; The second feature extraction module is used to extract the features The nearest neighbor classifier is used as the evaluation model for feature selection, and recursive feature elimination is performed to obtain the first feature subset; A third feature extraction module is used to calculate the intra-class scatter matrix and the inter-class scatter matrix of the first feature subset; determine the optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and project the first feature subset onto the optimal projection matrix to obtain a reduced-dimensional feature subset; A fourth feature extraction module, used for performing feature interaction operation on the dimension reduction feature subset to generate an interactive feature subset; A fifth feature extraction module, configured to screen the interactive feature subset using a packaged feature selection algorithm to obtain a final comprehensive feature set; The detection module is used to construct a thermal radiation anomaly detection model based on the final comprehensive feature set, and use the thermal radiation anomaly detection model to identify thermal radiation anomalies.

[0009] The recognition method of thermal radiation anomaly based on the fusion of multi-source heterogeneous data features has the following beneficial effects: Through data preprocessing and alignment, we ensure that multi-source heterogeneous data are consistent at the same time and spatial scale, and design multi-level feature extraction and fusion strategies to improve the comprehensive utilization efficiency of multi-source data.

[0010] Through methods such as packaged feature selection, feature dimensionality reduction and feature interaction, we can fully explore the potential correlation of multi-source heterogeneous data, generate a comprehensive feature set, and improve the accuracy of anomaly detection.

[0011] The fusion features of multi-source data are reduced in dimension and screened to ensure that the most discriminative features are retained while removing redundant and noisy features. Through further nonlinear feature interaction, the expressiveness between features is improved, and the computational efficiency and adaptability of the model are optimized.

[0012] Through feature interaction operations, nonlinear combination features are constructed, thereby improving the model's ability to express complex relationships between multi-source features.

[0013] Based on the final comprehensive feature set, machine learning or deep learning models are used to achieve efficient identification of thermal radiation anomalies and improve the sensitivity and detection accuracy of the model.

[0014] This solution improves the detection effect of thermal radiation anomalies and the robustness of the model by comprehensively optimizing the fusion, feature extraction, dimensionality reduction and model construction of multi-source heterogeneous data, providing an efficient and accurate solution for applications in fire warning, industrial equipment overheating detection, environmental monitoring and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0017] Below, the invention is preferably described in conjunction with the accompanying drawings and specific implementation methods.

[0018] This embodiment solves the above problem through the following steps: In one embodiment, reference Figure 1 The present invention provides a method for identifying thermal radiation anomalies based on the fusion of multi-source heterogeneous data features, aiming to achieve efficient detection and accurate identification of potential thermal radiation anomalies in the target area by extracting, analyzing and fusing multidimensional data from different sources and in different forms.

[0019] Multi-source heterogeneous data are acquired, where the multi-source heterogeneous data include infrared image data and meteorological data.

[0020] The infrared image data can be collected by remote sensing satellites, drones or ground thermal imaging equipment. The infrared image data contains the infrared radiation intensity distribution and thermal radiation characteristic information of each pixel point in the target area. The infrared image data can have a variety of spatial resolutions and temporal resolutions, and can reflect the temperature field distribution, temperature change trend and local thermal anomaly of the target area. The meteorological data may be collected by ground meteorological stations, meteorological satellites or mobile meteorological equipment. The meteorological data include meteorological elements such as ambient temperature, relative humidity, wind speed, wind direction, air pressure, precipitation, etc. The meteorological data is presented in the form of time series and can reflect the macro-climate conditions and micro-meteorological change trends of the target area; The process of acquiring multi-source heterogeneous data also includes time alignment and space alignment of data, ensuring that data from different sources are consistent in the same time window and the same geographical location, so as to facilitate subsequent feature extraction and data fusion.

[0021] The temperature distribution feature is extracted from the infrared image, and the ambient temperature, humidity and wind speed features are extracted from the meteorological data.

[0022] A temperature distribution feature is extracted from the infrared image, wherein the temperature distribution feature includes the following information: The infrared radiation intensity value of each pixel point in the target area is mapped to the corresponding temperature value through the pixel value to form the temperature field distribution of the target area; By analyzing the temperature field distribution of the infrared image, the temperature gradient features are further extracted, and the temperature gradient features include the local temperature change rate and the global temperature change trend, so as to identify the high temperature concentration area and its boundary in the target area; For the multi-frame sequence data of infrared images, the temperature time variation characteristics are extracted, and the temperature variation curve is generated by comparing the temperature changes in different time frames to reflect the dynamic changes of thermal radiation intensity in the target area.

[0023] Extracting environmental temperature, humidity and wind speed features from the meteorological data specifically includes: Extracting ambient temperature features from the temperature field in the meteorological data, the ambient temperature features are the air temperature information in the target area, and the fluctuation range and change trend of the ambient temperature can be calculated through time series analysis; Extracting relative humidity features from the humidity field in the meteorological data, the relative humidity features reflect the proportion of water vapor content in the air in the target area, and can be combined with the ambient temperature features to further analyze the impact of humid and hot conditions on thermal radiation changes; Wind speed features are extracted from the wind speed field in the meteorological data. The wind speed features include the magnitude of wind speed and the change of wind direction in the target area. By analyzing the wind speed features, the influence of wind on the diffusion and transmission of thermal radiation in the target area can be evaluated.

[0024] The above-mentioned temperature distribution characteristics, ambient temperature characteristics, humidity characteristics and wind speed characteristics constitute the basic feature set of multi-source heterogeneous data, providing a basis for subsequent feature fusion and thermal radiation anomaly identification.

[0025] Use the extracted features The nearest neighbor classifier is used as the evaluation model for feature selection, and recursive feature elimination is performed to obtain the first feature subset.

[0026] Set the extracted feature set to , in, is the temperature distribution characteristic of each pixel in the infrared image; is the temperature gradient feature in the infrared image; It is the temperature time variation feature in the infrared image; is the ambient temperature feature in the meteorological data; is the relative humidity feature in meteorological data; is the wind speed feature in meteorological data.

[0027] choose The nearest neighbor classifier is used as the evaluation model, setting the neighbor parameters , used to calculate the feature set The contribution of each feature to the classification of thermal radiation anomalies. The construction of the nearest neighbor classifier is based on the initial feature set , the performance of the classifier under different feature combinations is evaluated by cross-validation.

[0028] In the feature set Based on Nearest neighbor classifier and determine the relative importance of each feature by evaluating the accuracy of the model.

[0029] Specifically, for each feature , its importance score It is calculated by the change in model accuracy, defined as: , in Based on the entire feature set The accuracy of the trained model; To remove features The accuracy of the trained model.

[0030] For example, for the temperature distribution characteristics of infrared images ,calculate to evaluate its contribution to anomaly detection; similarly, the ambient temperature in meteorological features , relative humidity and wind speed Calculate their importance scores in turn.

[0031] In each iteration, the feature importance score , select the feature with the least importance (For example, the initial iteration might be the relative humidity feature ) and remove it from the current feature set Remove from the list to form a new feature set: .

[0032] Retraining Nearest neighbor classifier, calculate updated feature importance scores , and again select the least important feature to remove.

[0033] Continue to perform the above recursive feature elimination steps until the feature set is reduced to the preset feature subset size (For example, ), or the model accuracy no longer improves significantly.

[0034] After recursive feature elimination is completed, the retained feature set is the first feature subset , illustratively, its form is: , Among them, this feature subset retains the features that contribute most to the identification of thermal radiation anomalies, including temperature distribution and gradient features in infrared images, and wind speed features in meteorological data.

[0035] The first feature subset of the final output is through The set of key features screened out by the nearest neighbor classifier and recursive feature elimination algorithm provides the basis for subsequent feature dimensionality reduction, interaction and anomaly detection.

[0036] The above process combines the specific features of multi-source heterogeneous data (such as temperature distribution, temperature gradient, ambient temperature, humidity, wind speed, etc.) to achieve The feature selection and recursive feature elimination of the nearest neighbor classifier effectively extract the features that contribute most to the thermal radiation anomaly.

[0037] Calculate the intra-class scatter matrix and inter-class scatter matrix of the first feature subset; determine the optimal projection matrix based on the intra-class scatter matrix and inter-class scatter matrix, and project the first feature subset onto the optimal projection matrix to obtain a reduced-dimensional feature subset. This step combines the thermal radiation description of multi-source heterogeneous data features to improve the accuracy of anomaly recognition and feature separation capabilities.

[0038] Set the first feature subset obtained by recursive feature elimination (RFE) to ,in, Features extracted from multi-source heterogeneous data include but are not limited to temperature distribution features and temperature gradient features in infrared images, and wind speed features in meteorological data.

[0039] For each feature in the first feature subset, calculate the intra-class scatter matrix : , in, is the total number of categories; For the The set of samples in the class; For the The mean vector of class samples; is a feature vector, which represents the feature value of a specific sample.

[0040] Within-class scatter matrix It reflects the distribution of each feature in the first feature subset within the same class. For thermal radiation anomaly detection, this helps measure how the temperature distribution and environmental features are concentrated or dispersed in abnormal and normal samples, thereby ensuring the relative closeness of features in the same class.

[0041] Compute the between-class scatter matrix : , in, For the The number of samples in the class; For the The mean vector of class samples; is the global mean vector of all samples.

[0042] Maximize the inter-class scatter and minimize the intra-class scatter in the optimal projection space to find the optimal projection matrix : , in, represents the determinant of a matrix; is a linear projection matrix that can maximize the inter-class scatter relative to the intra-class scatter.

[0043] By solving the generalized eigenvalue problem, the optimal projection matrix can be obtained , whose column vector corresponds to Before The eigenvector corresponding to the largest eigenvalue is is the number of features after dimensionality reduction.

[0044] The first feature subset Project to the optimal projection matrix On the above, we get the dimension reduction feature subset : , in, It is a feature subset after dimensionality reduction, which retains the optimal discrimination ability for different categories of thermal radiation features.

[0045] By calculating the intra-class scatter matrix and the inter-class scatter matrix and using the optimal projection matrix for dimensionality reduction, the feature separation between different categories can be maximized, thereby improving the ability to distinguish between abnormal thermal radiation and normal thermal radiation. The infrared temperature distribution characteristics and meteorological environment characteristics in multi-source data are optimized during the dimensionality reduction process, making these features easier to be accurately identified by the classification model in the projected space. Dimensionality reduction can better adapt to the diverse features from infrared images and meteorological data through discriminative projection in the feature space, so that the features of different data sources are uniformly expressed in the space after dimensionality reduction, improving the model's adaptability to thermal radiation anomalies in complex environments.

[0046] Performing feature interaction operations on the dimension reduction feature subset to generate an interactive feature subset.

[0047] The reduced dimension feature subset obtained by dimensionality reduction is ,in, They respectively represent the features retained after dimensionality reduction from multi-source features such as temperature distribution features and temperature gradient features in infrared images and ambient temperature and wind speed in meteorological data.

[0048] Feature interaction refers to generating new interactive features through the combination, transformation and expansion of different dimensionality reduction features to enhance the nonlinear correlation expression between features. In the present invention, feature interaction includes feature product, feature square difference and weighted combination.

[0049] Characteristic product operation: For each pair of features in the dimension reduction feature subset and , perform feature product operation to generate new interactive features : , For example, for the feature representing the temperature distribution in the dimension reduction feature subset and the characteristic representing wind speed , its product characteristic It can be used to reflect the mutual influence between temperature change and wind speed, so as to better capture possible abnormal conditions in the process of thermal radiation diffusion.

[0050] Feature square difference operation: For each pair of features in the dimension reduction feature subset and , calculate the feature square difference and generate new interaction features : , For example, for the Temperature Gradient feature and ambient temperature characteristics , its square difference characteristic It can reflect the difference between the temperature gradient and the ambient temperature, and helps to detect local temperature anomalies and abnormal temperature gradient changes.

[0051] Feature weighted combination operation: For the features in the dimension reduction feature subset and Perform weighted combination to generate new interactive features : , in, and is the weighting coefficient, which is used to control the weight of different features in the combination.

[0052] For example, for the ambient temperature characteristic and humidity characteristics , through appropriate weighted combination, comprehensive features can be generated , reflecting the potential impact of comprehensive environmental conditions on thermal radiation.

[0053] Through the above feature interaction operation, all features in the dimension reduction feature subset are combined two by two to generate an interactive feature subset. : , The interactive feature subset includes new features generated by feature products, squared differences, and weighted combinations, which can capture potential relationships and abnormal patterns that are not apparent in the original features.

[0054] In this step, by performing feature interaction operations on the dimensionality reduction feature subsets to generate interactive feature subsets, the expressive power of multi-source heterogeneous data features can be effectively enhanced, the accuracy and robustness of thermal radiation anomaly detection can be improved, and the adaptability of the model to complex environments can be improved.

[0055] The interactive feature subset is screened using a packaged feature selection algorithm to obtain a final comprehensive feature set.

[0056] The interactive feature subset obtained through feature interaction operation is ,in Interaction features include features generated by feature product, feature square difference and weighted combination, such as the interaction term between infrared temperature distribution and meteorological environment characteristics.

[0057] Select a suitable machine learning model as the evaluation model for packaged feature selection, such as random forest, support vector machine or logistic regression. Preferably, a random forest model is used, which has strong feature selection ability and robustness.

[0058] Initialize the feature set to the interactive feature subset , and define the objective function , used to evaluate the performance of the model on the subset of interacting features. Objective function It can be accuracy, AUC, F1-score or other performance indicators.

[0059] In the interaction feature subset Train the random forest model and calculate each interaction feature Importance score : , in: Represents the model on all interaction feature subsets Performance on Removal of features The performance of the latter model.

[0060] Based on feature importance score The sorting of Remove the least important features , and obtain a new feature subset , Retrain the random forest model and update the feature importance scores of the remaining features Continue recursively performing feature elimination steps until the feature subset The size reaches the preset final number of features , or the performance index of the model No longer significantly improved.

[0061] After recursive feature elimination is completed, the retained feature set is the final comprehensive feature set, denoted as: ,in is the number of features in the final feature set.

[0062] The final comprehensive feature set contains the optimal features screened by packaged feature selection, which can show higher accuracy and robustness in thermal radiation anomaly detection of multi-source heterogeneous data features.

[0063] In this step, the packaged feature selection algorithm can effectively screen out the features that contribute most to the recognition of thermal radiation anomalies, remove redundant and noisy features, and thus improve the model's discriminative ability. In multi-source heterogeneous data, this selection mechanism helps to retain the key interaction terms between infrared images and meteorological features, thereby better capturing the characteristics of abnormal thermal radiation.

[0064] Based on the final comprehensive feature set, a thermal radiation anomaly detection model is constructed, and the thermal radiation anomaly detection model is used to identify thermal radiation anomalies.

[0065] According to the feature type and data distribution of the final comprehensive feature set, select a suitable machine learning or deep learning model for thermal radiation anomaly detection. The selectable models include but are not limited to: support vector machine, random forest, XGBoost model, deep neural network.

[0066] The final comprehensive feature set is used as input features and the selected model is trained using the training dataset.

[0067] During the training process, supervised learning is used to optimize the parameters of the classification model for abnormal thermal radiation samples and normal samples, and cross-validation and hyperparameter adjustment are used to improve the performance and stability of the model.

[0068] The model is evaluated and verified through model performance indicators such as accuracy, recall, F1-score, AUC, etc. to ensure that the model shows high anomaly detection accuracy on both the training data set and the verification data set.

[0069] According to the validation results of the model, the model structure and parameters are further optimized, such as adjusting the kernel function type of SVM, the number of decision trees of random forest, or the tree depth of XGBoost, so as to improve the model's adaptability to multi-source features and sensitivity to anomalies.

[0070] In practical applications, a comprehensive feature set is extracted from multi-source heterogeneous data monitored in real time and input into a trained thermal radiation anomaly detection model.

[0071] Through the forward reasoning of the model, the output results of the input comprehensive features in the model are calculated, and the model output results are anomaly scores or classification labels: Anomaly score: The model can output anomaly probability scores. The higher the value, the greater the possibility of thermal radiation anomaly. Classification label: The model can output a binary classification label, usually "1" for abnormality and "0" for normality.

[0072] Based on the output results of the model, determine whether the thermal radiation in the target area is abnormal: if the output anomaly score exceeds the preset threshold, or the output classification label is "1", it is considered that thermal radiation anomaly is detected; based on the anomaly detection results, trigger the corresponding alarm mechanism, including but not limited to sound alarm, video display alarm, SMS or email notification, etc.

[0073] Furthermore, in the real-time monitoring environment of multi-source data, comprehensive features are continuously extracted through the streaming data processing framework (such as Apache Kafka, Apache Flink, etc.) and input into the thermal radiation anomaly detection model to achieve real-time thermal radiation anomaly monitoring of the target area.

[0074] On the other hand, the present invention also provides a thermal radiation anomaly recognition system based on fusing multi-source heterogeneous data features, comprising: An acquisition module, used for acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes infrared image data and meteorological data; A first feature extraction module is used to extract temperature distribution features from the infrared image and to extract ambient temperature, humidity and wind speed features from the meteorological data; A second feature extraction module is used to use a K nearest neighbor classifier as an evaluation model for feature selection to perform recursive feature elimination on the extracted features to obtain a first feature subset; A third feature extraction module is used to calculate the intra-class scatter matrix and the inter-class scatter matrix of the first feature subset; determine the optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and project the first feature subset onto the optimal projection matrix to obtain a reduced-dimensional feature subset; A fourth feature extraction module, used for performing feature interaction operation on the dimension reduction feature subset to generate an interactive feature subset; A fifth feature extraction module, configured to screen the interactive feature subset using a packaged feature selection algorithm to obtain a final comprehensive feature set; The detection module is used to construct a thermal radiation anomaly detection model based on the final comprehensive feature set, and use the thermal radiation anomaly detection model to identify thermal radiation anomalies.

[0075] The prior art mentioned in the above background technology section and specific embodiment section of the present invention can be used as part of the present invention to understand the meaning of some technical features or parameters.

Claims

1. A method for identifying thermal radiation anomalies based on the fusion of multi-source heterogeneous data features, characterized in that: The method comprises the following steps: Acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes infrared image data and meteorological data; Extracting temperature distribution features from the infrared image, and extracting ambient temperature, humidity and wind speed features from the meteorological data; Use the extracted features The nearest neighbor classifier is used as the evaluation model for feature selection, and recursive feature elimination is performed to obtain the first feature subset; Calculating an intra-class scatter matrix and an inter-class scatter matrix of the first feature subset; determining an optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and projecting the first feature subset onto the optimal projection matrix to obtain a reduced-dimensional feature subset; Performing feature interaction operations on the dimension reduction feature subset to generate an interactive feature subset; Using a packaged feature selection algorithm to screen the interactive feature subset to obtain a final comprehensive feature set; Based on the final comprehensive feature set, a thermal radiation anomaly detection model is constructed, and the thermal radiation anomaly detection model is used to identify thermal radiation anomalies.

2. The method for identifying thermal radiation anomalies based on integrating multi-source heterogeneous data features according to claim 1 is characterized in that: The extracted features are treated as the evaluation model for feature selection using the K nearest neighbor classifier, and recursive feature elimination is performed to obtain the first feature subset including: Set the extracted feature set to , in, is the temperature distribution characteristic of each pixel in the infrared image; is the temperature gradient feature in the infrared image; is the temperature time variation feature in the infrared image; is the ambient temperature feature in the meteorological data; is the relative humidity feature in meteorological data; is the wind speed feature in the meteorological data; choose The nearest neighbor classifier is used as the evaluation model, setting the neighbor parameters , used to calculate the feature set The contribution of each feature to the classification of thermal radiation anomalies; For each feature , its importance score Calculated by the change in model accuracy: ， in: Based on the entire feature set The accuracy of the trained model; To remove features The accuracy of the post-training model; In each iteration, the feature importance score , select the feature with the least importance , and remove it from the current feature set Remove from the list to form a new feature set ; Retraining Nearest neighbor classifier, calculate updated feature importance scores , and again select the least important feature to remove; Continue to perform the above recursive feature elimination steps until the feature set is reduced to the preset feature subset size , or the model accuracy no longer improves; After recursive feature elimination is completed, the retained feature set is the first feature subset .

3. The method for identifying thermal radiation anomalies based on integrating multi-source heterogeneous data features according to claim 2 is characterized in that: Calculating the intra-class scatter matrix and the inter-class scatter matrix of the first feature subset; determining an optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and projecting the first feature subset onto the optimal projection matrix, to obtain a reduced-dimensional feature subset comprising: The first feature subset is ,in, Features extracted from multi-source heterogeneous data; For each feature in the first feature subset, calculate the intra-class scatter matrix : ， in, is the total number of categories; is the sample set in the 𝑖th class; For the The mean vector of class samples; is a feature vector, which represents the feature value of a specific sample; Compute the between-class scatter matrix : ， in, For the The number of samples in the class; For the The mean vector of class samples; is the global mean vector of all samples; Maximize the inter-class scatter and minimize the intra-class scatter in the optimal projection space to find the optimal projection matrix : ， in, represents the determinant of a matrix; is the linear projection matrix; The first feature subset Project to the optimal projection matrix On the above, we get the reduced dimension feature subset : 。 4. The method for identifying thermal radiation anomalies based on integrating multi-source heterogeneous data features according to claim 3 is characterized in that: Performing feature interaction operations on the dimension reduction feature subset to generate an interactive feature subset includes: feature product operation, feature square difference operation, and feature weighted combination operation.

5. The method for identifying thermal radiation anomalies based on integrating multi-source heterogeneous data features according to claim 4 is characterized in that: The packaged feature selection algorithm is used to screen the interactive feature subset to obtain the final comprehensive feature set including: The interaction feature subset is ,in is the interactive feature; Initialize the feature set to the interactive feature subset , and define the objective function ; In the interaction feature subset Train the random forest model and calculate each interaction feature Importance score : ， in: Represents the model on all interaction feature subsets Performance on Removal of features The performance of the post-model; Based on feature importance score The sorting of Remove the least important features , and obtain a new feature subset , Retrain the random forest model and update the feature importance scores of the remaining features Continue recursively performing feature elimination steps until the feature subset The size reaches the preset final number of features , or the performance index of the model No more promotion; After recursive feature elimination is completed, the retained feature set is the final comprehensive feature set, denoted as: ,in is the number of features in the final feature set.

6. A thermal radiation anomaly recognition system based on the fusion of multi-source heterogeneous data features, characterized in that: The system includes the following modules: An acquisition module, used for acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes infrared image data and meteorological data; A first feature extraction module is used to extract temperature distribution features from the infrared image and to extract ambient temperature, humidity and wind speed features from the meteorological data; A second feature extraction module is used to use a K nearest neighbor classifier as an evaluation model for feature selection to perform recursive feature elimination on the extracted features to obtain a first feature subset; A third feature extraction module is used to calculate the intra-class scatter matrix and the inter-class scatter matrix of the first feature subset; determine the optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and project the first feature subset onto the optimal projection matrix to obtain a reduced-dimensional feature subset; A fourth feature extraction module, used for performing feature interaction operation on the dimension reduction feature subset to generate an interactive feature subset; A fifth feature extraction module, configured to screen the interactive feature subset using a packaged feature selection algorithm to obtain a final comprehensive feature set; The detection module is used to construct a thermal radiation anomaly detection model based on the final comprehensive feature set, and use the thermal radiation anomaly detection model to identify thermal radiation anomalies.

7. The thermal radiation anomaly recognition system based on fusion of multi-source heterogeneous data features according to claim 6 is characterized in that: The extracted features are treated as the evaluation model for feature selection using the K nearest neighbor classifier, and recursive feature elimination is performed to obtain the first feature subset including: Set the extracted feature set to , in, is the temperature distribution characteristic of each pixel in the infrared image; is the temperature gradient feature in the infrared image; is the temperature time variation feature in the infrared image; is the ambient temperature feature in the meteorological data; is the relative humidity feature in meteorological data; is the wind speed feature in the meteorological data; choose The nearest neighbor classifier is used as the evaluation model, setting the neighbor parameters , used to calculate the feature set The contribution of each feature to the classification of thermal radiation anomalies; For each feature , its importance score Calculated by the change in model accuracy: ， in: Based on the entire feature set The accuracy of the trained model; To remove features The accuracy of the post-training model; In each iteration, the feature importance score , select the feature with the least importance , and remove it from the current feature set Remove from the list to form a new feature set ; Retraining Nearest neighbor classifier, calculate updated feature importance scores , and again select the least important feature to remove; Continue to perform the above recursive feature elimination steps until the feature set is reduced to the preset feature subset size , or the model accuracy no longer improves; After recursive feature elimination is completed, the retained feature set is the first feature subset .

8. The thermal radiation anomaly recognition system based on fusion of multi-source heterogeneous data features according to claim 7 is characterized in that: Calculating the intra-class scatter matrix and the inter-class scatter matrix of the first feature subset; determining an optimal projection matrix according to the intra-class scatter matrix and the inter-class scatter matrix, and projecting the first feature subset onto the optimal projection matrix, to obtain a reduced-dimensional feature subset comprising: The first feature subset is ,in, Features extracted from multi-source heterogeneous data; For each feature in the first feature subset, calculate the intra-class scatter matrix : ， in, is the total number of categories; is the sample set in the 𝑖th class; For the The mean vector of class samples; is a feature vector, which represents the feature value of a specific sample; Compute the between-class scatter matrix : ， in, is the number of samples in the 𝑖th class; For the The mean vector of class samples; is the global mean vector of all samples; Maximize the inter-class scatter and minimize the intra-class scatter in the optimal projection space to find the optimal projection matrix : ， in, represents the determinant of a matrix; is the linear projection matrix; The first feature subset Project to the optimal projection matrix On the above, we get the reduced dimension feature subset : 。 9. The thermal radiation anomaly recognition system based on fusion of multi-source heterogeneous data features according to claim 8 is characterized in that: The performing feature interaction operation on the dimension reduction feature subset to generate the interactive feature subset includes: feature product operation, feature square difference operation, and feature weighted combination operation.

10. The thermal radiation anomaly recognition system based on fusion of multi-source heterogeneous data features according to claim 9 is characterized in that: The packaged feature selection algorithm is used to screen the interactive feature subset to obtain the final comprehensive feature set including: The interaction feature subset is ,in is the interactive feature; Initialize the feature set to the interactive feature subset , and define the objective function ; In the interaction feature subset Train the random forest model and calculate each interaction feature Importance score : ， in: Represents the model on all interaction feature subsets Performance on Removal of features The performance of the post-model; Based on feature importance score The sorting of Remove the least important features , and obtain a new feature subset , Retrain the random forest model and update the feature importance scores of the remaining features Continue recursively performing feature elimination steps until the feature subset The size reaches the preset final number of features , or the performance index of the model No more promotion; After recursive feature elimination is completed, the retained feature set is the final comprehensive feature set, denoted as: ,in is the number of features in the final feature set.

Images