Non-contact infrared temperature measurement method combined with deep learning

Abstract

The invention relates to a non-contact infrared temperature measurement method combined with deep learning, belongs to the technical field of infrared radiation measurement, and aims to solve the technical problem that the prior art has limitation in processing a complex nonlinear relationship among multiple variables, and the method comprises the following steps: carrying out infrared temperature measurement on a measured object under different environmental conditions; obtaining various influence factor parameters and performing feature extraction; grouping the multiple influence factors by adopting a control variable method according to a feature extraction result to obtain original data; preprocessing the original data; using a processing result to construct a deep learning neural network, and training and optimizing to obtain an infrared temperature measurement compensation model; a high-precision measurement result is output by using an infrared temperature measurement compensation model, a complex nonlinear relationship of various influence factors is effectively processed by constructing and training a neural network model, and real-time measurement and compensation of infrared temperature measurement are realized at the same time.

Description

Technical Field

[0001] This invention relates to the field of infrared radiation measurement technology. Background Technology

[0002] Non-contact infrared thermometry is widely used in industrial monitoring, medical diagnostics, and environmental monitoring due to its advantages of speed, convenience, and non-invasiveness. Infrared temperature measurement is based on the infrared energy radiated by an object, calculating the object's temperature by measuring the intensity of infrared radiation on its surface. However, this measurement method is susceptible to external environmental conditions, such as ambient temperature, the emissivity of the object's surface, measurement distance, and angle, all of which can lead to measurement errors.

[0003] In existing infrared temperature measurement technologies, empirical formulas or simple linear compensation models are typically used to correct measurement errors. However, these methods have significant limitations when dealing with complex nonlinear relationships between multiple variables, especially in applications requiring high precision. For example, Chinese Patent CN117968863B proposes an infrared temperature measurement method, device, equipment, and storage medium. This method collects information on the detection environment, equipment, emissivity, and image data, and corrects the radiance after temperature measurement to obtain a temperature closer to the true value, thus solving the temperature measurement error problem caused by inaccurate or low emissivity. However, in complex measurement environments, the compensation effect remains limited and lacks adaptability, making it difficult to cope with the variable environmental conditions in practical applications.

[0004] In recent years, with the development of artificial intelligence technology, deep learning, especially neural networks, has been widely used in the modeling and prediction of nonlinear problems. Neural networks have a very broad application prospect in infrared temperature measurement. Summary of the Invention

[0005] To address the limitations of existing technologies in handling complex nonlinear relationships among multiple variables and to further achieve high-precision compensation for infrared temperature measurement errors, this invention proposes a non-contact infrared temperature measurement method incorporating deep learning, such as... Figure 1 As shown, it includes the following steps:

[0006] S1. Perform infrared temperature measurement on the object under different environmental conditions, obtain parameters of various influencing factors and extract features;

[0007] S2. Based on the feature extraction results of S1, the raw data of multiple influencing factors are obtained by grouping them using the control variable method.

[0008] S3. Raw data preprocessing;

[0009] S4. Use the processing results of S3 to construct a deep learning neural network, train and optimize it to obtain an infrared temperature measurement compensation model.

[0010] S5. High-precision measurement results are output using an infrared temperature compensation model.

[0011] Furthermore, the various influencing factors mentioned in S1 include ambient temperature and humidity, light intensity, distance between the infrared temperature sensor and the object being measured, and measurement angle.

[0012] Furthermore, the feature extraction described in S1 includes:

[0013] First, parameter preprocessing is performed to remove obvious outliers and noise. Outlier detection uses the three-standard-deviation method, and noise filtering uses a low-pass filter.

[0014] Parameter standardization involves normalizing all parameters to eliminate the influence of dimensions. The normalization formula is as follows:

[0015]

[0016] Where x represents various influencing factors, and x′ represents the normalized parameters;

[0017] Feature extraction involves using statistical analysis tools to analyze the normalized parameters and extract the feature values ​​of each influencing factor. These feature values ​​include the mean, variance, and correlation coefficient. Principal component analysis is then used to reduce the dimensionality of the high-dimensional feature space.

[0018] Furthermore, S3's raw data preprocessing includes data cleaning to remove outliers and noise, data normalization to standardize the data, and extraction of key features from the raw data. The preprocessed data will then be used as input to a deep learning neural network.

[0019] Furthermore, the input layer of the deep learning neural network described in S4 consists of preprocessed multidimensional feature data. Each time step includes the current ambient temperature and humidity, light intensity, the raw temperature value output by the infrared temperature sensor, the distance to the object being measured, and the measurement angle. The CNN layer can extract local features from the input layer; the pooling layer can reduce the feature dimensionality; the LSTM layer can receive the extracted local features and capture long-term dependencies in the time series through a gating mechanism; the attention mechanism layer can dynamically assign importance weights to different input features; the fully connected layer can further map high-dimensional features to a low-dimensional space; and the output layer is a linear regression unit that directly outputs the compensated temperature prediction value.

[0020] Technical effects:

[0021] This invention provides a non-contact infrared temperature measurement method combining deep learning. By constructing and training a neural network model, it effectively handles the complex nonlinear relationships of various influencing factors, reducing measurement errors to within ±2℃. Simultaneously, it achieves real-time measurement and compensation functions, resulting in an 80% increase in measurement efficiency and a reduction in energy consumption of approximately 33%.

[0022] 1. Introduction of Infrared Thermometry Compensation Model

[0023] This paper employs an architecture combining LSTM, CNN, and an attention mechanism. LSTM processes time-series data, CNN extracts local features from the input data, and the attention mechanism enhances the focusing ability of the deep neural network, automatically identifying the most important features for temperature measurement error compensation. This overcomes the limitations of traditional empirical formulas and linear compensation models in existing technologies. The neural network possesses powerful nonlinear modeling capabilities, enabling it to handle complex multivariate relationships and thus achieve accurate compensation for infrared temperature measurement errors. This method significantly improves measurement accuracy and stability, demonstrating remarkable advantages, especially in scenarios with complex environmental conditions or high measurement requirements. Through learning and training on a large amount of historical data, it achieves accurate compensation for measurement errors, thereby significantly improving the accuracy and stability of temperature measurement. It is particularly suitable for applications with high accuracy requirements, such as industrial monitoring, medical diagnosis, and scientific research experiments.

[0024] 2. Comprehensive consideration of multiple factors

[0025] This invention comprehensively considers various factors affecting infrared temperature measurement, including ambient temperature, humidity, measurement angle, distance, and surface characteristics of the object being measured. These factors are monitored in real time through a data acquisition system and jointly analyzed using an infrared temperature compensation model to ensure the accuracy of the measurement results. Compared to traditional single-factor compensation methods, this invention exhibits higher robustness and adaptability. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the overall process of this invention.

[0027] Figure 2 This is a schematic diagram of the infrared temperature measurement compensation model structure in an embodiment of the present invention. Detailed Implementation

[0028] To accurately analyze the impact of various influencing factors on infrared thermometry and achieve high-precision compensation for infrared temperature measurement errors, this invention provides a detailed description of the original data acquisition process and model construction, as follows:

[0029] S1. Perform infrared temperature measurement on the object under different environmental conditions, obtain parameters of various influencing factors, and extract features.

[0030] S101. In a controlled environment, a standard temperature measurement experimental platform is built. The indoor ambient temperature, humidity, and light intensity are adjustable to simulate various actual measurement conditions. Several common materials are selected for the object being measured, such as metal, glass, and plastic, which have different surface properties and different reflectivities and emissivities.

[0031] S102. Data Acquisition Equipment: A high-precision infrared temperature sensor with multi-band measurement capability is selected. The infrared temperature sensor is connected to the data acquisition system, which can record temperature data, influencing factor parameters, and other information in real time. All equipment is calibrated to ensure measurement accuracy.

[0032] S103. Set different environmental conditions for data acquisition. For example, conduct measurements under different ambient temperatures (e.g., -10℃, 25℃, 50℃) and different humidity levels (e.g., 20%, 50%, 80%). Set the distance between the infrared temperature sensor and the object being measured to 0.5m, 1m, 2m, etc., and gradually adjust the measurement angle from 0° to 90°. Under each combination of conditions, the amount of data collected must be sufficient to reflect the impact of that condition on the measurement results.

[0033] Before each data acquisition begins, ensure the infrared temperature sensor and related equipment are in stable condition. The data acquisition system automatically records the temperature data output by the infrared temperature sensor, along with ambient temperature, humidity, distance, and angle. Measurements are repeated multiple times under each condition to ensure data reliability and repeatability.

[0034] S2. Based on the feature extraction results, the raw data are obtained by grouping multiple influencing factors using the control variable method.

[0035] S201, Group Design

[0036] Based on the preliminary data collection and feature extraction results, the weight and importance of each influencing factor are determined. The importance of each factor is ranked using information gain or entropy methods, thus allowing for reasonable grouping. Grouping should cover all identified influencing factors, with only one or a few variables changed within each group to more clearly observe the impact of these factors on the measurement results.

[0037] S202, Group Data Acquisition

[0038] Strictly control variables and collect infrared temperature measurement data. The data volume for each group should be large enough to ensure data representativeness and the validity of statistical analysis. The collected data includes the temperature value output by the infrared temperature sensor, ambient temperature, humidity, measurement angle, measurement distance, and other information.

[0039] S3. Raw data preprocessing: This includes data cleaning to remove outliers and noise, data normalization to standardize the data, and extracting key features from the raw data. The preprocessed data will be used as input to the deep learning neural network.

[0040] S4. Construct a deep learning neural network, train and optimize it to obtain an infrared temperature measurement compensation model;

[0041] S401, Model Selection and Network Architecture Design

[0042] This invention employs an architecture combining LSTM (Long Short-Term Memory) network, CNN (Convolutional Neural Network), and an attention mechanism. LSTM is used to process time-series data, CNN is used to extract local features from the input data, and the attention mechanism is used to enhance the model's focusing ability, automatically identifying the features most important for temperature measurement error compensation, such as... Figure 2 As shown.

[0043] Input layer: Receives preprocessed multidimensional feature data. Each time step includes multiple input variables such as the current ambient temperature, humidity, measurement angle, and the raw temperature value output by the sensor.

[0044] CNN layer: Extracts local features from the input layer;

[0045] Pooling layer: Extracts local spatiotemporal features of the data by sliding convolutional kernels along the time dimension. The activation function is ReLU, and the feature maps after convolution are input into the pooling layer to further reduce the feature dimensionality.

[0046] LSTM Layer: The extracted features are input into the LSTM layer, which uses its gating mechanism to capture long-term dependencies in the time series. The LSTM layer uses a two-layer LSTM structure: the first LSTM layer captures short-term temperature change trends, and the second LSTM layer models long-term dependencies.

[0047] Attention Mechanism Layer: After the LSTM layer, an attention mechanism layer is added to dynamically assign importance weights to different input features.

[0048] Fully connected layer: After the attention mechanism layer, a fully connected layer is added to further map high-dimensional features to a low-dimensional space.

[0049] Output layer: The output layer is a linear regression unit that directly outputs the compensated temperature prediction value.

[0050] S402, Model Training and Optimization

[0051] In the model training and optimization process of this invention, data preparation and training settings are first implemented to ensure that the model can accurately capture key features in infrared thermometry. Specifically, the dataset is divided into training, validation, and test sets in a 7:2:1 ratio, and data augmentation techniques are used to improve the model's generalization ability. During model training, the Adam optimizer is used for gradient descent, and mean squared error is selected as the loss function to minimize the difference between predicted and actual values. Mini-batch gradient descent is used during training, with batch sizes typically set to 32 to 64. To prevent overfitting, L2 regularization and dropout techniques are introduced. Furthermore, a dynamic learning rate adjustment strategy is applied to ensure that the model can converge quickly and reach the global optimum during training.

[0052] During model optimization, hyperparameter tuning and cross-validation are used to improve model performance. Hyperparameter tuning employs grid search or random search to find the optimal parameter combination that allows the model to perform best on the validation set. K-fold cross-validation is used to evaluate the model's generalization ability, and an early stopping strategy is employed to prevent overfitting; training is terminated when the validation set loss stops decreasing. Finally, ensemble learning is used to integrate multiple models, further improving the model's robustness and accuracy.

[0053] S5. The infrared temperature sensor is compensated using an infrared temperature measurement compensation model to output high-precision measurement results.

[0054] The trained and optimized infrared temperature measurement compensation model is exported and deployed to edge computing devices or cloud platforms. It then begins receiving infrared temperature sensor data in real time, processing and compensating for it to output high-precision temperature measurement results. The entire process ensures the model can operate efficiently in real-world application scenarios, providing stable and reliable real-time data support.

[0055] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

Claims

1. A non-contact infrared temperature measurement method combining deep learning, characterized in that, Includes the following steps: S1. Perform infrared temperature measurement on the object under different environmental conditions, obtain parameters of various influencing factors and extract features; S2. Based on the feature extraction results of S1, the raw data of multiple influencing factors are obtained by grouping them using the control variable method. S3. Raw data preprocessing; S4. Use the processing results of S3 to construct a deep learning neural network, train and optimize it to obtain an infrared temperature measurement compensation model. S5. High-precision measurement results are output using an infrared temperature compensation model.

2. The non-contact infrared temperature measurement method combining deep learning according to claim 1, characterized in that, The various influencing factors mentioned in S1 include ambient temperature and humidity, light intensity, distance between the infrared temperature sensor and the object being measured, and measurement angle.

3. The non-contact infrared temperature measurement method combining deep learning according to claim 1, characterized in that, The feature extraction described in S1 includes: First, parameter preprocessing is performed to remove obvious outliers and noise. Outlier detection uses the three-standard-deviation method, and noise filtering uses a low-pass filter. Parameter standardization involves normalizing all parameters to eliminate the influence of dimensions. The normalization formula is as follows: Where x represents various influencing factors, and x′ represents the normalized parameters; Feature extraction involves using statistical analysis tools to analyze the normalized parameters and extract the feature values ​​of each influencing factor. These feature values ​​include the mean, variance, and correlation coefficient. Principal component analysis is then used to reduce the dimensionality of the high-dimensional feature space.

4. The non-contact infrared temperature measurement method combining deep learning according to claim 1, characterized in that, S3's raw data preprocessing includes data cleaning to remove outliers and noise, data normalization to standardize the data, and extraction of key features from the raw data. The preprocessed data will then be used as input to a deep learning neural network.

5. The non-contact infrared temperature measurement method combining deep learning according to claim 1, characterized in that, The input layer of the deep learning neural network described in S4 is preprocessed multidimensional feature data. Each time step includes the current ambient temperature and humidity, light intensity, the raw temperature value output by the infrared temperature sensor, the distance to the object being measured, and the measurement angle. The CNN layer can extract local features from the input layer. Pooling layers can reduce feature dimensionality; LSTM layers can receive extracted local features and capture long-term dependencies in time series through gating mechanisms; attention mechanism layers can dynamically assign importance weights to different input features. The fully connected layer can further map high-dimensional features to a low-dimensional space; the output layer is a linear regression unit that directly outputs the compensated temperature prediction value.

6. The non-contact infrared temperature measurement method combining deep learning according to claim 5, characterized in that, The LSTM layer uses a two-layer LSTM structure. The first LSTM layer captures short-term temperature change trends, and the second LSTM layer models long-term dependencies.

7. The non-contact infrared temperature measurement method combining deep learning according to claim 1, characterized in that, The infrared temperature measurement compensation model was trained and optimized in S4. The dataset was divided into training, validation and test sets in a ratio of 7:2:

1. Mini-batch gradient descent was performed using the Adam optimizer. The model performance was improved by adjusting hyperparameters and cross-validation during the optimization process.

Images