Small and medium-sized place fire early warning method based on low-resolution thermal imaging
By using a combination of low-resolution thermal imaging module and decision tree algorithm in small and medium-sized places, the problems existing in traditional fire detection methods in early hidden fire detection are solved, and a low-cost and efficient fire warning effect is achieved.
Patent Information
- Application Number
- CN202411859918.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-04-04
AI Technical Summary
Traditional fire detection methods have insufficient sensitivity and detection lag in early hidden fire detection. Especially in small and medium-sized places, low-cost but low-resolution thermal imaging equipment has low detection accuracy, making it difficult to achieve efficient fire warning.
The low-resolution thermal imaging module is used to collect temperature data, and the full resolution of the interpolation algorithm and Gaussian fuzzy are expanded, and the high-temperature area is calculated by combining the fill algorithm, and the decision tree algorithm is used to train the model for fire detection.
It realizes low-cost and efficient fire detection in small and medium-sized places, and can accurately identify fires in complex environments such as low light and smoke, and has good real-time and accuracy.
Smart Images
Figure CN119763263A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a fire early warning method for small and medium-sized places based on low-resolution thermal imaging, belonging to the technical field of fire prevention and control. Background Art
[0002] With the rapid development of society, large buildings, factories, warehouses, entertainment facilities, etc. are increasing day by day, and the accompanying fire problems are becoming more and more serious. Therefore, timely and accurate fire detection is very important to protect people's property safety.
[0003] Traditional fire detection methods mostly rely on smoke detectors or temperature sensors, which are often easily interfered with in some scenarios, and have problems with insufficient sensitivity or delayed detection, and may not be able to detect early fires in real time. Vision-based fire warning methods rely on cameras to capture images within the visible light range, and use image processing technology and computer vision algorithms to identify fire features such as flames and smoke. Usually, the system detects fire sources through visual features such as color, shape, and movement. For example, flames are usually orange-red, with high brightness and irregular shapes, and the algorithm can determine whether a fire has occurred by identifying these features. However, this method has limited detection range and capabilities. Especially for early hidden fires, since the characteristics of flames and smoke are not yet obvious, visual methods may be difficult to detect in time.
[0004] With the development of image processing and sensing technology, fire detection systems based on thermal imaging technology have emerged. Thermal imaging equipment generates temperature images by capturing the infrared heat radiated by objects, and can monitor in complex conditions such as no light and smoke. However, high-resolution thermal imaging equipment is expensive and has high deployment costs, which limits its application in some small and medium-sized places (such as dormitories, conference rooms, etc.). Low-resolution thermal imaging equipment is low-cost and suitable for large-scale deployment. However, its detection accuracy is low, and it is necessary to improve the accuracy of fire warning by optimizing the detection algorithm. Summary of the invention
[0005] In order to overcome the deficiencies of the prior art, the present invention provides a fire early warning method for small and medium-sized places based on low-resolution thermal imaging.
[0006] A fire early warning method for small and medium-sized places based on low-resolution thermal imaging, comprising the following steps:
[0007] Step 1: Use a low-resolution thermal imaging module to collect temperature data of common fire sources and non-fire source high-temperature objects in small and medium-sized places;
[0008] Step 2: Expand the resolution through interpolation algorithm and Gaussian blur, determine the adjacent high-temperature pixels through filling algorithm, and calculate the high-temperature area;
[0009] Step 3: Count the temperature characteristics of common fire sources and other high-temperature objects in the target environment, and use the decision tree algorithm to train and obtain a model;
[0010] Step 4: Use the trained model for actual fire warning and input the real-time collected temperature data into the model to achieve fire detection.
[0011] The step one specifically includes: using a non-contact infrared dot matrix temperature measurement module to obtain temperature data of common flames, recording temperature changes from the start of ignition to the vigorous burning of the flame, simulating the entire process of early combustion of the flame, and collecting temperature data of boiling water as an interference item to distinguish the temperature characteristics of normal high-temperature objects and flames. During the collection process, multiple experiments are conducted at different angles and distances to ensure that the temperature data at different spatial orientations covers a wider range.
[0012] In the step 1, the image size collected by the non-contact infrared dot matrix temperature measurement module is 32*24 pixels.
[0013] The step 2 specifically includes:
[0014] The temperature matrix is enlarged using the bilinear interpolation method, and the 32*24 pixel temperature data is expanded to an image with a resolution of 320*240 pixels according to a 10-fold ratio. The interpolation algorithm is based on the temperature value of each pixel. During the enlargement process, the high temperature area image is generated through the correspondence between temperature and color. The Gaussian blur algorithm is applied to the enlarged temperature image to smooth the image and eliminate noise and irregular temperature changes.
[0015] For the high-temperature points in the image, recursively check the surrounding adjacent points to determine whether these points meet the temperature threshold, and add the points that meet the conditions to the same high-temperature area to calculate the area. For the high-temperature points in the image, recursively check the surrounding adjacent points to determine whether these points meet the temperature threshold, and add the points that meet the conditions to the same high-temperature area to calculate the area.
[0016] Specifically,
[0017] The low-resolution temperature data obtained from the non-contact infrared dot matrix temperature measurement module is expanded by applying an interpolation algorithm. The original 32x24 pixel temperature data will be enlarged to 320x240 pixels according to a preset ratio. The interpolation process is based on the temperature values of adjacent pixels. New pixels are generated between these pixels through interpolation calculation to smoothly expand the original data.
[0018] Secondly, the Gaussian blur algorithm is applied to the expanded temperature image to smooth the image. Gaussian blur uses a weighted Gaussian kernel to calculate the surrounding pixel values of each pixel, making the temperature distribution smoother and more coherent, while reducing the interference of noise points on subsequent high-temperature area detection;
[0019] After interpolation and blurring, a filling algorithm is used to determine the boundaries of the high-temperature area. For high-temperature pixels in the image, the surrounding adjacent pixels are recursively checked to determine whether these pixels also belong to the high-temperature area. If the temperature value of the adjacent pixels also exceeds the threshold, they are added to the same high-temperature area. Through this recursive expansion method, the complete high-temperature area is identified and its area is calculated.
[0020] The step three specifically includes:
[0021] The temperature characteristics of common fire sources (such as alcohol fire, paper flame, etc.) and other high-temperature objects (such as boiling water, etc.) in the target environment are obtained to construct a data set. The collected temperature feature data is divided into flame data and non-flame data. Flame data indicates the situation with a fire source, and non-flame data indicates the situation without a fire source or a normal high-temperature object. Flame data and non-flame data are divided into training set and test set in a ratio of 4:1 to ensure that the model has strong generalization ability.
[0022] The decision tree algorithm is used to train the temperature data in the environment and construct a combined data set of flame and non-flame data. The features include the maximum temperature, the area of the high-temperature area, and the average temperature. The decision tree model automatically generates classification nodes and outputs the classification results by judging the maximum temperature value and the area of the high-temperature area of the input features.
[0023] Specifically,
[0024] The temperature characteristics of common fire sources and other high-temperature objects in the statistical environment, such as the maximum temperature, average temperature, and high-temperature area, are annotated and classified according to flame data and non-flame data. A training data set is constructed, and the decision tree algorithm is used to train the temperature data. The decision tree algorithm can automatically generate classification nodes based on the characteristics. During the training process, the algorithm will select the optimal features as the basis for classification, and gradually distinguish between flame data and non-flame data. The decision tree continuously refines the classification rules and finally outputs a model to determine whether the input data is a flame feature or a non-flame feature. During the training process, the data set is divided into a training set and a test set in a ratio of 4:1. The trained decision tree model can effectively distinguish high-temperature objects in the target environment.
[0025] The step 4 specifically includes:
[0026] The non-contact infrared dot matrix temperature measurement module collects temperature data in the environment in real time. The collected data is expressed in the form of a temperature matrix, including the temperature value corresponding to each pixel. The temperature data is collected at fixed time intervals through the module to achieve continuous monitoring of the ambient temperature.
[0027] The system first checks whether the temperature exceeds the preset safety temperature threshold. If the temperature is lower than the threshold, the system will judge it as a normal environment. If the temperature exceeds the threshold, the system will further analyze the area of the high-temperature area and the temperature characteristics of the area. Based on the characteristics in the temperature data, the decision tree model will output whether there is a flame. If flame characteristics are detected and the state lasts for more than multiple frames, the system will determine it as a fire.
[0028] Specifically, a thermal sensor is used to collect the target ambient temperature. When the temperature exceeds the threshold, the area of the high-temperature region is calculated using the method in step 2, and detection is started based on the temperature characteristics and the area of the high-temperature region. When it is determined that a flame is generated and this state lasts for more than 5 frames, it is determined that the environment is abnormal and a fire has occurred.
[0029] Compared with the prior art, the present invention has the following beneficial effects:
[0030] The present invention uses a low-resolution infrared temperature measurement module to collect temperature data. The equipment cost is greatly reduced compared to high-resolution thermal imaging equipment, and it is suitable for large-scale deployment in small and medium-sized places such as dormitories and conference rooms. Based on infrared thermal imaging technology, it can work normally in complex environments such as low light and smoke, avoiding the failure problem of traditional visual fire warning systems in insufficient light or smoke. The decision tree algorithm is used to efficiently classify flames and high-temperature objects. The system can process the input temperature data in real time, with good real-time performance and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] 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 or the description of the prior art 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.
[0032] Figure 1 The figure is a flow chart of the fire warning method for small and medium-sized places based on low-resolution thermal imaging of the present invention. DETAILED DESCRIPTION
[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0034] Reference Figure 1 , a fire early warning method for small and medium-sized places based on low-resolution thermal imaging, comprising the following steps:
[0035] Step 1: Use a low-resolution thermal imaging module to collect temperature data of common fire sources and non-fire source high-temperature objects in small and medium-sized places;
[0036] Specifically, the non-contact infrared dot matrix temperature measurement module is used to obtain the temperature data of common flames such as alcohol flames, oil fires, and paper fires, and the temperature changes from the beginning of ignition to the vigorous burning of the flame are recorded to simulate the entire process of early flame combustion. In order to improve the robustness and adaptability of the model, the temperature data of boiling water is also collected as an interference item to distinguish the temperature characteristics of normal high-temperature objects and flames. In addition, during the collection process, multiple experiments are conducted at different angles and distances to ensure that the temperature data at different spatial orientations covers a wider range, thereby improving the application effect and accuracy of the fire detection algorithm in actual scenarios.
[0037] Step 2: Expand the resolution through interpolation algorithm and Gaussian blur, determine the adjacent high-temperature pixels through filling algorithm, and calculate the high-temperature area;
[0038] Specifically, the low-resolution temperature data obtained from the non-contact infrared dot matrix temperature measurement module is first expanded by applying an interpolation algorithm. The original 32x24 pixel temperature data will be enlarged to 320x240 pixels according to a preset ratio (such as 10 times). The interpolation process is based on the temperature values of adjacent pixels, and new pixels are generated between these pixels through interpolation calculation, thereby smoothly expanding the original data and avoiding abrupt boundaries.
[0039] Secondly, the Gaussian blur algorithm is applied to the expanded temperature image to smooth the image. The purpose of Gaussian blur is to eliminate noise and irregular temperature fluctuations in the image, especially small-scale temperature anomalies that may appear after interpolation. Gaussian blur uses a weighted Gaussian kernel to calculate the surrounding pixel values of each pixel, making the temperature distribution smoother and more coherent, while reducing the interference of noise points on subsequent high-temperature area detection.
[0040] After interpolation and blurring, a filling algorithm is used to determine the boundaries of the high-temperature area. Specifically, for high-temperature pixels in the image (temperature values higher than the set fire threshold), recursively check the surrounding adjacent pixels to determine whether these pixels also belong to the high-temperature area. If the temperature value of the adjacent pixel also exceeds the threshold, it is added to the same high-temperature area. Through this recursive expansion method, the complete high-temperature area can be identified and its area calculated.
[0041] Step 3: Count the temperature characteristics of common fire sources and other high-temperature objects in the target environment, and use the decision tree algorithm to train and obtain a model;
[0042] Specifically, we first counted the temperature characteristics of common fire sources and other high-temperature objects in the environment, such as the highest temperature, average temperature, and high-temperature area. These temperature characteristics were labeled and classified according to flame data (indicating the presence of a fire source) and non-flame data (indicating the absence of a fire source or a normal high-temperature object) to construct a training data set. Then, we used the decision tree algorithm to train these temperature data. The decision tree algorithm can automatically generate classification nodes based on the characteristics. During the training process, the algorithm will select the optimal feature (such as the highest temperature or the area of the high-temperature area) as the basis for classification, and gradually distinguish between flame data and non-flame data. The decision tree continuously refines the classification rules and finally outputs a model to determine whether the input data is a flame feature or a non-flame feature. During the training process, the data set is divided into a training set and a test set in a ratio of 4:1. The trained decision tree model can effectively distinguish high-temperature objects in the target environment.
[0043] Step 4: Use the trained model for actual fire warning and input the real-time collected temperature data into the model to achieve fire detection.
[0044] Specifically, the temperature data in the environment is first collected in real time through the non-contact infrared dot matrix temperature measurement module. The collected data is represented in the form of a temperature matrix, which contains the temperature value corresponding to each pixel. These temperature data can be collected by the module at fixed time intervals (such as once per second), so as to realize continuous monitoring of the ambient temperature. The system first checks whether the temperature exceeds the preset safety temperature threshold. If the temperature is lower than the threshold, the system will judge it as a normal environment and no further processing is required. If the temperature exceeds the threshold, the system will further analyze the area of the high temperature area and the temperature characteristics of the area. According to the features in the temperature data (maximum temperature, average temperature, area of high temperature area, etc.), the decision tree model will output whether there is a flame. If the flame feature is detected and the state lasts for more than multiple frames (such as 5 frames), the system will determine it as a fire.
[0045] The following results were obtained based on the technical solution provided in the above example.
[0046] Table 1 Performance indicators of decision tree model on the test set
[0047]
[0048]
[0049] In this example, Table 1 shows the precision, recall, and F1 score of each category label obtained by inputting the test set into the decision tree model.
[0050] Table 2 Accuracy of the model under different flames
[0051] Model Alcohol Oil Paper RandomForest 99.78 99.65 97.07 Gradient_Boosting 100.00 99.48 97.30 Logistic_Regression 97.63 81.86 92.37 DecisionTree 99.56 99.65 98.61
[0052] In this example, Table 2 shows the accuracy of different machine learning models under different data sets. GradientBoosting, Random Forest, and Decision Tree all performed very well under the three types of flames, while Decision Tree performed stably under different fire sources.
[0053] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.
Claims
1. A fire early warning method for small and medium-sized places based on low-resolution thermal imaging, characterized by: The following steps are involved: Step 1: Use a low-resolution thermal imaging module to collect temperature data of common fire sources and non-fire source high-temperature objects in small and medium-sized places; Step 2: Expand the resolution through interpolation algorithm and Gaussian blur, determine the adjacent high-temperature pixels through filling algorithm, and calculate the high-temperature area; Step 3: Count the temperature characteristics of common fire sources and other high-temperature objects in the target environment, and use the decision tree algorithm to train and obtain a model; Step 4: Use the trained model for actual fire warning and input the real-time collected temperature data into the model to achieve fire detection.
2. A fire warning method for small and medium-sized places based on low-resolution thermal imaging according to claim 1, characterized in that: The step one specifically includes: using a non-contact infrared dot matrix temperature measurement module to obtain temperature data of common flames, recording temperature changes from the start of ignition to the vigorous burning of the flame, simulating the entire process of early combustion of the flame, and collecting temperature data of boiling water as an interference item to distinguish the temperature characteristics of normal high-temperature objects and flames. During the collection process, multiple experiments are conducted at different angles and distances to ensure that the temperature data at different spatial orientations covers a wider range.
3. The fire warning method for small and medium-sized places based on low-resolution thermal imaging according to claim 1 is characterized by: In the step 1, the image size collected by the non-contact infrared dot matrix temperature measurement module is 32*24 pixels.
4. The fire warning method for small and medium-sized places based on low-resolution thermal imaging according to claim 1 is characterized by: The step 2 specifically includes: The temperature matrix is enlarged using the bilinear interpolation method, and the 32*24 pixel temperature data is expanded to an image with a resolution of 320*240 pixels according to a 10-fold ratio. The interpolation algorithm is based on the temperature value of each pixel. During the enlargement process, the high temperature area image is generated through the correspondence between temperature and color. The Gaussian blur algorithm is applied to the enlarged temperature image to smooth the image and eliminate noise and irregular temperature changes. For the high-temperature points in the image, recursively check the surrounding adjacent points to determine whether these points meet the temperature threshold, and add the points that meet the conditions to the same high-temperature area to calculate the area.
5. The fire warning method for small and medium-sized places based on low-resolution thermal imaging according to claim 1 is characterized by: The step three specifically includes: The temperature characteristics of common fire sources and other high-temperature objects in the target environment are obtained, and a data set is constructed. The collected temperature characteristic data is divided into flame data and non-flame data. Flame data indicates the situation with a fire source, and non-flame data indicates the situation without a fire source or a normal high-temperature object. Flame data and non-flame data are divided into training set and test set in a ratio of 4:
1. The decision tree algorithm is used to train the temperature data in the environment and construct a combined data set of flame and non-flame data. The features include the maximum temperature, the area of the high-temperature area, and the average temperature. The decision tree model automatically generates classification nodes and outputs the classification results by judging the maximum temperature value and the area of the high-temperature area of the input features.
6. The fire warning method for small and medium-sized places based on low-resolution thermal imaging according to claim 1 is characterized by: The step 4 specifically includes: The non-contact infrared dot matrix temperature measurement module collects temperature data in the environment in real time. The collected data is expressed in the form of a temperature matrix, including the temperature value corresponding to each pixel. The temperature data is collected at fixed time intervals through the module to achieve continuous monitoring of the ambient temperature. The system first checks whether the temperature exceeds the preset safety temperature threshold. If the temperature is lower than the threshold, the system will judge it as a normal environment. If the temperature exceeds the threshold, the system will further analyze the area of the high-temperature area and the temperature characteristics of the area. Based on the characteristics in the temperature data, the decision tree model will output whether there is a flame. If flame characteristics are detected and the state lasts for more than multiple frames, the system will determine it as a fire.