Fire monitoring model, application method thereof and fire detection system
By using a computer vision-based fire monitoring model and image processing technology to extract local spatial features of flames and smoke, the problem of false alarms and missed alarms in outdoor scenarios of traditional fire monitoring has been solved, and efficient fire detection and early warning have been achieved.
Patent Information
- Application Number
- CN202511088163.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-05
- Publication Date
- 2025-11-21
AI Technical Summary
Traditional fire monitoring suffers from false alarms and missed alarms in outdoor scenarios, and it is particularly difficult to achieve real-time monitoring and accurate detection in complex environments.
A computer vision-based fire monitoring model is adopted. Images or videos are acquired through cameras, and fire identification is performed using an image data acquisition module, a data preprocessing and annotation module, a feature extraction module, and a fire sign output module. Multi-scale feature maps are generated by combining the backbone network to extract local spatial features of flames or smoke, and the category, location, and confidence score of candidate boxes are output.
It enables automatic detection and identification of fires, eliminates blind spots in manual inspections, and can quickly identify potential safety hazards, reducing the risk of accidents.
Smart Images
Figure CN120997525A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of valve plate gasket installation, and more specifically, to an outdoor scene fire detection system based on computer vision. Background Technology
[0002] Fire, as a serious natural disaster, poses a huge threat to human life and property. Especially in outdoor settings such as forests, industrial areas, oil fields, and storage areas, fires are often unpredictable and spread rapidly, with extremely serious consequences.
[0003] Currently, traditional fire monitoring mainly relies on traditional sensor monitoring or human patrols. When the monitoring range is large, manual patrols cannot achieve real-time monitoring and are prone to omissions. Moreover, the outdoor environment is complex and sensors are easily interfered with, making it difficult to accurately detect fires in the early stages. Therefore, false alarms and missed alarms are common problems. Summary of the Invention
[0004] The purpose of this invention is to provide a fire monitoring model that analyzes and processes images or videos of flames or smoke acquired by cameras, enabling automatic detection and identification of fires, eliminating blind spots in manual inspections, quickly identifying potential safety hazards, and effectively reducing the risk of accidents.
[0005] To achieve the above objectives, the present invention provides a fire monitoring model, comprising an image data acquisition module, a data preprocessing and annotation module, a feature extraction module, and a fire sign output module connected in sequence. The image data acquisition module is used to input image data into the data preprocessing and annotation module. The feature extraction module extracts local spatial features at different layers and splices multiple local spatial features in the channel dimension to generate the initial position and size of candidate boxes. The fire sign output module processes the candidate box data and outputs fire sign data including candidate box category, position, and confidence level.
[0006] This invention also provides a method for using a fire monitoring model, including:
[0007] Step 1: Input real-time images into the image data acquisition module;
[0008] Step 2: The data preprocessing and labeling module preprocesses the data input in Step 1 and marks the fire location;
[0009] Step 3: Based on Step 2, the feature extraction module generates the initial position and size of the predicted candidate boxes;
[0010] Step 4: Based on step 3, the fire sign output module outputs fire sign data including candidate box category, location, and confidence level;
[0011] Step 5: Train the data preprocessing and labeling module, feature extraction module, and fire sign output module to improve the accuracy of the fire monitoring model in judging fires.
[0012] Preferably, the training method in step 5 includes:
[0013] Step 0: Input image data into the image data acquisition module, mark the real value of the input fire image, manually mark the marker box and add a category note to the marker box;
[0014] Step 1: Based on the image data input in Step 0, the data preprocessing and annotation module preprocesses the image data and marks the fire location in the image;
[0015] Step 2: Generate the initial position and size of the prediction candidate box based on the feature extraction module in Step 1;
[0016] Step 3: Based on Step 2, the fire sign output module outputs predicted values, which include fire sign data such as candidate box category, location, and confidence level.
[0017] Step 4: Compare the results of Step 0 and Step 3, backpropagate the difference between the actual value and the predicted value, and train the fire monitoring model.
[0018] Preferably, in step 4, the difference between the predicted value and the true value is quantified by a loss function.
[0019] Preferably, in step 0, at least 10,000 fire images need to be entered.
[0020] Preferably, in step 3, the candidate box categories include flame and smoke.
[0021] Preferably, in Step 0, flame indicator frames and smoke indicator frames are respectively provided.
[0022] Preferably, in step 1, a multi-scale feature map is generated based on the image input in step 0, and in step 2, the flame candidate box and the smoke candidate box are identified respectively based on the multi-scale feature map in step 1.
[0023] Preferably, the multiple fire images input in step 0 are randomly divided into three sets: training set, validation set, and test set.
[0024] The present invention also provides a fire detection system, comprising: a sensor, a camera, a computing processing unit, a user terminal, and an alarm, wherein the sensor and the camera are electrically connected to the computing processing unit, and the computing processing unit is electrically connected to the user terminal and the alarm, and is used to send an alarm reminder to the user terminal or control the alarm to issue an alarm signal;
[0025] The computing processing unit includes a control system and the fire monitoring model as described in claim 1. The fire monitoring model receives data transmitted from the camera, identifies fire signs, and sends them to the control system. The control system sends alarm reminders to users or controls the alarm to issue alarm signals based on the data provided by the sensors and the fire detection model.
[0026] According to the above technical solution, the image data acquisition module of the present invention collects real-time images from the camera and inputs these images into the feature extraction module. The feature extraction module processes these image data, generates multi-scale feature maps through the backbone network, and extracts local spatial features of flames or smoke on different feature maps based on the multi-scale feature maps. These local spatial features are stitched together in the channel dimension to generate the initial position and size of candidate boxes. The fire sign output module processes the candidate box data and outputs fire sign data including candidate box category, position and confidence level.
[0027] To improve the accuracy of fire monitoring models in identifying fire information from image data, the data preprocessing and annotation module, feature extraction module, and fire sign output module can be trained by inputting fire images.
[0028] Image data preprocessing includes scaling the image to a uniform size, and can also apply rotation, flipping, brightness adjustment, and other data augmentation techniques.
[0029] Preferably, classification labels are set for image data. Based on the image content, images that show fire content are labeled as "fire," while images that do not show fire content are labeled as "non-fire."
[0030] Images tagged as "fire" also need to be backed up with images where flame and smoke areas are manually marked, such as... Figure 1 As shown.
[0031] After the image is input into the image data acquisition module, the data preprocessing and annotation module distinguishes whether the image belongs to a fire or not. If the image belongs to a fire, the feature extraction module and the fire sign output module need to post-process the fire features in the image until the output identifies the category, location and confidence level of the candidate box.
[0032] The candidate box category is set to smoke or flame to indicate whether smoke or flame is identified within the candidate box. The position of the candidate box includes the center point and size of the candidate box, and the confidence level is the probability of a fire occurring assigned by the fire monitoring model.
[0033] By comparing images of manually marked flame and smoke areas with the results output by the disaster sign output module, and through backpropagation of the comparison results, the fire monitoring model is trained, making the identification results of the fire monitoring model closer to the true values.
[0034] Therefore, this fire monitoring model can automatically detect and identify fires by analyzing and processing images or videos of flames or smoke acquired by cameras, eliminating blind spots in manual inspections, quickly identifying potential safety hazards, and effectively reducing the risk of accidents.
[0035] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description
[0036] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings:
[0037] Figure 1 This is a schematic diagram of a method for annotating fire images;
[0038] Figure 2 It is the core structure of the backbone network of a feature extraction module;
[0039] Figure 3 It is the core structure of a feature extraction module;
[0040] Figure 4 This is a schematic diagram of a fire indicator output module. Detailed Implementation
[0041] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0042] In this invention, unless otherwise stated, directional terms such as "circumferential" and "horizontal" in the terminology represent only the orientation of the term in its normal use or are common terms understood by those skilled in the art, and should not be regarded as limitations on the term.
[0043] A fire monitoring model includes an image data acquisition module, a data preprocessing and annotation module, a feature extraction module, and a fire sign output module connected in sequence. The image data acquisition module is used to input image data into the data preprocessing and annotation module. The feature extraction module extracts local spatial features at different layers and splices multiple local spatial features in the channel dimension to generate the initial position and size of candidate boxes. The fire sign output module processes the candidate box data and outputs fire sign data including candidate box category, position, and confidence level.
[0044] Through the implementation of the above technical solution, the image data acquisition module collects real-time images from the camera and inputs these images into the feature extraction module. The feature extraction module processes these image data, generates multi-scale feature maps through the backbone network, and extracts local spatial features of flames or smoke on different feature maps based on the multi-scale feature maps. These local spatial features are stitched together in the channel dimension to generate the initial position and size of candidate boxes. The fire sign output module processes the candidate box data and outputs fire sign data including candidate box category, position and confidence level.
[0045] To improve the accuracy of fire monitoring models in identifying fire information from image data, the data preprocessing and annotation module, feature extraction module, and fire sign output module can be trained by inputting fire images.
[0046] Image data preprocessing includes scaling the image to a uniform size, and can also apply rotation, flipping, brightness adjustment, and other data augmentation techniques.
[0047] Preferably, classification labels are set for image data. Based on the image content, images that show fire content are labeled as "fire," while images that do not show fire content are labeled as "non-fire."
[0048] Images tagged as "fire" also need to be backed up with images where flame and smoke areas are manually marked, such as... Figure 1 As shown.
[0049] After the image is input into the image data acquisition module, the data preprocessing and annotation module distinguishes whether the image belongs to a fire or not. If the image belongs to a fire, the feature extraction module and the fire sign output module need to post-process the fire features in the image until the output identifies the category, location and confidence level of the candidate box.
[0050] The candidate box category is set to smoke or flame to indicate whether smoke or flame is identified within the candidate box. The position of the candidate box includes the center point and size of the candidate box, and the confidence level is the probability of a fire occurring assigned by the fire monitoring model.
[0051] By comparing images of manually marked flame and smoke areas with the results output by the disaster sign output module, and through backpropagation of the comparison results, the fire monitoring model is trained, making the identification results of the fire monitoring model closer to the true values.
[0052] Therefore, this fire monitoring model can automatically detect and identify fires by analyzing and processing images or videos of flames or smoke acquired by cameras, eliminating blind spots in manual inspections, quickly identifying potential safety hazards, and effectively reducing the risk of accidents.
[0053] The present invention also provides a method for using a fire monitoring model, comprising:
[0054] Step 1: Train the data preprocessing and labeling module, feature extraction module, and fire sign output module to improve the accuracy of the fire monitoring model in identifying fires.
[0055] Step 2: Input real-time images into the image data acquisition module;
[0056] Step 3: The data preprocessing and labeling module preprocesses the image data input in Step 2 and marks the fire location;
[0057] Step 4: Based on step 3, the feature extraction module generates the initial position and size of the predicted candidate box;
[0058] Step 5: Based on step 4, the fire sign output module outputs fire sign data including candidate box category, location, and confidence level.
[0059] The camera captures images of the monitored target and transmits them back to the image data acquisition module of the fire monitoring model. The data preprocessing and labeling module preprocesses these images to identify and mark locations where fires may occur.
[0060] Multi-scale feature maps are generated through a backbone network, and local spatial features are extracted from different feature maps based on these multi-scale feature maps.
[0061] In one implementation, large flames are detected on a 20×20 feature map, while small smoke regions are detected on an 80×80 feature map, effectively adapting to both distant small flames and nearby large fires. Using this residual structure avoids gradient vanishing in deep networks and combines detailed (shallow) and semantic (deep) information to reduce missed detections.
[0062] Step 1 first trains the data preprocessing and annotation module, feature extraction module, and fire sign output module to improve the accuracy of the fire monitoring model in identifying fires. Preferably, during the use of the fire monitoring model, image data is continuously collected, which can also be used to continuously train the data preprocessing and annotation module, feature extraction module, and fire sign output module.
[0063] Through the above steps, the fire monitoring model can achieve high-precision real-time detection of outdoor fires.
[0064] In this embodiment, preferably, the training method in step 1 includes:
[0065] Step 0: Input image data into the image data acquisition module, mark the real value of the input fire image, manually mark the marker box and add a category note to the marker box;
[0066] Step 1: Based on the image data input in Step 0, the data preprocessing and annotation module preprocesses the image data and marks the fire location in the image;
[0067] Step 2: Generate the initial position and size of the prediction candidate box based on the feature extraction module in Step 1;
[0068] Step 3: Based on Step 2, the fire sign output module outputs predicted values, which include fire sign data such as candidate box category, location, and confidence level.
[0069] Step 4: Compare the results of Step 0 and Step 3, backpropagate the difference between the actual value and the predicted value, and train the fire monitoring model.
[0070] The image data input in step 0 is manually collected and input for training the fire monitoring model. Before input, the image data needs to be manually differentiated, with fire images labeled as "fire" and non-fire images labeled as "non-fire." For images labeled as "fire," flame and smoke markers are manually added and stored as backups for comparison.
[0071] After these image data are input into the fire monitoring model, the data preprocessing and labeling module preprocesses the images, distinguishes whether the images belong to fire labels or not, and marks the fire areas in the images that belong to fire labels.
[0072] In one implementation, the feature extraction module utilizes n Bootleneck structures, 3 convolutional modules, and Concat to extract flame and smoke features. The convolutional kernels are 1x1 with a stride of 2. The Bootleneck structure adds the feature map after two convolutional operations to the original feature map to obtain deeper image information. Concat directly concatenates the original features, allowing the deep learning network to learn how to fuse features and reduce information loss. Its structure is as follows: Figure 2 As shown.
[0073] The feature extraction module then needs to further process the extracted feature maps to provide higher-level semantic information and the ability to adapt to images at different scales. Its structure is as follows: Figure 3 As shown, (a) is the FPN backbone network, (b) is bottom-up path expansion, (c) is adaptive pooling, (d) Box represents branches, and (e) is fully connected fusion. By fusing semantic information from deep features and shallow features in feature maps at three scales, the model's detection accuracy for small targets can be improved.
[0074] Finally, the fire sign output module is used to generate the location and category information of the target from the feature map of the network. This model includes a total of 3 branches, which enables the model to effectively detect targets of different sizes and proportions. Figure 4 This is a schematic diagram of a fire indicator output module.
[0075] Step 4 includes the design of loss functions, which mainly include classification loss, bounding box regression loss, and confidence loss.
[0076] The classification loss and confidence score were calculated using a binary cross-entropy function.
[0077]
[0078] Where y is the true label of the sample, p is the predicted probability of the sample, and the lower the loss, the closer the model's prediction is to the actual label.
[0079] Bounding box regression loss is used to calculate the localization loss of flames and smoke in an image. This model uses CIoU as the localization loss.
[0080]
[0081] Where Iou is the intersection-union ratio of the predicted bounding box and the labeled bounding box, and ρ 2 (b,b gt ) is used to calculate the Euclidean distance between the center points of the predicted bounding box and the ground truth bounding box, where c is the diagonal length of the predicted bounding box and the ground truth bounding box, and w is the distance between the center points of the predicted bounding box and the ground truth bounding box. gt and h gtThe width and length of the annotation box are represented by w and t, respectively, which represent the width and length of the prediction box.
[0082] The final overall loss is shown below:
[0083] Loss=box_gain*box_loss+cls_gain*cls_loss+obj_gain*obj_loss
[0084] Here, box_gain, cls_gain, and obj_gain correspond to different loss weights. In one implementation, the default values for box_gain, cls_gain, and obj_gain are 0.05, 0.5, and 1.0, respectively.
[0085] The above details the model training and optimization. In one implementation, a batch size of 32 and an initial learning rate of 0.01 are used to train the model on two GeForce RTX 3090 TURBO 24G graphics cards. During training, the loss function and detection accuracy are monitored in real time, and hyperparameters such as the learning rate are adjusted until the model converges and reaches the expected performance target.
[0086] In this embodiment, preferably, the difference between the predicted value and the true value is quantified by a loss function in step 5.
[0087] The core principle of model training is to gradually adjust model parameters, such as convolution kernel weights, through backpropagation and gradient descent optimization, so that the model's prediction results gradually approach the real labeled data.
[0088] In the specific implementation process, the first step is forward propagation. The input image data is processed layer by layer by the network to obtain prediction results, including bounding box coordinates, confidence probabilities, etc.
[0089] Next, loss calculation is performed. The difference between the predicted value and the actual value is quantified using a loss function.
[0090] Finally, backpropagation is performed based on the loss calculation results. The chain rule is used to calculate the gradient of the loss with respect to the parameters of each layer to determine the direction and magnitude of parameter adjustments.
[0091] In one implementation, gradient calculation is used:
[0092] Where W is the weight of a certain layer.
[0093] A chain-like propagation method is used to backpropagate gradients layer by layer from the output layer. Finally, the model parameters are updated based on the gradients to reduce the loss.
[0094] In this embodiment, preferably, in step 0, no fewer than 10,000 fire images need to be input.
[0095] By combining on-site camera footage with online searches and downloads, no fewer than 10,000 fire-related images were collected, providing ample material for subsequent model training.
[0096] In this embodiment, the categories preferably include flames and smoke.
[0097] By identifying flames and smoke in the image, it is possible to effectively determine whether a fire has occurred.
[0098] In this embodiment, preferably, in Step 0, a flame indicator frame and a smoke indicator frame are respectively provided.
[0099] In this embodiment, preferably, in step 1, a multi-scale feature map is generated based on the image input in step 0, and in step 2, flame candidate boxes and smoke candidate boxes are identified respectively based on the multi-scale feature map in step 1.
[0100] In this embodiment, preferably, the multiple fire images input in step 0 are randomly divided into three sets: a training set, a validation set, and a test set.
[0101] The training set is used directly to train the model, and the model parameters are eventually tuned through backpropagation and gradient descent.
[0102] The validation set is used for model tuning. It is used for hyperparameter selection and model selection, but does not participate in parameter updates.
[0103] The validation set allows for real-time evaluation of the model's performance on unseen data, preventing overfitting.
[0104] In one implementation, if the training set accuracy is 95% but the validation set accuracy is only 70%, it indicates that the model may be overfitting to specific background or lighting conditions in the training set. In this case, it is necessary to adjust the Dropout rate or increase regularization (L2 weight decay) and then re-validate the results.
[0105] The test set serves as the final evaluation, simulating real-world scenarios and providing performance data for the unbiased estimation model.
[0106] The test set is used only after all training and tuning are complete, and must never be used for any training or hyperparameter tuning. The performance of the test set determines the model's performance in real-world surveillance cameras.
[0107] In one implementation, the training set:validation set:test set = 8:1:1. If there are 10,000 images, then the training set has 8,000 images, the validation set has 1,000 images, and the test set has 1,000 images. The image data are randomly assigned to different sets to avoid bias.
[0108] The present invention also provides a fire detection system, comprising: a sensor, a camera, a computing processing unit, a user terminal, and an alarm, wherein the sensor and the camera are electrically connected to the computing processing unit, and the computing processing unit is electrically connected to the user terminal and the alarm, and is used to send an alarm reminder to the user terminal or control the alarm to issue an alarm signal;
[0109] The computing processing unit includes a control system and a fire monitoring model as described in any one of claims 1-5. The fire monitoring model receives data transmitted from the camera, identifies fire signs, and sends them to the control system. The control system sends alarm reminders to users or controls the alarm to issue alarm signals based on the data provided by the sensors and the fire detection model.
[0110] This fire detection system can be deployed in any outdoor scene with a camera surveillance system, such as forests, industrial areas, and oil fields. The cameras and computing units (such as microcontrollers or embedded systems) in this system are installed in key areas of the outdoor monitoring scene. These cameras are used to capture real-time images of flames or smoke and transmit them to the computing unit for analysis and processing. The entire system can be combined with existing monitoring equipment or sensors to form a complete fire monitoring network. During installation, it is necessary to ensure that the camera's field of view covers potentially high-risk fire areas and that the camera equipment can operate normally under different weather conditions.
[0111] This invention provides two operating modes: fully automatic and semi-automatic. In fully automatic mode, the system can automatically detect fires based on images captured by the camera and immediately trigger an alarm or take other emergency measures upon detection. In semi-automatic mode, when the system detects suspected fire signs, it will prompt the user for confirmation via a user terminal. The user can view the fire scene through the real-time monitoring screen provided by the system and manually confirm whether to activate the alarm or take other actions.
[0112] This intelligent fire detection system can effectively reduce false alarms and missed alarms of traditional fire monitoring systems in complex outdoor environments, and improve the accuracy and real-time performance of fire early warning. Especially in the early stages of a fire, the system can quickly detect abnormalities and issue alarms in a timely manner, greatly reducing the losses caused by the fire.
[0113] 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.
[0114] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable way without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.
[0115] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.
Claims
1. A fire monitoring model, characterized in that, The system includes an image data acquisition module, a data preprocessing and annotation module, a feature extraction module, and a fire sign output module, which are connected in sequence. The image data acquisition module is used to input image data into the data preprocessing and annotation module. The feature extraction module extracts local spatial features at different layers and splices multiple local spatial features in the channel dimension to generate the initial position and size of candidate boxes. The fire sign output module processes the candidate box data and outputs fire sign data including candidate box category, position, and confidence level.
2. A method of using the fire monitoring model according to claim 1, characterized in that, include: Step 1: Input real-time images into the image data acquisition module; Step 2: The data preprocessing and labeling module preprocesses the data input in Step 1 and marks the fire location; Step 3: Based on Step 2, the feature extraction module generates the initial position and size of the predicted candidate boxes; Step 4: Based on step 3, the fire sign output module outputs fire sign data including candidate box category, location, and confidence level; Step 5: Train the data preprocessing and labeling module, feature extraction module, and fire sign output module to improve the accuracy of the fire monitoring model in judging fires.
3. The method of using the fire monitoring model according to claim 2, characterized in that, The training methods in step 5 include: Step 0: Input image data into the image data acquisition module, mark the real value of the input fire image, manually mark the marker box and add a category note to the marker box; Step 1: Based on the image data input in Step 0, the data preprocessing and annotation module preprocesses the image data and marks the fire location in the image; Step 2: Generate the initial position and size of the prediction candidate box based on the feature extraction module in Step 1; Step 3: Based on Step 2, the fire sign output module outputs predicted values, which include fire sign data such as candidate box category, location, and confidence level. Step 4: Compare the results of Step 0 and Step 3, backpropagate the difference between the actual value and the predicted value, and train the fire monitoring model.
4. The method of using the fire monitoring model according to claim 3, characterized in that, In step 4, the difference between the predicted value and the true value is quantified using a loss function.
5. The method of using the fire monitoring model according to claim 3, characterized in that, In step 0, you need to input no fewer than 10,000 fire images.
6. The method of using the fire monitoring model according to claim 5, characterized in that, In step 3, the candidate box categories include flame and smoke.
7. The method of using the fire monitoring model according to claim 6, characterized in that, In Step 0, flame and smoke indicator frames are set respectively.
8. The method of using the fire monitoring model according to claim 7, characterized in that, In step 1, a multi-scale feature map is generated based on the image input in step 0. In step 2, the flame candidate box and the smoke candidate box are labeled respectively based on the multi-scale feature map in step 1.
9. The method of using the fire monitoring model according to claim 5, characterized in that, The multiple fire images input in step 0 are randomly divided into three sets: training set, validation set, and test set.
10. A fire detection system, characterized in that, include: The system includes a sensor, a camera, a computing processing unit, a user terminal, and an alarm. The sensor and camera are electrically connected to the computing processing unit, which is electrically connected to the user terminal and the alarm. The computing processing unit is used to send alarm reminders to the user terminal or control the alarm to send alarm signals. The computing processing unit includes a control system and the fire monitoring model as described in claim 1. The fire monitoring model receives data transmitted from the camera, identifies fire signs, and sends them to the control system. The control system sends alarm reminders to users or controls the alarm to issue alarm signals based on the data provided by the sensors and the fire detection model.