Global Situation Awareness Method Based on Data Processing

Abstract

An embodiment of the present application provides a method for global situation awareness based on data processing. The method includes obtaining the actual current detected by a current sensor arranged on an electrical appliance device and the actual temperature detected by a temperature sensor arranged on a heating element. Determine whether there is a fire risk according to the actual current and the actual temperature. If so, obtain the arrangement position of the temperature sensor and the arrangement positions of each camera. According to the arrangement position of the temperature sensor and the arrangement positions of each camera, select a target camera from multiple cameras, control the target camera to capture a monitoring image of the electrical appliance device, obtain the light intensity at the shooting moment of the captured monitoring image, use a fire detection model to identify the monitoring image, and use the light intensity, actual current, and actual temperature at the shooting moment to correct the identification process, output a fire detection result, perform global situation awareness, and use data processing technology to process the sensed data to detect a fire as early as possible.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a global situation awareness method based on data processing. Background Art

[0002] With the increase of urbanization and building complexity, an efficient and reliable fire detection and alarm system has become an essential part of ensuring building safety.

[0003] Generally, smoke alarms or temperature sensors are arranged in the monitored area. When the smoke concentration monitored by the smoke alarm reaches a certain concentration, or when the temperature sensor senses a temperature rise, a fire alarm is issued. However, the smoke alarm usually requires the smoke concentration to reach the set concentration, which will delay the rescue time. The placement position of the temperature sensor is relatively important. If it is not arranged in the area where a fire is likely to occur, the fire cannot be detected in time.

[0004] Therefore, the existing solutions for detecting fires through smoke alarms or temperature sensors cannot detect smoke as early as possible, and the reliability of fire detection is not strong. Summary of the Invention

[0005] The embodiments of this application provide a global situation awareness method based on data processing, which uses current sensors, temperature sensors, and cameras for global situation awareness, and uses data processing technology to process the sensed data to achieve the technical effect of detecting fires as early as possible.

[0006] The embodiments of this application provide a global situation awareness method based on data processing. A plurality of cameras are arranged in the monitored area, current sensors are arranged on electrical equipment in the monitored area, and temperature sensors are arranged on the heating elements of the electrical equipment. The method is applied to a background server, and the method includes:

[0007] Obtain the actual current detected by the current sensor arranged on the electrical equipment and the actual temperature detected by the temperature sensor arranged on the heating element;

[0008] Determine whether there is a fire risk according to the actual current and the actual temperature. If so, obtain the arrangement position of the temperature sensor and the arrangement positions of each camera;

[0009] Select a target camera from multiple cameras according to the arrangement position of the temperature sensor and the arrangement positions of each camera;

[0010] Control the target camera to capture a monitoring image of the electrical equipment, and obtain the light intensity at the shooting moment when the monitoring image is captured;

[0011] The monitoring image is recognized using a fire detection model, and the recognition process is corrected using the light intensity, actual current, and actual temperature at the shooting moment, and the fire detection result is output.

[0012] In the above technical solution, cameras are arranged in the monitoring area, current sensors are arranged on the electrical equipment in the monitoring area, and temperature sensors are arranged on the heating elements of the electrical equipment. The background server obtains the actual current detected by the current sensor and the actual temperature detected by the temperature sensor, determines whether there is a fire risk in the monitoring area based on the actual current and actual temperature. When it is determined that there is a fire risk, the camera is called to capture the monitoring image, and the light intensity at the shooting moment is obtained. The monitoring image is recognized using the fire detection model, and the recognition process is corrected using the light intensity, actual current, and actual temperature at the shooting moment, so as to realize the all-region perception of fire, improve the accuracy of fire detection. And compared with the smoke alarm, the camera can detect the fire in time when the fire just occurs, without waiting until the smoke concentration reaches the set threshold. Compared with the sensing area of the temperature sensor, the shooting area of the camera is relatively large, and fewer cameras can achieve non-blind-spot monitoring of the monitoring area, improving the reliability of fire prevention and control.

[0013] In a possible implementation, the monitoring image is recognized using a fire detection model, and the recognition process is corrected using the light intensity, actual current, and actual temperature at the shooting moment, and the fire detection result is output, which specifically includes:

[0014] Input features are generated according to the monitoring image and the light intensity at the shooting moment, and classification features are generated according to the actual current and actual temperature;

[0015] The input features and the classification features are concatenated to generate concatenated features, the concatenated features are encoded using an encoder to output encoded features, and the classification data in the encoded features is decoded using a decoder to determine the fire detection result.

[0016] In the above technical solution, input features are generated according to the monitoring image and the light intensity at the shooting moment, the light intensity and the monitoring image are fused, classification features are generated according to the actual current and actual temperature, so that the classification features are initialized using the actual current and actual temperature. After the input features and the classification features are concatenated, concatenated features are obtained, so that the concatenated features contain more information, and then fire recognition is performed based on the concatenated features, which can improve the accuracy of the fire detection result.

[0017] In a possible implementation, generating input features according to the monitoring image and the light intensity at the shooting moment specifically includes:

[0018] The monitored image is segmented to obtain multiple image blocks and the positions of each image block, and the monitored sub-region corresponding to each image block is determined according to the shooting parameters of the camera;

[0019] The light intensity of each monitored sub-region at the shooting moment is obtained, and the light intensity of each monitored sub-region at the shooting moment is used as the light intensity of the corresponding image block;

[0020] Each image block, the position of each image block, and the light intensity of each image block are fused to obtain the fusion data of each image block; the fusion data of multiple image blocks are stitched together to generate the input feature.

[0021] In a possible implementation manner, a decoder is used to decode the encoded feature to determine the fire detection result, which specifically includes:

[0022] A first fully connected layer is used to linearly process the encoded feature, an activation function layer is used to non-linearly process the data output by the first fully connected layer, a second fully connected layer is used to linearly process the data output by the activation function layer, and a residual layer is used to superimpose and output the feature output by the second fully connected layer and the feature output by the activation function layer.

[0023] In the above technical solution, the decoder includes a fully connected layer for linear processing and an activation function layer for non-linear processing, which can improve the decoding performance of the decoder and the accuracy of fire monitoring.

[0024] In a possible implementation manner, it is determined whether there is a fire risk according to the actual current and the actual temperature, which specifically includes:

[0025] The rated current of the electrical equipment and the tolerated temperature of the electrical equipment are obtained;

[0026] If the actual current is greater than the rated current and the actual temperature is greater than the tolerated temperature, it is determined that there is a fire risk.

[0027] In the above technical solution, the actual temperature of the heating element is collected by a temperature sensor, the actual current of the electrical equipment is collected by a current sensor, and it is determined whether there is a fire risk based on the actual temperature and the actual current, which can improve the accuracy of the preliminary fire judgment.

[0028] In a possible implementation manner, obtaining the tolerated temperature of the electrical equipment specifically includes:

[0029] The actual currents at multiple moments detected by the current sensor and the actual temperatures at multiple moments detected by the temperature sensor are obtained;

[0030] A tolerated temperature prediction model is used to process the actual currents at multiple moments and the actual temperatures at multiple moments, and the tolerated temperature of the electrical equipment is output.

[0031] In the above technical solution, by obtaining the actual temperatures at multiple moments and the actual currents at multiple moments, and inputting the actual temperatures at multiple moments and the actual currents at multiple moments into a tolerance temperature prediction model for prediction, the tolerance temperature is updated in real time. Compared with using a fixed tolerance temperature, it can adapt to the situation where the performance of electrical components changes as the usage time increases, so that the fire risk can be detected in time.

[0032] In a possible implementation manner, before obtaining the actual current detected by the current sensor on the electrical device and the actual temperature detected by the temperature sensor arranged on the heating element, the method further includes:

[0033] Obtain the circuit topology diagram of the monitored area, and establish a power model in the power simulation software based on the circuit topology diagram;

[0034] Perform simulation analysis on the power model in the power simulation software to obtain the position information of the heating element; arrange a plurality of temperature sensors according to the position information of the heating element.

[0035] In the above technical solution, compared with arranging temperature sensors based on experience, by establishing a power model corresponding to the power topology diagram of the monitored area in the power simulation software, determining the operating temperature of each electrical device through simulation analysis, selecting the heating element based on the operating temperature, and arranging temperature sensors on the heating element, the temperature of the heating element can be monitored more accurately and abnormalities can be detected in time.

[0036] An embodiment of the present application provides a global situation awareness system based on data processing, including:

[0037] An acquisition module, configured to acquire the actual current detected by the current sensor arranged on the electrical device and the actual temperature detected by the temperature sensor arranged on the heating element;

[0038] A processing module, configured to determine whether there is a fire risk according to the actual current and the actual temperature. If so, acquire the arrangement positions of the temperature sensors and the arrangement positions of each camera;

[0039] The processing module is further configured to select a target camera from multiple cameras according to the arrangement positions of the temperature sensors and the arrangement positions of each camera;

[0040] The processing module is further configured to control the target camera to capture a monitoring image of the electrical device, and acquire the light intensity at the shooting moment when the monitoring image is captured;

[0041] The processing module is further configured to identify the monitoring image using a fire detection model, and correct the identification process using the light intensity, the actual current, and the actual temperature at the shooting moment, and output a fire detection result.

[0042] An embodiment of the present application provides a back-end server, including: a memory, a processor;

[0043] The memory stores computer-executable instructions;

[0044] The processor executes the computer-executable instructions stored in the memory, so that the processor executes various possible implementation manners as above.

[0045] An embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement various possible implementation manners as above.

[0046] An embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements various possible implementation manners as above. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0048] Figure 1 It is an application scenario diagram of the global situation awareness method based on data processing provided by the present application;

[0049] Figure 2 It is a schematic flowchart of the global situation awareness method based on data processing provided by the present application;

[0050] Figure 3 It is a schematic structural diagram of an encoder provided by an embodiment of the present application;

[0051] Figure 4 It is a schematic structural diagram of a decoder provided by an embodiment of the present application;

[0052] Figure 5 It is a schematic structural diagram of the back-end server provided by the present application.

[0053] Through the above accompanying drawings, the clear embodiments of the present application have been shown, and there will be more detailed descriptions hereinafter. These drawings and text descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0055] Generally, smoke alarms or temperature sensors are arranged in the monitoring area. When the smoke concentration monitored by the smoke alarm reaches a certain concentration, or when the temperature sensor senses a temperature rise, a fire alarm is issued. However, smoke alarms usually require the smoke concentration to reach the set concentration, which may delay the rescue time. The location of the temperature sensor is relatively important. If it is not arranged in an area prone to fire, the fire cannot be detected in time either.

[0056] Therefore, the existing solutions for detecting fires through smoke alarms or temperature sensors cannot detect smoke as early as possible, and the reliability of fire detection is not strong.

[0057] The method for global situation awareness based on data processing provided by the present application arranges cameras in the monitoring area, arranges current sensors on electrical equipment in the monitoring area, and arranges temperature sensors on the heating elements of electrical equipment. The current sensors, temperature sensors, and cameras are used for global situation awareness, and data processing technology is used to process the sensed data to achieve the technical effect of detecting fires as early as possible. More specifically, the background server obtains the actual current detected by the current sensor and the actual temperature detected by the temperature sensor, determines whether there is a fire risk in the monitoring area based on the actual current and the actual temperature. When it is determined that there is a fire risk, the camera is called to capture a monitoring image, and the light intensity at the capture moment is obtained. The fire detection model is used to identify the monitoring image, and the light intensity at the capture moment, the actual current, and the actual temperature are used to correct the identification process. In this way, global fire situation awareness is realized. Compared with smoke alarms, the camera can detect the fire in time when the fire just breaks out, without waiting for the smoke concentration to reach the set threshold. Compared with the sensing area of the temperature sensor, the capture area of the camera is relatively large, and fewer cameras can achieve a non-blind area monitoring of the monitoring area, improving the reliability of fire prevention and control.

[0058] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. These specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0059] Figure 1The application scenario diagram of the global situation awareness method based on data processing provided by this application is as follows. Figure 1 As shown in the figure, multiple cameras are arranged in the monitoring area. The cameras are used to capture the monitoring images in the monitoring area. There are electrical appliances in the monitoring area. Each electrical appliance is equipped with a current sensor for detecting the actual current of the electrical appliance. A temperature sensor is arranged on the heating element of the electrical appliance, and the temperature sensor detects the actual temperature of the heating element. The background server is communicatively connected to the cameras, current sensors, and temperature sensors in the monitoring area, receives the monitoring images uploaded by the cameras, the actual current uploaded by the current sensors, and the actual temperature uploaded by the temperature sensors, and the background server performs fire recognition based on the monitoring images, actual current, and actual temperature.

[0060] Figure 2 The flowchart of the global situation awareness method based on data processing provided by this application is as follows. Figure 2 As shown in the figure, the method includes the following steps:

[0061] S101. The background server obtains the actual current detected by the current sensor on the electrical appliance and the actual temperature detected by the temperature sensor arranged on the heating element.

[0062] Among them, the background server is communicatively connected to the current sensor. After the current sensor collects the actual current of the electrical appliance, it uploads the actual current of the electrical appliance to the background server, and the background server stores the actual current.

[0063] The background server is communicatively connected to the temperature sensor. After the temperature sensor collects the actual temperature of the heating element, it uploads the actual temperature of the heating element to the background server. The background server stores the actual temperature.

[0064] S102. The background server determines whether there is a fire risk based on the actual current and the actual temperature. If so, it obtains the arrangement positions of the temperature sensors and the arrangement positions of each camera. If not, it returns to S101.

[0065] Among them, the background server obtains the set temperature and set current of the electrical appliance, compares the set temperature of the electrical appliance with the actual temperature, compares the set current of the electrical appliance with the actual current, and determines whether there is a fire risk based on the two comparison results.

[0066] More specifically, if it is determined that the actual current is greater than the set current and the actual temperature is greater than the set temperature, it is determined that there is a fire risk. If it is determined that the actual current is less than or equal to the set current, but the actual temperature is greater than the set temperature, the actual temperature is continuously monitored. If the temperature rising rate is greater than the preset rate, it is determined that there is a fire risk. If it is determined that the actual current is less than or equal to the set current and the actual temperature is less than or equal to the set temperature, it is determined that there is no fire risk. If it is determined that the actual current is greater than the set current, but the actual temperature is less than or equal to the set temperature, it is determined that there is no fire risk.

[0067] S103. The background server selects a target camera from multiple cameras according to the arrangement positions of the temperature sensors and the arrangement positions of each camera.

[0068] Among them, the sensors that collect the actual temperature greater than the set temperature are obtained and marked as target sensors. The arrangement positions of the target sensors are obtained, and the cameras located near the target sensors are selected from multiple cameras according to the positions of the target sensors and the arrangement positions of each camera as the target cameras.

[0069] For example: There are 4 electrical devices in the monitoring area, and 1 temperature sensor is arranged on each electrical device. The actual temperatures detected by 3 temperature sensors are less than the set temperature, and only 1 temperature sensor detects the actual temperature greater than the set temperature. This temperature sensor is used as the target sensor. The arrangement position of the target sensor is obtained, and the camera located near the target sensor is selected as the target camera to monitor whether a fire occurs near the target sensor.

[0070] S104. The background server controls the target camera to capture the monitoring image of the electrical device and obtains the light intensity at the shooting moment of the captured monitoring image.

[0071] Among them, the background server sends a shooting instruction to the target camera to control the target camera to capture the monitoring image of the electrical device, and the target camera uploads the captured monitoring image to the background server.

[0072] In some examples, a light intensity sensor is arranged in the monitoring area, and the light intensity at the shooting moment is detected by the light intensity sensor. In other examples, the background server sends a meteorological data request to the meteorological system to make the meteorological system return the light intensity at the shooting moment.

[0073] S105. The background server uses the fire detection model to identify the monitoring image, and uses the light intensity, actual current and actual temperature at the shooting moment to correct the identification process, and outputs the fire detection result.

[0074] Among them, the fire detection model obtains feature data by extracting features from the monitoring image, and decodes the feature data to output the fire detection result. When extracting features from the monitoring image, the light intensity, actual current, and actual temperature at the shooting moment are used to correct the feature extraction process. In this way, the feature data retains the light intensity, actual current, actual temperature, and the information in the monitoring image. When using the decoder to decode the feature data, the classification accuracy is improved, and thus the fire detection accuracy is enhanced.

[0075] In the above technical solution, a camera is arranged in the monitoring area, a current sensor is arranged on the electrical equipment in the monitoring area, and a temperature sensor is arranged on the heating element of the electrical equipment. The background server obtains the actual current detected by the current sensor and the actual temperature detected by the temperature sensor, determines whether there is a fire risk in the monitoring area based on the actual current and the actual temperature. When it is determined that there is a fire risk, the camera is called to capture the monitoring image, and the light intensity at the shooting moment is obtained. The fire detection model is used to identify the monitoring image, and the light intensity, actual current, and actual temperature at the shooting moment are used to correct the identification process. In this way, the full-domain perception of fire is realized, the fire detection accuracy is improved, and compared with the smoke alarm, the camera can detect the fire in time when the fire just breaks out, without waiting until the smoke concentration reaches the set threshold. Compared with the sensing area of the temperature sensor, the shooting area of the camera is relatively large, and fewer cameras can monitor the area without dead angles, improving the reliability of fire prevention and control.

[0076] Optionally, the above solution obtains the actual current detected by the current sensor arranged on the electrical equipment and the actual temperature detected by the temperature sensor arranged on the heating element. By fusing these two types of data, a comprehensive monitoring of the working state of the electrical equipment can be achieved. When the actual current is greater than the rated current and the actual temperature is greater than the tolerance temperature, the system can quickly determine that there is a fire risk, and this judgment basis is more accurate and reliable than a single sensor.

[0077] Compared with traditional smoke alarms, the above solution can issue an early warning at the initial stage of a fire, that is, when the current and temperature are abnormal, without waiting for the smoke concentration to reach the set threshold, thus greatly shortening the fire response time. After determining the existence of a fire risk, the system will intelligently select the camera closest to the risk area as the target camera according to the layout positions of the temperature sensors and each camera. Compared with the limited sensing area of the temperature sensors, the shooting area of the camera is larger, and the monitoring area can be covered without dead angles by reasonably arranging a small number of cameras. This not only reduces the monitoring cost but also improves the reliability of fire prevention and control. While controlling the target camera to capture monitoring images, the system will also obtain the light intensity information at the shooting moment. This information is then used to correct the recognition process of the fire detection model to eliminate the influence of light changes on the image recognition results.

[0078] By combining the light intensity, actual current, and actual temperature at the shooting moment to correct the recognition process of the fire detection model, the system can more accurately identify fire characteristics and reduce the occurrence of false alarms and missed alarms. The background server uses data processing technology to perform real-time analysis on the acquired multi-source data, including current, temperature, images, and light intensity, etc., to achieve intelligent assessment of fire risks.

[0079] Based on the data analysis results, the system can output accurate fire detection results and provide intelligent decision-making support for subsequent emergency responses. For example, the system can automatically trigger corresponding warning mechanisms or emergency response processes according to the fire risk level.

[0080] The above solution is applicable to various monitoring scenarios, whether it is a home, commercial building, or industrial facility. Just arranging current sensors, temperature sensors, and cameras at the corresponding positions can achieve overall situation awareness.

[0081] As the monitoring area expands or the monitoring requirements increase, the system can easily expand the monitoring scope and functions by increasing the number of sensors and cameras. At the same time, the data processing ability of the background server can also be further improved by upgrading the hardware or optimizing the algorithm.

[0082] In summary, the overall situation awareness method based on data processing provided by the embodiments of this application realizes the early discovery, accurate positioning, and efficient monitoring of fire risks in the monitoring area through technical means such as integrating multi-source sensor data, intelligently selecting target cameras, light intensity correction, and big data processing and analysis. This method not only improves the accuracy and reliability of fire detection but also provides strong intelligent decision-making support for subsequent emergency responses.

[0083] In a possible implementation, in S105, the background server uses a fire detection model to identify the monitoring image, and uses the light intensity, actual current, and actual temperature at the shooting moment to correct the identification process, and outputs a fire detection result, specifically including:

[0084] S201. The background server generates input features based on the monitoring image and the light intensity at the shooting moment, and generates classification features based on the actual current and the actual temperature.

[0085] Among them, the background server processes the monitoring image using an image conversion function to generate image features. Processes the light intensity using a light intensity conversion function to generate light features. Calculates the sum of the image features and the light features to obtain input features. Performs feature conversion on the actual current and the actual temperature to obtain classification features.

[0086] The input feature is a matrix of m×n, and the classification feature is located in a matrix of 1×n, where m and n are positive integers.

[0087] S202. The background server splices the input feature and the classification feature to generate a spliced feature, uses an encoder to encode the spliced feature to output an encoded feature, and uses a decoder to decode the classification data in the encoded feature to determine the fire detection result.

[0088] Among them, the background server splices the input feature and the classification feature along the row direction to generate a spliced feature, and the dimension of the spliced feature is (m + 1)×n.

[0089] The encoder includes multiple levels of encoding modules. The encoding module at the first level encodes and outputs the spliced feature, the encoding module at the next level encodes and outputs the data output by the encoding module at the previous level, and the encoding module at the last level encodes and outputs the data output by the encoding module at the penultimate level to output the encoded feature.

[0090] The dimension of the encoded feature is the same as that of the spliced feature, which is also (m + 1)×n. Extracts the last row of data in the encoded feature, and uses the encoder to decode and process the last row of data to obtain the fire detection result.

[0091] The decoder includes a linear processing module and a non - linear processing module. The linear processing module performs linear processing on the data, and the non - linear processing module performs non - linear processing on the data. Combining the linear processing module and the non - linear processing module to process the classification data in the encoded feature can improve the decoding accuracy.

[0092] In the above technical solution, input features are generated based on the monitoring image and the light intensity at the shooting moment. The light intensity and the monitoring image are fused, and classification features are generated based on the actual current and the actual temperature. In this way, the classification features are initialized with the actual current and the actual temperature. After splicing the input features and the classification features, a spliced feature is obtained, so that the spliced feature contains more information. Then, fire recognition is performed based on the spliced feature, which can improve the accuracy of the fire detection result.

[0093] Optionally, input features are first generated based on the monitoring image and the light intensity at the shooting moment. By directly integrating the light intensity information into the monitoring image, the system can capture the fire features in the image more accurately. For example, in the case of low light intensity, the system can automatically adjust the image brightness or contrast to ensure that the fire features are clearly distinguishable. This fusion method effectively eliminates the influence of light changes on the image recognition result and improves the adaptability and accuracy of the fire detection model under different light conditions.

[0094] In the process of generating the input features, the monitoring image is processed in blocks and combined with the position information of each image block. This processing method not only reduces the amount of data processing and improves the algorithm calculation efficiency, but also enables the system to analyze the fire features in the image more precisely. For example, by identifying the flame shape, color and other features in the image block and combining its position information, the system can more accurately judge the occurrence location and spread trend of the fire.

[0095] At the same time, the system generates classification features based on the actual current and the actual temperature. These physical quantities, as important indicators of the occurrence of a fire, provide additional judgment bases for the fire detection model. For example, when the actual current rises abnormally and the actual temperature exceeds the tolerance temperature of the electrical equipment, the system can quickly judge that there is a fire risk and integrate this information into the classification features.

[0096] By initializing the classification features with the actual current and the actual temperature, the system can capture the fire features earlier in the fire detection process. This initialization method makes the classification features have a higher discrimination degree in the initial stage of the fire, which helps the system to more accurately identify the fire event.

[0097] After splicing the input features and the classification features, the system obtains a spliced feature containing more information. These features not only contain the fire features in the monitoring image, but also integrate physical quantity information such as light intensity, actual current and actual temperature. The fusion of this multi-source information enables the system to understand the monitoring scene more comprehensively and improve the accuracy of fire detection.

[0098] Compared with single features, the spliced features have higher robustness and accuracy in the fire detection process. For example, in the case of blurred images or insufficient lighting, the system can rely on the current and temperature information in the classification features to assist in judging fire incidents; conversely, in the case of insignificant changes in current and temperature, the system can rely on the image information in the input features to identify fire features.

[0099] After encoding the spliced features using an encoder, the system obtains encoded features containing key information. The encoder, through a multi-level encoding module structure, fully captures the key information in the spliced features and effectively suppresses noise interference. This information extraction method enables the system to more accurately identify fire incidents.

[0100] The decoder then decodes the classification data in the encoded features to determine the fire detection result. The decoder, through the combination of a linear processing layer and a non-linear processing layer, achieves precise decoding of the encoded features. This decoding method enables the system to more accurately output the fire detection result and provides strong support for subsequent emergency responses.

[0101] In summary, through the above-mentioned correction mechanism of the fire detection model and its recognition process, the data processing-based global situation awareness method provided by the embodiments of this application significantly improves the accuracy of fire detection results. This method realizes the precise identification and efficient response to fire risks in the monitored area through technical means such as fusing multi-source information, using light intensity correction, and adopting an encoding-decoding structure.

[0102] In one possible implementation, in S201, the background server generates input features based on the monitored image and the light intensity at the shooting moment, specifically including:

[0103] In S301, the background server divides the monitored image into multiple image blocks and the positions of each image block, and determines the monitored sub-region corresponding to each image block according to the shooting parameters of the camera.

[0104] Among them, as an example, the monitored image is divided into 2×3 image blocks, with 3 image blocks in each row and 2 image blocks in each column. The positions of the 3 image blocks in the first row are position 1, position 2, and position 3, and the positions of the 3 image blocks in the second row are position 4, position 5, and position 6.

[0105] The shooting parameters of the camera include internal parameters and external parameters. Based on the internal and external parameters of the camera, the shooting area is determined, and the shooting area is divided into blocks to obtain the sub-region corresponding to each image block. For example: if the monitored image is divided into 2×3 regions, then the shooting area can be divided into 2×3 shooting sub-regions, and the corresponding relationship between each image block and the shooting sub-region is established.

[0106] S302. The background server obtains the light intensity of each monitored sub - area at the shooting moment, and takes the light intensity of each monitored sub - area at the shooting moment as the light intensity of the corresponding image block.

[0107] Among them, as an example, the background server sends a meteorological data request to the meteorological server to obtain the light intensity and light angle of the monitored area at the shooting moment. Obtain the glass position of the monitored area, and determine the illuminated area and the non - illuminated area according to the glass position and angle of the monitored area. Set the light intensity of the illuminated area to the light intensity returned by the meteorological server, and set the light intensity of the non - illuminated area to the light intensity of the lighting lamp.

[0108] If the monitored sub - area is located in the illuminated area, the light intensity of the monitored sub - area is the light intensity returned by the meteorological server. If the monitored sub - area is located in the non - illuminated area, the light intensity of the monitored sub - area is the light intensity of the lighting lamp. If a part of the monitored sub - area is located in the illuminated area and another part is located in the non - illuminated area, take the average value of the light intensity and the light intensity of the lighting lamp as the light intensity of the monitored sub - area.

[0109] As another example, light intensity sensors are laid in the monitored area, and the light intensity sensors detect the light intensity of the area where they are located. If the monitored sub - area is within the detection range of 1 light intensity sensor, the light intensity of the monitored sub - area is the light intensity sensed by this light intensity sensor. If the monitored sub - area is within the detection range of 2 light intensity sensors, the light intensity of the monitored sub - area is the average value of the light intensities sensed by the 2 light intensity sensors. If the monitored sub - area is within the detection range of 3 light intensity sensors, the light intensity of the monitored sub - area is the average value of the light intensities sensed by the 3 light intensity sensors.

[0110] S303. The background server fuses each image block, the position of each image block, and the light intensity of each image block to obtain the fusion data of each image block; splices the fusion data of multiple image blocks to generate an input feature.

[0111] Among them, the background server converts each image block into an image vector, converts the position of each image block into a position vector, converts the light intensity of each image block into a light intensity vector, calculates the sum of the image vector, the position vector, and the light intensity vector to obtain the fusion vector of each image block. Splice the fusion vectors of each image block to generate an input feature. For example: the dimension of the fusion vector of each image block is n×1, each monitored image is divided into m image blocks, and after splicing and transposing the fusion vectors of the m image blocks along the row, an input feature with a size of m×n is generated.

[0112] In the above technical solution, the monitored image is divided into multiple image blocks, and the position and light intensity of each image block are obtained. Taking each image block as a unit, the image block, the position of the image block, and the light intensity of the image block are fused to obtain the fused data of the image block. Then, the fused data of each image block is stitched together to generate the input feature. Compared with the method of processing the entire image as a unit, the amount of data processing can be reduced, and the calculation efficiency of the algorithm can be improved.

[0113] Optionally, first perform block processing on the monitored image to obtain multiple image blocks and the specific position information of each image block. This step not only converts complex image data into smaller and more easily processed data blocks but also retains the spatial position relationship of the image blocks in the original image.

[0114] Next, according to the shooting parameters of the camera (such as focal length, viewing angle, etc.), determine the monitored sub-region corresponding to each image block. This mapping relationship provides an accurate spatial reference for subsequent light intensity acquisition and feature fusion.

[0115] By performing block processing on the monitored image, the originally large amount of image data can be decomposed into multiple small units for processing. This method significantly reduces the amount of data processed at one time, reduces the computational complexity, and thus improves the calculation efficiency of the algorithm.

[0116] For the monitored sub-region corresponding to each image block, obtain the light intensity information at the shooting moment. This step ensures the accurate correspondence between the light intensity and the image content and provides a reliable data basis for subsequent light correction.

[0117] After obtaining each image block, its position, and the corresponding light intensity, these information are fused to generate the fused data of each image block. This fusion process not only retains the visual features in the image block but also incorporates spatial position information and light conditions, providing more comprehensive information support for subsequent fire detection.

[0118] After stitching the fused data of multiple image blocks together, the final input feature is generated. This input feature not only contains rich visual information in the monitored image but also incorporates multi-dimensional information such as light intensity and image block position, providing more comprehensive input data for the fire detection model.

[0119] Compared with the method of processing the entire image as a unit, through steps such as image block division, light intensity acquisition, and fused data stitching, the amount of data processed at one time can be significantly reduced. This block processing method not only reduces the computational complexity but also improves the calculation efficiency of the algorithm, enabling the system to respond more quickly to the fire detection task.

[0120] In summary, through technical means such as monitoring image block processing, light intensity acquisition, and image block fusion, the global situation awareness method based on data processing provided by the embodiments of this application significantly optimizes the generation process of input features. This method not only reduces the amount of data processing and improves the computing efficiency, but also provides more comprehensive and accurate input data for the fire detection model, thereby enhancing the accuracy and reliability of the fire detection algorithm.

[0121] In a possible implementation manner, in S303, the background server fuses each image block, the position of each image block, and the light intensity of each image block to obtain the fusion data of each image block, which specifically includes:

[0122] S401, the background server uses multiple convolutional kernels to perform multiple convolutional processes on the image block to obtain the image vector of the image block.

[0123] Among them, when using multiple convolutional kernels to perform multiple convolutional processes on the image block, each convolutional process obtains one element of the image vector of the image block. After multiple convolutional processes, multiple elements of the image vector of the image block can be obtained.

[0124] S402, the background server uses a position conversion function to process the position of the image block to obtain the position vector of the image block.

[0125] Among them, the position conversion function includes multiple position conversion sub-functions, and each position conversion sub-function converts the position of the image block into one element of the position vector of the image block. Using multiple position conversion sub-functions to perform multiple processes on the position of the image block, multiple elements of the position vector of the image block are obtained.

[0126] The dimension of the image vector of the image block is the same as the dimension of the position vector of the image block. That is, the number of convolutional kernels is the same as the number of position conversion sub-functions, so that the number of convolutional calculations of the image block is the same as the number of position conversion processes of the image block. For example: if the dimension of the image vector of the image block is n×1, then the dimension of the position vector of the image block is also n×1.

[0127] S403, the background server uses a light conversion function to process the light intensity of the image block to obtain the light vector of the image block.

[0128] Among them, the light conversion function includes multiple light conversion sub-functions, and each light conversion sub-function converts the light intensity to obtain one element of the light vector of the image block. Using multiple light conversion sub-functions to perform multiple conversions on the light intensity, multiple elements of the light vector of the image block are obtained.

[0129] The dimension of the image vector of the image block is the same as that of the illumination vector of the image block, and the number of convolutional kernels is the same as the number of illumination conversion sub-functions. In this way, the number of times of convolution calculation for the image block is the same as the number of times of illumination intensity conversion. For example: if the dimension of the image vector of the image block is n×1, and the dimension of the illumination vector of the image block is n×1.

[0130] S404. For each image block, the background server calculates the sum of the image vector of the image block, the position vector of the image block, and the illumination vector of the image block to obtain the fusion vector of each image block.

[0131] Among them, for each image block, the background server calculates the sum of the i-th element of the image vector of the image block, the i-th element of the position vector of the image block, and the i-th element of the illumination vector of the image block as the i-th element of the fusion vector of the image block. By traversing i from 1 to n, each element in the fusion vector of the image block is obtained.

[0132] In the above technical solution, the image block, the position of the image block, and the illumination intensity of the image block are converted into vectors of the same dimension, and the sum of the image vector, the position vector, and the illumination vector is calculated, so as to realize the fusion of position information and illumination information into the image, make the fusion vector contain more information, and improve the accuracy of fire recognition.

[0133] Optionally, multiple convolutional kernels are used to perform multiple convolutional processes on the image block to obtain the image vector of the image block. As a feature extractor, the convolutional kernel can automatically learn the key features in the image, such as the shape, color, texture of the flame, etc. Through multiple convolutional processes, the fire features in the image block are effectively extracted and encoded into the image vector, providing a visual information basis for subsequent feature fusion.

[0134] The position conversion function is used to process the position of the image block to obtain the position vector of the image block. The position vector encodes the spatial position information of the image block in the original image, which is crucial for understanding the occurrence position and spread trend of the fire. By vectorizing the position information, the diffusion path and potential risk area of the fire can be judged more accurately.

[0135] The illumination conversion function is used to process the illumination intensity of the image block to obtain the illumination vector of the image block. As an important factor affecting the visual features of the image, the change of illumination intensity will directly affect the recognition effect of fire features. By vectorizing the illumination intensity, the recognition threshold for fire features in the image can be automatically adjusted, and the recognition accuracy under different illumination conditions can be improved.

[0136] For each image patch, calculate the sum of its image vector, position vector, and illumination vector to obtain the fusion vector for each image patch. This fusion process tightly combines visual features, spatial location information, and illumination conditions, forming a multi-dimensional feature representation rich in information. The fusion vector not only retains the original visual features in the image patch but also incorporates the influence of position information and illumination conditions on fire characteristics, providing a more comprehensive and accurate information basis for subsequent fire recognition.

[0137] By fusing multi-source information into a fusion vector, it is possible to more comprehensively understand the fire characteristics in the monitoring scene. For example, in the case of low light intensity, the fire can be identified by relying on the flame shape and color features in the image vector; at the same time, the incorporation of the position vector and illumination vector helps the system more accurately determine the location of the fire and its spread trend. This fusion of multi-source information significantly enhances the accuracy of fire recognition and improves the adaptability of the system in different environments.

[0138] The image vector, position vector, and illumination vector contained in the fusion vector together constitute a multi-dimensional feature representation rich in information. This feature representation not only improves the accuracy of fire recognition but also enhances the system's adaptability to complex monitoring scenes. For example, in the case of multiple potential fire sources in the monitoring area, the system can accurately distinguish different fire characteristics, avoiding false alarms and missed detections.

[0139] By processing the surveillance images in patches and generating fusion vectors, it is possible to improve the accuracy of fire recognition while maintaining computational efficiency. Compared with the method of processing the entire image as a unit, the patch processing method significantly reduces the amount of data processed in a single operation, improving the computational efficiency of the algorithm; at the same time, by fusing multi-source information to generate fusion vectors, the system can more accurately identify fire characteristics, improving the accuracy of recognition.

[0140] In summary, through the image patch fusion technology, the all-domain situation awareness method based on data processing provided by the embodiments of this application significantly improves the accuracy of fire recognition. Through means such as vector representation of multi-source information, multi-source information fusion, and feature enhancement, a comprehensive and accurate recognition of fire characteristics in the monitoring scene is achieved, providing strong support for subsequent emergency responses.

[0141] In one possible implementation, in S202, the background server generates classification features based on the actual current and actual temperature, specifically including:

[0142] In S501, the background server processes the actual current using a current conversion function to obtain a current vector, and processes the actual temperature using a temperature conversion function to obtain a temperature vector.

[0143] Among them, the current conversion function includes multiple current conversion sub-functions. One current conversion sub-function processes the actual current to obtain an element of the current vector, and multiple current conversion sub-functions process the actual current multiple times to obtain multiple elements of the current vector.

[0144] The dimension of the current vector is the same as that of the image vector of the image block. The number of convolution kernels is the same as the number of current conversion sub-functions. In this way, the number of times the image block performs convolution calculation is the same as the number of times the actual current is converted. For example: if the dimension of the image vector of the image block is n×1, then the dimension of the current vector is also n×1.

[0145] The temperature conversion function includes multiple temperature conversion sub-functions. One temperature conversion sub-function processes the actual temperature to obtain an element of the temperature vector, and multiple temperature conversion sub-functions process the actual temperature multiple times to obtain multiple elements of the temperature vector.

[0146] The dimension of the temperature vector is the same as that of the image vector of the image block. The number of convolution kernels is the same as the number of temperature conversion sub-functions. In this way, the number of times the image block performs convolution calculation is the same as the number of times the actual temperature is converted. For example: if the dimension of the image vector of the image block is n×1, then the dimension of the temperature vector is also n×1.

[0147] S502. The background server obtains the classification position according to the sorting results of multiple image blocks and classification features, and uses the position conversion function to process the classification position to obtain the classification position vector.

[0148] Among them, the classification features and multiple image blocks are numbered by position. To ensure the continuity of the image blocks, the classification features are arranged at the head, and multiple image blocks are sorted by position according to their positions in the image. For example: the monitoring image is divided into 2×3 image blocks, the classification position is set to position 0, the position of the image block in the first row and first column is set to position 1, the position of the image block in the first row and second column is set to position 2, the position of the image block in the first row and third column is set to position 3, the position of the image block in the second row and first column is set to position 4, the position of the image block in the second row and second column is set to position 5, and the position of the image block in the second row and third column is set to position 6.

[0149] One position conversion sub-function converts the classification position to obtain an element of the classification position vector, and multiple position conversion sub-functions convert the classification position multiple times to obtain multiple elements of the classification position vector.

[0150] S503. The background server superimposes the current vector, the temperature vector, and the classification position vector to generate the classification features.

[0151] Among them, the background server calculates the sum of the i-th element of the current vector, the i-th element of the temperature vector, and the i-th element of the classification position vector as the i-th element of the classification feature. By traversing i from 1 to n, each element in the classification feature is obtained.

[0152] In the above technical solution, the actual current and actual temperature are converted into vectors, the position of the classification feature is determined according to the classification feature and the sorting results of multiple image blocks, and the classification feature is initialized based on the temperature vector, current vector, and classification position feature, so that the initial value of the classification feature contains more information. In this way, when monitoring for fires based on the classification feature, the monitoring accuracy is improved.

[0153] Optionally, a current conversion function is used to process the actual current to obtain a current vector. This step converts continuously changing current values into a vector representation with clear physical meaning, facilitating subsequent feature fusion and calculation. A temperature conversion function is used to process the actual temperature to obtain a temperature vector. Similarly, the temperature vector maps temperature values to a high-dimensional space, retaining the trend and characteristics of temperature changes and providing important thermal information for fire monitoring.

[0154] By converting the current and temperature into vector forms, these multi-source information can be processed more flexibly. The vector representation not only retains the physical meaning of the original data but also facilitates mathematical operations and feature fusion, laying a foundation for the subsequent generation of classification features.

[0155] The classification position is obtained according to the classification feature and the sorting results of multiple image blocks. This step combines the classification feature with the spatial position information of the image blocks, and determines the specific position of the classification feature in the monitoring scene through the sorting results.

[0156] A position conversion function is used to process the classification position to obtain a classification position vector. The classification position vector encodes the spatial position information of the classification feature in the monitoring scene and provides an important spatial reference for subsequent fire monitoring.

[0157] The introduction of the classification position vector enables the classification feature to contain not only physical quantity information such as current and temperature but also spatial position information. This multi-dimensional feature representation helps the system to more accurately understand the occurrence location and spread trend of the fire, improving the accuracy and timeliness of fire monitoring.

[0158] The current vector, temperature vector, and classification position vector are superimposed to generate a classification feature. This step integrates multi-source information into a classification feature vector containing rich information, providing a comprehensive feature representation for subsequent fire monitoring.

[0159] The classification features initialized based on the temperature vector, current vector, and classification position features already contain information from multiple aspects such as current, temperature, and spatial position in their initial values. This initialization method enables the classification features to more accurately reflect the occurrence and development trends of fires during subsequent monitoring processes.

[0160] During the fire monitoring process, the system can determine the occurrence of a fire based on the change trends and feature patterns of the classification features. Since the classification features contain richer information, the system can more accurately identify fire features, reduce false alarms and missed alarms, and improve the accuracy of fire monitoring.

[0161] In summary, through the classification feature generation technology, the fire monitoring system provided by the embodiments of this application significantly improves the monitoring accuracy. This technology realizes the comprehensive and accurate identification of fire features through means such as multi-source information vector integration, introduction of classification position vectors, and generation of classification features, providing strong support for subsequent emergency responses.

[0162] In a possible implementation manner, in S202, the background server encodes the spliced features to output encoded features, which specifically includes:

[0163] S601: Use the encoding module at the first level to encode the spliced features, use the encoding module at the second level to perform encoding processing on the encoded features output by the encoding module at the first level, use the encoding module at the third level to process the encoded features output by the encoding module at the second level, and so on until the encoding module at the last level processes the encoded features output by the encoding module at the penultimate level.

[0164] Among them, each encoding module includes a multi-head attention mechanism module and a feed-forward neural network module. The multi-head attention mechanism module includes a multi-head attention mechanism layer and a residual normalization layer, and the feed-forward neural network module includes a feed-forward neural network layer and a residual normalization layer.

[0165] As Figure 3 shown, the encoder includes multiple cascaded encoding modules. The output end of the encoding module at the first level is connected to the input end of the encoding module at the second level, the output end of the encoding module at the second level is connected to the input end of the encoding module at the third level, and so on until the output end of the encoding module at the penultimate level is connected to the input end of the encoding module at the last level.

[0166] The encoding module at the first level processes the splicing features. More specifically, the splicing features first pass through the multi-head attention mechanism layer and then output the first intermediate data. The first intermediate data enters the residual normalization layer for processing and outputs the second intermediate data. The second intermediate data enters the feed-forward neural network layer for processing and outputs the third intermediate data. The third intermediate data passes through the residual normalization layer for processing and outputs the fourth intermediate data. The fourth intermediate data is used as the data output by the first encoding module.

[0167] The residual normalization layer includes a residual layer and a normalization layer, which first perform normalization processing on the input data and then perform residual calculation.

[0168] The encoding module at the second level encodes the encoding features output by the encoding module at the first level. The encoding process of the encoding module at the second level is the same as that of the encoding module at the first level, which will not be elaborated here. After multiple encoding modules perform multiple encoding processes, encoding features are output.

[0169] In the above technical solution, the encoder includes multiple cascaded encoding modules, and each encoding module includes a multi-head attention mechanism layer, a feed-forward neural network layer, and a residual normalization layer. Since the encoding module includes a multi-head attention mechanism layer, it can not only focus on the fusion vector itself, but also focus on the mutual influence between adjacent fusion vectors, so that the encoding features contain more information and improve the accuracy of fire detection.

[0170] Optionally, the encoder in the above technical solution is composed of multiple cascaded encoding modules, and each encoding module processes the input features in turn. This cascaded structure enables the feature information to be continuously refined and enhanced during the layer-by-layer transmission process, which helps to extract deeper feature representations.

[0171] Through the cascaded processing of multiple-level encoding modules, the complex patterns and correlation relationships in the splicing features can be gradually captured. Each level of encoding module further optimizes the feature representation on the basis of the previous level, thereby improving the quality and effectiveness of the features.

[0172] Each encoding module contains a multi-head attention mechanism module inside. By introducing multiple parallel attention heads, this module can simultaneously focus on the fusion vector itself and the mutual influence between adjacent fusion vectors. This mechanism enables a more comprehensive understanding of the correlation relationships between features and captures more subtle feature changes. The multi-head attention mechanism layer is followed by a residual normalization layer, which effectively alleviates the problem of gradient disappearance through residual connection and normalization processing, improving the stability and training efficiency of the model.

[0173] The feed-forward neural network module further enhances the expressive power of features through non-linear transformation. The feed-forward neural network layer further processes the output of the multi-head attention mechanism layer to extract higher-level feature representations. Similarly, the residual normalization layer is also included after the feed-forward neural network layer to ensure the stability and effectiveness of feature information during transmission.

[0174] The multi-head attention mechanism module and the feed-forward neural network module cooperate with each other to jointly complete the deep encoding process of the concatenated features. The multi-head attention mechanism module is responsible for capturing the correlation relationships between features, while the feed-forward neural network module is responsible for further refining and enhancing the feature representations. This synergy enables the system to extract richer and more effective feature information.

[0175] In addition, since the encoding module includes the multi-head attention mechanism layer, the system can simultaneously focus on the fusion vector itself and the mutual influence between adjacent fusion vectors. This mechanism enables the encoded features to contain more information about the correlation relationships between features, improving the richness and effectiveness of the features.

[0176] Based on the encoded features containing rich information, the system can more accurately identify fire features. During the fire monitoring process, the system can determine whether a fire has occurred based on the change trend and feature pattern of the encoded features. Since the encoded features contain more information about the correlation relationships between features, the system can more sensitively capture the subtle feature changes of the fire, thereby improving the accuracy and timeliness of fire detection.

[0177] In summary, the fire detection system proposed in the embodiment of the present application significantly improves the accuracy of fire detection by using a multi-level encoding module to perform deep encoding processing on the concatenated features. The encoder structure in this technical solution is unique, integrating the multi-head attention mechanism and the feed-forward neural network, and combining the residual normalization layer to achieve fine encoding and efficient processing of the concatenated features. This technical solution not only improves the quality and effectiveness of the features, but also provides more accurate and timely decision support for fire detection.

[0178] In a possible implementation, S202, using a decoder to decode the encoded features to determine the fire detection result, specifically includes:

[0179] S602, linearly processing the encoded features using a first fully connected layer, non-linearly processing the output features of the first fully connected layer using an activation function layer, linearly processing the output features of the activation function layer using a second fully connected layer, and outputting the superposition of the output features of the second fully connected layer and the output features of the activation function layer using a residual layer.

[0180] Among them, as Figure 4As shown in the figure, the decoder includes a first fully-connected layer, an activation function layer, a second fully-connected layer, and a residual layer. The output end of the first fully-connected layer is connected to the input end of the activation function layer, the output end of the activation function layer is connected to the input end of the second fully-connected layer, and the input end of the residual layer is connected to the output end of the second fully-connected layer.

[0181] The first fully-connected layer is used to linearly process the encoded features to output the fifth intermediate data, the activation function layer is used to non-linearly process the fifth intermediate data output by the first fully-connected layer to output the sixth intermediate data, the second fully-connected layer is used to linearly process the sixth intermediate data output by the activation function layer to output the seventh intermediate data, and the residual layer is used to superimpose the seventh intermediate data output by the second fully-connected layer and the sixth intermediate data output by the activation function layer and then output. A normalization exponentiation layer (softmax) can also be added after the residual layer to output the probabilities of each category.

[0182] In the above technical solution, the decoder includes a fully-connected layer for linear processing and an activation function layer for non-linear processing, which can improve the decoding performance of the decoder and the accuracy of fire monitoring.

[0183] Optionally, the decoder is mainly composed of a first fully-connected layer, an activation function layer, a second fully-connected layer, and a residual layer. This structure combines linear processing and non-linear processing, aiming to fully extract the effective information in the encoded features and make accurate fire judgments based on this.

[0184] Among them, for the first fully-connected layer, it linearly processes the encoded features and maps them to a higher-dimensional space so that subsequent non-linear processing can better capture the complex relationships between features. Through linear transformation, the first fully-connected layer effectively retains the key information in the encoded features and provides a good input basis for subsequent non-linear processing.

[0185] For the activation function layer, it non-linearly processes the output data of the first fully-connected layer, introducing non-linear factors, enabling the decoder to learn more complex feature patterns. The activation function layer enhances the expressive ability of the decoder by introducing non-linear factors, enabling it to handle more complex fire monitoring tasks. At the same time, non-linear processing also helps the decoder capture the subtle changes in the encoded features and improve the sensitivity of fire monitoring.

[0186] For the second fully-connected layer, it linearly processes the output data of the activation function layer and maps the non-linearly processed features back to the original space or a more appropriate decision space. The second fully-connected layer integrates the non-linearly processed features into the final decision basis through linear transformation. This step ensures that the decision result output by the decoder has both the flexibility brought by non-linear processing ability and the stability and interpretability brought by linear transformation.

[0187] In addition, for the residual layer, the output features of the second fully connected layer and the output features of the activation function layer are superimposed to form the final decoded output. By introducing residual connections, the residual layer effectively alleviates the common gradient vanishing problem in deep networks, improves the training stability and convergence speed of the model. At the same time, the residual connection also enables the decoder to better retain the original feature information and improves the accuracy of fire monitoring.

[0188] Through the collaborative action of the above components, the decoder realizes the efficient decoding of the encoded features. Specifically, the decoder effectively extracts the valid information in the encoded features by integrating linear processing and non-linear processing, and makes an accurate fire judgment based on this. This processing method not only improves the expression ability of the decoder, but also enhances its generalization ability and robustness. Since the decoder can more accurately capture the key information in the encoded features and make decisions based on this, the accuracy of fire monitoring is improved. This is of great significance for fire warning and emergency response in practical application scenarios.

[0189] In summary, the decoder structure in the fire monitoring system proposed in the embodiments of the present application effectively improves the decoding performance of the decoder and the accuracy of fire monitoring through carefully designed components and collaborative action mechanisms. This technical solution provides a new and efficient solution for the field of fire monitoring.

[0190] Another embodiment of the present application provides a global situation awareness method based on data processing, which includes the following steps:

[0191] S701. The background server obtains the circuit topology diagram of the monitoring area and establishes a power model in the power simulation software based on the circuit topology diagram.

[0192] Among them, the background server obtains the circuit topology diagram of the monitoring area based on the monitoring area design data, and establishes a power model corresponding to the circuit topology diagram of the monitoring area in the power simulation software, so that the working temperature of each electrical device in the monitoring area can be simulated using the simulation software.

[0193] S702. The background server performs simulation analysis on the power model in the power simulation software to obtain the position information of the heating elements. Arrange temperature sensors according to the position information of the heating elements.

[0194] Among them, the background server runs the power simulation software, performs simulation analysis on the power model using the power simulation software to obtain the working temperature of each electrical device. Select the heating elements from each electrical device according to the working temperature of each electrical device to obtain the position information of the heating elements. After obtaining the position information of the heating elements, arrange temperature sensors according to the position information of the heating elements.

[0195] Compared with arranging temperature sensors based on experience, by establishing a power model corresponding to the power topology diagram of the monitoring area in a power simulation software, determining the operating temperature of each electrical device through simulation analysis, selecting heating elements based on the operating temperature, and arranging temperature sensors on the heating elements, the temperature of the heating elements can be monitored more accurately and abnormalities can be detected in a timely manner.

[0196] Optionally, the above solution first obtains the circuit topology diagram of the monitoring area and constructs the power network structure model of this area. Subsequently, in the power simulation software, an accurate power model is established based on this topology diagram. Through simulation analysis of the power model, the operating state of each electrical device and the position of its heating element can be accurately predicted. Finally, according to the simulation results, multiple temperature sensors are arranged on the heating elements to realize real-time monitoring of the device temperature.

[0197] By obtaining the circuit topology diagram of the monitoring area, the basic structure of the power network is constructed. Based on this topology diagram, a power model is established in the power simulation software, providing a basis for subsequent simulation analysis. This step ensures the accuracy and integrity of the power model and provides a scientific basis for the subsequent arrangement of temperature sensors.

[0198] In the power simulation software, simulation analysis is carried out on the power model to predict the operating state of each electrical device and the position of its heating element. Through simulation analysis, the position of the heating element can be accurately identified, avoiding the blindness of arranging temperature sensors based on experience in traditional methods and improving the accuracy and efficiency of monitoring.

[0199] According to the position information of the heating element obtained from the simulation analysis, multiple temperature sensors are arranged on the heating element to realize real-time monitoring of the device temperature. By directly arranging temperature sensors on the heating element, the temperature change of the device can be monitored more accurately, abnormalities can be detected in a timely manner, and strong support is provided for the prevention of fire risks.

[0200] By determining the position of the heating element through simulation analysis and arranging temperature sensors on it, the temperature change of the device can be monitored more accurately. Compared with the method of arranging temperature sensors based on experience, this technical solution significantly improves the accuracy of monitoring and reduces the occurrence of false alarms and missed alarms.

[0201] The above technical solution determines the positions of all heating elements through simulation analysis at one time, avoiding the process of multiple attempts and adjustments in traditional methods. This not only improves the monitoring efficiency but also reduces the human and time costs.

[0202] In a situation where the environment in the monitoring area is complex and there are numerous electrical devices, it is very difficult to accurately deploy temperature sensors using traditional methods. However, through simulation analysis, this technical solution can easily handle complex environments and ensure the accurate deployment of temperature sensors.

[0203] This technical solution combines power simulation technology and temperature sensing technology to achieve intelligent monitoring of the temperature of electrical devices. This not only improves the accuracy and efficiency of monitoring, but also provides strong support for subsequent fire risk assessment and early warning.

[0204] By continuously monitoring the temperature of the device, this technical solution can promptly detect abnormal conditions of the device, providing a scientific basis for the maintenance and management of the device. This helps to extend the service life of the device, reduce maintenance costs, and improve the safety and reliability of the device.

[0205] S703. The background server obtains the actual current detected by the current sensor on the electrical device and the actual temperature detected by the temperature sensor arranged on the heating element.

[0206] S704. The background server obtains the rated current of the electrical device and the temperature tolerance of the electrical device.

[0207] Among them, when the electrical device operates above the rated current for a long time, it is likely to cause damage to the electrical device. When the electrical device operates above the temperature tolerance for a long time, it is likely to cause the insulation layer to melt due to excessive temperature, and then cause a short circuit and fire. By obtaining the rated current and temperature tolerance of the electrical device, it is possible to judge whether there is a risk of fire based on the rated current and temperature tolerance.

[0208] In some examples, obtaining the temperature tolerance of the electrical device specifically includes:

[0209] S81. The background server obtains the actual current at multiple moments detected by the current sensor and the actual temperature at multiple moments detected by the temperature sensor.

[0210] Among them, after the current sensor collects the actual current at each moment of the electrical device, it uploads the actual current at each moment to the background server. After the temperature sensor collects the actual temperature at each moment of the electrical device, it uploads the actual temperature at each moment to the background server. The background server saves the received actual current at each moment and the actual temperature at each moment.

[0211] S82. The background server uses the temperature tolerance prediction model to process the actual current at multiple moments and the actual temperature at multiple moments, and outputs the temperature tolerance of the electrical device.

[0212] Among them, the temperature tolerance of the insulating layer is lower than that of the conductive metal. Therefore, the temperature tolerance of the electrical equipment is mainly determined by the insulating layer. And the performance of the insulating layer will change with the increase of the service time.

[0213] By testing the electrical equipment, the actual current at multiple test times, the actual temperature at multiple test times, and the temperature tolerance at multiple test times are obtained. Using a regression model to perform regression analysis on the actual current at multiple test times, the actual temperature at multiple test times, and the temperature tolerance at multiple test times, a temperature tolerance prediction model is obtained.

[0214] The background server obtains the actual temperature at multiple times collected by the temperature sensor. The background server also obtains the actual current at multiple times collected by the current sensor. The actual temperature at multiple times and the actual current at multiple times are input into the temperature tolerance prediction model to obtain the temperature tolerance at the current time.

[0215] By obtaining the actual temperature at multiple times and the actual current at multiple times, inputting the actual temperature at multiple times and the actual current at multiple times into the temperature tolerance prediction model for prediction, and updating the temperature tolerance in real time. Compared with using a fixed temperature tolerance, it can adapt to the change of the performance of electrical components with the increase of the service time, so that the fire risk can be detected in time.

[0216] Optionally, the core of the above solution is to use the temperature tolerance prediction model to process the actual current and actual temperature data of the electrical equipment at multiple times to dynamically predict and update the temperature tolerance of the equipment. This approach fully considers the possible performance changes of electrical components with the increase of the service time, making the fire risk assessment more in line with the actual operating state of the equipment.

[0217] Among them, for the current sensor and the temperature sensor, the actual current and actual temperature data of the electrical equipment at multiple times are collected respectively. These data provide rich input information for the temperature tolerance prediction model, enabling the model to more accurately capture the change trend of the equipment performance.

[0218] For the temperature tolerance prediction model, it processes the actual current and actual temperature data at multiple times and outputs the temperature tolerance of the electrical equipment. By learning and analyzing historical data, the model can predict the temperature tolerance of the equipment under different working conditions in real time, so as to adapt to the change of the equipment performance.

[0219] As the service time of the electrical equipment increases, its internal components may deteriorate due to aging, wear, etc. The technical solution can accurately reflect the change of the equipment performance by updating the temperature tolerance in real time, thus avoiding the risk assessment error caused by the fixed temperature tolerance.

[0220] By comprehensively considering the actual current and actual temperature data at multiple moments, the tolerance temperature prediction model can more comprehensively evaluate the operating state of the equipment. This multi-dimensional and dynamic evaluation method can significantly improve the accuracy of fire risk assessment compared with the evaluation methods based on single factor or fixed threshold.

[0221] Since the tolerance temperature is updated in real time, the system can promptly detect the fire risk caused by the decline of equipment performance. Once the actual current or actual temperature approaches or exceeds the predicted tolerance temperature, the system can trigger the warning mechanism, thus gaining precious time for fire prevention.

[0222] The above technical solution supports customization according to different types of electrical equipment and different working environments. By adjusting the parameters or structure of the tolerance temperature prediction model, it can be made more adaptable to the performance change characteristics of specific equipment, thereby enhancing the flexibility and applicability of the system.

[0223] By introducing the tolerance temperature prediction model, the above technical solution realizes the intelligent perception and evaluation of the performance changes of electrical equipment. This intelligent monitoring method not only improves the accuracy of fire risk assessment, but also provides data support for equipment maintenance and performance optimization.

[0224] In summary, the above technical solution introduces the tolerance temperature prediction model, uses the actual current and actual temperature data at multiple moments to predict and update the tolerance temperature of electrical equipment in real time, and significantly improves the accuracy and timeliness of fire risk assessment. This technical solution provides a more intelligent and accurate solution for fire prevention and control of electrical equipment.

[0225] S705a: If the actual current is greater than the preset current difference threshold and the actual temperature is greater than the tolerance temperature, it is determined that there is a fire risk, and proceed to S706.

[0226] S705b: If it is determined that the actual current is less than or equal to the set current, but the actual temperature is greater than the set temperature, continue to monitor the actual temperature. If the temperature rising rate is greater than the preset rate, it is determined that there is a fire risk, and proceed to S706.

[0227] S705c: If it is determined that the actual current is less than or equal to the set current and the actual temperature is less than or equal to the set temperature, it is determined that there is no fire risk, and proceed to S701.

[0228] S705d: If it is determined that the actual current is greater than the set current, but the actual temperature is less than or equal to the set temperature, it is determined that there is no fire risk, and proceed to S701.

[0229] Collect the actual temperature of the heating element through a temperature sensor, collect the actual current of the electrical equipment through a current sensor, and determine whether there is a fire risk based on the actual temperature and the actual current, which can improve the accuracy of the preliminary fire judgment.

[0230] Optionally, the above solution first relies on high-precision temperature sensors and current sensors, which are respectively used to collect the actual temperature of the heating element in the electrical equipment and the actual current of the electrical equipment. These sensors can capture the key parameters of the equipment operation state in real time and accurately, providing a solid data basis for the subsequent risk assessment. After obtaining the actual current and actual temperature data, further compare and analyze these data with the rated current and the tolerance temperature of the electrical equipment. The rated current is the upper limit of the normal working current specified during the equipment design, while the tolerance temperature is the highest temperature that the equipment material can withstand during long-term operation. Through this comparison, it is possible to scientifically and reasonably evaluate whether the current working state of the equipment exceeds the limit of its safe operation. When the actual current is greater than the rated current and the actual temperature is greater than the tolerance temperature, the system immediately determines that there is a fire risk. This judgment logic is based on the fact that current overload and high temperature are two main factors leading to electrical fires, so it can quickly respond to the abnormal state of the equipment and provide a valuable time window for timely taking preventive measures (such as cutting off the power supply, starting an alarm, etc.).

[0231] Compared with single-parameter monitoring (such as only monitoring current or temperature), this solution effectively avoids false alarms or missed alarms caused by abnormal single parameters by comprehensively considering two dimensions of current and temperature. This multi-dimensional and comprehensive risk assessment method significantly improves the accuracy of the preliminary fire judgment and provides a more reliable basis for subsequent emergency response and disaster control.

[0232] In the long run, the implementation of this technical solution helps to timely discover and handle the safety hazards of electrical equipment, reduce fire accidents caused by equipment failures or overheating, thus ensuring the safe progress of production and life. At the same time, through continuous monitoring and analysis of the equipment operation state, it can also provide data support for the maintenance of the equipment, extend the service life of the equipment, and reduce the operation cost.

[0233] S706. Obtain the arrangement positions of the temperature sensors and the arrangement positions of each camera.

[0234] S707. The background server selects a target camera from multiple cameras according to the arrangement positions of the temperature sensors and the arrangement positions of each camera.

[0235] S708. The background server controls the target camera to capture the monitoring image of the electrical equipment and obtains the light intensity at the shooting moment when the monitoring image is captured.

[0236] S709. The background server uses a fire detection model to identify the monitoring images, and uses the light intensity, actual current, and actual temperature at the shooting moment to correct the identification process, and outputs the fire detection result.

[0237] In the above technical solution, a current sensor, a temperature sensor, and a camera are used for global situation awareness, and data processing technology is used to process the sensed data to achieve the technical effect of detecting a fire as early as possible.

[0238] In addition, on the basis of the above embodiments, when obtaining the data of the current sensor, the temperature sensor, and the camera, time synchronization is first performed to ensure that the timestamps of all data are consistent. Then, the sensor data is calibrated. For example, for the temperature sensor, the offset and gain of the sensor can be adjusted by comparing with a standard temperature source. Next, the obtained data is cleaned to remove obviously abnormal data points, such as data points with sudden jumps in current values or temperature values outside the normal range. Finally, through data filtering, only the key information related to fire detection is retained, such as the change rate of current, the rising rate of temperature, and the monitoring images of the camera.

[0239] After determining the existence of a fire risk, the system first calculates the weight of each camera according to the position of the temperature sensor and the fire risk level. The factors considered in the weight include the distance between the camera and the fire source, the viewing angle range of the camera, and the image quality of the camera. Then, the target camera is selected from high to low according to the weight value. If the fire risk level is high, the system may select multiple cameras for simultaneous monitoring to obtain more comprehensive fire information. In addition, the system will also dynamically adjust the selection strategy of the target camera according to the spread trend of the fire and the change of the monitoring area to ensure that the latest dynamics of the fire can always be captured.

[0240] Optionally, when obtaining the data of the current sensor, the temperature sensor, and the camera, time synchronization is performed to ensure the consistency of the timeliness of all data. At the same time, the sensor data is calibrated to eliminate the data deviation caused by sensor errors. The multi-source data obtained is cleaned to remove outliers and noise data to improve the data quality. Through data filtering, only the key information useful for fire detection is retained. Through data synchronization and calibration, data cleaning and filtering, the accuracy and reliability of the data are improved, providing a high-quality data basis for subsequent fire detection.

[0241] After determining that there is a fire risk, not only the location of the temperature sensor is considered, but also the fire risk level is combined to give priority to selecting cameras that are closer to the fire source and have a better viewing angle as target cameras. Dynamically adjust the camera selection strategy: According to the spread trend of the fire and changes in the monitored area, dynamically adjust the target camera selection strategy to ensure that the latest fire dynamics can always be captured. The intelligent target camera selection algorithm improves the pertinence and efficiency of fire monitoring, reduces the collection and processing of useless information, and speeds up the fire response speed.

[0242] In the fire detection model, not only the light intensity at the time of shooting is considered, but also the historical light data and weather forecast information are combined to dynamically compensate for the light intensity to eliminate the impact of light changes on image recognition results. According to the lighting environment of the monitored area, the light intensity threshold is automatically adjusted to meet the needs of fire detection under different lighting conditions. Through the light intensity adaptive adjustment technology, the adaptability and accuracy of the fire detection model under different lighting conditions are improved, reducing the occurrence of false alarms and missed alarms.

[0243] Continuously collect new fire and non-fire sample data to train and optimize the fire detection model to improve the generalization ability and accuracy of the model. Establish an online model update mechanism to automatically update model parameters when new fire features or patterns are detected to adapt to the changing fire detection needs. Through the optimization and update of the fire detection model, the advancement and accuracy of the model are maintained, and the reliability and real-time performance of fire detection are improved.

[0244] Integrate components such as current sensors, temperature sensors, cameras, backend servers, and fire detection models into a unified global situational awareness system to achieve seamless data transmission and sharing. The system design has good scalability and can easily add new sensors and cameras to expand the monitoring range and functions. Through system integration and expansion, the overall performance and flexibility of the global situational awareness system are improved, providing customized solutions for fire detection in different scenarios.

[0245] The present application embodiment provides a global situation awareness system based on data processing, including:

[0246] An acquisition module, used to acquire an actual current detected by a current sensor arranged on the electrical device, and an actual temperature detected by a temperature sensor arranged on the heating element;

[0247] A processing module, used to determine whether there is a fire risk according to the actual current and the actual temperature, and if so, obtain the arrangement position of the temperature sensor and the arrangement position of each camera;

[0248] The processing module is further configured to select a target camera from multiple cameras according to the arrangement positions of the temperature sensor and each camera;

[0249] The processing module is further configured to control the target camera to capture a monitoring image of the electrical equipment and obtain the light intensity at the shooting moment when the monitoring image is captured;

[0250] The processing module is further configured to use the fire detection model to identify the monitoring image, and use the light intensity, actual current, and actual temperature at the shooting moment to correct the identification process, and output a fire detection result.

[0251] Figure 5 It is a schematic structural diagram of the background server provided by this application. As Figure 5 shown, the background server 80 provided in this embodiment includes: at least one processor 801 and a memory 802. Optionally, the background server 80 further includes a communication component 803. Among them, the processor 801, the memory 802, and the communication component 803 are connected through a bus.

[0252] In the specific implementation process, at least one processor 801 executes the computer execution instructions stored in the memory 802, so that at least one processor 801 executes the above method.

[0253] For the specific implementation process of the processor 801, reference may be made to the above method embodiment, and its implementation principle and technical effects are similar, and will not be elaborated here in this embodiment.

[0254] In the above embodiment, it should be understood that the processor may be a central processing unit (English: Central Processing Unit, abbreviated as: CPU), and may also be other general-purpose processors, digital signal processors (English: Digital Signal Processor, abbreviated as: DSP), application specific integrated circuits (English: Application Specific Integrated Circuit, abbreviated as: ASIC), etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly embodied as being executed and completed by a hardware processor, or executed and completed by a combination of hardware and software modules in the processor.

[0255] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.

[0256] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the buses in the drawings of this application are not limited to only one bus or one type of bus.

[0257] This application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0258] This application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the processor executes the computer-executable instructions, the above method is implemented.

[0259] The above-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disc. The readable storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0260] An exemplary readable storage medium is coupled to the processor, enabling the processor to read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an Application Specific Integrated Circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in a device.

[0261] The division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces, and the indirect coupling or communication connection of devices or units can be in electrical, mechanical, or other forms.

[0262] The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0263] In addition, in each embodiment of the present invention, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0264] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present invention. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, etc., which can store program codes.

[0265] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disks, or optical discs, etc., which can store program codes.

[0266] Finally, it should be noted that: After considering the specification and practicing the invention disclosed herein, those skilled in the art will easily think of other implementation schemes of the present invention. The present invention aims to cover any variations, uses, or adaptive changes of the present invention. These variations, uses, or adaptive changes follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed in the present invention. It is not limited to the precise structures already described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A global situation awareness method based on data processing, characterized in that: A plurality of cameras are arranged in the monitoring area, a current sensor is arranged on the electrical equipment in the monitoring area, and a temperature sensor is arranged on the heating element of the electrical equipment. The method is applied to a background server, and the method includes: Acquire the actual current detected by the current sensor arranged on the electrical device, and the actual temperature detected by the temperature sensor arranged on the heating element; Determine whether there is a fire risk according to the actual current and the actual temperature, and if so, obtain the arrangement position of the temperature sensor and the arrangement position of each camera; Selecting a target camera from a plurality of cameras according to the arrangement position of the temperature sensor and the arrangement position of each camera; Controlling the target camera to capture a monitoring image of the electrical device, and obtaining the light intensity at the time of capturing the monitoring image; Using a fire detection model to identify the monitoring image, and using the light intensity at the time of shooting, the actual current and the actual temperature to correct the identification process, and outputting a fire detection result; The monitoring image is identified using a fire detection model, and the identification process is corrected using the light intensity at the shooting time, the actual current, and the actual temperature, and a fire detection result is output, specifically including: generating an input feature according to the monitoring image and the light intensity at the shooting time, and generating a classification feature according to the actual current and the actual temperature; Splicing the input feature and the classification feature to generate a spliced ​​feature, using an encoder to encode the spliced ​​feature to output a coded feature, and using a decoder to decode the classification data in the coded feature to determine the fire detection result; Encoding the splicing features and outputting encoding features specifically includes: Using the first-level encoding module to encode the splicing feature, using the second-level encoding module to process the encoding feature output by the first-level encoding module, using the third-level encoding module to process the encoding feature output by the second-level encoding module, until the last-level encoding module processes the encoding feature output by the penultimate-level encoding module; Among them, each encoding module includes a multi-head attention mechanism module and a feedforward neural network module; the multi-head attention mechanism module includes a multi-head attention mechanism layer and a residual normalization layer; the feedforward neural network module includes a feedforward neural network layer and a residual normalization layer.

2. The global situation awareness method according to claim 1, characterized in that: Generating input features according to the monitoring image and the light intensity at the shooting time specifically includes: Divide the monitoring image into blocks to obtain a plurality of image blocks and the position of each image block, and determine the monitoring sub-area corresponding to each image block according to the shooting parameters of the camera; Obtaining the light intensity of each monitoring sub-area at the time of shooting, and using the light intensity of each monitoring sub-area at the time of shooting as the light intensity of the corresponding image block; Each image block, the position of each image block and the light intensity of each image block are fused to obtain fused data of each image block; and the fused data of multiple image blocks are spliced ​​to generate the input feature.

3. The global situation awareness method according to claim 1, characterized in that: Decoding the coded features using a decoder to determine a fire detection result specifically includes: The first fully connected layer is used to linearize the encoding features, the activation function layer is used to nonlinearly process the output data of the first fully connected layer, the second fully connected layer is used to linearize the output data of the activation function layer, and the residual layer is used to superimpose the output features of the second fully connected layer and the output features of the activation function layer and then output them.

4. The global situation awareness method according to any one of claims 1 to 3, characterized in that: Determine whether there is a fire risk based on the actual current and actual temperature, including: Obtain the rated current and temperature tolerance of electrical equipment; If the actual current is greater than the rated current, and the actual temperature is greater than the tolerance temperature, it is determined that there is a fire risk.

5. The global situation awareness method according to claim 4, characterized in that: Get the tolerance temperature of electrical equipment, including: Acquire actual currents detected by the current sensor at multiple moments and actual temperatures detected by the temperature sensor at multiple moments; The actual current at multiple moments and the actual temperature at multiple moments are processed using a tolerance temperature prediction model to output the tolerance temperature of the electrical device.

6. The global situation awareness method according to any one of claims 1 to 3, characterized in that: Before obtaining the actual current detected by the current sensor on the electrical device and the actual temperature detected by the temperature sensor arranged on the heating element, the method further includes: Obtaining a circuit topology diagram of the monitoring area, and establishing a power model in power simulation software based on the circuit topology diagram; The power model is simulated and analyzed in the power simulation software to obtain the position information of the heating element; and the plurality of temperature sensors are arranged according to the position information of the heating element.

7. A global situation awareness system based on data processing, characterized in that: include: An acquisition module, used to acquire an actual current detected by a current sensor arranged on the electrical device, and an actual temperature detected by a temperature sensor arranged on the heating element; a processing module, configured to determine whether there is a fire risk according to the actual current and the actual temperature, and if so, to obtain a layout position of a temperature sensor and a layout position of each camera; The processing module is further used to select a target camera from a plurality of cameras according to the arrangement position of the temperature sensor and the arrangement position of each camera; The processing module is further used to control the target camera to capture the monitoring image of the electrical device and obtain the light intensity at the time of capturing the monitoring image; The processing module is further used to identify the monitoring image using a fire detection model, and to correct the identification process using the light intensity at the shooting time, the actual current and the actual temperature, and output a fire detection result; The processing module is specifically used for: generating an input feature according to the monitoring image and the light intensity at the shooting time, and generating a classification feature according to the actual current and the actual temperature; Splicing the input feature and the classification feature to generate a spliced ​​feature, using an encoder to encode the spliced ​​feature to output a coded feature, and using a decoder to decode the classification data in the coded feature to determine the fire detection result; The processing module is specifically used for: Using the first-level encoding module to encode the splicing feature, using the second-level encoding module to process the encoding feature output by the first-level encoding module, using the third-level encoding module to process the encoding feature output by the second-level encoding module, until the last-level encoding module processes the encoding feature output by the penultimate-level encoding module; Among them, each encoding module includes a multi-head attention mechanism module and a feedforward neural network module; the multi-head attention mechanism module includes a multi-head attention mechanism layer and a residual normalization layer; the feedforward neural network module includes a feedforward neural network layer and a residual normalization layer.

8. A backend server, characterized in that: include: processor; as well as, A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.

Images