An electrical fire detection and early warning method based on image processing

Abstract

The present invention discloses an electrical fire detection and early warning method based on image processing, relates to the technical field of fire early warning, and improves the accuracy of fire early warning. The present invention obtains the correlation between each local scene image model, splices each local scene image model according to the correlation to obtain a scene power distribution image model, divides a plurality of scene feature target areas in the scene power distribution image model, obtains the affinity value between each scene model voxel and each scene feature target area, and then divides a plurality of scene target models and scene background models according to the affinity value, obtains the standard thermal pattern distribution of each scene target model and scene background model, obtains the real-time thermal pattern distribution of the scene target model and the scene background model according to the real-time infrared scene video data, and judges whether the real-time thermal pattern distribution of each scene target model and the scene background model is abnormal through the standard thermal pattern distribution.

Description

Technical Field

[0001] The invention relates to the technical field of fire warning, and in particular to an electrical fire detection and warning method based on image processing. Background Art

[0002] With the continuous development of image processing technology, fire detection and early warning methods based on image processing have gradually attracted attention. This method uses cameras to monitor the electrical equipment area in real time, analyzes changes in the video stream through image processing algorithms, and can identify abnormal light spots in the early stages of a fire to achieve early warning.

[0003] However, this method has poor flexibility and cannot adapt to a variety of application scenarios. It also has problems such as slow response speed, insufficient sensitivity, and high false alarm rate. Therefore, an electrical fire detection and early warning method based on image processing is provided. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide an electrical fire detection and early warning method based on image processing.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] An electrical fire detection and early warning method based on image processing comprises the following steps:

[0007] Step S1, collecting infrared scene video data and laser signal spectrum of the application scene through infrared sensors and laser sensors, and then establishing a corresponding local scene image model according to the laser signal spectrum, taking each local scene image model as a matching center in turn, obtaining the correlation between each local scene image model, and splicing each local scene image model according to the correlation to obtain a scene power distribution image model;

[0008] Step S2, dividing the scene power consumption distribution image model into a number of scene model voxels of the same size and marking the spatial pixel values, and then averaging the spatial pixel values ​​of each scene model voxel, dividing a number of scene feature target areas in the scene power consumption distribution image model according to the spatial pixel value averaging result, obtaining the proximity value between each scene model voxel and each scene feature target area, and then dividing a number of scene target models and scene background models in the scene power consumption distribution image model according to the proximity value;

[0009] Step S3, obtaining the standard thermal pattern distribution of each scene target model and the scene background model through the infrared scene video data, obtaining the real-time thermal pattern distribution of the scene target model and the scene background model according to the real-time infrared scene video data, judging whether there is an abnormality in the real-time thermal pattern distribution of each scene target model and the scene background model through the standard thermal pattern distribution, and locating the fire pre-ignition position of the application scene according to the judgment result.

[0010] Furthermore, the acquisition process of infrared scene video data and laser signal spectrum includes:

[0011] Several groups of data acquisition units are installed in the application scene. Each group of data acquisition units consists of an infrared sensor, a laser sensor and a wireless data transmission device. The groups of data acquisition units are evenly distributed in the application scene, and the spatial distances between the data acquisition units at adjacent spatial positions are equal. Each data acquisition unit is numbered so that each data acquisition unit collects several copies of historical infrared video data and historical laser signal spectra within its data acquisition range through infrared sensors and laser sensors.

[0012] Furthermore, the process of obtaining the correlation between each local scene image model includes:

[0013] A local scene image model is established according to the historical laser signal spectrum, and each local scene image model is labeled with a corresponding data acquisition unit number;

[0014] Arrange each scene image model in sequence, and set the pixel point at the top center position of the local scene image model surface as the calibration center point, and according to the local scene image model arrangement result, take each local scene image model as the matching center in sequence, and select the calibration center point of the matching center as the starting point, and then select the calibration center points and the starting point of three local scene image models at adjacent positions in sequence to form a cube, and extract relevant parts from each associated local scene image model according to the model area contained in the cube to form a model matching area;

[0015] Establish a three-dimensional coordinate system, map each model matching area on the three-dimensional coordinate system, select the pixel points on the edge of each model area in the model matching area as the end point, and establish several straight line vectors with the calibration center point of the model area as the starting point;

[0016] A modulus value is set for each local straight line vector according to the pixel value of the surface of each model area in the model matching area, thereby obtaining the correlation between the straight line vectors of different model areas in the model matching area.

[0017] Furthermore, the process of splicing the local scene image models according to the correlation to obtain the scene power consumption distribution image model includes:

[0018] A unit similarity box is set, which is used to select a fixed number of straight line vectors, and a correlation threshold is set, and then the unit similarity box is used to traverse and select a fixed number of straight line vectors between the matching center and any model area in the model matching area, and the sum of the correlations between each straight line vector between two model areas at each selected position is counted;

[0019] If the sum of the correlations is greater than or equal to the correlation threshold, it is determined that the matching center of the position selected by the unit similarity box coincides with the corresponding model area, otherwise it is determined that they do not coincide;

[0020] The average value of the sum of the correlation degrees of each straight line vector at the coincident position is recorded as the geometric correction parameter γ, and then the coordinates of each pixel point on the coincident part of the model area and the matching center are corrected according to the geometric correction parameter γ;

[0021] Repeat the above process of correcting the pixel points in the model matching area, and use the corrected part as the connecting part to splice the local scene image model recorded as the matching center with the corresponding local scene image model to obtain the scene power distribution image model.

[0022] Furthermore, the process of averaging the spatial pixel values ​​of the scene model voxels includes:

[0023] Divide Num scene model voxels of the same size in the scene power consumption distribution image model, wherein the scene model voxels are in a cubic shape, and then obtain pixel values ​​of each surface of the scene model voxels and accumulate them to obtain an average value, and record the average value as the spatial pixel value of the scene model voxels, wherein Num is a natural number greater than 0;

[0024] Starting from the scene model voxels at the eight vertices of the scene power distribution image model, the scene model voxels and the scene model voxels at adjacent positions are averaged in turn, that is, the spatial pixel values ​​of each adjacent scene model voxel are accumulated and the average value is taken, and the average value is reassigned to each scene model voxel;

[0025] The above-mentioned spatial pixel value averaging process is repeated until all the scene model voxels selected from the eight vertices are one scene model voxel. After the spatial pixel value averaging is completed, several scene feature target areas are divided from the scene power distribution image model.

[0026] Furthermore, the process of obtaining the proximity value between each scene model voxel and each scene feature target area includes:

[0027] Obtain the average value J, covariance M and proportion weight K of the spatial pixel values ​​of all scene model voxels in each scene feature target area, where the calculation formula of the proportion weight K is;

[0028] Where K j and p j They represent the weight and spatial pixel value of the j-th scene model voxel respectively, and F is a natural number greater than 0 and less than Num;

[0029] Each scene feature target area is taken as the detection center in turn, and then the closeness value θ between each scene model voxel outside the detection center in the scene power distribution image model and the detection center is calculated. The calculation formula of the closeness value θ is:

[0030] θ x，y represents the closeness value between the x-th scene model voxel and the y-th scene feature target area, J y 、M y , T y represent the spatial pixel value mean, covariance and total number of scene model voxels in the y-th scene feature target area, G(p x |J y , M y ) represents the probability density function between the x-th scene model voxel and the y-th scene feature target area.

[0031] Furthermore, the process of dividing a plurality of scene target models and a scene background model in the scene power distribution image model according to the proximity value includes:

[0032] Classify each scene model voxel into the scene feature target area with the highest proximity value θ, and then divide a number of scene feature target models in the scene power consumption distribution image model;

[0033] Each scene feature target model is continuously updated according to the historical laser signal spectrum. If a scene feature target model undergoes local changes for three consecutive frames, the corresponding scene feature target model is recorded as the scene target model, otherwise it is recorded as the scene background model.

[0034] Furthermore, the process of locating the fire pre-occurrence position in the application scenario includes:

[0035] The historical thermal pattern distribution is marked on the model, and each historical thermal pattern is divided into several historical thermal pattern segments. Then, the historical thermal pattern segments generated by different historical infrared scene video data but corresponding to the same position are normally distributed, and the historical thermal pattern segment at the center of the normal distribution is selected as the standard thermal pattern segment at the corresponding position under normal operating conditions;

[0036] The standard thermal pattern fragments are sequentially spliced ​​to obtain the standard thermal pattern distribution of each scene target model and scene background model, and the standard temperature range corresponding to the standard thermal pattern distribution is marked;

[0037] Collect real-time infrared scene video data of the application scene, and generate real-time thermal pattern distribution of each scene target model and scene background model according to the real-time infrared scene video data;

[0038] Get the real-time temperature range corresponding to the real-time thermal pattern distribution, and then determine whether the real-time temperature range is greater than the corresponding standard temperature range. If not, do not perform any operation.

[0039] If it is greater than, the location of the real-time thermal pattern distribution is judged to be the fire prediction location of the application scenario.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The present invention divides a number of scene feature target areas in a scene power consumption distribution image model, obtains the proximity value between each scene model voxel and each scene feature target area, and then divides a number of scene target models and scene background models according to the proximity value, obtains the standard thermal pattern distribution of each scene target model and scene background model, obtains the real-time thermal pattern distribution of the scene target model and the scene background model according to the real-time infrared scene video data, and judges whether the real-time thermal pattern distribution of each scene target model and the scene background model is abnormal through the standard thermal pattern distribution, thereby effectively improving the response speed and sensitivity of fire warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention.

[0043] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0044] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0045] Example 1

[0046] like Figure 1 As shown, an electrical fire detection and early warning method based on image processing includes the following steps:

[0047] Step S1, collecting infrared scene video data and laser signal spectrum of the application scene through infrared sensors and laser sensors, and then establishing corresponding local scene image models according to the laser signal spectrum, taking each local scene image model as the matching center in turn, obtaining the correlation between the matching center area and the adjacent local scene image model, and splicing each local scene image model according to the correlation to obtain a scene power distribution image model;

[0048] Step S2, dividing the scene power consumption distribution image model into a number of scene model voxels of the same size and marking the spatial pixel values, and then averaging the spatial pixel values ​​of each scene model voxel, dividing a number of scene feature target areas in the scene power consumption distribution image model according to the spatial pixel value averaging result, obtaining the proximity value between each scene model voxel and each scene feature target area, and then dividing a number of scene target models and scene background models in the scene power consumption distribution image model according to the proximity value;

[0049] Step S3, obtaining the standard thermal pattern distribution of each scene target model and the scene background model through the infrared scene video data, obtaining the real-time thermal pattern distribution of the scene target model and the scene background model according to the real-time infrared scene video data, judging whether there is an abnormality in the real-time thermal pattern distribution of each scene target model and the scene background model through the standard thermal pattern distribution, and locating the fire pre-ignition position of the application scene according to the judgment result.

[0050] Example 2

[0051] This embodiment is a further limitation of Embodiment 1, and Step S1 is implemented by the following process:

[0052] Several groups of data acquisition units are installed in the application scene. Each group of data acquisition units consists of an infrared sensor, a laser sensor and a wireless data transmission device. Each group of data acquisition units is evenly distributed in the application scene, and the spatial distances between data acquisition units at adjacent spatial positions are equal.

[0053] It should be noted that the data collection ranges of the infrared sensor and the laser sensor are the same, and the data collection ranges of the data collection units at adjacent spatial positions partially overlap, and the data collection ranges of all data collection units are equal to the entire spatial volume of the application scene;

[0054] Set the numbers a1, a2, a3, ..., a for each data acquisition unit. n , where n is a natural number greater than 0, and each data acquisition unit collects several copies of historical infrared video data and historical laser signal spectra within its data acquisition range through infrared sensors and laser sensors;

[0055] A scene fire monitoring center is set up, and then each data acquisition unit sends all the collected historical infrared video data and historical laser signal spectrum to the scene fire monitoring center through a wireless data transmission device. It should be noted that during the data acquisition process, the angles of each infrared sensor and laser sensor are fixed;

[0056] The scene fire monitoring center marks the corresponding data acquisition unit number for each historical infrared video data and historical laser signal spectrum, and then establishes a scene power distribution image model based on the historical laser signal spectrum. The specific process includes:

[0057] The scene fire monitoring center establishes the corresponding local scene image model based on the historical laser signal spectrum generated at the same acquisition time, and marks the corresponding data acquisition unit number for each local scene image model;

[0058] Since there are related parts in the data collection ranges of the data collection units at adjacent spatial positions, there is also a correlation between the local scene image models corresponding to the relevant historical laser signal spectra. At the same time, since different laser sensors have different shooting angles for the same position in the application scene, there are geometric differences in the corresponding local scene image models for the same position in the application scene.

[0059] According to the actual geographical distribution of the data acquisition units in the application scene corresponding to the numbers of each local scene image model, the scene image models are arranged in sequence. At the same time, since the angles of the infrared sensor and the laser sensor are fixed during the acquisition process, the center position of the local scene image model is vertical to the laser sensor.

[0060] Then, the pixel point at the top center position of the surface of the local scene image model is set as the calibration center point, and according to the arrangement results of the local scene image models, each local scene image model is taken as the matching center in turn, and the calibration center point of the matching center is selected as the starting point, and then the calibration center points and the starting point of three local scene image models at adjacent positions are selected in turn to form a cube, and according to the model area contained in the cube, relevant parts are extracted from each associated local scene image model to form a model matching area;

[0061] A three-dimensional coordinate system is established, and each model matching area is mapped onto the three-dimensional coordinate system. Pixel points at the edge of each model area in the model matching area are selected as the end point, and several straight line vectors are established with the calibration center point of the model area as the starting point. It should be noted that the starting point of each straight line vector is the same but the end point is different, and the straight line vector is composed of several local straight line vectors. The starting point of each local straight line vector on the same straight line vector is the end point of the previous adjacent local straight line vector.

[0062] Setting a modulus value for each local straight line vector according to the pixel value of each model area surface in the model matching area;

[0063] Since each model region in the model matching region is correlated, the straight line vectors in different model regions have the same part, and then the correlation degree α between the straight line vectors of different model regions in the model matching region is obtained, where the calculation formula of the correlation degree α is: Where N represents the total number of local straight line vectors in the model area, and They represent the i-th local straight line vector in the straight line vector of different model regions respectively;

[0064] A unit similarity box is set, which is used to select a fixed number of straight line vectors, and a correlation threshold is set, and then the unit similarity box is used to traverse and select a fixed number of straight line vectors between the matching center and any model area in the model matching area, and the sum of the correlations between each straight line vector between two model areas at each selected position is counted;

[0065] If the sum of the correlations is greater than or equal to the correlation threshold, it is determined that the matching center of the position selected by the unit similarity box coincides with the corresponding model area, otherwise it is determined that they do not coincide;

[0066] The average value of the sum of the correlation degrees of each straight line vector at the coincident position is recorded as the geometric correction parameter γ, and then the coordinates of each pixel point on the coincident part of the model area and the matching center are corrected according to the geometric correction parameter γ;

[0067] Repeat the above process of correcting the pixel points in the model matching area, and use the corrected part as the connecting part to splice the local scene image model recorded as the matching center with the corresponding local scene image model to obtain the scene power distribution image model.

[0068] Example 3

[0069] This embodiment is a further limitation of Embodiment 1, and Step S2 is implemented by the following process:

[0070] Divide Num scene model voxels of the same size in the scene power consumption distribution image model, wherein the scene model voxels are in a cubic shape, and then obtain pixel values ​​of each surface of the scene model voxels and accumulate them to obtain an average value, and record the average value as the spatial pixel value of the scene model voxels, wherein Num is a natural number greater than 0;

[0071] Starting from the scene model voxels at the eight vertices of the scene power distribution image model, the scene model voxels and the scene model voxels at adjacent positions are averaged in turn, that is, the spatial pixel values ​​of each adjacent scene model voxel are accumulated and the average value is taken, and the average value is reassigned to each scene model voxel;

[0072] The above-mentioned spatial pixel value averaging process is repeated until all the scene model voxels selected from the eight vertices are one scene model voxel. After the spatial pixel value averaging is completed, several scene feature target areas are divided from the scene power distribution image model.

[0073] Furthermore, the average value J, covariance M and proportion weight K of the spatial pixel values ​​of all scene model voxels in each scene feature target area are obtained, wherein the calculation formula of the proportion weight K is:

[0074] Where K j and p j They represent the weight and spatial pixel value of the j-th scene model voxel respectively, and F is a natural number greater than 0 and less than Num;

[0075] Each scene feature target area is taken as the detection center in turn, and then the closeness value θ between each scene model voxel outside the detection center in the scene power distribution image model and the detection center is calculated. The calculation formula of the closeness value θ is:

[0076] θ x，y represents the closeness value between the x-th scene model voxel and the y-th scene feature target area, J y 、M y 、T y represent the spatial pixel value mean, covariance and total number of scene model voxels in the y-th scene feature target area, G(p x |J y , M y ) represents the probability density function between the x-th scene model voxel and the y-th scene feature target area.

[0077] Furthermore, each scene model voxel is classified into a scene feature target region with the highest proximity value θ, and then a number of scene feature target models are divided in the scene power consumption distribution image model;

[0078] Each scene feature target model is continuously updated according to the historical laser signal spectrum. If a scene feature target model undergoes local changes for three consecutive frames, the corresponding scene feature target model is recorded as the scene target model, otherwise it is recorded as the scene background model.

[0079] Example 4

[0080] This embodiment is a further limitation of Embodiment 1, and step S3 is implemented by the following process:

[0081] According to each historical infrared scene video data, the historical thermal pattern distribution is marked on each scene target model and the scene background model respectively, and each historical thermal pattern is divided into a number of historical thermal pattern segments, and then the historical thermal pattern segments generated by different historical infrared scene video data but corresponding to the same position are normally distributed, and the historical thermal pattern segment at the center of the normal distribution is selected and recorded as the corresponding position, which is the standard thermal pattern segment under normal operating conditions;

[0082] The standard thermal pattern fragments are sequentially spliced ​​to obtain the standard thermal pattern distribution of each scene target model and scene background model, and the standard temperature range corresponding to the standard thermal pattern distribution is marked;

[0083] Collect real-time infrared scene video data of the application scene, and generate real-time thermal pattern distribution of each scene target model and scene background model according to the real-time infrared scene video data;

[0084] Get the real-time temperature range corresponding to the real-time thermal pattern distribution, and then determine whether the real-time temperature range is greater than the corresponding standard temperature range. If not, do not perform any operation.

[0085] If it is greater than, the location of the real-time thermal pattern distribution is judged to be the fire prediction location of the application scenario.

[0086] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An electrical fire detection and early warning method based on image processing, characterized in that: The following steps are involved: Step S1, collecting infrared scene video data and laser signal spectrum of the application scene through infrared sensors and laser sensors, and then establishing a corresponding local scene image model according to the laser signal spectrum, taking each local scene image model as a matching center in turn, obtaining the correlation between each local scene image model, and splicing each local scene image model according to the correlation to obtain a scene power distribution image model; Step S2, dividing the scene power consumption distribution image model into a number of scene model voxels of the same size and marking the spatial pixel values, and then averaging the spatial pixel values ​​of each scene model voxel, dividing a number of scene feature target areas in the scene power consumption distribution image model according to the spatial pixel value averaging result, obtaining the proximity value between each scene model voxel and each scene feature target area, and then dividing a number of scene target models and scene background models in the scene power consumption distribution image model according to the proximity value; Obtain the average value J, covariance M and proportion weight K of the spatial pixel values ​​of all scene model voxels in each scene feature target area, where the calculation formula of the proportion weight K is; Where K j and p j They represent the weight and spatial pixel value of the j-th scene model voxel respectively, and F is a natural number greater than 0 and less than Num; Each scene feature target area is taken as the detection center in turn, and then the closeness value θ between each scene model voxel outside the detection center in the scene power distribution image model and the detection center is calculated. The calculation formula of the closeness value θ is: θ x，y represents the closeness value between the x-th scene model voxel and the y-th scene feature target area, J y 、M y , T y represent the spatial pixel value mean, covariance and total number of scene model voxels in the y-th scene feature target area, G(p x |J y , M y ) represents the probability density function between the x-th scene model voxel and the y-th scene feature target area; Step S3, obtaining the standard thermal pattern distribution of each scene target model and the scene background model through the infrared scene video data, obtaining the real-time thermal pattern distribution of the scene target model and the scene background model according to the real-time infrared scene video data, judging whether there is an abnormality in the real-time thermal pattern distribution of each scene target model and the scene background model through the standard thermal pattern distribution, and locating the fire pre-ignition position of the application scene according to the judgment result.

2. The electrical fire detection and early warning method based on image processing according to claim 1 is characterized in that: The acquisition process of infrared scene video data and laser signal spectrum includes: Several groups of data acquisition units are installed in the application scenario. Each group of data acquisition units consists of an infrared sensor, a laser sensor and a wireless data transmission device. Each data acquisition unit is numbered, and each data acquisition unit collects several copies of historical infrared video data and historical laser signal spectra within its data acquisition range through infrared sensors and laser sensors.

3. The electrical fire detection and early warning method based on image processing according to claim 2 is characterized in that: The process of obtaining the correlation between each local scene image model includes: A local scene image model is established according to the historical laser signal spectrum, and each local scene image model is labeled with a corresponding data acquisition unit number; Arrange each scene image model in sequence, and set the pixel point at the top center position of the local scene image model surface as the calibration center point, and according to the local scene image model arrangement result, take each local scene image model as the matching center in sequence, select the calibration center point of the matching center as the starting point, and then select the calibration center points and the starting point of three local scene image models at adjacent positions in sequence to form a cube, and extract relevant parts from each associated local scene image model according to the model area contained in the cube to form a model matching area; Establish a three-dimensional coordinate system, map each model matching area on the three-dimensional coordinate system, select the pixel points on the edge of each model area in the model matching area as the end point, and establish several straight line vectors with the calibration center point of the model area as the starting point; A modulus value is set for each local straight line vector according to the pixel value of the surface of each model area in the model matching area, thereby obtaining the correlation between the straight line vectors of different model areas in the model matching area.

4. The electrical fire detection and early warning method based on image processing according to claim 3 is characterized in that: The process of splicing each local scene image model according to the correlation to obtain the scene power distribution image model includes: A unit similarity box is set, which is used to select a fixed number of straight line vectors, and a correlation threshold is set, and then the unit similarity box is used to traverse and select a fixed number of straight line vectors between the matching center and any model area in the model matching area, and the sum of the correlations between each straight line vector between two model areas at each selected position is counted; If the sum of the correlations is greater than or equal to the correlation threshold, it is determined that the matching center of the position selected by the unit similarity box coincides with the corresponding model area, otherwise it is determined that they do not coincide; The average value of the sum of the correlation degrees of each straight line vector at the coincident position is recorded as the geometric correction parameter γ, and then the coordinates of each pixel point on the coincident part of the model area and the matching center are corrected according to the geometric correction parameter γ; The local scene image model recorded as the matching center is spliced ​​with the corresponding local scene image model to obtain the scene power consumption distribution image model.

5. The electrical fire detection and early warning method based on image processing according to claim 4 is characterized in that: The process of averaging the spatial pixel values ​​of the scene model voxels includes: Divide Num scene model voxels of the same size in the scene power consumption distribution image model, wherein the scene model voxels are in a cubic shape, and then obtain pixel values ​​of each surface of the scene model voxels and accumulate them to obtain an average value, and record the average value as the spatial pixel value of the scene model voxels, wherein Num is a natural number greater than 0; Starting from the scene model voxels at the eight vertices of the scene power distribution image model, the spatial pixel values ​​of the scene model voxels and the scene model voxels at their adjacent positions are averaged in turn, that is, the spatial pixel values ​​of each adjacent scene model voxel are accumulated and the average value is taken, and the average value is reassigned to each scene model voxel. When the spatial pixel value averaging is completed, several scene feature target areas are divided from the scene power distribution image model.

6. The electrical fire detection and early warning method based on image processing according to claim 5, characterized in that: The process of dividing a scene target model and a scene background model into several scene target models and a scene background model according to the close numerical values ​​includes: Classify each scene model voxel into the scene feature target area with the highest proximity value θ, and then divide a number of scene feature target models in the scene power consumption distribution image model; Each scene feature target model is continuously updated according to the historical laser signal spectrum. If a scene feature target model undergoes local changes for three consecutive frames, the corresponding scene feature target model is recorded as the scene target model, otherwise it is recorded as the scene background model.

7. The electrical fire detection and early warning method based on image processing according to claim 6, characterized in that: The process of locating the fire pre-occurrence position in the application scenario includes: The historical thermal pattern distribution is marked on each scene target model and scene background model, and each historical thermal pattern is divided into several historical thermal pattern segments. Then, the historical thermal pattern segments generated by different historical infrared scene video data but corresponding to the same position are normally distributed, and the historical thermal pattern segment at the center of the normal distribution is selected as the corresponding position, which is the standard thermal pattern segment under normal operating conditions; The standard thermal pattern fragments are sequentially spliced ​​to obtain the standard thermal pattern distribution of each scene target model and scene background model, and the standard temperature range corresponding to the standard thermal pattern distribution is marked; Collect real-time infrared scene video data of the application scene, and generate real-time thermal pattern distribution of each scene target model and scene background model according to the real-time infrared scene video data; Get the real-time temperature range corresponding to the real-time thermal pattern distribution, and then determine whether the real-time temperature range is greater than the corresponding standard temperature range. If not, do not perform any operation. If it is greater than, the location of the real-time thermal pattern distribution is judged to be the fire prediction location of the application scenario.

Images