An indoor fire source positioning method based on infrared thermal image of outer surface of glass curtain wall

Abstract

The application discloses a kind of indoor fire source positioning methods based on the outer surface infrared thermal image of glass curtain wall, steps include: database construction, mainly including constructing fire data set, the temperature distribution of the outer surface of glass curtain wall, image extraction and processing;Resnet deep learning model is constructed and trained;Fire source positioning is carried out using Resnet deep learning model.In the early stage of fire development, when the fire reconnaissance of super high-rise building fire is carried out, the reconnaissance unmanned aerial vehicle carrying infrared thermal imager flies to the glass curtain wall outside the fire floor, adjusts the position of reconnaissance unmanned aerial vehicle to obtain the complete infrared thermal image of the outer surface of glass curtain wall of fire floor, and inputs the infrared thermal image of the outer surface of glass curtain wall obtained by infrared thermal imager into trained Resnet-50 deep learning model, the position number of fire source is given by model analysis, accurate identification of fire source position is realized in early stage of fire development, real-time and reliability are strong, and the fire warning capability and rescue efficiency of super high-rise building are improved.

Description

TECHNICAL FIELD

[0001] The application relates to an indoor fire source positioning method based on an infrared thermal image of an outer surface of a glass curtain wall and belongs to the technical field of fire source positioning. BACKGROUND

[0002] With the acceleration of urbanization and urbanization, super high-rise buildings have emerged. Super high-rise buildings, with their large scale, complex structure and high density, provide convenience while also posing a huge fire safety hazard.

[0003] After a fire breaks out in a super high-rise building, the smoke layer in the burning room accumulates and gradually subsides, forming a certain thickness of smoke layer in a short time, greatly reducing the indoor visibility and making it impossible to accurately identify the fire source position. The outer surface of the existing super high-rise building is mainly constructed with glass curtain walls, which hinders the accurate identification of the indoor temperature field by the outdoor infrared thermal imager, making it difficult to carry out outdoor fire reconnaissance, affecting the rescue efficiency of the super high-rise building fire, leading to the expansion of the fire and causing serious casualties and property losses.

[0004] Currently, fire source early positioning technologies rely on equipment in the fire building for positioning. Room fire source positioning technologies mainly include visual fire positioning technology, wireless sensor network positioning technology and optical fiber sensor positioning technology. These positioning technologies all require fixed position sensors, which need to be installed during building construction. The fixed deployment has poor flexibility and has a high probability of failure during the fire process. In addition, these positioning technologies have high time costs and cannot meet the real-time requirements in actual application scenarios. SUMMARY

[0005] The application provides an indoor fire source positioning method based on an infrared thermal image of an outer surface of a glass curtain wall, which can accurately identify the fire source position in the early stage of fire development, has strong real-time reliability and improves the fire warning capability and rescue efficiency of super high-rise buildings.

[0006] To achieve the above purpose, the application provides an indoor fire source positioning method based on an infrared thermal image of an outer surface of a glass curtain wall, which comprises the following steps:

[0007] S1, database construction, mainly including construction of a fire data set, temperature distribution of the outer surface of the glass curtain wall, image extraction and processing;

[0008] S2, constructing and training a Resnet deep learning model;

[0009] S3, using the Resnet deep learning model for fire source positioning.

[0010] Further, the specific process of S1 is as follows:

[0011] S1.1, simulate fire data using fire science computational fluid dynamics simulation software FDS, build a fire data set, the specific process is:

[0012] S1.1.1, build a fire burning model: set parameters, including grid parameters, the geometric structure of the fire room (including glass curtain wall, ventilation opening), surface parameters, reaction parameters, initial environment temperature T0, fire source position p, fire source heat release rate q, combustible smoke production rate f, ventilation opening opening time l; A total of p x q x f x l sets of simulation conditions are set;

[0013] S1.1.2, set up n x n temperature measuring points in the form of equal rows and columns on the inner surface of the glass in the fire room;

[0014] S1.1.3, set the simulation time t to simulate, and obtain the data set {T nt} of n x n temperature measuring points, T nt represents the temperature data of the nth measuring point at the tth second, and the time-temperature data set of different temperature measuring points is as follows:

[0015]

[0016] S1.1.3, integrate the time-temperature data set into a complete FDS simulation data set as input data for subsequent finite element analysis;

[0017] S1.2, use the transient thermal analysis of computational fluid dynamics software ANSYS to simulate the glass curtain wall inner surface temperature load obtained by the fire science computational fluid dynamics simulation software FDS simulation, and conduct heat conduction simulation to obtain the temperature distribution of the outer surface of the glass curtain wall, the specific process is as follows:

[0018] S1.2.1, use SpaceCliam to build a glass model with the same geometric size as the FDS fire burning model;

[0019] S1.2.2, import the built glass model into transient thermal, complete the import of geometric structure;

[0020] S1.2.3, input the key material parameters of the glass model for subsequent heat conduction analysis, including: density, thermal conductivity, specific heat, thermal expansion coefficient;

[0021] S1.2.4, meshing, the parameters for meshing include: physical preference of element, order of element, element size, number of nodes and elements;

[0022] S1.2.5, set the initial temperature of the glass model to be consistent with the initial ambient temperature T0 in the FDS fire combustion model;

[0023] S1.2.6, set the time length of ANSYS finite element analysis to be consistent with the time length t of FDS simulation;

[0024] S1.2.7, set the convective heat transfer coefficient on the outer surface of the glass model to simulate the convective heat transfer between the hot glass and the external environment in the real high-rise building fire;

[0025] S1.2.8, input the glass curtain wall inner surface temperature data obtained by FDS simulation as "temperature" load to the inner surface of the ANSYS glass model, and pay special attention to: according to the resolution of grid division, each "temperature" load can cover multiple grid elements. The spatial position of each "temperature" load input is consistent with the spatial position of the temperature measuring point set in the FDS simulation, that is, the nth FDS simulation temperature measuring point data corresponds to the nth AYSYS temperature load input, and the nth temperature load input to the ANSYS glass model is:

[0026]

[0027] S1.2.9, perform ANSYS finite element analysis and obtain the temperature change of the outer surface of the glass model; this process is based on the heat conduction equation, and the heat conduction process is represented by the following integral equation:

[0028]

[0029] Where, ρ is the density of the glass model; c is the specific heat capacity of the glass model; T is a function of temperature change with time and space; t * is time; k is the thermal conductivity of the material; Q is the heat power density of the internal heat source, since the glass model has no internal heat source, Q = 0;

[0030] The temperature distribution at the initial time t * = 0 is T((x, y, z, 0) = T0, n x n temperature loads are applied on one side of the glass model, and the boundary condition is expressed as:

[0031] T(x i ,y i ,t) = T boundary (x i ,y i ,t);

[0032] Where, T boundary represents the n x n temperature loads applied on the inner surface of the glass model; x i and y iis the coordinate position of the temperature load on the inner surface of the glass model; the convection heat exchange boundary condition set on the outer surface of the glass model is expressed as:

[0033]

[0034] wherein, represents the temperature gradient of the glass model surface in the normal direction; h is the convection heat exchange coefficient; T * is the temperature of the glass model outer surface; T ∞ is the temperature of the environmental fluid at the glass outer surface;

[0035] In the finite element solving process, the convection boundary condition is discretized, and the solving equation is as follows:

[0036]

[0037] wherein, [C conv ] represents the contribution item of the convection heat exchange, [M] is the mass matrix, which is proportional to the specific heat capacity of the glass material and the volume of the unit, and describes the change of the overall system energy storage when the temperature of each model unit changes; [K] is the stiffness matrix, which represents the thermal stiffness of the model system, that is, the resistance of the glass model to heat transfer under unit temperature change; {Q} is the load vector, which represents the applied temperature load; {T} is the temperature changing with time and space;

[0038] S1.2.10, the color scheme selection of the glass model outer surface temperature solving result is "reverse gray scale";

[0039] S1.3, image extraction and processing:

[0040] S1.3.1, according to the frame number of the ANSYS finite element analysis result, the glass model outer surface temperature distribution diagram of all frame numbers is extracted;

[0041] S1.3.2, all glass model outer surface temperature distribution diagrams of p x q x f x l groups of working conditions are extracted, and an untreated original picture database is constructed;

[0042] S1.3.3, all pictures in the original picture database are subjected to Gaussian blur processing to eliminate the glass model outer surface temperature distribution regionalization phenomenon caused by multiple temperature load inputs in the process of S1.2. The Gaussian filter of the two-dimensional image of the glass model outer surface temperature image is expressed as:

[0043]

[0044] where x and y are the horizontal and vertical distances of the image from the center pixel, respectively; G(x, y) is the output value of the Gaussian filter, which is a discretized form of the two-dimensional Gaussian function; δ is the standard deviation; and σ is used to control the size of the Gaussian filter, and the larger the δ, the larger the size of the Gaussian filter, and the more obvious the blurring effect will be; when the Gaussian blurring processing is performed, the image is continuously convolved using the Gaussian filter to achieve the blurring effect. For a glass model outer surface temperature distribution image I(x, y) and a Gaussian filter G(x, y), the result I'(x, y) of the convolution is:

[0045]

[0046] where G(i, j) is a two-dimensional Gaussian filter kernel, representing the weight of the distance from the center pixel;

[0047] By weighted averaging of each pixel and its neighborhood, the sharp edges and details of the image are reduced; the image is made smoother, and the glass model outer surface temperature distribution regionalization phenomenon caused by the multi-temperature load input in the S1.2 process is eliminated;

[0048] S1.4, performing Laplace sharpening calculation on the glass model outer surface temperature distribution picture after the Gaussian blurring processing, to strengthen the image boundary between the high-temperature region and the low-temperature region in the temperature image. Laplace sharpening is an image enhancement technique used to enhance the edge details in the image and highlight the structure of the image, which combines the Laplace operator and the addition operation of the image, and the Laplace operator is a second derivative operator, which is defined in two-dimensional space as:

[0049]

[0050] where f(x, y) is the pixel value of the image at point (x, y), and represents the Laplace operator; in a discrete image, the Laplace operator is approximated by a filter, and a common Laplace filter is used to approximate the Laplace operator:

[0051]

[0052] The Laplace filter calculates the neighborhood difference of each pixel point in the image to highlight the edges and details of the image;

[0053] The Laplace sharpening combines the original pixel value of the image with the output of the Laplace operator, and its mathematical expression is as follows:

[0054] f sharp (x, y) = f(x, y) - a · Δf(x, y);

[0055] Wherein, f(x, y) is the pixel value of the original image; and Af(x, y) is the result of the image f(x, y) applying the Laplace operator;

[0056] Alpha is a parameter for controlling the sharpening intensity, and is a constant;

[0057] The Laplace sharpening step for the picture database is as follows:

[0058] S1.4.1, calculate the Laplace operator result of the image f(x, y) to obtain the edge information in the image;

[0059] S1.4.2, select the control parameter alpha according to the sharpening intensity; if alpha is too large, the image may become over-sharpened, resulting in the enhancement of noise; if alpha is too small, the sharpening effect is not obvious.

[0060] S1.4.3, combine the original image with the Laplace operator result to obtain the sharpened image f sharp (x, y);

[0061] S1.4.4, set a label of a fire source position for each picture in the image database after image processing, that is, a number between 0 and p, to complete the construction of the glass model outer surface temperature picture database.

[0062] Further, the specific process of S2 is as follows:

[0063] S2.1, use the Resnet-50 deep learning model to train the Resnet-50 deep learning model using the glass model outer surface temperature distribution image database constructed in S1, and extract the deep features in the temperature distribution image, the model structure includes:

[0064] S2.1.1, input layer: the image input from the image database is adjusted in size through this layer, and the pictures in the picture database are generated using the "reverse gray" color scheme, and the pictures input for training are all gray images and are all single channels;

[0065] S2.1.2, convolution layer: preliminary feature extraction is performed through this layer to reduce the spatial size of the output feature map;

[0066] S2.1.3, max pooling layer: reduces the spatial dimension of the image, takes the maximum value in each local area, retains the most significant features, and makes the feature extraction more robust to small changes in input;

[0067] S2.1.4, multiple groups of residual blocks: each residual block includes an identity block (input is directly passed to output, added with the result of convolution) and a convolution block (to make the dimensions of input and output consistent, the model uses a specific size of convolution to increase or decrease dimensions), to avoid gradient disappearance and degradation in deep neural networks; for ideal feature mapping:

[0068] Φ(x * )=f(x * )+x * ;

[0069] where f(x * )=Φ(x * )-x * is the difference between the desired learned features and the input features, i.e. the residual, generated by convolution layer stacking, residual learning is easier than directly learning the ideal feature mapping Φ(x * ):

[0070]

[0071] where, is the loss function, and the output of the residual block is y * , is the gradient of the loss with respect to the input x * , through the residual connection, the gradient directly contains a direct path The gradient is directly passed from the later layer to the former layer, avoiding gradient disappearance;

[0072] S2.1.5, global average pooling layer: global dimension reduction is performed on the feature map, and high-dimensional spatial features are mapped to low-dimensional global features, and a single feature value is output, which provides input for the final fire source position classification task;

[0073] S2.1.6, fully connected layer, softmax function: the output of the fully connected layer is passed through the softmax function, and the output is the class of the fire source identification task, i.e. the label of the fire source position, i.e. a number between 0 and p; Softmax function is a normalization function, which is used to map real number vectors to probability distribution; Here it is used as the last layer of the multi-classification task, which converts the input of the fully connected layer to a probability value in the range [0, 1], which is used to represent the probability of the fire source belonging to each position. For the output vector z i of the fully connected layer, the Softmax function converts it to a probability distribution P i , and satisfies:

[0074]

[0075] The formula of the Softmax function is as follows:

[0076]

[0077] The gradient calculation and back propagation of the Softmax function are based on the cross-entropy loss function:

[0078]

[0079] wherein, γ is the cross-entropy loss, y i is the target class, if it is the correct fire source position, it is 1, if it is the remaining fire source position, it is 0;

[0080] The gradient of the Softmax output is:

[0081]

[0082] The output gradient is used for back propagation to continuously optimize the model output result;

[0083] S2.2, divide the picture database into training set and validation set according to the proportion, train the Resnet-50 deep learning model, adjust the training times to train until the validation accuracy and loss reach the set value, and complete the Resnet-50 deep learning model training.

[0084] Further, the specific process of S3 is:

[0085] S3.1, the infrared thermal imager selects gray color matching, the lens of the infrared thermal imager is aimed at the outer surface of the glass of the fire room and keeps facing the center of the glass to collect gray infrared thermal image; in the case that the infrared thermal imager does not have gray color matching, the picture is first converted into a gray image and then the subsequent operation is performed, the pixel point of a color image is composed of R, G and B channels, and the gray value Y corresponding to the pixel point is:

[0086] Y = w R ·R + w G ·G + w B ·B;

[0087] wherein, w R is the weighting coefficient of R channel, w G is the weighting coefficient of G channel, and w B is the weighting coefficient of B channel;

[0088] S3.2, input the gray infrared thermal image obtained by the infrared thermal imager into the Resnet-50 deep learning model trained in S2;

[0089] S3.3, the Resnet-50 deep learning model outputs the label of the fire source position, that is, a number between 0 and p, to determine the fire source position.

[0090] This invention constructs an image database, assigning a specific "fire source location" label to each image. This database is then used to train a ResNet-50 deep learning model to perform subsequent image recognition tasks. In the early stages of a fire, during fire reconnaissance in high-rise buildings, a reconnaissance drone equipped with an infrared thermal imager is operated to fly to the outside of the glass curtain wall of the burning floor. The drone's position is adjusted to obtain a complete infrared thermal image of the outer surface of the glass curtain wall. This image is then input into the trained ResNet-50 deep learning model. The model analyzes the image and assigns a location number to the fire source, enabling accurate identification of the fire source location in the early stages of a fire. This real-time and reliable approach significantly improves the fire early warning capabilities and rescue efficiency for high-rise buildings. Attached Figure Description

[0091] Figure 1 This is a schematic diagram of the workflow of the ResNet-50 deep learning model of the present invention;

[0092] Figure 2 This is a diagram showing the training results of the ResNet-50 deep learning model of this invention;

[0093] Figure 3 This is an infrared thermal image of the outer surface of a glass curtain wall obtained by an infrared thermal imager facing the glass in an embodiment of the present invention;

[0094] Figure 4 This is an infrared thermal image of the outer surface of a glass curtain wall obtained by an infrared thermal imager from the side of the glass in an embodiment of the present invention. Detailed Implementation

[0095] The invention will now be further described with reference to the accompanying drawings.

[0096] A method for locating indoor fire sources based on infrared thermal images of the outer surface of a glass curtain wall includes the following steps:

[0097] S1. Database construction, mainly including the construction of fire dataset, temperature distribution of glass curtain wall outer surface, image extraction and processing;

[0098] S2. Construct and train the ResNet deep learning model; such as Figure 1 As shown, the ResNet-50 deep learning model includes an input layer, convolutional layers, max-pooling layers, multiple residual blocks, global average pooling layers, and fully connected layers containing the softmax function. The ResNet deep learning model is continuously optimized through these layers, and then divided into training and validation sets in a 7:3 ratio. Training continues until the validation accuracy and loss reach the set values, completing the ResNet-50 deep learning model training. Figure 2 As shown, where,Figure 2 (a) to (f) are respectively the loss value result graphs and the training, validation accuracy value result graphs of 3 batches (each batch contains 3 epochs) of training Resnet-50 deep learning model using the method of the application, a, b is the result graph of a batch, c, d is the result graph of a batch, e, f is the result graph of a batch;

[0099] S3, fire source positioning using Resnet deep learning model, as shown in Figure 3 and Figure 4 ;

[0100] Example: According to the existing full-size fire experiment device, the fire spread prediction experiment is carried out, the FDS fire simulation model is consistent with the actual full-size fire experiment arrangement, 9 groups of fire source positions are set, a plurality of full-size fire experiments are carried out, and 20 glass outer surface gray thermal images are selected in different fire source position experiment conditions and input into the trained Resnet-50 deep learning model to obtain the experimental data table as follows:

[0101]

[0102]

[0103] Through the above examples, the following conclusions are obtained: in the test of 20 glass outer surface gray thermal images, the prediction accuracy reaches 95%, therefore the fire source position prediction technology of the application which fuses computational fluid dynamics, finite element analysis simulation technology, deep learning technology and infrared thermal imaging technology is suitable for fire reconnaissance tasks in high-rise building fires. Compared with other fire source positioning methods, the application has higher universality, timeliness and efficiency, and has higher level of accuracy.

Claims

1. A method for locating an indoor fire source based on an infrared thermal image of an outer surface of a glass curtain wall, characterized in that, Comprise the following steps: S1, database construction, including building fire data set, glass curtain wall outer surface temperature distribution, image extraction and processing; S2, build and train Resnet deep learning model; S3, using Resnet deep learning model for fire source positioning; S1.2, using computational fluid dynamics software ANSYS transient thermal analysis fire science computational fluid dynamics simulation software FDS simulation to obtain the glass curtain wall inner surface temperature load and conduct heat conduction simulation, get the temperature distribution of the outer surface of the glass curtain wall, the specific process is as follows: S1.2.1, using SpaceCliam to build a glass model with the same geometric size as the FDS fire burning model; S1.2.2, import the built glass model into transient thermal, complete the import of geometric structure; S1.2.3, input the key material parameters of the glass model for subsequent heat conduction analysis, including: density, thermal conductivity, specific heat, thermal expansion coefficient; S1.2.4, meshing, the parameters for meshing include: physical preference of element, element order, element size, number of nodes and elements; S1.2.5, set the initial temperature of the glass model to keep consistent with the initial ambient temperature T0 in the FDS fire burning model; S1.2.6, set the time length of ANSYS finite element analysis to keep consistent with the time length t of FDS simulation; S1.2.7, set the convective heat transfer coefficient on the outer surface of the glass model; S1.2.8, input the glass curtain wall inner surface temperature data obtained by FDS simulation as "temperature" load into the inner surface of the ANSYS glass model, the spatial position of each "temperature" load input is consistent with the spatial position of the temperature measuring point set in FDS simulation, that is, the nth FDS simulation temperature measuring point data corresponds to the nth ANSYS temperature load input, the nth temperature load input to the ANSYS glass model is: T nt Tn(t) represents the temperature data of the nth measuring point at the tth second; S1.2.9, ANSYS finite element analysis and solution, and obtain the temperature change of the outer surface of the glass model; The process is based on the heat conduction equation, and the heat conduction process is represented by the following integral equation: Where, ρ is the density of the glass model; C is the specific heat capacity of the glass model; T is a function of temperature change with time and space; t * t is time; k is the thermal conductivity of the material; Q is the heat power density of the internal heat source; At the initial time t * The temperature distribution with T(x,y,z,0) = To at the initial time t = 0, n x n temperature loads are applied on one side of the glass model, and the boundary conditions are expressed as: T(x i ,y i ,t)=T boundary (x i ,y i ,t); where T boundary represents n x n temperature loads applied on the inner surface of the glass model; x i and y i are the coordinate positions of the temperature loads on the inner surface of the glass model; and the convection heat exchange boundary condition set on the outer surface of the glass model is expressed as: wherein, represents the temperature gradient of the glass model surface in the normal direction; h is the heat transfer coefficient of convection; T * is the temperature of the glass model outer surface; T ∞ is the temperature of the ambient fluid at the glass outer surface; In the finite element solution process, the convective boundary condition will be discretized, and the solution equation is as follows: where [C conv ] represents the contribution term of convective heat transfer, [M] is the mass matrix, which is proportional to the specific heat capacity of the glass material and the volume of the unit; [K] is the stiffness matrix, which represents the thermal stiffness of the model system, that is, the resistance of the glass model to heat transfer under unit temperature change; {Q} is the load vector, which represents the applied temperature load; {T} is the temperature varying with time and space; S1.2.10, the color scheme selection of the glass model outer surface temperature solution result is "reverse gray".

2. The method for indoor fire source positioning based on the infrared thermogram of the outer surface of a glass curtain wall according to claim 1, characterized in that, The specific process of S1 is: S1.1, using fire science computational fluid dynamics simulation software FDS to simulate fire data, building fire data set, the specific process is: S1.1.1, build a fire burning model: set parameters, including grid parameters, geometric structure of the fire room, Surface parameters, Reaction parameters, initial ambient temperature T0, fire source position p, fire source heat release rate q, flammable smoke production rate f, ventilation opening opening time l; A total of p×q×f×l simulation conditions are set; S1.1.2, setting temperature measuring points n x n in an equi-row equi-column square array on the inner surface of the fire room glass; S1.1.3, set the simulation time t to simulate, and obtain the data set {T nt} of n x n temperature measuring points nt , T nt represents the temperature data of the nth measuring point at the tth second, and the time-temperature data set of different temperature measuring points is as follows: S1.1.4, integrating the time-temperature data set into a complete FDS simulation data set as input data for subsequent finite element analysis; S1.3, image extraction and processing: S1.3.1, extracting the glass model outer surface temperature distribution map of all frame numbers according to the frame number of the ANSYS finite element analysis result; S1.3.2, extracting the glass model outer surface temperature distribution map of all p x q x f x l groups of working conditions to construct an untreated original picture database; S1.3.3, performing Gaussian blur processing on all pictures in the original picture database, and the Gaussian filter of the two-dimensional image of the glass model outer surface temperature image is expressed as: wherein x and y are the horizontal and vertical distances of the image and the center pixel; G(x, y) is the output value of the Gaussian filter, which is a discrete form of the two-dimensional Gaussian function; δ is the standard deviation; when performing Gaussian blur processing, the image is continuously convolved using the Gaussian filter; for a glass model outer surface temperature distribution image I(x, y) and a Gaussian filter G(x, y), the convolution result I'(x, y) is: wherein G(i, j) is a two-dimensional Gaussian filter kernel, indicating the weight distance from the center pixel; Through the weighted average of each pixel and its neighborhood, the sharp edges and details of the image are reduced; S1.4, performing Laplace sharpening calculation on the glass model outer surface temperature distribution picture after Gaussian blur processing, and the Laplace sharpening combines the Laplace operator and the addition operation of the image; the Laplace operator is a second derivative operator, which is defined in two-dimensional space as: wherein f(x, y) is the pixel value of the image at point (x, y), and represents the Laplace operator; in a discrete image, the Laplace operator is approximated by a filter, and a common Laplace filter is used to approximate the Laplace operator: The Laplace filter calculates the neighborhood difference of each pixel point in the image to highlight the edges and details of the image; The mathematical expression of the Laplace sharpening combining the original pixel value of the image with the output of the Laplace operator is as follows: f sharp (x,y) = f(x,y) - a - Af(x,y); wherein f(x, y) is the pixel value of the original image; Δf(x, y) is the result of applying the Laplace operator to the image f(x, y); α is a parameter for controlling the sharpening intensity, which is a constant; The steps of Laplace sharpening on the picture database are as follows: S1.4.1, calculating the Laplace operator result of the image f(x, y) to obtain the edge information in the image; S1.4.2, selecting the control parameter α according to the sharpening intensity; S1.4.3, combine the original image with the Laplacian result to get the sharpened image f sharp (x,y); S1.4.4, setting a label of a fire source position for each picture in the image database after image processing, i.e. a number between 0 and p, to complete the construction of the glass model outer surface temperature picture database.

3. The method for locating indoor fire sources based on infrared thermographs of the outer surface of a glass curtain wall according to claim 1 or 2, characterized in that, The specific process of S2 is as follows: S2.1, using Resnet-50 deep learning model, using S1 to construct the glass model outer surface temperature distribution image database to train Resnet-50 deep learning model, and extract deep features in the temperature distribution image, the model structure includes: S2.1.1, input layer: the image input from the image database is adjusted in size through this layer, and the images in the image database are generated by using the "reverse gray scale" color scheme, and the images input for training are all gray images and are all single channels; S2.1.2, convolution layer: preliminary feature extraction is performed through this layer to reduce the spatial size of the output feature map; S2.1.3, maximum pooling layer: reduces the spatial dimension of the image, takes the maximum value in each local area, and retains the most significant features, while making the feature extraction more robust to small changes in input; S2.1.4, multiple residual blocks: each residual block includes an identity block and a convolution block, which is used to avoid gradient disappearance and degradation in deep neural networks; for feature mapping: Φ(x * ) = f(x * ) + x * ; where f(x * ) = Φ(x * ) - x * is the difference between the desired learned feature and the input feature, i.e., the residual, which is generated by a stack of convolutional layers, and residual learning is easier than learning the feature map Φ(x * ) directly: where ζ is the loss function, and the output of the residual block is y * , is the gradient of the loss with respect to the input x * Through the residual connection, the gradient directly contains a direct path The gradient directly passes from the later layer to the former layer, avoiding the vanishing gradient S2.1.5, global average pooling layer: global dimension reduction is performed on the feature map to map high-dimensional spatial features to low-dimensional global features, and output a single feature value to provide input for the final fire source location classification task; S2.1.6, full connection layer, softmax function: the output result of the full connection layer is output as the class of the fire source identification task after the softmax function, that is, the label of the fire source position, that is, a number between 0 and p; the softmax function is a normalization function, which is used to map a real number vector to a probability distribution; for the output vector z of the full connection layer i , the softmax function converts it into a probability distribution P i , and satisfies: The gradient calculation and back propagation of the Softmax function are based on the cross-entropy loss function: where γ is the cross-entropy loss, y i is the target class, 1 if the correct fire location, and 0 otherwise. The gradient of the Softmax output is: The output gradient is used for back propagation to continuously optimize the model output result; S2.2, divide the picture database into training set and validation set according to the proportion, train the Resnet-50 deep learning model, adjust the training times to train until the validation accuracy and loss reach the set value, and complete the Resnet-50 deep learning model training.

4. The method for indoor fire source positioning based on the infrared thermograph of the outer surface of glass curtain wall according to claim 3, characterized in that, The specific process of S3 is: S3.1, infrared thermal imager selects gray color, and the lens of the infrared thermal imager is aimed at the outer surface of the glass of the fire room and keeps facing the center of the glass to collect gray infrared thermal images; in the case that the infrared thermal imager does not have gray color, the picture is first converted into a gray image and then the subsequent operation is performed, and the pixel point of a color image is composed of R, G and B channels, and the gray value Y corresponding to the pixel point is: Y = w R • R + w G • G + w B • B; where w R is a weighting factor for the R channel, w G is a weighting factor for the G channel, and w B is a weighting factor for the B channel. S3.2, input the gray infrared thermal image obtained by the infrared thermal imager into the Resnet-50 deep learning model trained in S2; S3.3, the Resnet-50 deep learning model outputs the label of the fire source position, that is, a number between 0 and p, to determine the fire source position.

Images