Method for identifying indoor fire source location based on infrared thermal imaging of the outer surface of a glass curtain wall

Abstract

A method for identifying the location of an indoor fire source based on infrared thermal images of the outer surface of a glass curtain wall is provided. [Solution] The steps include: data structure construction: mainly including steps of constructing a fire dataset, temperature distribution of the outer surface of the glass curtain wall, image extraction and processing; construction and training of a Resnet deep learning model; and fire source location determination using the Resnet deep learning model. In the early stages of fire development, a reconnaissance drone equipped with an infrared thermal imaging camera is operated to fly outside the glass curtain wall of the fire floor, and the position of the reconnaissance drone is adjusted to obtain a complete infrared thermal image of the outer surface of the glass curtain wall of the fire floor. The infrared thermal image of the outer surface of the glass curtain wall acquired by the infrared thermal imaging camera is input to a trained Resnet-50 deep learning model, and the model outputs the fire source location number after analysis.

Description

[Technical Field]

[0001] The present invention relates to a method for identifying the location of an indoor fire source based on an infrared thermal image of the outer surface of a glass curtain wall, and belongs to the field of fire source location identification technology. [Background technology]

[0002] As urbanization and city development accelerate, skyscrapers are becoming increasingly common. While their immense scale, complex structure, and high density offer convenience, they also pose significant fire safety risks.

[0003] After a fire breaks out in a skyscraper, a layer of smoke accumulates inside the fire chamber and gradually settles. A smoke layer of a certain thickness forms in a short time, significantly reducing visibility inside the room and making it impossible to accurately identify the location of the fire source. The exterior surfaces of many existing skyscrapers employ glass curtain wall structures, which hinder the accurate identification of the temperature field inside by infrared thermal imaging cameras outside the building. This makes it difficult to conduct fire reconnaissance from outside, affecting the efficiency of rescue operations in skyscraper fires and leading to the spread of the fire and serious human and property damage.

[0004] Currently, initial fire source location technologies rely on equipment within the burning building for localization. Indoor fire source location technologies primarily include visual fire location technologies, wireless sensor network location technologies, and fiber optic sensor location technologies. All of these localization technologies require fixed-position sensors that must be installed during building construction. Fixed placement lacks flexibility and has a high probability of malfunction during the fire process. Furthermore, these localization technologies are time-consuming and cannot meet the real-time requirements of actual application scenarios. [Overview of the Initiative] [Problems that the invention aims to solve]

[0005] This invention provides a method for locating an indoor fire source based on infrared thermal imaging of the outer surface of a glass curtain wall. This location identification method can accurately identify the fire source location in the early stages of fire development, offers excellent real-time performance and reliability, and improves fire alarm capabilities and rescue efficiency in high-rise buildings. [Means for solving the problem]

[0006] To achieve the above objective, the present invention provides a method for identifying the location of an indoor fire source based on an infrared thermal image of the outer surface of a glass curtain wall. S1, Data structure construction: This mainly involves steps including the construction of a fire dataset, temperature distribution of the outer surface of a glass curtain wall, image extraction, and processing. S2, the steps for building and training the ResNet deep learning model, S3 includes a step of locating the fire source using a Resnet deep learning model.

[0007] Furthermore, the specific process of S1 is as follows: S1.1. Using the fire science computational fluid dynamics simulation software FDS, fire data was simulated to construct a fire dataset. The specific process is as follows: S1.1.1, Construct a fire combustion model: Set parameters including mesh parameters, geometric structure of the fire chamber (including glass curtain wall and ventilation openings), surface parameters, reaction parameters, initial ambient temperature T0, p ignition source locations, q ignition source heat release rates, f combustible smoke generation rates, and l ventilation opening opening times, and set a total of p × q × f × l sets of simulation conditions. S1.1.2, n × n temperature measurement points are set on the inner surface of the glass in the fire chamber in an equal-row, equal-column square matrix. S1.1.3, set the simulation time t and perform the simulation, and obtain the dataset {T} of n × n temperature measurement points. nt Obtain} and T nt represents the temperature data at the t-second mark at the nth measurement point, and the time-temperature dataset for different temperature measurement points is given by the following equation:

number

Number

Number

Number

Number

Number

Number

number

number

number

number

[0008] Furthermore, the specific process of S2 is as follows: In S2.1, a Resnet-50 deep learning model was adopted. The Resnet-50 deep learning model was trained using the glass model outer surface temperature distribution image database constructed in S1. Deep features were extracted from the temperature distribution images, and the model structure was as follows: S2.1.1, This layer adjusts the applicability of image size based on images input from the image database. All images in the image database use an "inverted grayscale" color scheme during generation, and all images fed into training are grayscale images, and all input layers are single-channel. S2.1.2, This layer performs initial feature extraction and a convolutional layer reduces the spatial size of the output feature map. S2.1.3, A maximum pooling layer reduces the spatial dimension of the image, takes the maximum value in each local region, preserves the most prominent features, and makes feature extraction more robust to small changes in the input. S2.1.4, Each residual block includes an identity block (which directly transmits the input to the output and adds it to the convolution result) and a convolution block (which ensures the input and output dimensions match, and this model uses a convolution of a specific size to perform dimensionality expansion or reduction), and is used to avoid gradient vanishing and degeneration in deep neural networks, for an ideal feature mapping:

number

number

number

number

number

number

number

[0009] Furthermore, the specific process of S3 is as follows: S3.1 Select a grayscale color scheme for the infrared thermal imaging camera, point the lens of the infrared thermal imaging camera towards the outer surface of the glass in the fire room, and with the camera facing directly towards the center of the glass, collect a grayscale infrared thermal image. If the infrared thermal imaging camera does not have a grayscale color scheme, first convert the image to a grayscale image before proceeding with the subsequent operations. The pixel points of the color image are composed of R, G, and B of the RGB channels, and the grayscale value Y corresponding to the pixel point is given by the following formula:

number

[0010] This invention constructs an image database, associates each image in the database with a specific "fire source location" label, and uses the constructed image database to train a residual neural network Resnet-50 deep learning model to complete subsequent image recognition tasks. In the early stages of a fire outbreak, when conducting reconnaissance of a fire in a high-rise building, a reconnaissance drone equipped with an infrared thermal imaging camera is operated to fly outside the glass curtain wall of the fire floor, and the drone's position is adjusted to acquire a complete infrared thermal image of the outer surface of the glass curtain wall on the fire floor. The infrared thermal image of the outer surface of the glass curtain wall acquired by the infrared thermal imaging camera is input into the trained Resnet-50 deep learning model, and after analysis, the model outputs the fire source location number. This enables accurate identification of the fire source location in the early stages of a fire outbreak, offering excellent real-time performance and reliability, and significantly improving the fire alarm capabilities and rescue efficiency of high-rise buildings. [Brief explanation of the drawing]

[0011] [Figure 1] This is a schematic diagram of the workflow of the Resnet-50 deep learning model of the present invention. [Figure 2(a)-2(f)] This is a diagram showing the training results of the ResNet-50 deep learning model of the present invention. [Figure 3] In an embodiment of the present invention, this is an infrared thermal image of the outer surface of a glass curtain wall acquired by an infrared thermal imaging camera facing directly at the glass. [Figure 4] In an embodiment of the present invention, this is an infrared thermal image of the outer surface of a glass curtain wall acquired by an infrared thermal imaging camera pointed obliquely at the glass. [Modes for carrying out the invention]

[0012] The present invention will be further described below with reference to the drawings.

[0013] A method for identifying the location of an indoor fire source based on infrared thermal images of the outer surface of a glass curtain wall is: S1, Data structure construction: This mainly involves steps including the construction of a fire dataset, temperature distribution of the outer surface of a glass curtain wall, image extraction, and processing. S2. Construction and Training of the Resnet Deep Learning Model: As shown in Figure 1, the Resnet-50 deep learning model includes an input layer, a convolutional layer, a max pooling layer, multiple sets of residual blocks, a global mean pooling layer, and a fully connected layer including a softmax function. The Resnet deep learning model is continuously optimized through each of these layers, divided into a training set and a validation set in a 7:3 ratio, and trained until both the validation accuracy and loss reach the setpoints, completing the training of the Resnet-50 deep learning model, as shown in Figure 2. Here, Figures 2(a) to (f) are loss value result figures and training and validation accuracy value result figures for three batches (each batch containing three epochs) in which the Resnet-50 deep learning model was trained using the method of the present invention, where a and b are result figures for one batch, c and d are result figures for one batch, and e and f are result figures for one batch. S3 includes a fire source location step using a Resnet deep learning model, as shown in Figures 3 and 4.

[0014] Example: A flashover prediction experiment is conducted based on an existing full-scale fire experiment setup. The FDS fire simulation model is matched to the actual full-scale fire experiment setup, a total of nine fire source locations are set, and multiple full-scale fire experiments are conducted. Twenty grayscale thermal images of the outer surface of glass are selected under experimental conditions with different fire source locations and input into a trained Resnet-50 deep learning model to obtain the following experimental data table.

[0015] [Table 1]

[0016] Through the above embodiments, the following conclusions are reached: In a test of 20 grayscale thermal images of the outer surface of glass, the prediction accuracy reached 95%. Therefore, the fire source location prediction technology, which integrates computational fluid dynamics, finite element analysis simulation technology, deep learning technology, and infrared thermal imaging technology, is applicable to fire situation reconnaissance missions in high-rise building fires. Compared to other fire source location methods, the present invention has high versatility, immediacy, and efficiency, while simultaneously possessing a high level of accuracy.

Claims

1. A method for identifying the location of an indoor fire source based on an infrared thermal image of the outer surface of a glass curtain wall, S1, Database construction: Steps including construction of a fire dataset, temperature distribution of the outer surface of the glass curtain wall, image extraction and processing, S2, Steps for building and training the Renet deep learning model, S3 includes a step of identifying the location of the fire source using a Resnet deep learning model, S1.

2. Using the ANSYS transient thermal computational fluid dynamics software, the temperature load on the inner surface of the glass curtain wall obtained from the FDS fire science computational fluid dynamics simulation software was analyzed, and a heat conduction simulation was performed to obtain the temperature distribution on the outer surface of the glass curtain wall. The specific process is as follows: S1.2.1, Using SpaceClean, construct a glass model with the same geometric dimensions as the FDS fire combustion model. S1.2.2 Import the constructed glass model into transient thermal and complete the import of the geometric structure. S1.2.

3. Input the main material parameters of the glass model, including density, thermal conductivity, specific heat, and thermal expansion coefficient, and use them for subsequent thermal conduction analysis. S1.2.4, Mesh division is performed, and the parameters of interest in mesh division include the physical preference of the element, the order of the element, the size of the element, the number of nodes, and the number of elements. S1.2.5, Set the initial temperature of the glass model and the initial ambient temperature T in the FDS fire combustion model. ０ To match, S1.2.6, Set the time length of the ANSI finite element analysis to match the time length t of the FDS simulation. S1.2.7, set the convective heat transfer coefficient on the outer surface of the glass model. S1.2.8, The glass curtain wall inner surface temperature data obtained by FDS simulation is input as a "temperature" load to the inner surface of the ANSYS glass model, and the spatial position of each input "temperature" load is matched with the spatial position of the temperature measurement point set in the FDS simulation. That is, the nth temperature measurement point data from the FDS simulation corresponds to the nth ANSYS temperature load input, and the nth temperature load input to the ANSYS glass model is given by the following equation: [Math 1] T ｎｔ This represents the temperature data at the t-second mark at the nth measurement point. S1.2.9, ANSYS finite element analysis was performed to 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 expressed by the following differential and integral equation: [Math 2] Here, ρ is the density of the glass model, c is the specific heat capacity of the glass model, T is a function of temperature that changes with time and space, and t ＊ is time, k is the thermal conductivity of the material, and Q is the heat output density of the internal heat source. initial time t ＊ The temperature distribution at = 0 is T(x, y, z, 0) = T ０ Therefore, when n × n temperature loads are applied to one side of the glass model, the boundary conditions are expressed as follows: [Math 3] Here, T 境界 x represents n × n temperature loads applied to the inner surface of the glass model, and x ｉ and y ｉ is the coordinate position of the temperature load on the inner surface of the glass model, and the convective heat transfer boundary conditions set on the outer surface of the glass model are expressed by the following equation: [Math 4] Here, ∂T / ∂n represents the temperature gradient in the normal direction of the glass model surface, h is the convective heat transfer coefficient, T ＊ is the temperature of the outer surface of the glass model, T ∞ is the temperature of the environmental fluid on the outer surface of the glass, and In the finite element solution process, the convective boundary conditions are discretized, and the solution equation is as follows: [Math 5] Here [C] ｃｏｎｖ {Q} 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 element, {K} is the stiffness matrix, which represents the thermal stiffness of the model system, i.e., the heat transfer resistance capacity of the glass model for a unit temperature change, {Q} is the load vector, which represents the applied temperature load, and {T} is the temperature that changes with time and space. S1.2.10, A method for determining the location of an indoor fire source based on an infrared thermal image of the outer surface of a glass curtain wall, characterized in that the color scheme for the temperature calculation results of the outer surface of the glass model is selected as "inverted grayscale".

2. The specific process of S1 is as follows: S1.

1. Fire data was simulated using the fire science computational fluid dynamics simulation software FDS, and a fire dataset was constructed. The specific process is as follows: S1.1.

1. Construct a fire combustion model: mesh parameters, fire chamber geometry, surface parameters, reaction parameters, initial ambient temperature T ０ By setting parameters including p ignition source locations, q ignition source heat release rates, f combustible material smoke generation rates, and l ventilation opening opening times, a total of p × q × f × l simulation conditions are set. S1.1.2, n × n temperature measurement points are set on the inner surface of the fire chamber glass in an equal-row, equal-column square matrix. S1.1.3, set the simulation time t and perform the simulation to obtain a dataset of n × n temperature measurement points {T ｎｔ Obtain} and T ｎｔ represents the temperature data at the t-second mark at the nth measurement point, and the time-temperature dataset for different temperature measurement points is given by the following equation: [Math 6] S1.1.4, The time-temperature dataset is integrated into a complete FDS simulation dataset and used as input data for subsequent finite element analysis. S1.3 Image extraction and processing: S1.3.1, Based on the number of frames in the ANSI finite element analysis results, extract the outer surface temperature distribution diagram of the glass model for all frames. S1.3.2, extract all glass model outer surface temperature distribution maps for the p×q×f×l set conditions, and construct an unprocessed raw image database. S1.3.

3. Gaussian blurring is applied to all images in the original image database, and the Gaussian filter for the two-dimensional image of the glass model outer surface temperature image is expressed as follows: [Number 7] Here, x and y are the horizontal and vertical distances between the image and the central pixel, respectively, G(x,y) is the output value of the Gaussian filter, which is the discretized form of the two-dimensional Gaussian function, and σ is the standard deviation. When performing Gaussian blurring, the convolution operation is continued on the image using a Gaussian filter, and for one glass model outer surface temperature distribution image I(x,y) and one Gaussian filter G(x,y), the convolution result I'(x,y) is given by the following equation: [Number 8] In the formula, G(i,j) is a two-dimensional Gaussian filter kernel, representing the weights for the distance from the central pixel. By weighting each pixel and its neighboring region, sharp edges and details of the image are reduced. S1.

4. Laplace sharpening is performed on the glass model outer surface temperature distribution image after Gaussian blurring. Laplace sharpening combines the Laplace operator with image addition, and the Laplace operator is a second-order differential operator, defined in two-dimensional space as follows: [Number 9] Here, f(x,y) is the pixel value of the image at point (x,y), and represents the Laplace operator. In discrete images, the Laplace operator is approximated by a filter, and a general Laplace filter is used to approximate the Laplace operator. [Number 10] The Laplace filter calculates the neighborhood difference of each pixel point in an image, highlighting the edges and details of the image. Laplace sharpening combines the original pixel values ​​of an image with the output of the Laplace operator, and its mathematical expression is as follows: [Math 11] Here, f(x,y) is the pixel value of the original image, and Δf(x,y) is the result of applying the Laplace operator to the image f(x,y). α is a constant parameter that controls the intensity of sharpening. The steps for Laplace sharpening of an image database are as follows: S1.4.1, Calculate the Laplace operator result of the image f(x,y) to obtain edge information in the image. S1.4.2, Select the control parameter α based on the sharpening intensity. S1.4.3, The original image and the result of the Laplace operator are combined to form the sharpened image f 鋭い Obtain (x, y), S1.4.4, A method for identifying an indoor fire source location based on an infrared thermal image of the outer surface of a glass curtain wall, characterized in that a fire source location label, i.e., a number from 0 to p, is assigned to each image in the image database after image processing, thereby completing the construction of a glass model outer surface temperature image database.

3. The specific process of S2 is as follows: S2.1, the Renet-50 deep learning model is adopted, and the Renet-50 deep learning model is trained using the glass model outer surface temperature distribution image database constructed in S1. Deep features are extracted from the temperature distribution images, and the model structure is as follows: S2.1.1, This layer adjusts the applicability of image size based on the image input from the image database. All images in the image database use an "inverted grayscale" color scheme during generation, and all images fed into training are grayscale images, and all input layers are single-channel. S2.1.2, This layer performs initial feature extraction and a convolutional layer reduces the spatial size of the output feature map, S2.1.3, A maximum pooling layer reduces the spatial dimension of the image, takes the maximum value in each local region, preserves the most prominent features, and makes feature extraction more robust to small changes in the input. S2.1.4, Each residual block includes an identity block and a convolutional block, used to avoid gradient vanishing and degeneration in deep neural networks, for feature mapping: [Math 12] Here, [Number 13] The residual is the difference between the features expected to be learned and the input features, and is generated by stacking convolutional layers. Residual learning involves feature mapping Φ(x). ＊ This is easier than learning it directly. [Number 14] Here, ζ is the loss function, and the output of the residual block is y. ＊ , ∂ζ / ∂x ＊ is input x ＊ The gradient of the loss with respect to is the direct path ∂ζ / ∂y through the residual connection. ＊ It directly includes multiple sets of residual blocks in which the gradient is directly transmitted from the later layer to the earlier layer, and gradient vanishing is avoided. S2.1.5, A global mean pooling layer that performs global dimensionality reduction on the feature map, maps high-dimensional spatial features to low-dimensional global features, outputs a single feature value, and provides input to the final fire source location classification task. S2.1.6 includes a fully connected layer and a softmax function: The output of the fully connected layer, after passing through the softmax function, outputs the category of the fire source identification task, i.e., the fire source location label, a number from 0 to p, where the softmax function is a normalization function used to map the real vector to a probability distribution, and the output vector of the fully connected layer z ｉ In contrast, the Softmax function uses the probability distribution P to represent it. ｉ Convert to satisfy the following equation: [Number 15] The gradient calculation and backpropagation of the Softmax function are based on the cross-entropy loss function: [Number 16] Here, γ is the cross-entropy loss, y ｉ This is the target category, where 1 indicates the correct ignition source location, and 0 indicates any other ignition source location. The gradient of the Softmax output is given by the following equation: [Number 17] The output gradient is used for backpropagation, enabling continuous optimization of the model output results. S2.2, A method for determining the location of an indoor fire source based on an infrared thermal image of the outer surface of a glass curtain wall, as described in claim 1 or 2, characterized in that the image database is divided proportionally into a training set and a validation set, the Renesnet-50 deep learning model is trained, the number of training iterations is adjusted until both the validation accuracy and loss reach set values, and the training of the Renesnet-50 deep learning model is completed.

4. The specific process of S3 is as follows: S3.

1. Select a grayscale color scheme with the infrared thermal imaging camera, point the lens of the infrared thermal imaging camera towards the outer surface of the glass in the fire room, and with the camera facing directly towards the center of the glass, and collect a grayscale infrared thermal image. If the infrared thermal imaging camera does not have a grayscale color scheme, first convert the image to a grayscale image before performing the subsequent operations. The pixel points of the color image are composed of R, G, and B of the RGB channels, and the grayscale value Y corresponding to the pixel point is given by the following formula: [Number 18] Here, lol Ｒ w is the weighting coefficient of the R channel. Ｇ w is the weighting coefficient of the G channel. Ｂ This is the weighting coefficient for the B channel, S3.2 The grayscale infrared thermal image acquired by the infrared thermal imaging camera is input into the Resnet-50 deep learning model trained in S2. S3.3, A method for determining the location of an indoor fire source based on an infrared thermal image of the outer surface of a glass curtain wall, characterized in that the Resnet-50 deep learning model outputs a label for the fire source location, i.e., a number from 0 to p, thereby determining the fire source location.

Images