Gyon deflagration fire prediction system based on target area monitoring

Abstract

The invention relates to the technical field of deflagration fire prediction, and discloses a canyon deflagration fire prediction system based on target area monitoring, and the system comprises a sub-area division module which divides a target area into M sub-areas; the data acquisition module is used for acquiring monitoring data of each sub-region after the fire disaster occurs according to a preset time interval t in a first preset time period T; the data processing module is used for preprocessing the monitoring data to construct a feature matrix; the deflagration fire prediction module inputs the characteristic matrix into a deflagration fire prediction model and an early warning measure generation module; the method comprises the following steps: dividing sub-regions, collecting remote sensing images and meteorological data in real time, and combining atmospheric correction and data normalization processing to construct a feature matrix; a multi-layer neural network model is utilized to accurately predict the occurrence probability of detonation fire, and early warning measures are generated according to the risk level; according to the system, the accuracy and timeliness of detonation fire prediction are improved, and the loss caused by detonation fire is reduced.

Description

Technical Field

[0001] The present invention relates to the field of deflagration fire prediction, and more particularly, to a canyon deflagration fire prediction system based on target area monitoring. Background Art

[0002] The "deflagration" phenomenon in forest fires is an extreme combustion behavior characterized by a sudden high-intensity forest fire burning in a short period of time, spreading rapidly at a speed of several meters per second or even faster. This phenomenon has a significant impact in forest fires because it can cause huge casualties and losses.

[0003] The existing methods for predicting deflagration fires have the following problems: one existing prediction method includes using drones to detect the gases generated by the combustion of the fire for prediction, but due to the influence of factors such as topography, the prediction accuracy of the deflagration fire is low; a feasible prediction method includes using ANSYS (finite element analysis) software for high-precision prediction. ANSYS software can simulate the process of fire spread through its fluid dynamics module, and then predict the time and range of the deflagration; however, the simulation of fire spread and deflagration usually requires high-precision grids and complex physical models, which places high demands on computing resources and takes too long to simulate. Summary of the invention

[0004] The present invention provides a canyon deflagration fire prediction system based on target area monitoring, which solves the problems raised in the background technology.

[0005] The present invention provides a canyon deflagration fire prediction system based on target area monitoring, comprising:

[0006] A sub-region division module, which is used to divide the target region into M sub-regions;

[0007] A data collection module, which is used to collect monitoring data of each sub-area after a fire occurs within a first preset time period T and at a preset time interval t;

[0008] Monitoring data include: remote sensing images and meteorological data;

[0009] Meteorological data include: temperature, humidity, air pressure, wind speed, wind direction, rainfall, nitrogen dioxide gas concentration, ozone gas concentration, PM1.0 concentration, PM2.5 concentration, PM10 concentration;

[0010] A data processing module is used to pre-process the monitoring data and construct a feature matrix;

[0011] The element in the jth column of the i-th row of the feature matrix represents the preprocessed monitoring data of the i-th sub-region at the jth time point, where 1≤i≤M, 1≤j≤N, N=T / t;

[0012] A deflagration fire prediction module, which is used to input the feature matrix into the deflagration fire prediction model, and the output value represents the probability value of the deflagration fire occurring in the target area within the second preset time period G;

[0013] The early warning measure generation module is used to generate early warning measures according to the probability value of the target area within the second preset time period G.

[0014] Furthermore, M, T, t, and G are all custom parameters.

[0015] Furthermore, the monitoring data is preprocessed to construct a feature matrix, including the following steps:

[0016] Step S201, filling missing values ​​in the meteorological data with the average value of the meteorological data of two adjacent time points corresponding to the missing value;

[0017] Step S202, performing atmospheric correction on the remote sensing image through an atmospheric correction model, wherein the atmospheric correction model is an ATCOR atmospheric correction model;

[0018] Step S303: normalize the meteorological data using the Min-Max method.

[0019] Further, the deflagration fire prediction model includes: M hidden layers, each hidden layer includes: N first units, N splicing units and N second units;

[0020] N first units share weight parameters and bias parameters;

[0021] The jth first unit of the i-th hidden layer inputs the preprocessed remote sensing image of the i-th sub-region of the feature matrix at the j-th time point, outputs the remote sensing image features, and expands them into a vector representation;

[0022] The jth concatenation unit of the i-th hidden layer is used to concatenate the vector representation of the remote sensing image feature output by the j-th first unit of the i-th hidden layer to the preprocessed meteorological data of the i-th sub-region of the feature matrix at the j-th time point to obtain a combined vector;

[0023] The j-th second unit of the i-th hidden layer inputs the combined vector output by the j-th concatenated unit of the i-th hidden layer and outputs the updated feature;

[0024] The updated features output by the Nth second unit of the Mth hidden layer are input into the first classifier, and the classification space of the first classifier represents the probability value of the occurrence of a deflagration fire in the target area within the second preset time period G.

[0025] Furthermore, the first unit is constructed based on the CNN convolutional neural network model.

[0026] Furthermore, the calculation formula of the j-th second unit of the i-th hidden layer includes:

[0027]

[0028] Among them, 1≤j≤M, 1≤i≤N, represents the updated features of the output of the jth second unit of the i-th hidden layer, represents the combined vector of the output of the jth concatenated unit of the i-th hidden layer and the input of the jth second unit of the i-th hidden layer, and They represent the forget gate, input gate, candidate cell state, cell state and output gate of the jth second unit of the i-th hidden layer, respectively. represents the cell state of the j-1th second unit of the i-th hidden layer, Assign a value of 0, and They represent the first weight parameter, the second weight parameter and the bias parameter of the forget gate of the jth second unit of the i-th hidden layer, respectively. and They respectively represent the first weight parameter, the second weight parameter and the bias parameter of the input gate of the jth second unit of the i-th hidden layer, and They respectively represent the first weight parameter, the second weight parameter and the bias parameter of the candidate cell state of the jth second unit of the i-th hidden layer, and denote the first weight parameter, the second weight parameter and the bias parameter of the output gate of the jth second unit of the i-th hidden layer, respectively. ⊙ denotes point-by-point multiplication. sigmoid denotes the sigmoid activation function. tanh denotes the hyperbolic tangent activation function. and They represent the updated features and the number of dimensions of the updated features output by the j-1th second unit of the i-th hidden layer, respectively. and They represent the updated features and the number of dimensions of the updated features output by the jth second unit of the i-1th hidden layer, respectively. and All are assigned a value of 0, and MLP stands for multi-layer perceptron.

[0029] Further, the collected monitoring data are equally divided into Q groups in chronological order, where Q is a custom parameter, preferably, Q is set to 20, and each group includes 10 continuous monitoring data; the monitoring data is input into the simulation platform, and the fire scenario simulation is run to generate the result of the fire occurring in the target area within the second preset time period G; according to the simulation results, a classification label for training the deflagration fire prediction model is generated; if the simulation results show that a deflagration fire occurs within the second preset time period G, the corresponding classification label is set to 1; if a deflagration fire does not occur, the classification label is set to 0, and the sample label of the training sample for training the deflagration fire prediction model is obtained; the deflagration fire prediction model inputs each group of data in sequence and outputs 10 prediction results S, and calculates the risk level P, and the calculation formula is as follows:

[0030]

[0031] Among them, S T It indicates the number of consistent results between the output of the deflagration fire prediction model and the output of the simulation platform.

[0032] Furthermore, the first unit is pre-trained before the deflagration fire prediction model is trained. During the pre-training process, the feature map output by the first unit is expanded into a vector representation and then input into the second classifier. The classification space of the second classifier represents the area of ​​the fire in the target area.

[0033] Furthermore, the probability value of the target area within the second preset time period G is divided into three levels. The probability value of 0 to 0.3 is defined as a low risk level, and no early warning measures are taken at the low risk level; the probability value of 0.3 to 0.7 is defined as a medium risk level, and the early warning measures for the medium risk level are to immediately send early warning text messages and emails to relevant responsible persons; the probability value of 0.7 to 1 is defined as a high risk level, and the early warning measures for the high risk level are to immediately evacuate people in the target area and prohibit entry, send an alarm signal through the public broadcasting system in the target area, and simultaneously send early warning text messages and emails to relevant responsible persons.

[0034] Furthermore, the cross entropy loss function is used to evaluate the performance of the deflagration fire prediction model; the cross entropy loss function formula is as follows:

[0035]

[0036] Where s represents the number of samples; y i Represents the true label of the i-th sample, with a value of 0 or 1; Represents the predicted probability of the i-th sample, with a value between 0 and 1

[0037] The beneficial effects of the present invention are as follows: the present invention constructs a feature matrix by dividing sub-areas, collecting remote sensing images and meteorological data in real time, and combining atmospheric correction and data normalization processing. The multi-layer neural network model is used to accurately predict the probability of occurrence of deflagration fire, and early warning measures are generated according to the risk level. This system improves the accuracy and timeliness of deflagration fire prediction and reduces the losses caused by deflagration fire. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic diagram of a canyon deflagration fire prediction system based on target area monitoring of the present invention;

[0039] Figure 2 It is a flow chart of the present invention for preprocessing monitoring data to generate a feature matrix.

[0040] In the figure: a sub-area division module 101, a data acquisition module 102, a data processing module 103, a deflagration fire prediction module 104, and a prediction measure generation module 105. DETAILED DESCRIPTION

[0041] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0042] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0043] like Figure 1-2 As shown, a canyon flash fire prediction system based on target area monitoring includes:

[0044] A sub-region division module 101 is used to divide the target region into M sub-regions;

[0045] A data collection module 102, which is used to collect monitoring data of each sub-area after a fire occurs within a first preset time period T and at a preset time interval t;

[0046] Monitoring data include: remote sensing images and meteorological data;

[0047] Meteorological data include: temperature, humidity, air pressure, wind speed, wind direction, rainfall, nitrogen dioxide gas concentration, ozone gas concentration, PM1.0 concentration, PM2.5 concentration, PM10 concentration;

[0048] The data processing module 103 is used to pre-process the monitoring data and construct a feature matrix;

[0049] The element in the jth column of the i-th row of the feature matrix represents the preprocessed monitoring data of the i-th sub-region at the jth time point, where 1≤i≤M, 1≤j≤N, N=T / t;

[0050] A deflagration fire prediction module 104, which is used to input the feature matrix into the deflagration fire prediction model, and the output value represents the probability value of the deflagration fire occurring in the target area within the second preset time period G;

[0051] The early warning measure generation module is used to generate early warning measures according to the probability value of the target area within the second preset time period G.

[0052] In one embodiment of the present invention, M, T, t and G are all custom parameters. Preferably, M is set according to the area of ​​the target area. For example, if M is set to 20, T is set to 30 minutes, and t is set to 2 minutes, then N is 15 and G is set to 5 minutes.

[0053] In one embodiment of the present invention, Figure 2 As shown, the monitoring data is preprocessed to construct a feature matrix, including the following steps:

[0054] Step S201, filling missing values ​​in the meteorological data with the average value of the meteorological data of two adjacent time points corresponding to the missing value;

[0055] Step S202, performing atmospheric correction on the remote sensing image through an atmospheric correction model, where the atmospheric correction model may be an ATCOR atmospheric correction model;

[0056] Step S303: normalize the meteorological data using the Min-Max method.

[0057] In one embodiment of the present invention, the deflagration fire prediction model includes: M hidden layers, each hidden layer includes: N first units, N splicing units and N second units;

[0058] N first units share weight parameters and bias parameters;

[0059] The jth first unit of the i-th hidden layer inputs the preprocessed remote sensing image of the i-th sub-region of the feature matrix at the j-th time point, outputs the remote sensing image features, and expands them into a vector representation;

[0060] The jth concatenation unit of the i-th hidden layer is used to concatenate the vector representation of the remote sensing image feature output by the j-th first unit of the i-th hidden layer to the preprocessed meteorological data of the i-th sub-region of the feature matrix at the j-th time point to obtain a combined vector;

[0061] The j-th second unit of the i-th hidden layer inputs the combined vector output by the j-th concatenated unit of the i-th hidden layer and outputs the updated feature;

[0062] The updated features output by the Nth second unit of the Mth hidden layer are input into the first classifier, and the classification space of the first classifier represents the probability value of the occurrence of a deflagration fire in the target area within the second preset time period G.

[0063] In one embodiment of the present invention, the first unit is constructed based on CNN (Convolutional Neural Network model).

[0064] In one embodiment of the present invention, the calculation formula of the j-th second unit of the i-th hidden layer includes:

[0065]

[0066]

[0067] Where 1≤j≤M, 1≤i≤N, represents the updated features of the output of the jth second unit of the i-th hidden layer, represents the combined vector of the output of the jth concatenated unit of the i-th hidden layer and the input of the jth second unit of the i-th hidden layer, and They represent the forget gate, input gate, candidate cell state, cell state and output gate of the jth second unit of the i-th hidden layer, respectively. represents the cell state of the j-1th second unit of the i-th hidden layer, Assign a value of 0, and They represent the first weight parameter, the second weight parameter and the bias parameter of the forget gate of the jth second unit of the i-th hidden layer, respectively. and They respectively represent the first weight parameter, the second weight parameter and the bias parameter of the input gate of the jth second unit of the i-th hidden layer, and They respectively represent the first weight parameter, the second weight parameter and the bias parameter of the candidate cell state of the jth second unit of the i-th hidden layer, and denote the first weight parameter, the second weight parameter and the bias parameter of the output gate of the jth second unit of the i-th hidden layer, respectively. ⊙ denotes point-by-point multiplication. sigmoid denotes the sigmoid activation function. tanh denotes the hyperbolic tangent activation function. and They represent the updated features and the number of dimensions of the updated features output by the j-1th second unit of the i-th hidden layer, respectively. and They represent the updated features and the number of dimensions of the updated features output by the jth second unit of the i-1th hidden layer, respectively. and All are assigned a value of 0, and MLP stands for multi-layer perceptron.

[0068] In one embodiment of the present invention, the collected monitoring data is input into a simulation platform; a fire scenario simulation is run to generate the result of a fire occurring in a target area within a second preset time period G; based on the simulation results, a classification label for training a deflagration fire prediction model is generated; if the simulation results show that a deflagration fire occurs within the second preset time period G, the corresponding classification label is set to 1; if a deflagration fire does not occur, the classification label is set to 0, and sample labels for training samples for training a deflagration fire prediction model are obtained.

[0069] The collected monitoring data are equally divided into Q groups in chronological order, where Q is a custom parameter. Preferably, Q is set to 20, and each group includes 10 continuous monitoring data; the monitoring data is input into the simulation platform, and the fire scenario simulation is run to generate the result of the fire occurring in the target area within the second preset time period G; according to the simulation results, a classification label for training the deflagration fire prediction model is generated; if the simulation results show that a deflagration fire occurs within the second preset time period G, the corresponding classification label is set to 1; if a deflagration fire does not occur, the classification label is set to 0, and the sample label of the training sample for training the deflagration fire prediction model is obtained; the deflagration fire prediction model inputs each group of data in sequence and outputs 10 prediction results S, and calculates the risk level P, and the calculation formula is as follows:

[0070]

[0071] Among them, S TIt indicates the number of consistent results between the output of the deflagration fire prediction model and the output of the simulation platform.

[0072] In one embodiment of the present invention, the first unit is pre-trained before the deflagration fire prediction model is trained. During the pre-training process, the feature map output by the first unit is expanded into a vector representation and then input into the second classifier. The classification space of the second classifier represents the area of ​​the fire in the target area.

[0073] In one embodiment of the present invention, the first unit is pre-trained before the deflagration fire prediction model is trained. During the pre-training process, the feature map output by the first unit is expanded into a vector representation and then input into the third classifier. The classification space of the third classifier represents the longitude and latitude of the center point of the fire in the target area.

[0074] It should be noted that pre-training can speed up the training of the model, extract the key features of the image, and significantly reduce the computational burden and time cost of custom training; and when the new task has something in common with the initial task, through the transfer learning strategy, the features extracted by pre-training can be fully utilized to achieve a significant leap in performance in the new task scenario; in addition, the parameters of the pre-trained model have a strong generalization ability for common patterns due to the experience of being trained on massive data sets. Therefore, in the process of fine-tuning the new data set, the model is more focused on capturing subtle changes unique to the task rather than reconstructing global features. This mechanism effectively reduces the risk of overfitting.

[0075] In one embodiment of the present invention, the probability value of the target area within the second preset time period G is divided into three levels, and the probability value of 0 to 0.3 is defined as a low risk level, and no early warning measures are taken at the low risk level; the probability value of 0.3 to 0.7 is defined as a medium risk level, and the early warning measures for the medium risk level are to immediately send early warning text messages and emails to relevant responsible persons; the probability value of 0.7 to 1 is defined as a high risk level, and the early warning measures for the high risk level are to immediately evacuate the personnel in the target area and prohibit entry, send an alarm signal through the public broadcasting system in the target area, and simultaneously send early warning text messages and emails to relevant responsible persons.

[0076] In one embodiment of the present invention, the loss function of the risk scoring model adopts a cross entropy loss function; the cross entropy loss function formula is as follows:

[0077]

[0078] Where: S represents the number of samples; y i Represents the true label of the i-th sample, with a value of 0 or 1; Represents the predicted probability of the i-th sample, with a value between 0 and 1.

[0079] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are protected by the present embodiment.

Claims

1. A canyon fire prediction system based on target area monitoring, characterized in that: include: A sub-region division module, which is used to divide the target region into M sub-regions; A data collection module, which is used to collect monitoring data of each sub-area after a fire occurs within a first preset time period T and at a preset time interval t; Monitoring data include: remote sensing images and meteorological data; Meteorological data include: temperature, humidity, air pressure, wind speed, wind direction, rainfall, nitrogen dioxide gas concentration, ozone gas concentration, PM1.0 concentration, PM2.5 concentration, PM10 concentration; A data processing module is used to pre-process the monitoring data and construct a feature matrix; The element in the jth column of the i-th row of the feature matrix represents the preprocessed monitoring data of the i-th sub-region at the jth time point, where 1≤i≤M, 1≤j≤N, N=T / t; A deflagration fire prediction module, which is used to input the feature matrix into the deflagration fire prediction model, and the output value represents the probability value of the deflagration fire occurring in the target area within the second preset time period G; The early warning measure generation module is used to generate early warning measures according to the probability value of the target area within the second preset time period G.

2. The canyon deflagration fire prediction system based on target area monitoring according to claim 1 is characterized in that: M, T, t, and G are custom parameters.

3. The canyon deflagration fire prediction system based on target area monitoring according to claim 1, characterized in that: Preprocessing the monitoring data to construct a feature matrix includes the following steps: Step S201, filling missing values ​​in the meteorological data with the average value of the meteorological data of two adjacent time points corresponding to the missing value; Step S202, performing atmospheric correction on the remote sensing image through an atmospheric correction model; Step S303: normalize the meteorological data using the Min-Max method.

4. The canyon deflagration fire prediction system based on target area monitoring according to claim 1, characterized in that: The deflagration fire prediction model includes: M hidden layers, each hidden layer includes: N first units, N splicing units and N second units; N first units share weight parameters and bias parameters; The jth first unit of the i-th hidden layer inputs the preprocessed remote sensing image of the i-th sub-region of the feature matrix at the j-th time point, outputs the remote sensing image features, and expands them into a vector representation; The jth concatenation unit of the i-th hidden layer is used to concatenate the vector representation of the remote sensing image feature output by the j-th first unit of the i-th hidden layer to the preprocessed meteorological data of the i-th sub-region of the feature matrix at the j-th time point to obtain a combined vector; The j-th second unit of the i-th hidden layer inputs the combined vector output by the j-th concatenated unit of the i-th hidden layer and outputs the updated feature; The updated features output by the Nth second unit of the Mth hidden layer are input into the first classifier, and the classification space of the first classifier represents the probability value of the occurrence of a deflagration fire in the target area within the second preset time period G.

5. The canyon deflagration fire prediction system based on target area monitoring according to claim 4 is characterized in that: The first unit is built based on the CNN convolutional neural network model.

6. The canyon deflagration fire prediction system based on target area monitoring according to claim 4, characterized in that: The calculation formula for the jth second unit of the i-th hidden layer is include: Among them, 1≤j≤M, 1≤i≤N, represents the updated features of the output of the jth second unit of the i-th hidden layer, represents the combined vector of the output of the jth concatenated unit of the i-th hidden layer and the input of the jth second unit of the i-th hidden layer, and They represent the forget gate, input gate, candidate cell state, cell state and output gate of the jth second unit of the i-th hidden layer, respectively. represents the cell state of the j-1th second unit of the i-th hidden layer, Assign a value of 0, and They represent the first weight parameter, the second weight parameter and the bias parameter of the forget gate of the jth second unit of the i-th hidden layer, respectively. and They respectively represent the first weight parameter, the second weight parameter and the bias parameter of the input gate of the jth second unit of the i-th hidden layer, and They respectively represent the first weight parameter, the second weight parameter and the bias parameter of the candidate cell state of the jth second unit of the i-th hidden layer, and denote the first weight parameter, the second weight parameter and the bias parameter of the output gate of the jth second unit of the i-th hidden layer, respectively. ⊙ denotes point-by-point multiplication. sigmoid denotes the sigmoid activation function. tanh denotes the hyperbolic tangent activation function. and They represent the updated features and the number of dimensions of the updated features output by the j-1th second unit of the i-th hidden layer, respectively. and They represent the updated features and the number of dimensions of the updated features output by the jth second unit of the i-1th hidden layer, respectively. and All are assigned a value of 0, and MLP stands for multi-layer perceptron.

7. The canyon deflagration fire prediction system based on target area monitoring according to claim 1, characterized in that: The collected monitoring data are equally divided into Q groups in chronological order, where Q is a custom parameter and each group includes 10 continuous monitoring data; the monitoring data are input into the simulation platform, and the fire scenario simulation is run to generate the result of the fire occurring in the target area within the second preset time period G; According to the simulation results, a classification label for training the deflagration fire prediction model is generated; if the simulation results show that a deflagration fire occurs within the second preset time period G, the corresponding classification label is set to 1; if a deflagration fire does not occur, the classification label is set to 0, and the sample labels of the training samples for training the deflagration fire prediction model are obtained; the deflagration fire prediction model inputs each set of data in sequence and outputs 10 prediction results S, and calculates the risk level P, and the calculation formula is as follows: Among them, S T It indicates the number of consistent results between the output of the deflagration fire prediction model and the output of the simulation platform.

8. The canyon fire prediction system based on target area monitoring according to claim 4, characterized in that: Before training the flash fire prediction model, the first unit is pre-trained. During the pre-training process, the feature map output by the first unit is expanded into a vector representation and then input into the second classifier. The classification space of the second classifier represents the area of ​​the fire in the target area.

9. The canyon deflagration fire prediction system based on target area monitoring according to claim 1, characterized in that: The probability value of the target area in the second preset time period G is divided into three levels, and the probability value of 0 to 0.3 is defined as a low risk level, and no early warning measures are taken at the low risk level; A probability value of 0.3 to 0.7 is defined as a medium risk level. The early warning measures for the medium risk level are to immediately send early warning text messages and emails to the relevant responsible persons; a probability value of 0.7 to 1 is defined as a high risk level. The early warning measures for the high risk level are to immediately evacuate people in the target area and prohibit entry, send an alarm signal through the public broadcasting system in the target area, and simultaneously send early warning text messages and emails to the relevant responsible persons.

10. The canyon deflagration fire prediction system based on target area monitoring according to claim 1, characterized in that: The cross entropy loss function is used to evaluate the performance of the deflagration fire prediction model; the cross entropy loss function formula is as follows: Where S represents the number of samples; y i Represents the true label of the i-th sample, with a value of 0 or 1; Represents the predicted probability of the i-th sample, with a value between 0 and 1.

Images