Fire identification method, system, device and medium for charging pile scenarios

Abstract

The present invention relates to the field of fire identification technology, and discloses a fire identification method, system, device, and medium for use in a charging pile scenario, comprising obtaining time-series images and time-series environmental parameters of a charging pile area to be identified, wherein the time-series images include bitmap images and infrared images, and the time-series environmental parameters include time-series smoke data; inputting the bitmap images into a target recognition model to obtain a target image, and selecting an infrared target image from the infrared images based on the target image; extracting temperature data from the infrared target images of each infrared image to obtain time-series temperature data; inputting the target image, the smoke data, and the temperature data into a fire recognition model to obtain a fire identification result for the charging pile area. The present invention can accurately and efficiently identify flames and charging piles through the target recognition model, and through the fire recognition model, fuses multiple features, thereby improving the accuracy of model recognition and further ensuring the safety of the charging environment.

Description

Technical Field

[0001] The present invention relates to the field of fire identification technology, and in particular to a fire identification method, system, device and medium for use in a charging pile scenario. Background Art

[0002] With the development of electric vehicles, electric cars and electric bicycles have become a large part of people's daily lives. Due to the characteristics of electric cars, they need to be charged frequently, which is why charging piles have emerged. Currently, charging piles on the market are generally installed in basements, carports, and parking lots. Due to the large number of wiring in existing charging piles and the high charging current when charging electric vehicles, the equipment is prone to aging and damage. Furthermore, most charging piles are generally located in open-air environments, making it difficult to avoid high temperatures, low temperatures, and humid working environments. This makes charging piles extremely prone to fires. Furthermore, most charging piles are deployed in relatively remote locations or are unmanned. Once a fire occurs, timely warning and disposal are difficult, and the fire develops and spreads quickly, resulting in serious consequences and significant economic losses.

[0003] Existing charging stations typically integrate temperature sensors to monitor the vehicle's temperature during charging. However, temperature sensors alone cannot monitor smoke, sparks, and other potential fire hazards that may be generated during charging, making it difficult to detect fire hazards immediately. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a fire identification method, system, device and medium for use in a charging pile scenario, so as to solve the problem of untimely discovery of fire hazards and achieve the effect of accurately and efficiently identifying fire hazards in charging piles.

[0005] In a first aspect, the present invention provides a fire identification method for a charging pile scenario, the method comprising:

[0006] Acquire time-series images and time-series environmental parameters of the charging pile area to be identified, wherein the time-series images include a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data;

[0007] Inputting the bitmap image into a target recognition model to obtain a target image, and selecting a corresponding infrared target image from the infrared image based on the target image;

[0008] extracting temperature data from the infrared target image of each of the infrared images, and obtaining time series temperature data based on the extracted temperature data;

[0009] The target image, the time-series smoke data, and the time-series temperature data are input into a fire recognition model to obtain a fire recognition result of the charging pile area.

[0010] Furthermore, the step of inputting the bitmap image into the target recognition model to obtain the target image includes:

[0011] Preprocessing the bitmap data, and inputting the preprocessed bitmap data into a pre-trained target recognition model to recognize the flame and the charging pile to obtain a target image, wherein the target image includes a flame image and a charging pile image;

[0012] Among them, the target recognition model is constructed using the YOLO model based on GhostNet, and an attention module and a multi-layer receptive field module are set in the backbone network of the YOLO model. The attention module is a hybrid attention module based on the convolutional block attention mechanism and the coordinate attention mechanism.

[0013] Furthermore, the step of extracting temperature data from the infrared target image of each of the infrared images and obtaining time series temperature data based on the extracted temperature data includes:

[0014] Obtaining calibration data of an infrared camera corresponding to the infrared image, and extracting pixel values ​​corresponding to an infrared target image of the infrared image;

[0015] Converting the pixel values ​​into radiation intensity values ​​according to the calibration data, and converting the radiation intensity values ​​into temperature data according to Planck's law;

[0016] Time series temperature data is obtained according to the temperature data corresponding to each of the infrared images.

[0017] Furthermore, the fire recognition model is constructed using a neural network model, including a first recurrent neural network model, a second recurrent neural network model, a convolutional neural network model, a feature fusion layer and a fully connected layer;

[0018] The input data of the first recurrent neural network model is time series smoke data, and the output data of the first recurrent neural network model is smoke feature data;

[0019] The input data of the second recurrent neural network model is time series temperature data, and the output data of the second recurrent neural network model is temperature feature data;

[0020] The input data of the convolutional neural network model is the target image, and the output data of the convolutional neural network model is image feature data;

[0021] The input data of the feature fusion layer is the smoke feature data, temperature feature data and image feature data, and the output data of the feature fusion layer is fused feature data;

[0022] The input data of the fully connected layer is the fused feature data, and the output data of the fully connected layer is the fire recognition result.

[0023] In a second aspect, the present invention provides a fire identification system for a charging pile scenario, the system comprising:

[0024] A data acquisition module, configured to acquire time-series images and time-series environmental parameters of the charging pile area to be identified, wherein the time-series images include a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data;

[0025] a target recognition module, configured to input the bitmap image into a target recognition model to obtain a target image, and select a corresponding infrared target image from the infrared image based on the target image;

[0026] a temperature acquisition module, configured to extract temperature data from the infrared target image of each of the infrared images, and obtain time series temperature data based on the extracted temperature data;

[0027] The fire identification module is used to input the target image, the time-series smoke data and the time-series temperature data into a fire identification model to obtain a fire identification result of the charging pile area.

[0028] Furthermore, the target recognition module is further configured to pre-process the bitmap data and input the pre-processed bitmap data into a pre-trained target recognition model to recognize the flame and the charging pile to obtain a target image, wherein the target image includes a flame image and a charging pile image;

[0029] Among them, the target recognition model is constructed using the YOLO model based on GhostNet, and an attention module and a multi-layer receptive field module are set in the backbone network of the YOLO model. The attention module is a hybrid attention module based on the convolutional block attention mechanism and the coordinate attention mechanism.

[0030] Furthermore, the temperature acquisition module is further configured to acquire calibration data of an infrared camera corresponding to the infrared image, and extract pixel values ​​corresponding to the infrared target image of the infrared image;

[0031] Converting the pixel values ​​into radiation intensity values ​​according to the calibration data, and converting the radiation intensity values ​​into temperature data according to Planck's law;

[0032] Time series temperature data is obtained according to the temperature data corresponding to each of the infrared images.

[0033] Furthermore, the fire identification module also includes a fire identification model, which is constructed using a neural network model, including a first recurrent neural network model, a second recurrent neural network model, a convolutional neural network model, a feature fusion layer and a fully connected layer;

[0034] The input data of the first recurrent neural network model is time series smoke data, and the output data of the first recurrent neural network model is smoke feature data;

[0035] The input data of the second recurrent neural network model is time series temperature data, and the output data of the second recurrent neural network model is temperature feature data;

[0036] The input data of the convolutional neural network model is the target image, and the output data of the convolutional neural network model is image feature data;

[0037] The input data of the feature fusion layer is the smoke feature data, temperature feature data and image feature data, and the output data of the feature fusion layer is fused feature data;

[0038] The input data of the fully connected layer is the fused feature data, and the output data of the fully connected layer is the fire recognition result.

[0039] In a third aspect, an embodiment of the present invention further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0040] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0041] The present invention provides a fire identification method, system, device, and medium for use in charging pile scenarios. Through a target recognition model, the present invention can accurately and efficiently identify flames and charging piles, and based on the recognition results, obtain corresponding temperature data, thereby providing accurate data support for subsequent data analysis. Furthermore, through a multi-input fire recognition model, the present invention integrates multiple features, effectively improving the accuracy of model recognition. The identification method provided by the present invention can accurately and efficiently identify fire hazards in charging piles, further ensuring the safety of the charging environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 1 is a flow chart of a fire identification method according to an embodiment of the present invention;

[0043] Figure 2 Schematic diagram of the structure of the target recognition model in an embodiment of the present invention;

[0044] Figure 3 yes Figure 2 Schematic diagram of the module structure of the target recognition model;

[0045] Figure 4 yes Figure 2 Schematic diagram of the structure of the attention module of the target recognition model;

[0046] Figure 5 is a structural diagram of a fire identification system according to an embodiment of the present invention;

[0047] Figure 6 1 is a diagram showing the internal structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0049] See also Figure 1 The first embodiment of the present invention proposes a fire identification method for a charging pile scenario, which includes steps S10 to S40:

[0050] Step S10, obtaining time-series images and time-series environmental parameters of the charging pile area to be identified, wherein the time-series images include a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data;

[0051] Step S20, inputting the bitmap image into a target recognition model to obtain a target image, and selecting a corresponding infrared target image from the infrared images based on the target image;

[0052] Step S30, extracting temperature data from the infrared target image of each of the infrared images, and obtaining time series temperature data based on the extracted temperature data;

[0053] Step S40: Input the target image, the time-series smoke data, and the time-series temperature data into a fire recognition model to obtain a fire recognition result of the charging pile area.

[0054] The present invention adopts a method based on the combination of image analysis and data analysis to identify fire hazards in charging piles. The images used include bitmap images extracted from surveillance videos and infrared images taken by infrared cameras. The data uses environmental parameters of the charging pile area. Specifically, smoke data collected by smoke sensors can be used. In order to reduce the amount of calculation and improve computing efficiency, the bitmap images and infrared images used in the present invention are in a corresponding relationship, and each bitmap image corresponds to an infrared image.

[0055] In the actual application scenario of charging piles, fire hazard areas can be divided into two types: one is a fire caused by the combustion of the charging pile itself, and the other is a fire caused by the presence of combustible materials in the area around the charging pile. In order to accurately identify multiple fire hazards, in a preferred embodiment, first, images within the range of the charging pile area are identified and screened. The specific steps include:

[0056] Preprocessing the bitmap data, and inputting the preprocessed bitmap data into a pre-trained target recognition model to recognize the flame and the charging pile to obtain a target image, wherein the target image includes a flame image and a charging pile image;

[0057] Among them, the target recognition model is constructed using the YOLO model based on GhostNet, and an attention module and a multi-layer receptive field module are set in the backbone network of the YOLO model. The attention module is a hybrid attention module based on the convolutional block attention mechanism and the coordinate attention mechanism.

[0058] In this embodiment, the images of the charging pile area used are a group of images sorted in time series, also known as time series images. The following takes a single image in the time series image as an example to illustrate the target recognition steps in this embodiment.

[0059] For a single bitmap image, image preprocessing is first performed, including operations such as cropping, padding, and normalization. Then, the preprocessed bitmap image is input into a pre-established target recognition model for target recognition. Target recognition includes fire target recognition and charging pile target recognition, thereby obtaining target images, namely flame images and charging pile images.

[0060] The target recognition model in this embodiment is built using the YOLOv5 model based on GhostNet. Figure 2The target recognition model is mainly composed of the backbone BackBone, neck and head. The BackBone part is composed of the GhostCBS module and the GSP1 module, which is mainly used for sufficient feature extraction. The GhostCBS module replaces the traditional convolution Conv module to achieve a significant lightweight backbone network. At the same time, the CBAM-CA attention mechanism and ASPP multi-layer receptive field are added to the tail of the backbone network to increase the feature extraction ability of the backbone network; the Neck part is composed of the feature pyramid network and the path aggregation network, which reduces the number of model parameters while improving the feature fusion ability of the model. The Head part is composed of 3 detectors, whose function is to complete the detection of target position and category.

[0061] The module structure of the target recognition model is as follows Figure 3 As shown, specifically, the GhostCBS module consists of Ghost Conv (Ghost convolution), BN (normalization) and SILU (activation function). The Ghost convolution unit extracts the preliminary feature map through the standard convolution layer, and then further generates the Ghost feature map through several deep convolution layers. These deep convolution layers perform channel-by-channel convolution operations on the preliminary feature map. The generated Ghost feature map can maintain the feature expression capability while reducing the computational workload and the number of parameters. Compared with the conventional YOLOv5 model, this embodiment uses the Ghost Conv unit to replace the conventional Conv unit. Therefore, the structure and function of other modules in the model, such as ResUnit and GSP, will not be repeated.

[0062] Another difference from the conventional YOLOv5 model is that in this embodiment, the CBAM-CA attention mechanism and ASPP multi-layer receptive field are added to the tail of the backbone network. The CBAM-CA attention mechanism module consists of the convolutional block attention mechanism CBAM and the coordinate attention mechanism CA, as shown in Figure 4 As shown in the figure, the convolutional block attention mechanism CBAM includes a channel attention mechanism and a spatial attention mechanism. First, the feature map is globally averaged pooled through the channel attention mechanism, and the feature map of each channel is compressed into a scalar value to obtain the global spatial features of each channel. The fully connected layer is used to perform a nonlinear transformation on the features after global average pooling to extract the correlation between channels. The transformation process usually includes a ReLU activation function to increase nonlinearity, and a sigmoid activation function to compress the output value to between 0 and 1, generate channel attention weights, and multiply the channel attention weights by the original feature map element by element to obtain the channel-weighted feature map. This step emphasizes the channels that contribute most to the task and suppresses irrelevant or redundant channels.

[0063] The channel-weighted feature map is then processed by the coordinate attention mechanism, that is, the channel-weighted feature map is globally average pooled in the width and height directions respectively to capture the global features in the two directions. This step generates two feature maps, representing the global features in the width and height directions respectively. The global feature maps in the two directions are convolved using convolutional layers and merged to match the number of channels of the input feature map, and the sigmoid activation function is applied to generate attention scores in the width and height dimensions, that is, the coordinate attention weights. Finally, the coordinate attention weights are element-wise multiplied with the channel-weighted feature map to obtain the coordinate-weighted feature map. This step emphasizes the areas in the feature map that contribute most to the task and suppresses irrelevant or redundant areas.

[0064] Finally, the spatial attention mechanism processes the spatial dimension of the coordinate-weighted feature map, including channel average pooling and channel maximum pooling, thereby compressing the channel dimension while retaining spatial information. The results of channel maximum pooling and channel average pooling are spliced ​​to generate a feature descriptor containing spatial information. The spliced ​​feature descriptor is further processed using a convolutional layer to generate spatial attention weights. The spatial attention weights are element-wise multiplied with the coordinate-weighted feature map to obtain the final hybrid attention weighted feature map. Through the hybrid attention mechanism in this embodiment, the features of the input data can be captured more comprehensively, thereby improving the accuracy of target recognition.

[0065] Furthermore, the ASPP module in this embodiment includes multiple parallel convolutional layers and maximum pooling layers for acquiring features of different receptive fields, and fusing the features of different receptive fields through merging layers and splicing layers, thereby capturing a wider range of context and feature information in the image.

[0066] After obtaining the target recognition model of the above structure, the bitmap image can be input into the target recognition model to identify the flame and the charging pile, thereby obtaining the flame image and the charging pile image. The reason for identifying the charging pile is that when no fire hazard is found, the charging pile image is combined with the subsequent temperature processing, so that the situation where the charging pile temperature is too high and spontaneous combustion occurs can be included in the scope of fire monitoring. For the target images identified by the target recognition model, these target areas are used as the corresponding regions of interest in the infrared image, and temperature data is extracted from these regions of interest. The specific steps include:

[0067] Obtaining calibration data of an infrared camera corresponding to the infrared image, and extracting pixel values ​​corresponding to an infrared target image of the infrared image;

[0068] Converting the pixel values ​​into radiation intensity values ​​according to the calibration data, and converting the radiation intensity values ​​into temperature data according to Planck's law;

[0069] Time series temperature data is obtained according to the temperature data corresponding to each of the infrared images.

[0070] In this embodiment, the calibration data of the infrared camera is first obtained. Generally, the calibration data of the infrared camera will provide a lookup table or a mathematical equation to describe the relationship between the pixel value and the radiation intensity value. For example, if the camera provides a lookup table, the radiation intensity value corresponding to the pixel value can be directly obtained through the lookup table. If the camera provides a mathematical equation, the pixel value can be substituted into the equation for calculation to obtain the radiation intensity value. After obtaining the radiation intensity value, the radiation intensity value can be converted into a temperature value using Planck's law. Planck's law describes the relationship between the spectral radiance of blackbody radiation and the wavelength and temperature. However, for infrared imaging, what is usually of interest is the radiation intensity within a specific wavelength range. Therefore, a simplified form of Planck's law or a formula at a specific wavelength can be used. For example, for the medium-wave or long-wave infrared bands commonly used by infrared cameras, a simplified formula in the following form can be used:

[0071]

[0072] Where T is the temperature (unit: Kelvin or Celsius), C1 and C2 are the calibration constants of the camera, and R is the radiation intensity value. represents the target emissivity.

[0073] Repeat the temperature data extraction steps for each time point image to obtain the temperature data for that time point. Arrange these data in chronological order to construct the time series temperature data.

[0074] Through the above steps, the target image identified by the target recognition model, the time-series temperature data corresponding to the target image, and the time-series smoke data collected from the smoke sensor can be obtained. For these images and data, the present invention adopts a multi-input neural network model to construct a fire recognition model, and processes these data through the fire recognition model to identify fire hazards. Specifically, the fire recognition model is composed of a convolutional neural network and a recurrent neural network, wherein the convolutional neural network CNN is used to extract features of the target image, and the recurrent neural network RNN ​​includes a first recurrent neural network and a second recurrent neural network. These two recurrent neural networks are used to extract features of the time-series smoke data and the time-series temperature data, respectively, that is, the three neural networks are in a parallel structure. It should be noted here that the convolutional neural network model is used to extract features of the flame image. When the target recognition model does not recognize the flame image, the convolutional neural network model will directly output the default value feature value, because the features of the charging pile image are not of great significance for the identification of fire hazards. The neural network model structure used in this embodiment can refer to the conventional model structure, and will not be described one by one here.

[0075] The image features, temperature features, and smoke features extracted by the model will be input into the feature fusion layer, and the features will be fused by merging and splicing. The obtained fusion features will be input into the fully connected layer. There are multiple fully connected layers here. ReLU, ELU and other activation functions can be used after each fully connected layer. The last fully connected layer uses the softmax activation function for classification, thereby obtaining the fire recognition result.

[0076] This embodiment provides a fire identification method for a charging pile scenario. The present invention uses a target recognition model to accurately and efficiently identify flames and charging piles, and obtains corresponding temperature data based on the recognition results, thereby providing accurate data support for subsequent data analysis. In addition, the present invention uses a multi-input fire recognition model to fuse multiple features, effectively improving the accuracy of model recognition. The identification method provided by the present invention can accurately and efficiently identify fire hazards in charging piles, further ensuring the safety of the charging environment.

[0077] See also Figure 5 Based on the same inventive concept, a second embodiment of the present invention proposes a fire identification system for a charging pile scenario, including:

[0078] A data acquisition module 10 is configured to acquire time-series images and time-series environmental parameters of the charging pile area to be identified, wherein the time-series images include a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data;

[0079] a target recognition module 20 for inputting the bitmap image into a target recognition model to obtain a target image, and selecting a corresponding infrared target image from the infrared image based on the target image;

[0080] a temperature acquisition module 30 for extracting temperature data from the infrared target image of each of the infrared images, and obtaining time series temperature data based on the extracted temperature data;

[0081] The fire identification module 40 is configured to input the target image, the time-series smoke data, and the time-series temperature data into a fire identification model to obtain a fire identification result for the charging pile area.

[0082] Furthermore, the target recognition module 20 is further configured to pre-process the bitmap data and input the pre-processed bitmap data into a pre-trained target recognition model to recognize the flame and the charging pile to obtain a target image, wherein the target image includes a flame image and a charging pile image;

[0083] Among them, the target recognition model is constructed using the YOLO model based on GhostNet, and an attention module and a multi-layer receptive field module are set in the backbone network of the YOLO model. The attention module is a hybrid attention module based on the convolutional block attention mechanism and the coordinate attention mechanism.

[0084] Furthermore, the temperature acquisition module 30 is further configured to acquire calibration data of an infrared camera corresponding to the infrared image, and extract pixel values ​​corresponding to the infrared target image of the infrared image;

[0085] Converting the pixel values ​​into radiation intensity values ​​according to the calibration data, and converting the radiation intensity values ​​into temperature data according to Planck's law;

[0086] Time series temperature data is obtained according to the temperature data corresponding to each of the infrared images.

[0087] Furthermore, the fire identification module 40 further includes a fire identification model, which is constructed using a neural network model, including a first recurrent neural network model, a second recurrent neural network model, a convolutional neural network model, a feature fusion layer, and a fully connected layer;

[0088] The input data of the first recurrent neural network model is time series smoke data, and the output data of the first recurrent neural network model is smoke feature data;

[0089] The input data of the second recurrent neural network model is time series temperature data, and the output data of the second recurrent neural network model is temperature feature data;

[0090] The input data of the convolutional neural network model is the target image, and the output data of the convolutional neural network model is image feature data;

[0091] The input data of the feature fusion layer is the smoke feature data, temperature feature data and image feature data, and the output data of the feature fusion layer is fused feature data;

[0092] The input data of the fully connected layer is the fused feature data, and the output data of the fully connected layer is the fire recognition result.

[0093] The technical features and technical effects of the fire identification system for charging pile scenarios proposed in the embodiment of the present invention are the same as those of the method proposed in the embodiment of the present invention and are not described in detail here. The various modules in the above-mentioned fire identification system for charging pile scenarios can be implemented in whole or in part through software, hardware, or a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0094] In addition, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0095] See also Figure 6 , an internal structure diagram of a computer device in one embodiment, the computer device can specifically be a terminal or a server. The computer device includes a processor, a memory, a network interface, a display and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a fire identification method for a charging pile scenario. The display screen of the computer device can be a liquid crystal display or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the computer device housing, or an external keyboard, touchpad or mouse.

[0096] It can be understood by those skilled in the art that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have the same component arrangement.

[0097] In addition, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when the computer program is executed by a processor.

[0098] In summary, the embodiments of the present invention propose a fire identification method, system, device and medium for a charging pile scenario. The method obtains a time-series image and time-series environmental parameters of the charging pile area to be identified, wherein the time-series image includes a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data; inputs the bitmap image into a target recognition model to obtain a target image, and selects a corresponding infrared target image from the infrared image based on the target image; extracts temperature data from the infrared target image of each infrared image, and obtains time-series temperature data based on the extracted temperature data; inputs the target image, the time-series smoke data and the time-series temperature data into a fire identification model to obtain a fire identification result for the charging pile area. The present invention can accurately and efficiently identify flames and charging piles through the target recognition model, and obtain corresponding temperature data based on the recognition result, thereby providing accurate data support for subsequent data analysis. In addition, the present invention uses a multi-input fire recognition model to fuse multiple features, effectively improving the accuracy of model recognition. The identification method provided by the present invention can accurately and efficiently identify fire hazards in charging piles, further ensuring the safety of the charging environment.

[0099] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the various technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] The above-described embodiments merely represent several preferred implementations of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art could make several improvements and substitutions without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be based on the scope of protection of the claims.

Claims

1. A fire identification method for a charging pile scenario, characterized in that: include: Acquire time-series images and time-series environmental parameters of the charging pile area to be identified, wherein the time-series images include a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data; Inputting the bitmap image into a target recognition model to obtain a target image, and selecting a corresponding infrared target image from the infrared image based on the target image; extracting temperature data from the infrared target image of each of the infrared images, and obtaining time series temperature data based on the extracted temperature data; Inputting the target image, the time-series smoke data, and the time-series temperature data into a fire recognition model to obtain a fire recognition result of the charging pile area; The target recognition model is constructed using the YOLO model based on GhostNet. The backbone network consists of the GhostCBS module and the GSP1 module. The GhostCBS module replaces the convolution Conv module to achieve lightweight backbone network. At the same time, an attention module and a multi-layer receptive field module are added to the tail of the backbone network of the YOLO model. The attention module is a hybrid attention module based on the convolution block attention mechanism and the coordinate attention mechanism. The convolution block attention mechanism includes a channel attention mechanism and a spatial attention mechanism. The channel attention mechanism is used to perform global average pooling and global maximum pooling on the input feature map, perform nonlinear transformation on the features of the global average pooling and the features of the global maximum pooling, merge them, generate channel attention weights, and multiply the channel attention weights by the feature map element by element to obtain the channel-weighted feature map; The coordinate attention mechanism is used to perform global average pooling on the input channel-weighted feature map in the width and height directions respectively, and merge the two obtained feature maps after convolution to generate coordinate attention weights, and then multiply the coordinate attention weights by the channel-weighted feature map element-wise to obtain the coordinate-weighted feature map; The spatial attention mechanism is used to perform channel average pooling and channel maximum pooling on the input coordinate-weighted feature map, concatenate and convolve the results of channel maximum pooling and channel average pooling to generate spatial attention weights, and multiply the spatial attention weights by the coordinate-weighted feature map element-by-element to obtain the final hybrid attention weighted feature map.

2. The fire identification method for charging pile scenarios according to claim 1 is characterized in that: The step of inputting the bitmap image into the target recognition model to obtain the target image comprises: The bitmap data is preprocessed, and the preprocessed bitmap data is input into a pre-trained target recognition model to recognize the flame and the charging pile to obtain a target image, wherein the target image includes a flame image and a charging pile image.

3. The fire identification method for charging pile scenarios according to claim 1 is characterized in that: The step of extracting temperature data from the infrared target image of each of the infrared images and obtaining time series temperature data based on the extracted temperature data includes: Obtaining calibration data of an infrared camera corresponding to the infrared image, and extracting pixel values ​​corresponding to an infrared target image of the infrared image; Converting the pixel values ​​into radiation intensity values ​​according to the calibration data, and converting the radiation intensity values ​​into temperature data according to Planck's law; Time series temperature data is obtained according to the temperature data corresponding to each of the infrared images.

4. The fire identification method for charging pile scenarios according to claim 1, characterized in that: The fire recognition model is constructed using a neural network model, including a first recurrent neural network model, a second recurrent neural network model, a convolutional neural network model, a feature fusion layer and a fully connected layer; The input data of the first recurrent neural network model is time series smoke data, and the output data of the first recurrent neural network model is smoke feature data; The input data of the second recurrent neural network model is time series temperature data, and the output data of the second recurrent neural network model is temperature feature data; The input data of the convolutional neural network model is the target image, and the output data of the convolutional neural network model is image feature data; The input data of the feature fusion layer is the smoke feature data, temperature feature data and image feature data, and the output data of the feature fusion layer is fused feature data; The input data of the fully connected layer is the fused feature data, and the output data of the fully connected layer is the fire recognition result.

5. A fire identification system for charging pile scenarios, characterized in that: include: A data acquisition module, configured to acquire time-series images and time-series environmental parameters of the charging pile area to be identified, wherein the time-series images include a bitmap image and a corresponding infrared image, and the time-series environmental parameters include time-series smoke data; a target recognition module, configured to input the bitmap image into a target recognition model to obtain a target image, and select a corresponding infrared target image from the infrared image based on the target image; a temperature acquisition module, configured to extract temperature data from the infrared target image of each of the infrared images, and obtain time series temperature data based on the extracted temperature data; a fire identification module, configured to input the target image, the time-series smoke data, and the time-series temperature data into a fire identification model to obtain a fire identification result for the charging pile area; The target recognition model is constructed using the YOLO model based on GhostNet. The backbone network consists of the GhostCBS module and the GSP1 module. The GhostCBS module replaces the convolution Conv module to achieve lightweight backbone network. At the same time, an attention module and a multi-layer receptive field module are added to the tail of the backbone network of the YOLO model. The attention module is a hybrid attention module based on the convolution block attention mechanism and the coordinate attention mechanism. The convolution block attention mechanism includes a channel attention mechanism and a spatial attention mechanism. The channel attention mechanism is used to perform global average pooling and global maximum pooling on the input feature map, perform nonlinear transformation on the features of the global average pooling and the features of the global maximum pooling, merge them, generate channel attention weights, and multiply the channel attention weights by the feature map element by element to obtain the channel-weighted feature map; The coordinate attention mechanism is used to perform global average pooling on the input channel-weighted feature map in the width and height directions respectively, and merge the two obtained feature maps after convolution to generate coordinate attention weights, and then multiply the coordinate attention weights by the channel-weighted feature map element-wise to obtain the coordinate-weighted feature map; The spatial attention mechanism is used to perform channel average pooling and channel maximum pooling on the input coordinate-weighted feature map, concatenate and convolve the results of channel maximum pooling and channel average pooling to generate spatial attention weights, and multiply the spatial attention weights by the coordinate-weighted feature map element-by-element to obtain the final hybrid attention weighted feature map.

6. The fire identification system for charging pile scenarios according to claim 5 is characterized in that: The target recognition module is further used to preprocess the bitmap data and input the preprocessed bitmap data into a pre-trained target recognition model to recognize the flame and the charging pile to obtain a target image, which includes a flame image and a charging pile image.

7. The fire identification system for charging pile scenarios according to claim 5 is characterized in that: The temperature acquisition module is further used to obtain calibration data of the infrared camera corresponding to the infrared image, and extract pixel values ​​corresponding to the infrared target image of the infrared image; Converting the pixel values ​​into radiation intensity values ​​according to the calibration data, and converting the radiation intensity values ​​into temperature data according to Planck's law; Time series temperature data is obtained according to the temperature data corresponding to each of the infrared images.

8. The fire identification system for charging pile scenarios according to claim 5 is characterized in that: The fire identification module also includes a fire identification model, which is constructed using a neural network model and includes a first recurrent neural network model, a second recurrent neural network model, a convolutional neural network model, a feature fusion layer, and a fully connected layer; The input data of the first recurrent neural network model is time series smoke data, and the output data of the first recurrent neural network model is smoke feature data; The input data of the second recurrent neural network model is time series temperature data, and the output data of the second recurrent neural network model is temperature feature data; The input data of the convolutional neural network model is the target image, and the output data of the convolutional neural network model is image feature data; The input data of the feature fusion layer is the smoke feature data, temperature feature data and image feature data, and the output data of the feature fusion layer is fused feature data; The input data of the fully connected layer is the fused feature data, and the output data of the fully connected layer is the fire recognition result.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images