Ultrahigh-voltage line forest fire early warning method and device based on two-stage confirmation large model, computer equipment and medium

Abstract

The invention provides an ultrahigh-voltage line forest fire early warning method and device based on a two-stage confirmation large model, computer equipment and a medium. According to the two-stage confirmation large model method, accurate recognition of forest fire hidden dangers is achieved through a hierarchical model verification mechanism. The method specifically comprises the steps that after a shot image of an ultra-high voltage line camera is input, NAS Net is adopted for first-layer training, the classification probability is output, and if the probability exceeds a preset fault threshold value, an early warning result is directly output; and when the probability does not reach a threshold value, starting a second-layer converter-based bidirectional encoder representation BERT model to carry out semantic feature retraining, and improving the recognition confidence through secondary probability judgment. According to the method, the problem of high misjudgment rate of a single detection model is effectively solved through two-stage model cooperative verification, multi-dimensional feature cross verification of forest fire hidden dangers is realized, the anti-noise capability of a traditional early warning system on complex environment interference is optimized, and meanwhile, the redundant consumption of computing resources is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of cross - technologies of power system security monitoring, computer vision and natural language processing, application of deep learning models, and high - voltage transmission line fault warning, and relates to a control method of a dual - level confirmation model based on a deep neural network, which is applicable to the intelligent identification and control of hidden fire hazards on extra - high - voltage transmission lines of power systems. Background Art

[0002] When the existing deep network has low accuracy in predicting wildfires, most researchers adopt the following methods:

[0003] First, modify the parameters or the trainer and run repeatedly many times, wasting a lot of research time, but still unable to obtain high accuracy.

[0004] Second, adopt a deeper or wider network structure, but these always belong to a single network, and the representation ability of the complex input - output relationship of a single network is always limited, and it is difficult to greatly improve the accuracy.

[0005] Third, connect multiple methods in series, such as connecting a convolutional neural network in series with a convolutional neural network, or connecting a time - series network in series with a convolutional neural network, or connecting a convolutional neural network in series with a time - series network. However, in the end, a single network is still formed, and the accuracy is still limited, and it is difficult to achieve ultra - high accuracy.

[0006] Fourth, a framework of two parallel networks is used, and a large amount of memory and time are required, but the effect of fusing the two networks still cannot reach a particularly high level. Summary of the Invention

[0007] Based on this, in view of the above - mentioned technical problems, it is necessary to provide a method, device, computer device, computer - readable storage medium, and computer program product for early warning of wildfires on extra - high - voltage lines based on a dual - level confirmation large model.

[0008] In a first aspect, the present application provides a method for early warning of wildfires on extra - high - voltage lines based on a dual - level confirmation large model. The method includes:

[0009] Receiving image data captured by a camera of an extra - high - voltage power line, and performing size normalization and channel standardization processing on the original image by using the methods of bilinear interpolation, mean - zeroing the RGB three channels, and standard deviation scaling to generate a standardized image that meets the input specifications;

[0010] Performing a first - layer prediction by using a NAS Net model to obtain the classification probability of wildfires on the extra - high - voltage line;

[0011] Performing a second - layer prediction by using a BERT model to obtain the verification probability of wildfires on the extra - high - voltage line.

[0012] In a second aspect, the present application also provides an ultra-high voltage line wildfire warning device based on a dual-level confirmation large model. The device includes:

[0013] An acquisition and initialization data module, configured to receive image data captured by a camera of an ultra-high voltage power line, and perform size normalization and channel standardization processing on the original image by using methods such as bilinear interpolation, mean-zeroing of RGB three channels, and standard deviation scaling to generate a standardized image that conforms to the input specification;

[0014] A first-layer prediction module, configured to perform a first-layer prediction by using a NAS Net model to obtain the classification probability of an ultra-high voltage line wildfire;

[0015] A second-layer prediction module, configured to perform a second-layer prediction by using a BERT model to obtain the verification probability of an ultra-high voltage line wildfire.

[0016] In a third aspect, the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0017] Receive image data captured by a camera of an ultra-high voltage power line, and perform size normalization and channel standardization processing on the original image by using methods such as bilinear interpolation, mean-zeroing of RGB three channels, and standard deviation scaling to generate a standardized image that conforms to the input specification;

[0018] Perform a first-layer prediction by using a NAS Net model to obtain the classification probability of an ultra-high voltage line wildfire;

[0019] Perform a second-layer prediction by using a BERT model to obtain the verification probability of an ultra-high voltage line wildfire.

[0020] In a fourth aspect, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0021] Receive image data captured by a camera of an ultra-high voltage power line, and perform size normalization and channel standardization processing on the original image by using methods such as bilinear interpolation, mean-zeroing of RGB three channels, and standard deviation scaling to generate a standardized image that conforms to the input specification;

[0022] Perform a first-layer prediction by using a NAS Net model to obtain the classification probability of an ultra-high voltage line wildfire;

[0023] Perform a second-layer prediction by using a BERT model to obtain the verification probability of an ultra-high voltage line wildfire.

[0024] Fifth aspect, the present application also provides a computer program product. The computer program product includes a computer program which, when executed by a processor, implements the following steps:

[0025] Receive image data captured by a camera of an extra-high voltage power line, and perform size normalization and channel standardization processing on the original image by using the methods of bilinear interpolation, mean-zeroing of the RGB three channels, and standard deviation scaling to generate a standardized image that meets the input specifications;

[0026] Perform the first-layer prediction by using the NAS Net model to obtain the classification probability of wildfires on the extra-high voltage line;

[0027] Perform the second-layer prediction by using the BERT model to obtain the verification probability of wildfires on the extra-high voltage line.

[0028] The above-mentioned early warning method, device, computer device, storage medium, and computer program product for wildfires on extra-high voltage lines based on a two-stage confirmation large model receive image data captured by a camera of an extra-high voltage power line, and perform size normalization and channel standardization processing on the original image by using the methods of bilinear interpolation, mean-zeroing of the RGB three channels, and standard deviation scaling to generate a standardized image that meets the input specifications; perform the first-layer prediction by using the NAS Net model to obtain the classification probability of wildfires on the extra-high voltage line; perform the second-layer prediction by using the BERT model to obtain the verification probability of wildfires on the extra-high voltage line; and can achieve accurate identification and hierarchical early warning of wildfire hazards, effectively improving the accuracy and robustness of the early warning system. Description of the Drawings

[0029] Figure 1 It is a wildfire early warning framework diagram of an extra-high voltage line in an embodiment.

[0030] Figure 2 It is a network framework diagram of the BERT model in an embodiment.

[0031] Figure 3 It is an internal structure diagram of a computer device in an embodiment. Detailed Embodiments

[0032] In order to make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0033] In one embodiment, as Figure 1As shown, a super high-voltage line wildfire warning method based on a two-stage confirmation large model is provided. In this embodiment, taking the application of this method to a terminal as an example, it can be understood that this method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is realized through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0034] In one embodiment, as Figure 1 shown, a super high-voltage line wildfire warning framework diagram is provided. First, input the images captured in real time by the super high-voltage line camera; then, extract the deep features of the images through the primary image classification model based on NAS Net in the first layer, and output the preliminary classification probability of fire hazards ; when the classification probability does not reach the preset threshold , enter the deep image classification link based on the BERT model in the second layer, construct a multi-scale feature pyramid for the images, and accordingly output the classification probability of fire hazards ; finally, trigger a wildfire warning or eliminate false alarms according to the collaborative determination results of the two-level models.

[0035] In one embodiment, as Figure 2 shown, a BERT network framework diagram is provided. First, input the images processed by the first layer of NASNet, and perform size normalization on the images; then extract multi-scale features, and respectively extract the spatial details of the shallow feature extraction layer, the texture features of the middle feature extraction layer, and the semantic features of the deep feature extraction layer through a convolutional neural network, and thereby construct a multi-scale feature pyramid. Subsequently, pass through the Transformer encoder group, through multi-head self-attention, feed-forward network, residual connection, and layer normalization, finally generate a verification probability and make a secondary decision, and output the classification probability.

[0036] In one embodiment, the specific steps of the super high-voltage line wildfire warning method based on the two-stage confirmation large model are as follows:

[0037] Step (1), receive the image data captured by the super high-voltage power line camera, and perform size normalization and channel standardization processing on the original image by using the methods of bilinear interpolation, mean zeroing of the RGB three channels, and standard deviation scaling, to generate a standardized image that meets the input specifications:

[0038] Receive the image data captured by the super high-voltage power line camera, and perform bilinear interpolation scaling on the input image with any resolution to 256×256 pixels to eliminate the size fluctuations caused by device differences. Perform mean zeroing on each of the RGB three channels, take the mean , and perform standard deviation scaling on them, take the standard deviation , suppress the interference of light changes and generate a standardized image that meets the input specifications , providing a stable data basis for subsequent feature extraction. The standardized image is:

[0039]

[0040] In the formula, the Resize function is a bilinear interpolation scaling function; the ChannelNorm function is a channel normalization function, and its mathematical expression is:

[0041]

[0042] Step (2): Use the NAS Net model for the first-layer prediction. The NAS Net model includes separable convolution, pooling, spatial downsampling convolution, and global average pooling GAP to obtain the classification probability of wildfires on ultra-high voltage transmission lines:

[0043] Construct the NAS Net model to extract deep features of the image: At a resolution of 32×32, use a combination of separable convolution and pooling operations to extract local features of flame color and smoke morphology; in the spatial downsampling stage , compress the size of the feature map through a convolution with a stride of 2, and gradually expand the receptive field to capture the trend of fire spread. Thus, a composite feature map with a resolution of 32×32 and 4032 channels is generated , which contains the key spatial information for fire situation discrimination. The composite feature map is:

[0044]

[0045] In the formula, represents the independent combination of multiple mathematical objects according to the structure to form a new composite object; represents the structured independent stacking of the results of K branches; represents function combination.

[0046] Compress the feature map and calculate the probability of a wildfire occurrence: Compress the spatial dimension of from 32×32 to 1×1, retain the global response intensity of each channel, and generate a 4032-dimensional feature vector; through the weight matrix perform a linear transformation, and then through the Sigmoid function map it to a probability value . The output probability quantifies the confidence level of the existence of wildfire hazards in the image, and the higher the value, the greater the fire risk.

[0047]

[0048] In the formula, the GAP function is the global average pooling function, and its mathematical expression is:

[0049]

[0050] Preset threshold Trigger the fast warning mechanism: Through receiver operating characteristic curve analysis, select to ensure that the recall rate of the primary model is greater than or equal to 95%, and give priority to excluding obvious non-fire samples. If , directly output the warning signal ; Otherwise, transfer it to the second-level model for in-depth verification. The warning signal is:

[0051]

[0052] Step (3): Use the BERT model for the second-layer prediction. The BERT model includes multi-spectral decomposition, adaptive histogram equalization, multi-scale feature pyramid, feature fusion and serialization, sine position encoding, multi-head self-attention mechanism Transformer encoder, layer normalization LN, and global average pooling GAP to obtain the verification probability of ultra-high voltage line wildfires:

[0053] Perform multi-spectral decomposition and contrast enhancement on the original image: Separate the near-infrared and thermal-infrared bands from the original RGB image to generate 5-channel data, namely the red band, green band, blue band, near-infrared band, and thermal-infrared band, enhance the flame thermal radiation characteristics, and perform adaptive histogram stretching on each channel to improve the contrast between the smoke texture and the background. Thus, a multi-spectral image with a resolution of 512×512 is generated , providing rich spectral and texture information for multi-scale feature extraction. The multi-spectral image is:

[0054]

[0055] In the formula, the MSD function is the multi-spectral decomposition function; the HistEq function is the adaptive histogram equalization function.

[0056] Extract features at different abstraction levels through a convolutional neural network to construct a multi-scale feature group , and thus construct a multi-scale feature pyramid to comprehensively describe the fire characteristics.

[0057]

[0058] In the formula, the shallow features Use a 3×3 convolutional kernel to perform two-dimensional convolutional operation Conv to extract edges and color distribution, with a resolution of 256×256 and 64 channels; the middle-level features Capture the smoke diffusion texture through a 5×5 residual block ResBlock, reduce the resolution to 128×128, and the number of channels is 256; deep features Adopt a dense connection block DenseBlock to aggregate high-level semantic information, with a resolution of 64×64 and the number of channels being 1024.

[0059] Fuse multi-scale features and transform them into sequence data: Perform upsampling and lateral connection to generate fused features with a unified resolution , retaining the discriminative information of each level. The fused features with a unified resolution are:

[0060]

[0061] In the formula, the FPN function is the feature pyramid network function.

[0062] Flatten into a vector sequence, and add sine position encoding to mark the spatial position relationship of the feature space. Generate serialized features with spatial perception ability , which is used to adapt to the sequence processing mechanism of the Transformer. The serialized feature is:

[0063]

[0064] In the formula, the Flatten function is the feature flattening function, that is, the three-dimensional feature map is unfolded into a two-dimensional matrix in the order of spatial positions.

[0065] Model the long-range dependencies between features through the self-attention mechanism: Split the input sequence into 8 heads, calculate the attention weights in a 64-dimensional subspace respectively, and capture the cross-region associations between flames and smoke. Adopt two layers of fully connected layers, that is, 2048→512 dimensions, to enhance the non-linear expression ability of the features. The output of each layer is:

[0066]

[0067] In the formula, the MHAtt function is the multi-head self-attention function; the LN function is the layer normalization function.

[0068] Output the encoded features , which contains global context information and can be used to distinguish light source interference similar to wildfires. The output encoded feature is:

[0069]

[0070] In the formula, the FFN function is a feed-forward network function.

[0071] Generating a verification probability based on the encoded features: For the encoded features perform global average pooling (GAP) to compress the sequence dimension and generate a 512-dimensional feature vector. Output the verification probability through a 512-dimensional fully connected layer and Sigmoid activation as:

[0072]

[0073] Set , and trigger the verification flag if and only if . Thus, high-precision review is performed on the samples suspected by the primary model to exclude interference from similar images such as clouds and heat source devices. The verification flag is:

[0074]

[0075] Generating the final warning signal by integrating the verification results of the two levels as:

[0076]

[0077] In the formula, is logical OR.

[0078] Once any model triggers a warning, it is determined as a fire, ensuring the fault tolerance of the system. Output as the wildfire warning instruction to drive the emergency response system.

[0079] In one embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 3As shown in the figure. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. 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 computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for early warning of wildfires on extra-high voltage lines based on a two-stage confirmation large model. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0080] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of some structures 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 computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0081] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0082] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0083] In one embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0084] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions.

[0085] Those of ordinary skill in the art can understand that all or part of the processes of the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., and are not limited thereto.

[0086] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0087] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for ultra-high voltage line wildfire warning based on a dual-level confirmation large model, characterized in that, The method includes: Receiving the image data captured by the camera of the extra-high voltage power line, and performing size normalization and channel standardization processing on the original image by using the methods of bilinear interpolation, mean-zeroing the RGB three channels, and standard deviation scaling to generate a standardized image that meets the input specifications; Performing the first-layer prediction by using the NAS Net model to obtain the classification probability of the wildfire on the extra-high voltage line; Performing the second-layer prediction by using the BERT model to obtain the verification probability of the wildfire on the extra-high voltage line.

2. The ultra-high voltage line wildfire warning method based on a double-level confirmation large model according to claim 1, wherein, The performing the first-layer prediction by using the NAS Net model is: constructing the NAS Net model, extracting the deep features of the image, compressing the feature map, calculating the wildfire occurrence probability, and triggering the fast warning mechanism with a preset threshold.

3. The ultra-high voltage line wildfire warning method based on a dual-level confirmation large model according to claim 1, wherein, The performing the second-layer prediction by using the BERT model is: performing multi-spectral decomposition and contrast enhancement on the original image, fusing multi-scale features and converting them into sequence data, modeling the long-range dependence between features through the self-attention mechanism, and generating the verification probability based on the encoded features.

4. The ultra-high voltage line wildfire warning method based on a double-level confirmation large model according to claim 1, wherein The NAS Net model includes separable convolution, pooling, spatial downsampling convolution, and global average pooling GAP.

5. The ultra-high voltage line wildfire warning method based on a double-level confirmation large model according to claim 1, wherein The BERT model includes multi-spectral decomposition, adaptive histogram equalization, multi-scale feature pyramid, feature fusion and serialization, sine position encoding, multi-head self-attention mechanism Transformer encoder, layer normalization LN, and global average pooling GAP.

6. An ultra-high voltage line wildfire warning device based on a double-level confirmation large model, characterized in that, The device includes: An acquiring and initializing data module, which is used to receive the image data captured by the camera of the extra-high voltage power line, and perform size normalization and channel standardization processing on the original image by using the methods of bilinear interpolation, mean-zeroing the RGB three channels, and standard deviation scaling to generate a standardized image that meets the input specifications; A first-layer prediction module, which is used to perform the first-layer prediction by using the NAS Net model to obtain the classification probability of the wildfire on the extra-high voltage line; A second-layer prediction module, which is used to perform the second-layer prediction by using the BERT model to obtain the verification probability of the wildfire on the extra-high voltage line.

7. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the extra-high voltage line wildfire warning method based on the two-stage confirmation large model described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the extra-high voltage line wildfire warning method based on the two-stage confirmation large model described in any one of claims 1 to 5.

Images