Electricity utilization safety diagnosis method based on convolutional neural network

Abstract

The invention provides an electrical safety diagnosis method based on a convolutional neural network, and the method comprises the steps: constructing a denoising correction module, carrying out the denoising correction of a temperature value at a current diagnosis moment, and obtaining a corrected temperature value at the current diagnosis moment; constructing a comparison module, comparing the corrected temperature value at the current diagnosis moment with a preset temperature threshold value, and generating a selection signal; constructing a preprocessing module, performing contrast enhancement on the target area image, and generating a target area enhanced image; constructing a feature extraction module, selecting a high receptive field feature extraction channel or a low receptive field feature extraction channel according to the selection signal, performing feature extraction on the enhanced image of the target area, and generating target features; and constructing a classifier, classifying the target features, and generating a diagnosis result.

Description

Technical Field

[0001] The present invention relates to the field of smart electricity use, and in particular to a method for diagnosing electricity safety based on a convolutional neural network. Background Art

[0002] In recent years, with the addition of more and more electrical equipment, the probability of electrical fires has continued to increase, and the issue of safe electricity use has become increasingly important. Existing technologies mainly use wireless sensing technologies such as the Internet of Things and mobile terminals to diagnose the operation of the monitored lines or the environmental conditions of the monitored scenes in real time, analyze changes in electrical parameters, temperature, smoke, and combustible gas concentrations, and identify possible fires or impending fires to improve the level of electricity safety.

[0003] However, the electrical parameters, temperature, smoke and flammable gas concentrations in the existing diagnostic methods are easily affected by the surrounding electronic noise and environment, and the false alarm rate is extremely high. Compared with wireless sensing technology, the power safety diagnostic method based on visual information has the advantages of shorter response, wider monitoring range, and more information acquisition.

[0004] Therefore, providing an electricity safety diagnosis method with high accuracy and fast response time has become a technical problem that urgently needs to be solved. Summary of the invention

[0005] In view of the above problems, the present application is proposed, which provides an electricity safety diagnosis method based on a convolutional neural network, which uses the calibration temperature value at the current diagnosis moment to select a high receptive field feature extraction channel or a low receptive field feature extraction channel, and performs feature extraction on the target area image at the current diagnosis moment to prevent the loss or blurring of the target features, improve the response speed, reduce the response time, and achieve accurate and rapid diagnosis of electricity safety in the target area.

[0006] The present application provides a power safety diagnosis method based on a convolutional neural network, comprising steps S1 to S6:

[0007] S1: Obtain the temperature value of the target area at the current diagnosis time, the target area image, and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, where N is greater than or equal to 2;

[0008] S2: construct a denoising and correction module, receive the temperature value at the current diagnosis time and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, perform denoising and correction processing on the temperature value at the current diagnosis time, and obtain a corrected temperature value at the current diagnosis time;

[0009] S3: constructing a comparison module, receiving the corrected temperature value at the current diagnosis moment, comparing the corrected temperature value at the current diagnosis moment with a preset temperature threshold, and generating a selection signal;

[0010] S4: construct a preprocessing module, receive the target area image of the target area at the current diagnosis moment, perform contrast enhancement on the target area image, and generate an enhanced image of the target area;

[0011] S5: construct a feature extraction module, receive the selection signal and the target area enhanced image, select a high receptive field feature extraction channel or a low receptive field feature extraction channel according to the selection signal, perform feature extraction on the target area enhanced image, and generate target features;

[0012] S6: Build a classifier, receive target features, classify the target features, and generate diagnosis results.

[0013] The denoising and correction processing of the temperature value at the current diagnosis time is performed to obtain the corrected temperature value at the current diagnosis time, including steps S21 to S23:

[0014] S21: Based on the temperature value at the current diagnosis time and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, a maximum temperature value and a minimum temperature value are obtained by comparison;

[0015] S22: Calculate the optimal correction coefficient according to the maximum temperature value, the minimum temperature value and the temperature values ​​at N diagnosis moments;

[0016] S23: Based on the optimal correction coefficient α opt , the corrected temperature value at the previous diagnosis moment and the temperature value at the current diagnosis moment, and calculate the corrected temperature value at the current diagnosis moment.

[0017] Among them, the optimal correction coefficient is specifically:

[0018]

[0019] Among them, α opt is the optimal correction coefficient, X max is the maximum temperature value, X min is the minimum temperature value, X t is the temperature value at the diagnosis time t.

[0020] Among them, the corrected temperature value at the current diagnosis time is specifically:

[0021] S N =S N-1 +α opt (X N -S N-1 )

[0022] Among them, S N is the corrected temperature value at the current diagnosis time, S N-1 is the corrected temperature value at the previous diagnosis time, X NIt is the temperature value at the current diagnosis time.

[0023] The step of performing contrast enhancement on the target area image to generate the target area enhanced image includes steps S41 to S42:

[0024] S41: adjusting the grayscale values ​​of the pixels of the target area image;

[0025] S42: Normalizing the grayscale values ​​of the adjusted pixels to obtain an enhanced image of the target area.

[0026] The grayscale values ​​of the pixels in the target area image are adjusted, specifically:

[0027]

[0028] Among them, P is the gray value of the pixel in the target area image, P min is the low gray threshold, P max is the high grayscale threshold, and P1 is the grayscale value of the pixel after adjustment.

[0029] The normalization process is as follows:

[0030]

[0031] Wherein, P2 is the gray value of the normalized pixel.

[0032] Among them, according to the selection signal, a high receptive field feature extraction channel or a low receptive field feature extraction channel is selected, and feature extraction is performed on the target area enhanced image to generate target features, which specifically includes steps S51 to S52:

[0033] S51: when the selection signal indicates that the corrected temperature value at the current diagnosis moment is less than the preset temperature threshold, a low receptive field feature extraction channel is selected, wherein the low receptive field feature extraction channel is sequentially connected by four convolutional layers, and feature extraction is performed on the enhanced image of the target area to generate target features;

[0034] S52: When the selection signal indicates that the corrected temperature value at the current diagnosis moment is not less than the preset temperature threshold, a high receptive field feature extraction channel is selected, wherein the high receptive field feature extraction channel is connected in sequence by 8 convolutional layers, and feature extraction is performed on the enhanced image of the target area to generate target features.

[0035] Among them, the classifier uses the Softmax function to generate diagnosis results.

[0036] The beneficial effects of this application are:

[0037] (1) The present application provides a method for diagnosing electricity safety based on a convolutional neural network, which uses the calibration temperature value at the current diagnosis time to select a high receptive field feature extraction channel or a low receptive field feature extraction channel, and performs feature extraction on the target area image at the current diagnosis time to prevent the loss or blurring of target features, improve the response speed, reduce the response time, and achieve accurate and rapid diagnosis of electricity safety in the target area.

[0038] (2) The present application also constructs a denoising correction module and designs a denoising calibration method. According to the temperature value of the target area at the current diagnosis moment and the corrected temperature value at the previous diagnosis moment, the burrs and noise of the temperature value are removed, and the corrected temperature value at the current diagnosis moment is obtained by comprehensive calculation, thereby improving the accuracy of temperature value acquisition and improving the subsequent power safety diagnosis results.

[0039] (3) The present application also constructs a preprocessing module to receive the target area image at the current diagnosis moment of the target area, perform contrast enhancement preprocessing on the target area image, enhance the target features, and improve the accuracy of subsequent feature extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 A flow chart of the power safety diagnosis method based on convolutional neural network provided in this application;

[0042] Figure 2 A flow chart for obtaining a corrected temperature value provided for this application;

[0043] Figure 3 Flowchart of contrast enhancement provided for this application;

[0044] Figure 4 The overall block diagram of the power safety diagnosis method based on convolutional neural network provided in this application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms of "a", "said", and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings, and "multiple" generally includes at least two, but does not exclude the inclusion of at least one.

[0047] The present application provides a method for diagnosing electricity safety based on a convolutional neural network, which uses the calibration temperature value at the current diagnosis moment to select a high receptive field feature extraction channel or a low receptive field feature extraction channel, and performs feature extraction on the target area image at the current diagnosis moment to prevent the loss or blurring of target features, improve the response speed, reduce the response time, and achieve accurate and rapid diagnosis of electricity safety in the target area.

[0048] At the same time, this application also constructs a denoising correction module and designs a denoising calibration method. According to the temperature value of the target area at the current diagnostic moment and the corrected temperature value at the previous diagnostic moment, the burrs and noise of the temperature value are removed, and the corrected temperature value at the current diagnostic moment is obtained by comprehensive calculation, which improves the accuracy of temperature value acquisition and improves the subsequent power safety diagnosis results.

[0049] Finally, the present application also constructs a preprocessing module to receive the target area image at the current diagnosis moment of the target area, perform contrast enhancement preprocessing on the target area image, enhance the target features, and improve the accuracy of subsequent feature extraction.

[0050] The present application is further described below in conjunction with the accompanying drawings and specific embodiments.

[0051] like Figure 1 As shown, in one embodiment, the present application provides a power safety diagnosis method based on a convolutional neural network. The power safety diagnosis method specifically includes the following steps S1 to S6:

[0052] S1: Obtain the temperature value of the target area at the current diagnosis time, the target area image, and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, where N is greater than or equal to 2;

[0053] Specifically, a temperature sensor is selected to monitor the temperature of the target area, and a temperature value of the target area at a current diagnosis time and a temperature value at N-1 diagnosis times before the current diagnosis time are obtained.

[0054] A visual sensor is selected to perform image monitoring on the target area, and an image of the target area at a current diagnosis moment of the target area is obtained.

[0055] S2: Construct a denoising and correction module, receive the temperature value at the current diagnosis moment and the temperature values ​​at N-1 diagnosis moments before the current diagnosis moment, perform denoising and correction processing on the temperature value at the current diagnosis moment, and obtain a corrected temperature value at the current diagnosis moment.

[0056] In the process of acquiring the temperature of the target area, due to the interference of various environmental factors, such as electromagnetic noise, temperature sensor instability, etc., the directly acquired temperature value often contains many glitches and noise. These glitches and noise not only lead to inaccurate temperature values, but also lead to inappropriate selection of subsequent feature extraction channels, affecting the diagnosis results of power safety. By constructing a denoising and correction processing module, the temperature value of the target area at the current diagnosis time is received, and the glitches and noise of the temperature value are removed to obtain the corrected temperature value at the current diagnosis time.

[0057] Specifically, Figure 2 As shown, the temperature value at the current diagnosis moment is denoised and corrected to obtain the corrected temperature value at the current diagnosis moment, including the following steps:

[0058] S21: Based on the temperature value at the current diagnosis time and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, a maximum temperature value and a minimum temperature value are obtained by comparison.

[0059] S22: Calculate the optimal correction coefficient α based on the maximum temperature value, the minimum temperature value and the temperature values ​​at N diagnosis times opt , optimal correction coefficient α opt Specifically:

[0060]

[0061] Among them, X max is the maximum temperature value, X min is the minimum temperature value, X t is the temperature value at the diagnosis time t. The optimal correction coefficient α is calculated by this formula opt , which can ensure that the average temperature is within the temperature variation range X max -X min Within, and can ensure 0≤α opt ≤1.

[0062] S23: Based on the optimal correction coefficient α opt, the corrected temperature value at the previous diagnosis moment and the temperature value at the current diagnosis moment, and the corrected temperature value at the current diagnosis moment is calculated. The corrected temperature value at the current diagnosis moment is specifically:

[0063] S N =S N-1 +α opt (X N -S N-1 )

[0064] Among them, S N is the corrected temperature value at the current diagnosis time, S N-1 is the corrected temperature value at the previous diagnosis time, X N It is the temperature value at the current diagnosis time.

[0065] Specifically, the present application uses the temperature value X at the current diagnosis time N and the corrected temperature value S at the previous diagnosis time N-1 Construct the corrected temperature value S at the current diagnosis time N , set the correction coefficient α to adjust the temperature value X at the current diagnosis time N and the corrected temperature value S at the previous diagnosis time N-1 The weight of

[0066] S N =αX N +(1-α)S N-1

[0067]

[0068] The correction coefficient α at each diagnosis moment is different, 0≤α≤1. When the correction coefficient α is closer to 1, the correction temperature value S N The temperature value X at the current diagnosis time is equal to N , correction temperature value S N Mainly determined by the temperature value X at the current diagnosis time N When the correction coefficient α is closer to 0, the correction temperature value S N The more equal to the corrected temperature value S at the previous diagnosis time N-1 , correction temperature value S N Mainly composed of the corrected temperature value S at the previous diagnosis moment N-1 Decide.

[0069] Further, in step S22 of the present application, the temperature variation interval is defined by the temperature difference between the maximum temperature value and the minimum temperature value, according to Determine the degree of deviation between the average temperature and the maximum temperature value in the temperature variation range, and then use the degree of deviation to obtain the real-time optimal correction coefficient α opt, used to correct the temperature value X at the current diagnosis time N When the average temperature is close to the maximum temperature value of the temperature change range, the indication deviation is large, and the burrs and noise in the temperature value are serious. Therefore, the optimal correction coefficient α opt Close to 0, the corrected temperature value at the current diagnosis time is mainly determined by the corrected temperature value at the previous diagnosis time. On the contrary, when the temperature average value is far away from the maximum temperature value of the temperature change range, the indication deviation is small, and the burrs and noise in the temperature value are weak, so the optimal correction coefficient α opt Close to 1, the corrected temperature value at the current diagnosis time is mainly based on the temperature value X at the current diagnosis time N Decide.

[0070] This application constructs a denoising correction module based on the temperature value X of the target area at the current diagnosis time. N The corrected temperature value at the previous diagnosis moment is combined with the burrs and noise of the temperature value, and the corrected temperature value at the current diagnosis moment is obtained by comprehensive calculation, which improves the accuracy of temperature value acquisition and improves the subsequent power safety diagnosis results.

[0071] As a preferred embodiment, when the current diagnosis time is the initial diagnosis time, the correction temperature value is set to the temperature value of the target area at the current diagnosis time.

[0072] S3: construct a comparison module, receive the corrected temperature value at the current diagnosis moment, compare the corrected temperature value at the current diagnosis moment with a preset temperature threshold, and generate a selection signal.

[0073] When the corrected temperature value at the current diagnosis moment is less than the preset temperature threshold, it indicates that the possibility of a fire occurring in the target area at the current diagnosis moment is low; conversely, when the corrected temperature value at the current diagnosis moment is not less than the preset temperature threshold, it indicates that the possibility of a fire occurring in the target area at the current diagnosis moment is high.

[0074] S4: construct a preprocessing module, receive the target area image of the target area at the current diagnosis moment, perform contrast enhancement on the target area image, and generate a target area enhanced image.

[0075] In the early stage of a fire, the fire target in the target area image obtained by the visual sensor is relatively small and the fire target features are relatively blurred. Therefore, the present application performs contrast enhancement preprocessing on the target area image to enhance the target features and improve the accuracy of subsequent feature extraction.

[0076] Specifically, Figure 3 As shown, contrast enhancement is performed on the target area image to generate a target area enhanced image, which specifically includes the following steps:

[0077] S41: adjusting the grayscale values ​​of the pixels in the target area image, specifically:

[0078]

[0079] Among them, P is the gray value of the pixel in the target area image, P min is the low gray threshold, P max is the high grayscale threshold, and P1 is the grayscale value of the pixel after adjustment.

[0080] S42: normalizing the grayscale values ​​of the adjusted pixels to obtain an enhanced image of the target area. The normalization process is specifically as follows:

[0081]

[0082] Wherein, P2 is the gray value of the normalized pixel.

[0083] S5: construct a feature extraction module, receive the selection signal and the target area enhanced image, select a high receptive field feature extraction channel or a low receptive field feature extraction channel according to the selection signal, perform feature extraction on the target area enhanced image, and generate target features.

[0084] There is a direct relationship between the temperature value of the target area at the current diagnosis moment and whether a fire occurs. When a fire is about to occur or after a fire occurs, the temperature value of the target area will be relatively large. At this time, since the fire has just occurred, the fire target is relatively small and the fire target features are relatively vague. It is necessary to enhance the feature extraction capability of the feature extraction module at the current diagnosis moment. Therefore, a high receptive field feature extraction channel is selected through the high temperature value at the current diagnosis moment, and relatively more convolutional layers are used to enhance the feature extraction capability, prevent missed diagnosis, and improve the accuracy of the target area's power safety diagnosis. On the contrary, when there is no fire or under normal circumstances, the temperature value of the target area will be relatively small, and the possibility of a fire is relatively small. Therefore, a low receptive field feature extraction channel is selected through the low temperature value at the current diagnosis moment, and relatively fewer convolutional layers are used to improve the response speed of the diagnosis, reduce the response time, and achieve rapid diagnosis of power safety in the target area.

[0085] Specifically, the overall block diagram of the power safety diagnosis method based on convolutional neural network is as follows: Figure 4 As shown, a high receptive field feature extraction channel or a low receptive field feature extraction channel is selected according to a selection signal, and feature extraction is performed on the enhanced image of the target area to generate target features, specifically including the following steps:

[0086] S51: when the selection signal indicates that the corrected temperature value at the current diagnosis moment is less than the preset temperature threshold, a low receptive field feature extraction channel is selected, wherein the low receptive field feature extraction channel is sequentially connected by four convolutional layers, and feature extraction is performed on the enhanced image of the target area to generate target features;

[0087] S52: When the selection signal indicates that the corrected temperature value at the current diagnosis moment is not less than the preset temperature threshold, a high receptive field feature extraction channel is selected, wherein the high receptive field feature extraction channel is connected in sequence by 8 convolutional layers, and feature extraction is performed on the enhanced image of the target area to generate target features.

[0088] S6: Build a classifier, receive target features, classify the target features, and generate diagnosis results.

[0089] Specifically, the classifier uses the Softmax function to generate a diagnosis result.

[0090] The above description shows and describes the preferred embodiments of the present application, but as mentioned above, it should be understood that the present application is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified within the scope of the invention concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not depart from the spirit and scope of the present application, and should be within the scope of protection of the claims attached to the present application.

Claims

1. A power safety diagnosis method based on convolutional neural network, characterized in that: The method comprises steps S1 to S6: S1: Obtain the temperature value of the target area at the current diagnosis time, the target area image, and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, where N is greater than or equal to 2; S2: construct a denoising and correction module, receive the temperature value at the current diagnosis time and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, perform denoising and correction processing on the temperature value at the current diagnosis time, and obtain a corrected temperature value at the current diagnosis time; S3: constructing a comparison module, receiving the corrected temperature value at the current diagnosis moment, comparing the corrected temperature value at the current diagnosis moment with a preset temperature threshold, and generating a selection signal; S4: construct a preprocessing module, receive the target area image of the target area at the current diagnosis moment, perform contrast enhancement on the target area image, and generate an enhanced image of the target area; S5: construct a feature extraction module, receive the selection signal and the target area enhanced image, select a high receptive field feature extraction channel or a low receptive field feature extraction channel according to the selection signal, perform feature extraction on the target area enhanced image, and generate target features; S6: Build a classifier, receive target features, classify the target features, and generate diagnosis results.

2. The power safety diagnosis method according to claim 1, characterized in that: The denoising and correction processing of the temperature value at the current diagnosis time to obtain the corrected temperature value at the current diagnosis time includes steps S21 to S23: S21: Based on the temperature value at the current diagnosis time and the temperature values ​​at N-1 diagnosis times before the current diagnosis time, a maximum temperature value and a minimum temperature value are obtained by comparison; S22: Calculate the optimal correction coefficient according to the maximum temperature value, the minimum temperature value and the temperature values ​​at N diagnosis moments; S23: Based on the optimal correction coefficient α opt , the corrected temperature value at the previous diagnosis moment and the temperature value at the current diagnosis moment, and calculate the corrected temperature value at the current diagnosis moment.

3. The power safety diagnosis method according to claim 2, characterized in that: The optimal correction coefficient is specifically: Among them, α opt is the optimal correction coefficient, X max is the maximum temperature value, X min is the minimum temperature value, X t is the temperature value at the diagnosis time t.

4. The power safety diagnosis method according to claim 3, characterized in that: The corrected temperature value at the current diagnosis time is specifically: S N =S N-1 +α opt (X N -S N-1 ) Among them, S N is the corrected temperature value at the current diagnosis time, S N-1 is the corrected temperature value at the previous diagnosis time, X N It is the temperature value at the current diagnosis time.

5. The method for diagnosing power safety according to claim 1, characterized in that: The step of performing contrast enhancement on the target area image to generate the target area enhanced image includes steps S41 to S42: S41: adjusting the grayscale values ​​of the pixels of the target area image; S42: Normalizing the grayscale values ​​of the adjusted pixels to obtain an enhanced image of the target area.

6. The power safety diagnosis method according to claim 5, characterized in that: The grayscale value of the pixel of the target area image is adjusted, specifically: Among them, P is the gray value of the pixel in the target area image, P min is the low gray threshold, P max is the high grayscale threshold, and P1 is the grayscale value of the pixel after adjustment.

7. The power safety diagnosis method according to claim 6, characterized in that: The normalization process is specifically as follows: Wherein, P2 is the gray value of the normalized pixel.

8. The power safety diagnosis method according to claim 1, characterized in that: The method selects a high receptive field feature extraction channel or a low receptive field feature extraction channel according to the selection signal, extracts features from the target area enhanced image, and generates target features, specifically including steps S51 to S52: S51: when the selection signal indicates that the corrected temperature value at the current diagnosis moment is less than the preset temperature threshold, a low receptive field feature extraction channel is selected, wherein the low receptive field feature extraction channel is sequentially connected by four convolutional layers, and feature extraction is performed on the enhanced image of the target area to generate target features; S52: When the selection signal indicates that the corrected temperature value at the current diagnosis moment is not less than the preset temperature threshold, a high receptive field feature extraction channel is selected, wherein the high receptive field feature extraction channel is connected in sequence by 8 convolutional layers, and feature extraction is performed on the enhanced image of the target area to generate target features.

9. The power safety diagnosis method according to claim 1, characterized in that: The classifier uses a Softmax function to generate a diagnosis result.

Images