Electric energy meter abnormal phenomenon detection method and system based on image recognition

Abstract

The invention discloses an electric energy meter abnormal phenomenon detection method and system based on image recognition. The method comprises the following steps: monitoring an electric energy meter in real time through a high-speed camera, preprocessing an image by using a multistage filtering and edge fusion algorithm, extracting texture, shape and color features through frequency domain transformation, an edge gradient direction histogram and a color histogram, and obtaining a feature image after multi-feature fusion; and comparing a feature image after multi-feature fusion with a template image feature, detecting abnormality by adopting a composite difference method and composite correlation analysis, performing comprehensive judgment by combining local and global detection, and triggering dynamic video capture abnormality. According to the invention, by fusing multi-scale edge information, the edge features of the image are enhanced, and subsequent feature extraction and anomaly detection are facilitated; a comprehensive feature image is formed by fusing various features, so that the accuracy and robustness of feature extraction are improved; the two detection results are combined through comprehensive judgment, so that the accuracy and reliability of the detection results are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of electric energy meter detection, and in particular to an abnormal phenomenon detection method and system of an electric energy meter based on image recognition. Background Art

[0002] The accuracy and reliability of electric energy meters are crucial to power companies and users, because it is directly related to the fairness of electricity billing and the stable operation of the power system. Therefore, the detection and maintenance of electric energy meters are particularly important.

[0003] During the actual operation of the energy meter, abnormal phenomena such as sparks and screen flashing may occur, which will not only affect the metering accuracy of the energy meter, but also may cause equipment damage or even safety accidents. Therefore, developing a technology that can detect abnormal phenomena of the energy meter in real time and accurately is of great significance to ensure the normal operation of the energy meter and improve the reliability of power measurement.

[0004] As an advanced computer vision technology, image recognition technology has been widely used in various fields in recent years. Through high-speed cameras and advanced image processing algorithms, image recognition technology can realize real-time monitoring and analysis of target objects, with the advantages of high precision, high efficiency and high degree of automation. Therefore, the application of image recognition technology in the detection of abnormal phenomena in electric energy meters has broad application prospects.

[0005] The existing technology has at least the following technical problems: the existing technology has low accuracy and efficiency, and is easily affected by human factors. It is difficult to achieve real-time and accurate detection of abnormal phenomena such as sparks and screen flashes. It is prone to missed detections and false detections. It is difficult to maintain consistent detection effects in complex and changeable environments, and cannot adapt to the dynamic changes in the working state of the electricity meter, resulting in unreliable detection results; the response speed is slow when abnormal phenomena occur, and detailed information at the moment of the abnormality cannot be captured in real time, resulting in a lack of key data when analyzing abnormal phenomena. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the defects of the above-mentioned prior art and provide an abnormal phenomenon detection method for electric energy meters based on image recognition, which is convenient for feature extraction and anomaly detection, and can improve the accuracy and robustness of feature extraction to ensure the accuracy and reliability of the detection results.

[0007] To this end, the present invention adopts the following technical solution.

[0008] In a first aspect, the present invention provides a method for detecting abnormal phenomena of an electric energy meter based on image recognition, which comprises:

[0009] Step S1, real-time monitoring of the electric energy meter by a high-speed camera, preprocessing the image by using a multi-level filtering and edge fusion algorithm, extracting texture, shape and color features by frequency domain transformation, edge gradient direction histogram and color histogram, and obtaining a feature image after multi-feature fusion;

[0010] Step S2, compare the feature image after multi-feature fusion with the template image features, use composite difference method and composite correlation analysis to detect anomalies, combine local and global detection to make a comprehensive judgment, and trigger dynamic video to capture anomalies.

[0011] Furthermore, in multi-stage filtering, the image is processed by combining Gaussian filters of different scales, and the multi-scale features in the image are extracted by combining Gaussian functions of different standard deviations. The specific formula is as follows:

[0012]

[0013] Among them, G k (x, y) represents the output result of the filter of the kth scale at the image coordinate (x, y), a i is the weight of the i-th filter, σ i is the standard deviation of the ith filter, μ i and v i is the center coordinate of the i-th filter, and n is the number of filters.

[0014] Furthermore, the edge information of feature images of different scales is fused through the edge fusion algorithm to enhance the edge features of the image and form a fused edge-enhanced image. The specific formula is as follows:

[0015]

[0016] Where E(x, y) represents the value of the fused edge-enhanced image at the image coordinate (x, y), and w k is the fusion weight of the kth scale, is the gradient image of the kth scale, m is the number of fusion scales, yes The absolute value of represents the edge strength.

[0017] Furthermore, after completing the preprocessing of the image, feature extraction and anomaly detection are started. By analyzing the preprocessed image, key feature points and feature areas are extracted, and a multi-feature fusion method is adopted to achieve it through texture analysis, shape recognition and color feature fusion.

[0018] Furthermore, in order to obtain the texture features of the image, a complex method based on frequency domain transformation is used to convert the image into the frequency domain, and the high-frequency and low-frequency information is captured through Fourier transform. The specific formula is as follows:

[0019]

[0020] Among them, T(u, v) represents the representation of the preprocessed image E(x, y) in the frequency domain, is the Fourier transform operator, M and N are the width and height of the image, u and v are the coordinates in the frequency domain, and j is the ordinal unit;

[0021] A method based on edge gradient direction histogram is used to calculate the gradient histogram of the preprocessed image in different directions, and shape features are extracted through shape recognition. The specific formula is as follows:

[0022]

[0023] Among them, H r Represents the gradient histogram value in the rth direction, R r is the rth sub-region of the preprocessed image, is the gradient amplitude of the preprocessed image at point (x, y), is the gradient of the preprocessed image at point (x, y), θ(x, y) is the gradient direction, θ r is the central angle of the rth direction, δ is the Dirac function;

[0024] Convert the image to HSV color space, capture color distribution information through color histogram, and extract color features. The specific formula is as follows:

[0025]

[0026] Among them, C h It represents the color histogram value of the hth color channel of the image in the HSV space, and H(x, y) is the hue value of the preprocessed image E(x, y) at the point (x, y).

[0027] Furthermore, by extracting texture, shape and color features, a multi-feature fused feature image is formed. The specific formula is as follows:

[0028] F(x, y) = β 1 T(u,v)+β 2 H r +β 3 C h

[0029] Among them, F(x, y) represents the fused feature image, β1, β2 and β3 are weight coefficients, which control the influence proportion of texture, shape and color features respectively.

[0030] Furthermore, in the composite difference method, a feature image is obtained in advance by a multi-feature fusion method, and the fused feature image F(x, y) and the feature image F of the template image are compared. m (x, y) is compared to obtain the differential image D(x, y). The specific formula is:

[0031]

[0032] Where D(x, y) is the value of the difference image at the image coordinate (x, y), F m (x, y) is the feature map of the template image, α 1 and α 2 are weight coefficients, and ∈ is a constant to avoid the denominator being zero.

[0033] Furthermore, a composite correlation analysis is performed on the difference image to calculate the composite correlation coefficient between the feature map of the difference image and the template image. The specific formula is as follows:

[0034]

[0035] Among them, ρ c is the composite correlation coefficient, H D，l and H m，l are the values ​​of the lth feature point of the difference image and the template image, μ D and μ m are the mean values ​​of feature points of the difference image and template image respectively, L is the number of feature points, and ∈ is a constant to avoid the denominator being zero.

[0036] Furthermore, anomaly determination is performed based on the differential image and the composite correlation coefficient, and the determination rule is:

[0037] Local anomaly detection: Calculate the difference values ​​of all pixels in the differential image and compare them with the local threshold. If the number of pixels whose difference values ​​exceed the local threshold exceeds the set ratio p, it is considered that there is a local anomaly.

[0038] Global anomaly detection: Calculate the composite correlation coefficient and compare it with the global threshold; if the composite correlation coefficient is less than the global threshold, it is considered that there is a global anomaly;

[0039] Comprehensive judgment: Combine the local anomaly detection and global anomaly detection results. If any one of the detection results is judged as abnormal, it is considered that an abnormal phenomenon has been detected.

[0040] In a second aspect, the present invention provides an abnormal phenomenon detection system for an electric energy meter based on image recognition, which comprises:

[0041] Feature image acquisition unit with multi-feature fusion: The unit monitors the electric energy meter in real time through a high-speed camera, pre-processes the image using multi-level filtering and edge fusion algorithms, extracts texture, shape and color features through frequency domain transformation, edge gradient direction histogram and color histogram, and obtains a feature image after multi-feature fusion;

[0042] Anomaly comprehensive judgment unit: compares the feature image after multi-feature fusion with the template image features, uses composite difference method and composite correlation analysis to detect anomalies, combines local and global detection for comprehensive judgment, and triggers dynamic video to capture anomalies.

[0043] The present invention has the following beneficial effects:

[0044] 1. Use multi-level filtering and edge fusion algorithms to pre-process the real-time collected image data, and extract multi-scale feature information in the image through a combination of Gaussian filters of different scales, providing basic data for subsequent edge fusion and feature extraction; edge fusion technology enhances the edge features of the image by fusing multi-scale edge information, which facilitates subsequent feature extraction and anomaly detection.

[0045] 2. Extract the key feature points and feature areas of the image through texture analysis, shape recognition and color feature fusion methods. Use the texture analysis method based on frequency domain transformation to convert the image to the frequency domain, capture high-frequency and low-frequency information through Fourier transform, and identify texture features; use the method based on edge gradient direction histogram to calculate the gradient histogram of the image in different directions and extract shape features; capture color distribution information and extract color features by converting the image to HSV color space; and form a comprehensive feature image by fusing multiple features to improve the accuracy and robustness of feature extraction.

[0046] 3. Detect abnormal changes in images through composite difference method and composite correlation analysis. In the composite difference method, the feature map of the template image is obtained in advance through the method of multi-feature fusion, and compared with the comprehensive feature image of the real-time image to obtain a differential image; the differential image identifies possible abnormal phenomena by calculating the composite difference between the fused feature image and the template image; the composite correlation analysis further confirms abnormal phenomena by calculating multiple correlation features between images.

[0047] 4. Combine local anomaly detection and global anomaly detection to comprehensively determine abnormal phenomena. Local anomaly detection determines whether there is a local anomaly by calculating the difference values ​​of all pixels in the differential image and comparing them with the local threshold. Global anomaly detection determines whether there is a global anomaly by calculating the composite correlation coefficient and comparing it with the global threshold. Comprehensive judgment combines the two detection results to ensure the accuracy and reliability of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flow chart of a method for detecting abnormal phenomena of an electric energy meter based on image recognition according to the present invention;

[0049] Figure 2 The present invention is a structural diagram of an abnormal phenomenon detection system of an electric energy meter based on image recognition. DETAILED DESCRIPTION

[0050] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0052] Example 1

[0053] See attached Figure 1 , which shows a flow chart of a method for detecting abnormal phenomena of an electric energy meter based on image recognition provided by an embodiment of the present invention, the method comprising the following steps:

[0054] S1. The electric energy meter is monitored in real time by a high-speed camera. The image is preprocessed using a multi-level filtering and edge fusion algorithm. The texture, shape and color features are extracted through frequency domain transformation, edge gradient direction histogram and color histogram to obtain the fused feature image.

[0055] Real-time monitoring of the electric energy meter test area is achieved by configuring a high-speed camera. A high-speed camera with a frame rate of 120 frames per second is selected and installed above the electric energy meter test area to ensure that the camera's field of view can fully cover the entire test area and avoid monitoring blind spots. The high-speed camera collects image data of the electric energy meter's working status in real time and transmits the image data to the visual software processing module for processing; the visual software has image acquisition, preprocessing, feature extraction and anomaly recognition functions, and uses multi-level filtering and edge fusion algorithms to preprocess the real-time collected image data.

[0056] In multi-stage filtering, the image is processed by combining Gaussian filters of different scales, and the multi-scale features in the image are extracted by combining Gaussian functions of different standard deviations. The specific formula is as follows:

[0057]

[0058] Among them, G k(x, y) represents the output result of the filter of the kth scale at the image coordinate (x, y), a i is the weight of the i-th filter, σ i is the standard deviation of the ith filter, μ i and v i is the center coordinate of the ith filter, and n is the number of filters. Through Gaussian filters of different scales, multi-scale feature information in the image is extracted to provide basic data for subsequent edge fusion and feature extraction.

[0059] Through edge fusion technology, feature images of different scales are fused with edge information to enhance the edge features of the image and form a fused edge-enhanced image. The specific formula is as follows:

[0060]

[0061] Where E(x, y) represents the value of the fused edge-enhanced image at the image coordinate (x, y), and w k is the fusion weight of the kth scale, is the gradient image of the kth scale, m is the number of fusion scales, yes The absolute value of represents the edge strength. By fusing multi-scale edge information, the edge features of the image are enhanced, which facilitates subsequent feature extraction and anomaly detection.

[0062] After completing the preprocessing of the image, feature extraction and anomaly detection are started. By analyzing the preprocessed image, key feature points and feature areas are extracted, and the multi-feature fusion method is adopted to achieve it through texture analysis, shape recognition and color feature fusion. In order to obtain the texture features of the image, a complex method based on frequency domain transformation is adopted to convert the image to the frequency domain and capture high-frequency and low-frequency information through Fourier transform. The formula is as follows:

[0063]

[0064] Among them, T(u, v) represents the representation of the preprocessed image E(x, y) in the frequency domain, is the Fourier transform operator, M and N are the width and height of the image, u and v are the coordinates in the frequency domain, and j is the ordinal unit. By converting the preprocessed image to the frequency domain, the frequency domain features of the preprocessed image are extracted, high-frequency and low-frequency information are captured, and texture features are identified.

[0065] A method based on edge gradient direction histogram is used to calculate the gradient histogram of the preprocessed image in different directions, and shape features are extracted through shape recognition. The specific formula is as follows:

[0066]

[0067] Among them, H r Represents the gradient histogram value in the rth direction, R r is the rth sub-region of the preprocessed image, is the gradient amplitude of the preprocessed image at point (x, y), is the gradient of the preprocessed image at point (x, y), θ(x, y) is the gradient direction, θ r is the central angle of the rth direction, and δ is the Dirac function.

[0068] Convert the image to HSV color space, capture color distribution information through color histogram, and extract color features. The specific formula is as follows:

[0069]

[0070] Among them, C h It represents the color histogram value of the hth color channel of the image in the HSV space, and H(x, y) is the hue value of the preprocessed image E(x, y) at the point (x, y).

[0071] By extracting texture, shape and color features, a comprehensive feature image of multi-feature fusion is formed. The specific formula is as follows:

[0072] F(x, y) = β 1 T(u,v)+β 2 H r +β 3 C h

[0073] Among them, F(x, y) represents the fused feature image, β 1 , β 2 and β 3 are weight coefficients, which control the influence ratio of texture, shape and color features respectively. The purpose of this step is to form a comprehensive feature image by fusing multiple features to improve the accuracy and robustness of feature extraction.

[0074] S2. Compare the fused feature image with the template image features, use the composite difference method and composite correlation analysis to detect anomalies, combine local and global detection for comprehensive judgment, and trigger dynamic video to capture anomalies.

[0075] The fused feature image is compared with the image template under normal working conditions, and the composite difference method and composite correlation analysis are used to detect abnormal changes in the image. In the composite difference method, the feature map of the template image is obtained in advance by the above-mentioned multi-feature fusion method, and the fused feature image F(x, y) and the feature map of the template image F are compared. m(x, y) is compared to obtain the differential image D(x, y). By calculating the composite difference between the fused feature image and the feature map of the template image, possible abnormal phenomena can be identified. The specific formula is:

[0076]

[0077] Where D(x, y) is the value of the difference image at the image coordinate (x, y), F m (x, y) is the feature map of the template image, α 1 and α 2 is the weight coefficient, and ∈ is a constant to avoid the denominator being zero.

[0078] Perform composite correlation analysis on the differential image to calculate the composite correlation coefficient between the feature map of the differential image and the template image. Composite correlation analysis further confirms abnormal phenomena by calculating multiple correlation features between images. The specific formula is as follows:

[0079]

[0080] Among them, ρ c is the composite correlation coefficient, H D，l and H m，l are the values ​​of the lth feature point of the difference image and the template image, μ D and μ m are the mean values ​​of feature points of the difference image and template image respectively, L is the number of feature points, and ∈ is a constant to avoid the denominator being zero.

[0081] The abnormality is judged based on the differential image and the composite correlation coefficient. The judgment rules are as follows:

[0082] Local anomaly detection: Calculate the difference value d(x, y) of all pixels in the differential image D(x, y) and compare it with the local threshold; if the number of pixels whose difference value exceeds the local threshold exceeds the set proportion p, it is considered that there is a local anomaly. The formula is expressed as:

[0083]

[0084] Among them, A represents the local anomaly, and ε is the local threshold, which is set in advance.

[0085] Global anomaly detection: Calculate the composite correlation coefficient and compare it with the global threshold; if the composite correlation coefficient is less than the global threshold, it is considered that a global anomaly exists.

[0086] If ρ c <τ, then B

[0087] Among them, B represents the global anomaly, and τ is the global threshold, which is set in advance.

[0088] Comprehensive judgment: Combine the local anomaly detection and global anomaly detection results. If any one of the detection results is judged as abnormal, it is considered that an abnormal phenomenon has been detected.

[0089] When an abnormal phenomenon is detected, the dynamic video capture function is automatically triggered to save 10 seconds of dynamic video before and after the abnormality occurs for detailed analysis; and metadata such as timestamp, abnormality type and detection equipment number are added to each captured video to facilitate later data management and backtracking.

[0090] In terms of data management, a data management system is established to classify and store the captured abnormal phenomenon videos by abnormal type, and an index database is established to support fast retrieval and query. Efficient retrieval of video data based on time, abnormal type and other conditions facilitates testers to conduct later analysis and verification.

[0091] In summary, a method for detecting abnormal phenomena in electric energy meters based on image recognition has been completed.

[0092] Example 2

[0093] This embodiment provides an abnormal phenomenon detection system for electric energy meters based on image recognition, such as Figure 2 As shown, it consists of a feature image acquisition unit for multi-feature fusion and an abnormal comprehensive judgment unit.

[0094] The multi-feature fused feature image acquisition unit monitors the electric energy meter in real time through a high-speed camera, pre-processes the image using a multi-level filtering and edge fusion algorithm, extracts texture, shape and color features through frequency domain transformation, edge gradient direction histogram and color histogram, and obtains a multi-feature fused feature image.

[0095] The abnormal comprehensive judgment unit compares the feature image after multi-feature fusion with the template image feature, uses the composite difference method and composite correlation analysis to detect abnormalities, combines local and global detection to make a comprehensive judgment, and triggers dynamic video to capture abnormalities.

[0096] The feature image acquisition unit of the multi-feature fusion processes the image by combining Gaussian filters of different scales in multi-stage filtering, and extracts multi-scale features in the image by combining Gaussian functions of different standard deviations. The specific formula is as follows:

[0097]

[0098] Among them, G k (x, y) represents the output result of the filter of the kth scale at the image coordinate (x, y), a i is the weight of the i-th filter, σ i is the standard deviation of the ith filter, μi and v i is the center coordinate of the ith filter, and n is the number of filters. Through Gaussian filters of different scales, multi-scale feature information in the image is extracted to provide basic data for subsequent edge fusion and feature extraction.

[0099] The multi-feature fusion feature image acquisition unit fuses edge information of feature images of different scales through an edge fusion algorithm, enhances the edge features of the image, and forms a fused edge-enhanced image. The specific formula is as follows:

[0100]

[0101] Where E(x, y) represents the value of the fused edge-enhanced image at the image coordinate (x, y), and w k is the fusion weight of the kth scale, is the gradient image of the kth scale, m is the number of fusion scales, yes The absolute value of represents the edge strength. By fusing multi-scale edge information, the edge features of the image are enhanced, which facilitates subsequent feature extraction and anomaly detection.

[0102] After completing the preprocessing of the image, feature extraction and anomaly detection begin. By analyzing the preprocessed image, key feature points and feature areas are extracted. The multi-feature fusion method is adopted to achieve this through texture analysis, shape recognition and color feature fusion.

[0103] In order to obtain the texture features of the image, a complex method based on frequency domain transformation is used to convert the image to the frequency domain, and the high-frequency and low-frequency information is captured through Fourier transform. The specific formula is as follows:

[0104]

[0105] Among them, T(u, v) represents the representation of the preprocessed image E(x, y) in the frequency domain, is the Fourier transform operator, M and N are the width and height of the image, u and v are the coordinates in the frequency domain, and j is the ordinal unit;

[0106] A method based on edge gradient direction histogram is used to calculate the gradient histogram of the preprocessed image in different directions, and shape features are extracted through shape recognition. The specific formula is as follows:

[0107]

[0108] Among them, H r Represents the gradient histogram value in the rth direction, R r is the rth sub-region of the preprocessed image, is the gradient amplitude of the preprocessed image at point (x, y), is the gradient of the preprocessed image at point (x, y), θ(x, y) is the gradient direction, θ r is the central angle of the rth direction, δ is the Dirac function;

[0109] Convert the image to HSV color space, capture color distribution information through color histogram, and extract color features. The specific formula is as follows:

[0110]

[0111] Among them, C h It represents the color histogram value of the hth color channel of the image in the HSV space, and H(x, y) is the hue value of the preprocessed image E(x, y) at the point (x, y).

[0112] By extracting texture, shape and color features, a feature image of multi-feature fusion is formed. The specific formula is as follows:

[0113] F(x, y) = β 1 T(u,v)+β 2 H r +β 3 C h

[0114] Among them, F(x, y) represents the fused feature image, β 1 , β 2 and β 3 They are all weight coefficients, which control the influence proportion of texture, shape and color features respectively.

[0115] In the abnormal comprehensive judgment unit, the fused feature image is compared with the image template under normal working conditions, and the composite difference method and composite correlation analysis are used to detect abnormal changes in the image.

[0116] In the composite difference method, the feature map of the template image is obtained in advance by the above-mentioned multi-feature fusion method, and the fused feature image F(x, y) and the feature map of the template image F m (x, y) is compared to obtain the differential image D(x, y). The specific formula is:

[0117]

[0118] Where D(x, y) is the value of the difference image at the image coordinate (x, y), F m (x, y) is the feature map of the template image, α 1 and α 2 are weight coefficients, and ∈ is a constant to avoid the denominator being zero.

[0119] Perform composite correlation analysis on the difference image and calculate the composite correlation coefficient between the feature map of the difference image and the template image. The specific formula is as follows:

[0120]

[0121] Among them, ρ c is the composite correlation coefficient, H D，l and H m，l are the values ​​of the lth feature point of the difference image and the template image, μ D and μ m are the mean values ​​of feature points of the difference image and template image respectively, L is the number of feature points, and ∈ is a constant to avoid the denominator being zero.

[0122] In the abnormal comprehensive judgment unit, abnormal judgment is performed based on the differential image and the composite correlation coefficient, and the judgment rule is:

[0123] Local anomaly detection: Calculate the difference values ​​of all pixels in the differential image and compare them with the local threshold. If the number of pixels whose difference values ​​exceed the local threshold exceeds the set ratio p, it is considered that there is a local anomaly.

[0124] Global anomaly detection: Calculate the composite correlation coefficient and compare it with the global threshold; if the composite correlation coefficient is less than the global threshold, it is considered that there is a global anomaly;

[0125] Comprehensive judgment: Combine the local anomaly detection and global anomaly detection results. If any one of the detection results is judged as abnormal, it is considered that an abnormal phenomenon has been detected.

[0126] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0127] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0128] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormal phenomena in electric energy meters based on image recognition, characterized in that: include: Step S1, real-time monitoring of the electric energy meter by a high-speed camera, preprocessing the image by using a multi-level filtering and edge fusion algorithm, extracting texture, shape and color features by frequency domain transformation, edge gradient direction histogram and color histogram, and obtaining a feature image after multi-feature fusion; Step S2, compare the feature image after multi-feature fusion with the template image features, use composite difference method and composite correlation analysis to detect anomalies, combine local and global detection to make a comprehensive judgment, and trigger dynamic video to capture anomalies.

2. The method for detecting abnormal phenomena of electric energy meters according to claim 1, characterized in that: In multi-stage filtering, the image is processed by combining Gaussian filters of different scales, and the multi-scale features in the image are extracted by combining Gaussian functions of different standard deviations. The specific formula is as follows: Among them, G k (x, y) represents the output result of the filter of the kth scale at the image coordinate (x, y), a i is the weight of the i-th filter, σ i is the standard deviation of the ith filter, μ i and v i is the center coordinate of the i-th filter, and n is the number of filters.

3. The method for detecting abnormal phenomena of electric energy meters according to claim 2, characterized in that: Through the edge fusion algorithm, the edge information of feature images of different scales is fused to enhance the edge features of the image and form a fused edge-enhanced image. The specific formula is as follows: Where E(x, y) represents the value of the fused edge-enhanced image at the image coordinate (x, y), and w k is the fusion weight of the kth scale, is the gradient image of the kth scale, m is the number of fusion scales, yes The absolute value of represents the edge strength.

4. The method for detecting abnormal phenomena of electric energy meters according to claim 1, characterized in that: After completing the preprocessing of the image, feature extraction and anomaly detection begin. By analyzing the preprocessed image, key feature points and feature areas are extracted. The multi-feature fusion method is adopted to achieve this through texture analysis, shape recognition and color feature fusion.

5. The method for detecting abnormal phenomena of electric energy meters according to claim 4, characterized in that: In order to obtain the texture features of the image, a complex method based on frequency domain transformation is used to convert the image to the frequency domain and capture high-frequency and low-frequency information through Fourier transform. The specific formula is as follows: Among them, T(u, v) represents the representation of the preprocessed image E(x, y) in the frequency domain, is the Fourier transform operator, M and N are the width and height of the image, u and v are the coordinates in the frequency domain, and j is the ordinal unit; A method based on edge gradient direction histogram is used to calculate the gradient histogram of the preprocessed image in different directions, and shape features are extracted through shape recognition. The specific formula is as follows: Among them, H r Represents the gradient histogram value in the rth direction, R r is the rth sub-region of the preprocessed image, is the gradient amplitude of the preprocessed image at point (x, y), is the gradient of the preprocessed image at point (x, y), θ(x, y) is the gradient direction, θ r is the central angle of the rth direction, δ is the Dirac function; Convert the image to HSV color space, capture color distribution information through color histogram, and extract color features. The specific formula is as follows: Among them, C h It represents the color histogram value of the hth color channel of the image in the HSV space, and H(x, y) is the hue value of the preprocessed image E(x, y) at the point (x, y).

6. The method for detecting abnormal phenomena of electric energy meters according to claim 5, characterized in that: By extracting texture, shape and color features, a multi-feature fused feature image is formed. The specific formula is as follows: F(x,y)=β1T(u,v)+β2H r +β3C h Among them, F(x, y) represents the fused feature image, β1, β2 and β3 are weight coefficients, which control the influence proportion of texture, shape and color features respectively.

7. The method for detecting abnormal phenomena of electric energy meters according to claim 1, characterized in that: In the composite difference method, the feature map of the template image is obtained in advance by the method of multi-feature fusion, and the fused feature image F(x, y) and the feature map F of the template image are compared. m (x, y) is compared to obtain the differential image D(x, y). The specific formula is: Where D(x, y) is the value of the difference image at the image coordinate (x, y), F m (x, y) is the feature map of the template image, α1 and α2 are weight coefficients, and ∈ is a constant to avoid the denominator being zero.

8. The method for detecting abnormal phenomena in an electric energy meter according to claim 7, characterized in that: Perform composite correlation analysis on the difference image and calculate the composite correlation coefficient between the feature map of the difference image and the template image. The specific formula is as follows: Among them, ρ c is the composite correlation coefficient, H D，l and H m，l are the values ​​of the lth feature point of the difference image and the template image, μ D and μ m are the mean values ​​of feature points of the difference image and template image respectively, L is the number of feature points, and ∈ is a constant to avoid the denominator being zero.

9. The method for detecting abnormal phenomena in an electric energy meter according to claim 6, characterized in that: The abnormality is judged based on the differential image and the composite correlation coefficient. The judgment rules are as follows: Local anomaly detection: Calculate the difference values ​​of all pixels in the differential image and compare them with the local threshold. If the number of pixels whose difference values ​​exceed the local threshold exceeds the set ratio p, it is considered that there is a local anomaly. Global anomaly detection: Calculate the composite correlation coefficient and compare it with the global threshold; if the composite correlation coefficient is less than the global threshold, it is considered that there is a global anomaly; Comprehensive judgment: Combine the local anomaly detection and global anomaly detection results. If any one of the detection results is judged as abnormal, it is considered that an abnormal phenomenon has been detected.

10. An abnormal phenomenon detection system for electric energy meters based on image recognition, used to implement the abnormal phenomenon detection method for electric energy meters according to any one of claims 1 to 9, characterized in that: include: Feature image acquisition unit with multi-feature fusion: The unit monitors the electric energy meter in real time through a high-speed camera, pre-processes the image using multi-level filtering and edge fusion algorithms, extracts texture, shape and color features through frequency domain transformation, edge gradient direction histogram and color histogram, and obtains a feature image after multi-feature fusion; Anomaly comprehensive judgment unit: compares the feature image after multi-feature fusion with the template image features, uses composite difference method and composite correlation analysis to detect anomalies, combines local and global detection for comprehensive judgment, and triggers dynamic video to capture anomalies.

Images