A method for detecting appearance anomalies of power equipment based on texture features

Abstract

The present invention relates to a method for detecting appearance anomalies of power equipment based on texture features, comprising: obtaining image information of the power equipment and preprocessing the information; comparing the current captured image with historical images of normal appearance under the same environmental factors, and judging whether the historical images are the same image or similar images; when the number of the same image and the number of similar images are both greater than or equal to one, judging whether the current captured image is normal or abnormal; when there is only one same image and one similar image, judging whether the current captured image is normal or abnormal; if there are no same images and similar images, recording relevant information. The present invention proposes to compare the similarity of the same or similar images with the current captured image by detecting environmental variables, so as to achieve the purpose of detecting the appearance anomalies of the monitored power equipment; the method can monitor equipment in a variety of different application scenarios, has stronger adaptability, and can effectively improve the accuracy of appearance anomaly detection.

Description

Technical Field

[0001] The present invention relates to the technical field of power system equipment detection, and in particular to a method for detecting appearance anomalies of power equipment based on texture features. Background Art

[0002] The power transmission and transformation equipment of the power system is widespread and most of them are outdoors for a long time. They are very vulnerable to various natural disasters, meteorological conditions and external damage, which to a certain extent brings hidden dangers to the stability and safety of the operation of the power system. According to user statistics, tripping events caused by external hidden dangers such as fire, ice damage, dancing, flashover, degradation, and external damage are increasing year by year. With the large-scale application of video surveillance and the theoretical research and application of image processing and recognition technology, good application effects have been achieved, such as transformers, high-voltage insulators, knife switch status in substations, and foreign body intrusion in transmission lines, etc. However, the existing technology still has problems such as limited detection function and low recognition accuracy.

[0003] The patent application with application number 201610207490.1 and invention name “Method and system for identifying abnormal power equipment images” records that the power equipment features are extracted from the power equipment image, and the power equipment and the image area where it is located are identified based on the power equipment features, and then the image of the image area is compared with the corresponding comparison image to determine whether there are abnormal changes, thereby identifying the abnormal power equipment image, but the method does not mention the specific method steps for extracting image features.

[0004] The patent application with application number 201510229248.X and invention name “A method for detecting abnormalities in the appearance of power equipment based on image comparison” proposes to compare two images taken by the inspection robot at the same stop point and the same angle at different times, and to detect abnormalities such as damage to power equipment and hanging foreign objects by judging the regional changes between the current image and the historical inspection image with the same content. However, the method does not mention the specific method of how to eliminate image recognition detection in different environmental backgrounds. Summary of the invention

[0005] The purpose of the present invention is to provide a method for detecting appearance anomalies of electric equipment based on texture features, which obtains image information of electric equipment, performs condition comparison according to environmental factor information, and compares the similarity of identical or similar images, thereby determining whether the electric equipment in the current captured image is abnormal.

[0006] To achieve the above object, the present invention adopts the following technical solution: a method for detecting appearance anomalies of power equipment based on texture features, the method comprising the following steps in order:

[0007] (1) acquiring image information of the electric power equipment and preprocessing the image information thereof, wherein the preprocessed image information of the electric power equipment includes environmental factor information;

[0008] (2) Compare the current captured image with historical images with normal appearance under the same environmental factors to determine whether the historical images are the same or similar images;

[0009] (3) If the number of identical images and similar images is greater than or equal to 1, the current captured image and one of the identical images are segmented into multiple parts, and the cosine similarity calculation is performed on the parts at the same position after segmentation to obtain multiple cosine similarities, and the multiple cosine similarities are averaged to obtain a cosine similarity mean value; by analogy, the cosine similarity calculation is performed on the current captured image and all identical images. If all the cosine similarity means are greater than or equal to 0.9, the current captured image is considered normal, otherwise, the current captured image is considered abnormal; the cosine similarity calculation is performed on the segmented part of the current captured image with a cosine similarity less than 0.9 and the part at the same position after segmentation of a similar image to obtain multiple cosine similarities, and the multiple cosine similarities are averaged to obtain a cosine similarity mean value; by analogy, the cosine similarity calculation is performed on the current captured image and all the similar images. If all the cosine similarity means are greater than or equal to 0.8, the current captured image is considered normal, otherwise, the current captured image is considered abnormal;

[0010] (4) If there is only one identical image or similar image, the current image and the identical image or similar image are segmented into several parts, and the cosine similarity of the segments at the same position is calculated. If the mean values ​​of all cosine similarities are greater than or equal to 0.8, the current image is considered normal; otherwise, the current image is considered abnormal.

[0011] (5) If there are no identical or similar images, record the relevant information.

[0012] The step (1) specifically refers to: acquiring image information of the power equipment, and processing it according to environmental factors so that its data information carries environmental factor information, wherein the environmental factor information includes light adaptability information and natural weather information, the light adaptability information includes light sensor data, the longitude and latitude of the location where the image is taken, and the date and time of the image being taken, and the natural weather information includes rain and snow sensor data, environmental dust and haze sensor data, and temperature, humidity and wind sensor data.

[0013] The step (2) specifically refers to: comparing the current captured image with the historical image with normal appearance under the same environmental factors; if all five conditions from condition 1 to condition 5 are met, then the current captured image is judged to be the same as the historical image with normal appearance under the same environmental factors, and the historical image is judged to be the same image; if at least two of the five conditions from condition 1 to condition 5 are met and condition 2 is included, then the current captured image is judged to be similar to the historical image with normal appearance under the same environmental factors, and the historical image is judged to be a similar image;

[0014] Condition 1: retrieve the latitude and longitude of the current image shooting location and the date and time of the image shooting, and compare them with the historical image. If the longitude and latitude of the image shooting location are the same and the date and time of the image shooting are within 10 days, the historical image is considered to meet the conditions;

[0015] Condition 2: retrieve the illumination sensor data of the current image capture location and compare it with the historical image. If the illumination difference is less than ±10%, the historical image is considered to meet the condition.

[0016] Condition 3: retrieve the rain and snow sensor data at the location where the current image is taken and compare it with the historical image. If the data is the same, the historical image is considered to meet the conditions;

[0017] Condition 4: retrieve the environmental dust and haze sensor data of the current image shooting location and compare it with the historical image. If the data difference is less than ±20%, the historical image is considered to meet the condition;

[0018] Condition 5: retrieve the temperature, humidity and wind sensor data of the current image shooting location and compare it with the historical image. If the data difference is less than ±10%, the historical image is considered to meet the conditions.

[0019] The step (3) specifically refers to:

[0020] If the number of identical images and similar images is greater than or equal to 1, the current image and the same image are divided into small squares by area. Each small square is named (nm), where n represents the column and m represents the row. The cosine similarity is calculated for the identical parts of the current image and the same image (nm). A represents the vector of the small square (nm) in the current image, and B represents the vector of the small square (nm) in the same image. The cosine value between A and B is calculated by using the Euclidean dot product formula:

[0021] A·B=∣∣A∣∣∣∣B∣cosθ

[0022] Among them, θ represents the angle between the two phase quantities A and B. The cosine value of the angle θ is calculated by the following formula:

[0023]

[0024] Here A i , B i Represent the components of vectors A and B respectively;

[0025] The similarity given ranges from -1 to 1: -1 means that the two vectors point in opposite directions, 1 means that they point in exactly the same direction, 0 generally means that they are independent, and values ​​in between indicate intermediate similarities or dissimilarities.

[0026] It can be seen from the above technical scheme that the beneficial effects of the present invention are: First, the present invention proposes to compare the similarity of the same or similar images with the currently captured image by detecting environmental variables, so as to achieve the purpose of detecting the appearance abnormality of the monitored power equipment; Second, this method can monitor equipment in a variety of different application scenarios, has stronger adaptability, and can effectively improve the accuracy of appearance abnormality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the method of the present invention;

[0028] Figure 2 Schematic diagram of the corresponding positions of image segmentation. DETAILED DESCRIPTION

[0029] like Figure 1 As shown, a method for detecting appearance anomalies of electric power equipment based on texture features comprises the following steps in order:

[0030] (1) acquiring image information of the electric power equipment and preprocessing the image information thereof, wherein the preprocessed image information of the electric power equipment includes environmental factor information;

[0031] (2) Compare the current captured image with historical images with normal appearance under the same environmental factors to determine whether the historical images are the same or similar images;

[0032] (3) If the number of identical images and similar images is greater than or equal to 1, the current captured image and one of the identical images are segmented into multiple parts, and the cosine similarity calculation is performed on the parts at the same position after segmentation to obtain multiple cosine similarities, and the multiple cosine similarities are averaged to obtain a cosine similarity mean value; by analogy, the cosine similarity calculation is performed on the current captured image and all identical images. If all the cosine similarity means are greater than or equal to 0.9, the current captured image is considered normal, otherwise, the current captured image is considered abnormal; the cosine similarity calculation is performed on the segmented part of the current captured image with a cosine similarity less than 0.9 and the part at the same position after segmentation of a similar image to obtain multiple cosine similarities, and the multiple cosine similarities are averaged to obtain a cosine similarity mean value; by analogy, the cosine similarity calculation is performed on the current captured image and all the similar images. If all the cosine similarity means are greater than or equal to 0.8, the current captured image is considered normal, otherwise, the current captured image is considered abnormal;

[0033] (4) If there is only one identical image or similar image, the current image and the identical image or similar image are segmented into several parts, and the cosine similarity of the segments at the same position is calculated. If the mean values ​​of all cosine similarities are greater than or equal to 0.8, the current image is considered normal; otherwise, the current image is considered abnormal.

[0034] (5) If there are no identical or similar images, record the relevant information.

[0035] The step (1) specifically refers to: acquiring image information of the power equipment, and processing it according to environmental factors so that its data information carries environmental factor information, wherein the environmental factor information includes light adaptability information and natural weather information, the light adaptability information includes light sensor data, the longitude and latitude of the location where the image is taken, and the date and time of the image being taken, and the natural weather information includes rain and snow sensor data, environmental dust and haze sensor data, and temperature, humidity and wind sensor data.

[0036] The step (2) specifically refers to: comparing the current captured image with the historical image with normal appearance under the same environmental factors; if all five conditions from condition 1 to condition 5 are met, then the current captured image is judged to be the same as the historical image with normal appearance under the same environmental factors, and the historical image is judged to be the same image; if at least two of the five conditions from condition 1 to condition 5 are met and condition 2 is included, then the current captured image is judged to be similar to the historical image with normal appearance under the same environmental factors, and the historical image is judged to be a similar image;

[0037] Condition 1: retrieve the latitude and longitude of the current image shooting location and the date and time of the image shooting, and compare them with the historical image. If the longitude and latitude of the image shooting location are the same and the date and time of the image shooting are within 10 days, the historical image is considered to meet the conditions;

[0038] Condition 2: retrieve the illumination sensor data of the current image capture location and compare it with the historical image. If the illumination difference is less than ±10%, the historical image is considered to meet the condition.

[0039] Condition 3: retrieve the rain and snow sensor data at the location where the current image is taken and compare it with the historical image. If the data is the same, the historical image is considered to meet the conditions;

[0040] Condition 4: retrieve the environmental dust and haze sensor data of the current image shooting location and compare it with the historical image. If the data difference is less than ±20%, the historical image is considered to meet the condition;

[0041] Condition 5: retrieve the temperature, humidity and wind sensor data of the current image shooting location and compare it with the historical image. If the data difference is less than ±10%, the historical image is considered to meet the conditions.

[0042] The step (3) specifically refers to:

[0043] like Figure 2 As shown in the figure, if the number of identical images and similar images is greater than or equal to 1, the current captured image and the identical image are divided into small blocks evenly by area. Each small block is named (nm), where n represents the column and m represents the row. The cosine similarity between the current captured image and the identical image (nm) is calculated. A represents the vector of the small block (nm) in the current captured image, and B represents the vector of the small block (nm) in the identical image. The cosine value between A and B is calculated by using the Euclidean dot product formula:

[0044] A·B=∣∣A∣∣∣∣B∣cosθ

[0045] Among them, θ represents the angle between the two phase quantities A and B. The cosine value of the angle θ is calculated by the following formula:

[0046]

[0047] Here A i , B i Represent the components of vectors A and B respectively;

[0048] The similarity given ranges from -1 to 1: -1 means that the two vectors point in opposite directions, 1 means that they point in exactly the same direction, 0 generally means that they are independent, and values ​​in between indicate intermediate similarities or dissimilarities.

[0049] To sum up, the present invention proposes to compare the similarity of the same or similar images with the currently captured image by detecting environmental variables, so as to achieve the purpose of detecting appearance abnormalities of the monitored power equipment; secondly, this method can monitor equipment in a variety of different application scenarios, has stronger adaptability, and can effectively improve the accuracy of appearance abnormality detection.

Claims

1. A method for detecting appearance anomalies of power equipment based on texture features, characterized in that: The method comprises the following steps in order: (1) acquiring image information of the electric power equipment and preprocessing the image information thereof, wherein the preprocessed image information of the electric power equipment includes environmental factor information; (2) Compare the current captured image with historical images with normal appearance under the same environmental factors to determine whether the historical images are the same or similar images; (3) If the number of identical images and similar images is greater than or equal to 1, the current captured image and one of the identical images are segmented into multiple parts, and the cosine similarity calculation is performed on the parts at the same position after segmentation to obtain multiple cosine similarities, and the multiple cosine similarities are averaged to obtain a cosine similarity mean value; by analogy, the cosine similarity calculation is performed on the current captured image and all identical images. If all the cosine similarity means are greater than or equal to 0.9, the current captured image is considered normal, otherwise, the current captured image is considered abnormal; the cosine similarity calculation is performed on the segmented part of the current captured image with a cosine similarity less than 0.9 and the part at the same position after segmentation of a similar image to obtain multiple cosine similarities, and the multiple cosine similarities are averaged to obtain a cosine similarity mean value; by analogy, the cosine similarity calculation is performed on the current captured image and all the similar images. If all the cosine similarity means are greater than or equal to 0.8, the current captured image is considered normal, otherwise, the current captured image is considered abnormal; (4) If there is only one identical image or similar image, the current image and the identical image or similar image are segmented into several parts, and the cosine similarity of the segments at the same position is calculated. If the mean values ​​of all cosine similarities are greater than or equal to 0.8, the current image is considered normal; otherwise, the current image is considered abnormal. (5) If there are no identical or similar images, record the relevant information; The step (1) specifically refers to: acquiring image information of the power equipment, and processing it according to environmental factors so that the data information carries environmental factor information, wherein the environmental factor information includes light adaptability information and natural weather information, wherein the light adaptability information includes light sensor data, the longitude and latitude of the location where the image is taken, and the date and time of the image being taken, and the natural weather information includes rain and snow sensor data, environmental dust and haze sensor data, and temperature, humidity and wind sensor data; The step (2) specifically refers to: comparing the current captured image with the historical image with normal appearance under the same environmental factors; if all five conditions from condition 1 to condition 5 are met, then the current captured image is judged to be the same as the historical image with normal appearance under the same environmental factors, and the historical image is judged to be the same image; if at least two of the five conditions from condition 1 to condition 5 are met and condition 2 is included, then the current captured image is judged to be similar to the historical image with normal appearance under the same environmental factors, and the historical image is judged to be a similar image; Condition 1: retrieve the latitude and longitude of the current image shooting location and the date and time of the image shooting, and compare them with the historical image. If the longitude and latitude of the image shooting location are the same and the date and time of the image shooting are within 10 days, the historical image is considered to meet the conditions; Condition 2: retrieve the illumination sensor data of the current image capture location and compare it with the historical image. If the difference in illumination is less than ±10%, the historical image is considered to meet the condition. Condition 3: retrieve the rain and snow sensor data at the location where the current image is taken and compare it with the historical image. If the data is the same, the historical image is considered to meet the conditions; Condition 4: retrieve the environmental dust and haze sensor data of the current image shooting location and compare it with the historical image. If the data difference is less than ±20%, the historical image is considered to meet the condition; Condition 5: retrieve the temperature, humidity and wind sensor data of the current image shooting location and compare it with the historical image. If the data difference is less than ±10%, the historical image is considered to meet the conditions.

2. The method for detecting appearance anomalies of electric power equipment based on texture features according to claim 1 is characterized in that: The step (3) specifically refers to: If the number of identical images and similar images is greater than or equal to 1, the current image and the same image are divided into small squares by area. Each small square is named (nm), where n represents the column and m represents the row. The cosine similarity is calculated for the identical parts of the current image and the same image (nm). A represents the vector of the small square (nm) in the current image, and B represents the vector of the small square (nm) in the same image. The cosine value between A and B is calculated by using the Euclidean dot product formula: A·B=∣∣A∣∣∣∣B∣cosθ Among them, θ represents the angle between the two phase quantities A and B. The cosine value of the angle θ is calculated by the following formula: Here A i , B i Represent the components of vectors A and B respectively; The similarity given ranges from -1 to 1: -1 means that the two vectors point in opposite directions, 1 means that they point in exactly the same direction, 0 means that they are independent, and values ​​in between indicate intermediate similarities or dissimilarities.

Images