A method for intelligent identification of power safety violations

Abstract

The present application relates to the technical field of electric power violation identification, and specifically to a method for intelligent identification of electric power safety violations, the method comprising: collecting images in electric power operation scenes, recorded as images to be detected; obtaining background feature areas and each operation scene area of ​​the image to be detected; determining the pixel distribution uniformity of the entire operation scene area according to the distance between different operation scene areas of the image to be detected and the grayscale difference of pixel points, as well as the distribution number of pixel points; determining the welding scene confidence according to the distribution characteristics of the electric spark boundary box and the smoke boundary box in the operation scene area; determining the electric power safety violation confidence according to the overlap of the area between each equipment boundary box and the staff boundary box in the background feature area, and intelligently identifying electric power safety violations. The present application can detect the standardization of workers wearing welding equipment and improve the accuracy of safety violation identification in electric power operation welding scenes.

Description

Technical Field

[0001] The present application relates to the technical field of power violation identification, and in particular to a method for intelligently identifying power safety violations. Background Art

[0002] The power industry is a high-risk industry that involves high voltage electricity, large equipment, and complex operating procedures. At power construction sites or during power operations, any violation of regulations may lead to serious safety accidents. Therefore, ensuring operational safety is an important task for the power industry. Power safety violations include behavioral violations, device violations, and management violations. Among them, device violations refer to production equipment, facilities, environment, and tools and safety protection supplies used in operations that do not meet regulations and cannot reliably guarantee personal and equipment safety.

[0003] In the electric welding scenario of power operations, workers need to wear special welding masks to ensure safety. However, the high brightness of the welding arc and sparks during welding will cause overexposure of the image, and the smoke generated will also cause interference in target recognition. Directly using the existing target detection algorithm to detect whether there are any power safety violations in the equipment worn by the workers will lead to inaccurate recognition results. Summary of the invention

[0004] In order to solve the above technical problems, a method for intelligent identification of power safety violations is provided to solve the existing problems.

[0005] The solution to the technical problem of this application is to provide a method for intelligently identifying power safety violations, including the following steps:

[0006] Collect images of power operation scenes and record them as images to be detected;

[0007] According to the local grayscale change characteristics and local texture information characteristics of each pixel point in the image to be detected, and the similarity of their characteristics, the background feature area and each operation scene area of ​​the image to be detected are obtained;

[0008] Determine the pixel distribution uniformity of the entire working scene area according to the distance between different working scene areas of the image to be detected, the grayscale difference of the pixel points, and the distribution number of the pixel points;

[0009] Determine the welding scene confidence according to the distribution characteristics of the spark boundary box and the smoke boundary box in the working scene area and the pixel distribution uniformity;

[0010] According to the overlap of the areas between the bounding boxes of each equipment and the bounding boxes of the staff in the background feature area, and the confidence of the welding scene, the confidence of the power safety violation is determined, and the power safety violation is intelligently identified.

[0011] Preferably, the method for acquiring the background feature area and each operation scene area of ​​the image to be detected is:

[0012] Analyze the local grayscale changes and texture information of each pixel in the image to be detected, and determine the local contrast factor and local texture factor of each pixel;

[0013] The local contrast factor and the local texture factor are combined to form a feature vector of each pixel;

[0014] The feature vectors of all pixels are clustered, and the cluster containing the most pixels is recorded as the background feature area, and the other clusters are recorded as various operation scene areas.

[0015] Preferably, determining the local contrast factor and the local texture factor of each pixel point includes:

[0016] Construct a local window centered on each pixel in the image to be detected;

[0017] The ratio of the mean and variance of the grayscale values ​​of all pixels in the local window of each pixel is used as the local contrast factor of each pixel;

[0018] Obtaining a gray level co-occurrence matrix of a local window of each pixel, and taking the product of the contrast and energy of the gray level co-occurrence matrix as a texture weight factor of each pixel;

[0019] The product of the LBP value of the local window of each pixel and the texture weight factor is used as the local texture factor of each pixel.

[0020] Preferably, determining the pixel distribution uniformity of the entire operation scene area includes:

[0021] Analyze the distance between each work scene area and the grayscale difference of the pixel points to determine the aggregation factor between each work scene area and other work scene areas;

[0022] Merge each operation scene area with its maximum aggregation area to obtain all merged operation scene areas, wherein the maximum aggregation area of ​​each operation scene area is the operation scene area corresponding to the maximum value of the aggregation factor between each operation scene area and other operation scene areas;

[0023] Perform image enhancement processing on all merged work scene areas, and record the average grayscale value of all pixels in each work scene area as the average grayscale;

[0024] The product of the number of all pixels in each operation scene area and the average grayscale is recorded as the uniformity coefficient of each operation scene area;

[0025] The normalized result of the sum of the uniformity coefficients of all the working scene areas is taken as the pixel distribution uniformity of the entire working scene area.

[0026] Preferably, the determining of the aggregation factor between each operation scene area and other operation scene areas includes:

[0027] Calculate the distance between each operation scene area and the cluster center of other operation scene areas, and record it as the first distance;

[0028] Calculate the mean of the grayscale values ​​of all pixels in each operation scene area, and record it as the first mean;

[0029] The difference between each operation scene area and the first mean value of other operation scene areas is recorded as the regional grayscale difference;

[0030] The ratio of the regional grayscale difference to the first distance is used as an aggregation factor between each work scene area and other work scene areas.

[0031] Preferably, the determining of the welding scene confidence level comprises:

[0032] A multi-target detection model is used to extract the spark bounding box, smoke bounding box and welding arc bounding box in the working scene area;

[0033] Calculate the normalized result of the sum of the areas of all EDM bounding boxes, and record it as the EDM area weight;

[0034] Calculate the normalized result of the sum of the areas of all smoke bounding boxes, and record it as the smoke area weight;

[0035] Analyze the spatial distribution characteristics of sparks and smoke relative to welding arc light, and determine the spark distribution index and smoke distribution index;

[0036] The product of the inverse of the spark distribution index and the spark area weight is recorded as the first product;

[0037] The product of the inverse of the smoke distribution index and the smoke area weight is recorded as the second product;

[0038] The sum of the first product and the second product is recorded as a first sum value; and the product of the first sum value and the pixel distribution uniformity is used as the welding scene confidence.

[0039] Preferably, the determining of the spark distribution index and the smoke distribution index comprises:

[0040] Taking the center pixel of the welding arc boundary box as the starting point, all vectors pointing to the center pixel of the electric spark boundary box are recorded as photoelectric vectors; all vectors pointing to the center pixel of the smoke boundary box are recorded as light fog vectors;

[0041] The sine value of the angle between the direction of the sum of all photoelectric vectors and the positive direction of the X-axis is taken as the spark distribution index;

[0042] The sine value of the angle between the direction of the sum vector of all light fog vectors and the positive direction of the X-axis is recorded as the first sine value;

[0043] The absolute value of the difference between the first sine value and the value 1 is used as the smoke distribution index.

[0044] Preferably, the determining of the confidence level of the power safety violation includes:

[0045] A multi-target detection model is used to extract the equipment bounding box, welding gun bounding box and worker bounding box in the background feature area;

[0046] If there is a welding gun bounding box in the background feature area, the welding scene weight is assigned to 1; if there is no welding gun bounding box, the welding scene weight is the welding scene confidence;

[0047] Analyze the degree of overlap between the area of ​​each equipment boundary box and the staff boundary box to determine the safety standardization of the equipment;

[0048] The equipment safety specification is integrated with the welding scene weight to determine the confidence level of power safety violation.

[0049] Preferably, the determining of the equipment safety standardization degree comprises:

[0050] The area of ​​the intersection of each equipment bounding box and the staff bounding box is recorded as the first area;

[0051] The area of ​​the union of each equipment bounding box and the staff bounding box is recorded as the second area;

[0052] Recording a ratio of the second area to the first area as a first ratio of each equipment bounding box;

[0053] The confidence of each equipment bounding box is obtained through the multi-target detection model, and the product of the first ratio of each equipment bounding box and the confidence of the corresponding equipment bounding box in the multi-target detection process is recorded as the equipment specification coefficient of each equipment bounding box;

[0054] The equipment specification coefficients of all equipment bounding boxes are fused as the equipment safety specification.

[0055] Preferably, the intelligent identification of power safety violations includes: normalizing the power safety violation confidence level; if the normalized result is greater than a preset safety threshold, the staff is not dressed properly and there is a power safety violation; otherwise, there is no power safety violation.

[0056] This application has at least the following beneficial effects:

[0057] The present application obtains the background feature area and each operation scene area of ​​the image to be detected by analyzing the local grayscale change characteristics and local texture information characteristics of each pixel point in the image to be detected, and the similarity of their characteristics; its beneficial effect is to distinguish the areas in the image where welding scenes may exist, so as to facilitate the subsequent analysis of power safety violations in the working scene of electric welding operations; secondly, after merging each operation scene area, image enhancement is performed to determine the uniformity of pixel distribution, and its beneficial effect is to merge two operation scene areas that are close in distance but have a large grayscale difference, thereby improving the recognizability of each operation scene area in the image, and performing image enhancement on the area Strengthen, prevent the subsequent recognition effect of power safety violations from being affected by excessive grayscale differences, consider the grayscale distribution of pixels in the working scene after image enhancement processing, and reflect the degree of influence on target recognition; determine the confidence of the welding scene, its beneficial effect is to analyze the distribution characteristics of arc, electric sparks and smoke generated by electric welding in the image to be detected, so as to reflect the possibility that the identified image to be detected belongs to the welding power work scene; determine the confidence of power safety violations, realize the intelligent recognition of power safety, its beneficial effect is to analyze the standardization of welding equipment that needs to be worn in the welding scene, and improve the accuracy of safety violation recognition in the welding scene of power operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The following is a further detailed description of an intelligent identification method for power safety violations of the present application in conjunction with the accompanying drawings.

[0059] Figure 1 A flowchart of the steps of a method for intelligent identification of power safety violations provided in this application;

[0060] Figure 2 Schematic diagram of the power operation scenario provided for this application;

[0061] Figure 3 A flowchart of the steps of the method for obtaining the confidence level of a welding scene provided in this application;

[0062] Figure 4 A flowchart of the steps of the method for obtaining the confidence level of power safety violations provided in this application;

[0063] Figure 5 A flow chart for identifying power safety violations in images to be detected provided in this application. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of this application more clear, the following is a further detailed description of an intelligent identification method for power safety violations proposed in this application in conjunction with the accompanying drawings and implementation examples. It should be understood that the specific embodiments described here are only used to explain this application and are not used to limit this application.

[0065] 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 application belongs.

[0066] See also Figure 1 , which shows a flowchart of a method for intelligently identifying power safety violations provided by an embodiment of the present application, the method comprising the following steps:

[0067] Step 1: collect images of power operation scenes and record them as images to be detected.

[0068] The images of workers performing work in the collected power operation scenes are used as images to be detected, and the images are subjected to grayscale processing and noise reduction processing.

[0069] Preferably, in this embodiment, the image to be detected is grayed out for grayscale processing, and Gaussian filtering is used to reduce the noise of the image, wherein image graying and Gaussian filtering are well-known technologies, and the specific process will not be repeated. As other implementation methods, the implementer can use other existing calculation methods for noise reduction processing, and this embodiment does not impose any special restrictions on this.

[0070] At this point, the image to be detected is obtained.

[0071] Step 2: According to the local grayscale change characteristics and local texture information characteristics of each pixel in the image to be detected, and the similarity of their characteristics, the background feature area and each operation scene area of ​​the image to be detected are obtained.

[0072] When performing electrical operations, compliance with safety regulations is the key to ensuring the safety of workers and the stable operation of equipment. Violations of regulations, especially unqualified equipment, greatly increase operational risks.

[0073] Based on the above analysis, it is necessary to identify the wearing status of the operator's safety equipment. In the welding scene, when using a welding gun for welding, welding arc, electric sparks and smoke will be generated, which is a feature that is clearly different from other power operation scenes. The scene can be judged by these phenomena in the image. However, the welding arc and electric sparks are bright, which may form overexposed areas in the image, and the electric sparks are small and numerous, and the smoke has a special shape and texture, which will affect the recognition of the image to be detected.

[0074] Furthermore, the welding arc light is white or blue-white, with high brightness and clustered distribution; the electric spark is yellow or orange, with high brightness and large number, and is widely distributed; while the smoke is mainly white, gray or black, and the texture is obviously different from others. Therefore, the local grayscale change characteristics and texture characteristics are analyzed, specifically:

[0075] Construct a local window centered on each pixel in the image to be detected;

[0076] Calculate the mean and variance of the grayscale values ​​of all pixels within the local window of each pixel;

[0077] The ratio of the mean to the variance is used as a local contrast factor for each pixel;

[0078] Obtaining a gray level co-occurrence matrix of a local window of each pixel, and taking the product of the contrast and energy of the gray level co-occurrence matrix as a texture weight factor of each pixel;

[0079] Calculate the LBP value of the local window of each pixel;

[0080] The product of the texture weight factor and the LBP value is used as the local texture factor of each pixel;

[0081] Preferably, in this embodiment, a 3×3 local window is constructed with each pixel as the center. If the local window of a pixel cannot be complete, the incomplete local window is expanded and filled with the mean of the grayscale value in the local window. The grayscale co-occurrence matrix and LBP value are well-known technologies and will not be described in detail here.

[0082] It should be noted that the local contrast factor represents the grayscale change relationship between each pixel and other pixels in its local window, where the larger the mean, the more likely the pixel is to belong to the welding arc and spark area, and the smaller the variance, the closer the pixel is to other pixels in its local window, and the more likely it is to be a pixel of the same type of phenomenon. Therefore, the closer the local contrast factor, the more likely it is to be a pixel of welding arc or spark. The local texture factor represents the texture feature of each pixel, where the more similar the texture, the more likely it is to be a pixel of the same type. In addition, the texture complexity of smoke is usually greater than that of other things, so the larger the local texture factor, the more likely it is to be a pixel of smoke.

[0083] Furthermore, based on the local contrast factor and the local texture factor, the image to be detected is segmented to distinguish the background pixels and the welding scene pixels. Specifically:

[0084] The local contrast factor and the local texture factor of each pixel in the image to be detected are used to form a feature vector of each pixel;

[0085] Clustering the feature vectors of all pixels in the image to be detected, recording the cluster containing the most pixels as the background feature area, and recording the other clusters as the operation scene areas;

[0086] Preferably, in this embodiment, the Mean Shift clustering algorithm is used to cluster the feature vectors, wherein the search window size of the Mean Shift clustering algorithm is set to 3×3. As other implementation methods, implementers can adopt other methods of the prior art, such as fuzzy C-means clustering, etc. This embodiment does not impose any special restrictions on this.

[0087] It should be noted that each operation scene area may be an electric welding scene or other scenes, and further judgment is needed on each operation scene area.

[0088] At this point, the background feature area and each operation scene area of ​​the image to be detected are obtained.

[0089] Step 3: Determine the pixel distribution uniformity of the entire operation scene area according to the distance between different operation scene areas of the image to be detected, the grayscale difference of the pixels, and the distribution number of the pixels.

[0090] In the work scene area, the grayscale of the welding arc and spark area is large, which is easy to cause overexposure, while the grayscale of the smoke area is small, which is easy to cause unclear texture and affect the subsequent recognition effect. Therefore, in order to increase the recognizability of each work scene area, the work scene areas with close distance but large grayscale differences are merged. Specifically:

[0091] Calculate the distance between each operation scene area and the cluster center of other operation scene areas, and record it as the first distance;

[0092] Calculate the mean of the grayscale values ​​of all pixels in each operation scene area, and record it as the first mean;

[0093] The difference between each operation scene area and the first mean value of other operation scene areas is recorded as the regional grayscale difference;

[0094] Using the ratio of the regional grayscale difference to the first distance as an aggregation factor between each operation scene area and other operation scene areas;

[0095] Merge each operation scene area with its maximum aggregation area to obtain all merged operation scene areas, wherein the maximum aggregation area of ​​each operation scene area is the operation scene area corresponding to the maximum value of the aggregation factor between each operation scene area and other operation scene areas;

[0096] Preferably, in this embodiment, the aggregation factor is calculated as follows: Among them, B r,s is the aggregation factor between the rth operation scene area and the hth operation scene area, μ r is the mean gray value of all pixels in the rth operating scene area, μ h is the mean gray value of all pixels in the hth operating scene area, d(q r ,q h ) represents the Euclidean distance between the cluster centers of the rth operation scene area and the hth operation scene area, i.e., the first distance, μ r -μ h is the regional grayscale difference.

[0097] It can be understood that, in this embodiment, the Euclidean distance is used to calculate the distance between two pixels. As other implementations, the implementer can use other methods of the prior art, such as Manhattan distance, etc. This embodiment does not impose any special restrictions on this.

[0098] It should be noted that the larger the aggregation factor is, the greater the contrast between the two work scene areas is and the closer the distance is, and the more suitable it is for the two work scene areas to be merged into one area. If only one cluster remains in the end, it will no longer be merged.

[0099] Furthermore, in order to prevent the grayscale difference between the working scene areas from being too large, the merged area is processed and the distribution uniformity is obtained to reflect the grayscale distribution of the pixel points, specifically:

[0100] Perform image enhancement processing on all merged work scene areas, and record the average grayscale value of all pixels in each work scene area as the average grayscale;

[0101] The product of the number of all pixels in each operation scene area and the average grayscale is recorded as the uniformity coefficient of each operation scene area;

[0102] The normalized result of the sum of the uniformity coefficients of all the work scene areas is used as the pixel distribution uniformity of the entire work scene area;

[0103] Preferably, in this embodiment, the image enhancement is performed by using limited contrast adaptive histogram equalization; wherein, limited contrast adaptive histogram equalization is a well-known technology and will not be described in detail here. As other implementation methods, the implementer may use other methods for image enhancement, and this embodiment does not impose any special restrictions on this.

[0104] It should be noted that the purpose of image enhancement is to prevent excessive brightness differences from affecting the recognition effect; secondly, the pixel distribution uniformity reflects the uniformity of the grayscale distribution of pixels in the working scene area. The greater the pixel distribution uniformity, the more uniform the grayscale distribution of pixels in the working scene area after image enhancement processing, and the smaller the impact on target recognition; on the contrary, the greater the impact on the recognition effect.

[0105] At this point, the pixel distribution uniformity of the operation scene area is obtained.

[0106] Step 4: Determine the welding scene confidence level according to the distribution characteristics of the spark boundary box and the smoke boundary box in the working scene area and the pixel distribution uniformity.

[0107] Based on the above analysis, the power operation scene area in the image to be detected may be a welding work scene or other power work scenes. Therefore, it is necessary to further determine whether the power work scene in the image to be detected is a welding work scene. The distribution of welding arc is concentrated, appearing around the welding gun and the welding point, with a large area; electric sparks appear around the welding arc, with a relatively scattered distribution and a large number; smoke usually appears above the welding arc and sparks, with a large area. Secondly, since the electric sparks are usually evenly distributed around the welding arc during welding, and the smoke is usually distributed above the welding arc, the confidence of the welding scene is analyzed to characterize the possibility that the scene in the picture is a welding scene, specifically:

[0108] In this embodiment, 1000 images of workers performing welding in welding scenes are collected, wherein the images are collected from multiple angles, grayscaled and denoised, and the images are annotated to mark the workers, welding guns, safety helmets, insulating gloves, and work clothes that appear in the images. In addition, the welding arc, electric sparks, and smoke that appear in the welding scene images are also annotated. All the annotated images are used as input, and the YOLOv5 target detection algorithm is used for model training to obtain a trained multi-target detection model and category labels.

[0109] A multi-target detection model is used to extract the spark bounding box, smoke bounding box and welding arc bounding box in the working scene area;

[0110] Calculate the normalized result of the sum of the areas of all EDM bounding boxes, and record it as the EDM area weight;

[0111] Calculate the normalized result of the sum of the areas of all smoke bounding boxes, and record it as the smoke area weight;

[0112] Taking the center pixel of the welding arc boundary box as the starting point, all vectors pointing to the center pixel of the electric spark boundary box are recorded as photoelectric vectors; all vectors pointing to the center pixel of the smoke boundary box are recorded as light fog vectors;

[0113] In this embodiment, the schematic diagram of the power operation scene is as follows: Figure 2 As shown, 1 represents the welding arc bounding box, 2 and 3 represent the smoke bounding boxes, and 4, 5 and 6 represent the spark bounding boxes.

[0114] The sine value of the angle between the direction of the sum of all photoelectric vectors and the positive direction of the X-axis is taken as the spark distribution index;

[0115] Similarly, the sine value of the angle between the direction of the sum vector of all light fog vectors and the positive direction of the X-axis is recorded as the first sine value;

[0116] The absolute value of the difference between the first sine value and the value 1 is used as the smoke distribution index;

[0117] It should be noted that the smaller the spark distribution index is, the more uniform the spark distribution is, and the greater the possibility that it conforms to the welding scene; the smaller the smoke distribution index is, the closer the smoke distribution is to the top, and the greater the possibility that it conforms to the welding scene.

[0118] The product of the inverse of the spark distribution index and the spark area weight is recorded as the first product;

[0119] The product of the inverse of the smoke distribution index and the smoke area weight is recorded as the second product;

[0120] The sum of the first product and the second product is recorded as a first sum value; the product of the first sum value and the pixel distribution uniformity is used as the welding scene confidence;

[0121] It should be noted that the greater the confidence of the welding scene, the more the position distribution of various bounding boxes in the image to be detected conforms to the position distribution of the welding power work scene, and the greater the possibility that the image to be detected belongs to the welding power work scene; conversely, the smaller the possibility that the image to be detected belongs to the welding power work scene.

[0122] Further, the step flow chart of the method for obtaining the confidence of the welding scene provided in this embodiment is as follows: Figure 3 shown.

[0123] At this point, the confidence of the welding scene is obtained.

[0124] Step 5, determining the confidence of power safety violation based on the overlap of the area between each equipment boundary box and the worker boundary box in the background feature area, and the confidence of the welding scene, and performing intelligent identification of power safety violation.

[0125] In the process of collecting images, due to the image angle, the image to be detected may not necessarily contain the welding arc, sparks and smoke that appear in the welding scene, so it is necessary to determine the welding gun boundary box detected in the background feature area. Specifically:

[0126] A multi-target detection model is used to extract the equipment bounding box, welding gun bounding box and worker bounding box in the background feature area;

[0127] If there is a welding gun bounding box in the background feature area, the welding scene weight is assigned to 1; if there is no welding gun bounding box, the welding scene weight is the welding scene confidence;

[0128] It should be noted that the equipment bounding box is represented by a welding mask bounding box, a safety helmet bounding box, a work clothes bounding box, and an insulating glove bounding box.

[0129] Furthermore, by analyzing the degree of overlap between the equipment of the workers and the borders of the workers, the standardization of the equipment worn by the workers can be judged, so as to improve the safety of power work, specifically:

[0130] The area of ​​the intersection of each equipment bounding box and the staff bounding box is recorded as the first area;

[0131] The area of ​​the union of each equipment bounding box and the staff bounding box is recorded as the second area;

[0132] Recording a ratio of the second area to the first area as a first ratio of each equipment bounding box;

[0133] The confidence of each equipment bounding box is obtained through the multi-target detection model, and the product of the first ratio of each equipment bounding box and the confidence of the corresponding equipment bounding box in the multi-target detection process is recorded as the equipment specification coefficient of each equipment bounding box;

[0134] The equipment specification coefficients of all equipment boundary boxes are merged to obtain equipment safety specification;

[0135] The equipment safety standard degree is integrated with the welding scene weight to determine the confidence level of power safety violation;

[0136] It is understandable that the fusion can be expressed as a multiplication relationship, an addition relationship, etc., which is determined by the actual application and the present application does not impose any special restrictions on this.

[0137] Preferably, in this embodiment, the calculation method of the power safety violation confidence is: Among them, A is the confidence level of power safety violation, ω is the weight of welding scene, S iS is the area of ​​the intersection of the i-th equipment bounding box and the staff bounding box in the background feature area, that is, the first area; i ′ is the area of ​​the union of the i-th equipment bounding box and the staff bounding box in the background feature area, that is, the second area; ρ i is the confidence of the i-th equipment bounding box in the background feature area, K is the number of equipment bounding boxes in the background feature area, is the first ratio, To ensure the safety of the equipment.

[0138] It should be noted that the greater the confidence of power safety violation, the smaller the overlap between the worker boundary box and the welding mask boundary box, the safety helmet boundary box, the work clothes boundary box and the insulating gloves boundary box in the power welding scenario, that is, the more irregular the worker's work equipment is.

[0139] Further, the step flow chart of the method for obtaining the confidence level of power safety violation provided in this embodiment is as follows: Figure 4 shown.

[0140] Further, based on the power safety violation confidence level, the power safety violation is intelligently identified, specifically:

[0141] The confidence level of the power safety violation is normalized. If the normalized result is greater than a preset safety threshold, the staff is not dressed properly and there is a power safety violation. If the normalized result is less than the preset safety threshold, there is no power safety violation.

[0142] Preferably, in this embodiment, the preset safety threshold value is 0.5. As other implementation modes, the implementer can set it according to the actual situation, and this embodiment does not impose any special restrictions on this.

[0143] Furthermore, the flow chart of power safety violation identification of the image to be detected provided in this embodiment is as follows: Figure 5 shown.

[0144] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0145] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0146] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the present application. It should be pointed out that, for those of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application without departing from the content of the technical solution of the present application, shall fall within the protection scope of the technical solution of the present application.

Claims

1. A method for intelligently identifying power safety violations, characterized in that: The method comprises the following steps: Collect images of power operation scenes and record them as images to be detected; According to the local grayscale change characteristics and local texture information characteristics of each pixel point in the image to be detected, and the similarity of their characteristics, the background feature area and each operation scene area of ​​the image to be detected are obtained; Determine the pixel distribution uniformity of the entire working scene area according to the distance between different working scene areas of the image to be detected, the grayscale difference of the pixel points, and the distribution number of the pixel points; Determine the welding scene confidence according to the distribution characteristics of the spark boundary box and the smoke boundary box in the working scene area and the pixel distribution uniformity; According to the overlap between the area of ​​each equipment boundary box and the worker boundary box in the background feature area, and the welding scene confidence, the power safety violation confidence is determined, and the power safety violation is intelligently identified; Determining the pixel distribution uniformity of the entire operation scene area includes: Analyze the distance between each work scene area and the grayscale difference of the pixel points to determine the aggregation factor between each work scene area and other work scene areas; Merge each operation scene area with its maximum aggregation area to obtain all merged operation scene areas, wherein the maximum aggregation area of ​​each operation scene area is the operation scene area corresponding to the maximum value of the aggregation factor between each operation scene area and other operation scene areas; Perform image enhancement processing on all merged work scene areas, and record the average grayscale value of all pixels in each work scene area as the average grayscale; The product of the number of all pixels in each work scene area and the average grayscale is recorded as the uniformity coefficient of each work scene area; The normalized result of the sum of the uniformity coefficients of all the working scene areas is taken as the pixel distribution uniformity of the entire working scene area.

2. The method for intelligently identifying power safety violations according to claim 1, characterized in that: The method for acquiring the background feature area and each operation scene area of ​​the image to be detected is: Analyze the local grayscale changes and texture information of each pixel in the image to be detected, and determine the local contrast factor and local texture factor of each pixel; The local contrast factor and the local texture factor are combined to form a feature vector of each pixel; The feature vectors of all pixels are clustered, and the cluster containing the most pixels is recorded as the background feature area, and the other clusters are recorded as various operation scene areas.

3. A method for intelligently identifying power safety violations as claimed in claim 2, characterized in that: The determining of the local contrast factor and the local texture factor of each pixel includes: Construct a local window centered on each pixel in the image to be detected; The ratio of the mean and variance of the grayscale values ​​of all pixels in the local window of each pixel is used as the local contrast factor of each pixel; Obtaining a gray level co-occurrence matrix of a local window of each pixel, and taking the product of the contrast and energy of the gray level co-occurrence matrix as a texture weight factor of each pixel; The product of the LBP value of the local window of each pixel and the texture weight factor is used as the local texture factor of each pixel.

4. The method for intelligently identifying power safety violations according to claim 1, characterized in that: The determining of the aggregation factor between each operation scene area and other operation scene areas includes: Calculate the distance between each operation scene area and the cluster center of other operation scene areas, and record it as the first distance; Calculate the mean of the grayscale values ​​of all pixels in each operation scene area, and record it as the first mean; The difference between each operation scene area and the first mean value of other operation scene areas is recorded as the regional grayscale difference; The ratio of the regional grayscale difference to the first distance is used as an aggregation factor between each work scene area and other work scene areas.

5. The method for intelligently identifying power safety violations according to claim 1, characterized in that: Determining the confidence of the electric welding scene includes: A multi-target detection model is used to extract the spark bounding box, smoke bounding box and welding arc bounding box in the working scene area; Calculate the normalized result of the sum of the areas of all EDM bounding boxes, and record it as the EDM area weight; Calculate the normalized result of the sum of the areas of all smoke bounding boxes, and record it as the smoke area weight; Analyze the spatial distribution characteristics of sparks and smoke relative to welding arc light, and determine the spark distribution index and smoke distribution index; The product of the inverse of the spark distribution index and the spark area weight is recorded as the first product; The product of the inverse of the smoke distribution index and the smoke area weight is recorded as the second product; The sum of the first product and the second product is recorded as a first sum value; and the product of the first sum value and the pixel distribution uniformity is used as the welding scene confidence.

6. A method for intelligently identifying power safety violations as claimed in claim 5, characterized in that: The determining of the spark distribution index and the smoke distribution index comprises: Taking the center pixel of the welding arc boundary box as the starting point, all vectors pointing to the center pixel of the electric spark boundary box are recorded as photoelectric vectors; all vectors pointing to the center pixel of the smoke boundary box are recorded as light fog vectors; The sine value of the angle between the direction of the sum of all photoelectric vectors and the positive direction of the X-axis is taken as the spark distribution index; The sine value of the angle between the direction of the sum vector of all light fog vectors and the positive direction of the X-axis is recorded as the first sine value; The absolute value of the difference between the first sine value and the value 1 is used as the smoke distribution index.

7. The method for intelligently identifying power safety violations according to claim 1, characterized in that: Determining the confidence level of power safety violation includes: A multi-target detection model is used to extract the equipment bounding box, welding gun bounding box and worker bounding box in the background feature area; If there is a welding gun bounding box in the background feature area, the welding scene weight is assigned to 1; if there is no welding gun bounding box, the welding scene weight is the welding scene confidence; Analyze the degree of overlap between the area of ​​each equipment boundary box and the staff boundary box to determine the safety standardization of the equipment; The equipment safety specification is integrated with the welding scene weight to determine the confidence level of power safety violation.

8. The method for intelligently identifying power safety violations according to claim 7, characterized in that: Determining the safety standard of the equipment includes: The area of ​​the intersection of each equipment bounding box and the staff bounding box is recorded as the first area; The area of ​​the union of each equipment bounding box and the staff bounding box is recorded as the second area; Recording a ratio of the second area to the first area as a first ratio of each equipment bounding box; The confidence of each equipment bounding box is obtained through the multi-target detection model, and the product of the first ratio of each equipment bounding box and the confidence of the corresponding equipment bounding box in the multi-target detection process is recorded as the equipment specification coefficient of each equipment bounding box; The equipment specification coefficients of all equipment bounding boxes are fused as the equipment safety specification.

9. The method for intelligently identifying power safety violations according to claim 1, characterized in that: The intelligent identification of power safety violations includes: normalizing the confidence of the power safety violations. If the normalized result is greater than a preset safety threshold, the staff is not dressed properly and there is a power safety violation. Otherwise, there is no power safety violation.

Images