Illegal behavior identification and early warning method, system and device for intelligent power distribution network, and medium

Abstract

The invention discloses an intelligent power distribution network violation behavior identification and early warning method, system, equipment and medium, and belongs to the technical field of power distribution network violation behavior identification, and the method comprises the steps: obtaining continuous frame images of power distribution network field image data through a monitoring device, building an initialized background model based on the continuous frame images, and carrying out the early warning of the power distribution network violation behavior; separating a static area and a foreground target area; performing inter-frame difference processing in the foreground target area based on a gray difference value between continuous frame images, and judging whether an abnormal target image exists in combination with a preset threshold value; carrying out feature extraction processing on the abnormal target image, inputting an extraction result into an illegal behavior recognition model fused with an attention mechanism, and carrying out illegal behavior judgment; and judging whether an illegal behavior exists based on an identification result, and starting an early warning response mechanism when the illegal behavior is identified. According to the invention, high-precision identification and rapid early warning of small target violation behaviors in the power distribution network are realized, and the monitoring reliability and response efficiency in a complex environment are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network violation identification, and in particular to a smart distribution network violation identification and early warning method, system, equipment and medium. Background Art

[0002] In the existing technologies of smart distribution network violation identification and early warning systems, there are already a variety of methods and systems for improving the monitoring and early warning capabilities of violations. In addition, the distribution network construction personnel violation detection method and system based on motion recognition uses skeletal feature information to determine the type of construction work behavior of the workers, and uses target detection technology to detect whether the safety protection equipment has violated the regulations.

[0003] The features of small targets are not obvious: near the distribution network, the features of small targets such as fishing and kite flying are relatively small, and the detection results are easily affected by ambient light, surrounding objects, etc., resulting in detection errors; low recognition accuracy and poor efficiency: existing methods often have problems with low recognition accuracy and poor efficiency when detecting violations, especially in detecting straying into dangerous areas; insufficient model robustness: in the detection of target violations in smart substations, existing methods may have insufficient robustness for small data sets, affecting detection performance. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for identifying and warning of violations in a smart distribution network to solve the problems of unclear small target features in existing methods: near the distribution network, the features of small targets such as fishing and kite flying are relatively small, and the detection results are easily affected by ambient light, surrounding objects, etc., resulting in detection errors; low recognition accuracy and poor efficiency: existing methods often have problems of low recognition accuracy and poor efficiency when detecting violations, especially in the detection of straying into dangerous areas; insufficient model robustness: in the detection of target violations in smart substations, existing methods may have insufficient robustness for small data sets, affecting detection performance.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for identifying and warning of violations in a smart distribution network, comprising: obtaining continuous frame images of distribution network field image data through a monitoring device, and constructing an initialization background model based on the continuous frame images to separate static areas and foreground target areas; performing inter-frame difference processing in the foreground target area based on the grayscale difference between continuous frame images, and judging whether there is an abnormal target image in combination with a preset threshold; performing feature extraction processing on the abnormal target image, including extracting spatial features and motion features of the image, and inputting the extraction results into a violation recognition model that integrates an attention mechanism to judge the violation; judging whether there is a violation based on the recognition result, and when a violation is identified, starting a warning response mechanism.

[0008] As a preferred solution of the method for identifying and warning violations in a smart distribution network described in the present invention, the monitoring device is configured in the distribution network monitoring area, including a first image acquisition device and a second image acquisition device; the first image acquisition device is a fixedly deployed wide dynamic range image acquisition device with infrared night vision function, and the second image acquisition device is an image acquisition device with mobility.

[0009] Compared with the existing technology, the present invention significantly improves the recognition accuracy of small targets (such as helmets, fishing tools, kites, etc.), making the detection results in the complex environment of the distribution network more reliable.

[0010] As a preferred solution of the method for identifying and warning violations in a smart distribution network described in the present invention, the initialization background model is constructed based on continuous multi-frame images, and a static area and foreground target area separation structure is formed through image grayscale and binarization processing.

[0011] As a preferred solution of the method for identifying and warning violations in a smart distribution network described in the present invention, the violation identification model includes a classification model obtained by training based on image features and behavior patterns; the warning response mechanism includes grading the recognition results and generating warning prompt information in different forms accordingly.

[0012] As a preferred solution of the method for identifying and warning violations in a smart distribution network described in the present invention, the method comprises: constructing an initialization background model based on continuous frame images, including collecting a continuous five-frame image sequence and performing weighted averaging processing on the k-5, k-4, k-3, k-2, and k-1 frame images to obtain a grayscale image; converting the grayscale image into a binary image and performing pixel processing to create a background template with a full pixel value of 1 and an initialization background image with a pixel value of 0; comparing the k-th frame image with the initialization background image to separate the foreground target area and the static area, and constructing an initialization background model; based on the constructed initialization background model, performing a differential operation on temporally continuous image frames using an inter-frame difference method to determine whether the corresponding pixel grayscale difference exceeds a preset threshold, and identifying moving targets; performing feature extraction on the identified moving target image, the features including the center of mass difference and the rectangle aspect ratio; extracting abnormal target images based on the feature results, and generating corresponding abnormal target image photos.

[0013] As a preferred solution of the method for identifying and warning violations in a smart distribution network described in the present invention, the method includes: inputting the violation recognition model to perform target recognition and behavior judgment, including inputting the abnormal target image into a violation recognition model based on YOLOV8, including an input end, a backbone network structure, a neck structure and a head structure; extracting the basic features of the image in the backbone network structure, fusing the multi-scale image features in the neck structure, and generating a bounding box and corresponding classification results in the head structure; embedding a triple attention mechanism between the backbone network structure and the neck structure, performing attention calculations in the channel dimension, vertical dimension and horizontal dimension respectively to enhance the image feature expression capability; and judging whether there is a violation in the image based on the output result.

[0014] The present invention employs a multi-scale attention mechanism, fusing feature maps of different scales during model training. This approach effectively improves the model's recognition capabilities for small sample datasets and enhances its generalization performance. Compared to existing technologies that lack robustness for small datasets, the violation recognition model of the present invention is more adaptable to violation detection in diverse scenarios, reducing fluctuations in detection performance due to data sample characteristics and ensuring stable and accurate violation identification in all situations.

[0015] As a preferred solution of the method for identifying and warning violations in a smart distribution network described in the present invention, the warning response mechanism includes on-site broadcasts and system prompts; for on-site broadcasts, when the violation identification model deems the behavior to be a violation, a warning prompt is issued through the loudspeaker of the on-site monitoring equipment, prompting the violator to stop the violation and leave; for system prompts, in the distribution network violation system, when a violation is identified, a prompt will be issued in the system, and the corresponding staff will continue to track it.

[0016] In a second aspect, the present invention provides a smart distribution network violation identification and early warning system, comprising: an image acquisition module, an image processing module, a violation identification model module, and a violation early warning module; the image acquisition module is used to obtain continuous frame images of distribution network field image data through a monitoring device, and to construct an initialization background model based on the continuous frame images to separate static areas and foreground target areas;

[0017] The image processing module is used to perform inter-frame difference processing in the foreground target area based on the grayscale difference between consecutive frame images, and to determine whether there is an abnormal target image in combination with a preset threshold; the violation recognition model module is used to perform feature extraction processing on the abnormal target image, including extracting the spatial features and motion features of the image, and inputting the extraction results into the violation recognition model that integrates the attention mechanism to perform violation judgment; the violation warning module is used to determine whether there is a violation based on the recognition results, and to activate the warning response mechanism when a violation is identified.

[0018] In a third aspect, the present invention provides an electronic device, comprising:

[0019] memory and processor;

[0020] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of a method for identifying and warning violations in a smart distribution network are implemented.

[0021] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for identifying and warning violations in a smart distribution network.

[0022] Compared with existing technologies, the present invention has the following beneficial effects: The invention introduces wide dynamic range (WDR) cameras and drone aerial photography technology, greatly enhancing the ability to capture small targets. Under varying lighting conditions, such as in environments with high contrast or in low light conditions at night, the WDR camera can provide clear images, effectively avoiding errors in small target detection caused by ambient light. Compared with existing technologies, the recognition accuracy of small targets (such as helmets, fishing tools, kites, etc.) is significantly improved, making detection results more reliable in the complex environment of the distribution network.

[0023] A multi-scale attention mechanism is employed to fuse feature maps of different scales during model training. This approach effectively improves the model's recognition capabilities for small sample datasets and enhances its generalization performance. Compared to existing technologies that are less robust on small datasets, the proposed model is more adaptable to detecting violations in diverse scenarios, reducing fluctuations in detection performance due to data sample characteristics and ensuring stable and accurate identification of violations in all situations.

[0024] By combining manual feature extraction with the YOLOv8 feature extraction framework, the feature extraction process is optimized. This allows for more refined extraction of violation features. Compared to existing feature extraction methods that may be less refined, this method can more accurately identify complex and diverse violation patterns, improving the system's ability to distinguish between various violations and reducing the probability of misjudgments and missed detections.

[0025] By building a database through data augmentation and utilizing data enhancement techniques such as random flipping, rotation, scaling, color conversion, cropping, and inpainting (RICAP), the original 1,000 violation photos were expanded to 3,000, and the training, test, and validation sets were rationally divided. This abundant and rich data sample provided a solid foundation for model training, effectively improving the model's training effectiveness and recognition accuracy, overcoming the existing problem of poor model training results due to insufficient data samples.

[0026] Optimizations in algorithm and hardware selection enhance the system's real-time processing capabilities, ensuring timely image acquisition, analysis, and judgment of the monitored area. Furthermore, an early warning mechanism, integrating on-site broadcasts and system prompts, quickly warns the offender via a loudspeaker upon identification of a violation, and prompts staff within the system for ongoing tracking. Compared to existing systems that lack real-time monitoring and early warning capabilities, this invention can more quickly and effectively address violations within the distribution network, ensuring its safe operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 The present invention is a flowchart of a method for identifying and warning violations in a smart distribution network according to an embodiment of the present invention.

[0029] Figure 2 The figure is a schematic diagram of the overall process of a method for identifying and warning violations in a smart distribution network according to an embodiment of the present invention.

[0030] Figure 3 This is a YOLOV8-based feature recognition framework diagram of a smart distribution network violation identification and early warning method described in an embodiment of the present invention.

[0031] Figure 4 This is a YOLOV8-based feature recognition framework diagram of a smart distribution network violation identification and early warning method described in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0033] Example 1, with reference to Figure 1 , as an embodiment of the present invention, provides a method for identifying and warning violations in a smart distribution network, comprising:

[0034] S1. Obtain continuous frame images of distribution network field image data through a monitoring device, and construct an initialization background model based on the continuous frame images to separate the static area and the foreground target area.

[0035] S2. Perform inter-frame difference processing in the foreground target area based on the grayscale difference between consecutive frame images, and determine whether there is an abnormal target image in combination with a preset threshold.

[0036] S3. Perform feature extraction on the abnormal target image, including extracting the spatial features and motion features of the image, and input the extraction results into the violation recognition model integrated with the attention mechanism to make violation judgments.

[0037] S4. Determine whether there is any violation based on the identification results, and activate the early warning response mechanism when a violation is identified.

[0038] At the operation and maintenance site of the smart distribution network, there are behaviors such as people arbitrarily approaching dangerous areas, illegally building, and violating construction regulations. These behaviors often occur in remote areas, and traditional manual inspection methods are difficult to detect in real time, resulting in monitoring blind spots. At the same time, due to complex weather and lighting conditions, small target recognition is difficult, the model is not robust enough, and data samples are insufficient, which can easily lead to misjudgments and missed judgments.

[0039] Insufficient small target detection capability: In existing systems, the recognition accuracy of small targets such as helmets is low and easily affected by environmental factors. This invention enhances the ability to capture small targets under different lighting conditions, thereby improving recognition accuracy.

[0040] Insufficient model robustness: To address the insufficient robustness of existing models on small sample data sets, this paper adopts an attention mechanism to improve the model's recognition ability for small samples and enhance the model's generalization performance by fusing feature maps of different scales during model training.

[0041] Optimization of the feature extraction framework: Feature extraction in the existing technology may not be precise enough. This invention is based on the feature extraction framework of YOLOv8 and improves the model's recognition accuracy of violation characteristics by optimizing the feature extraction process.

[0042] Insufficient data samples: In order to solve the problem of poor model training effect caused by insufficient data samples, the present invention combines data augmentation to construct a database, expands the training samples through data enhancement technology, and improves the model training effect and recognition accuracy.

[0043] Real-time and early warning capabilities: Existing systems may have deficiencies in real-time monitoring and early warning. This invention improves the system's real-time processing capabilities by optimizing algorithms and hardware selection, and integrates an effective early warning mechanism to ensure that violations can be discovered and warned in a timely manner.

[0044] Example 2, reference Figures 1 to 3 , which is an embodiment of the present invention, provides a method for identifying and warning violations in a smart distribution network based on the above embodiment.

[0045] A method for identifying and warning violations in smart distribution networks Figure 1 and Figure 2 shown.

[0046] In an embodiment of the present application, in step S1, a monitoring device is configured in the distribution network monitoring area. The monitoring device includes a first image acquisition device and a second image acquisition device. The first image acquisition device is a fixedly deployed image acquisition device, and the second image acquisition device is an image acquisition device with mobility capabilities.

[0047] In the embodiment of the present application, the first image acquisition device is a fixedly deployed image acquisition device, which is set up in areas of the distribution network where violations are prone to occur, including typical high-risk points such as the edge of a pond, construction sites, and around towers; the fixed image acquisition device uses a camera with wide dynamic range (WDR) image acquisition function and infrared night vision function, which can provide stable images under strong light contrast and night conditions, providing high-quality image data for subsequent detection.

[0048] Wide dynamic range cameras can provide clear images in environments with strong light contrast. Cameras with infrared night vision functions can solve the problems at night or in low light conditions. They can be installed in areas where the distribution network needs to be monitored, such as pond fishing, construction areas, towers and other typical unsafe behavior areas, to achieve fixed-point monitoring and collection.

[0049] The second image acquisition device is a mobile one, specifically a camera-equipped drone. It can follow pre-set patrol routes and is suitable for covering high-altitude areas and remote monitoring points where deployment is unavailable. For monitoring areas such as power towers, which are difficult for humans to reach, drones are used for aerial photography.

[0050] In an optional embodiment, the first image acquisition device is a thermal imaging camera, which is deployed in the area surrounding high-voltage equipment to monitor the body temperature of personnel or the heating status of electrical equipment through infrared thermal signals to assist in image recognition; the second image acquisition device can be an automatic cruise vehicle, which is suitable for narrow venues or ground inspection areas, and obtains low-angle image data through an on-board vision module.

[0051] In another optional embodiment, the first image acquisition device is embedded in a fence or bracket through a fiber optic monitoring system to obtain images of nearby personnel activities in a low-visibility form; the second image acquisition device is a balloon launch platform or a controllable pan-tilt system, which has the ability to move in the air over a small range and can implement flexible monitoring of key short-term construction areas.

[0052] The steps of the present invention can significantly improve the spatiotemporal coverage and illumination adaptability of image acquisition, effectively solve problems such as nighttime monitoring blind spots, image overexposure or underexposure in distribution network sites, and ensure that subsequent target detection has a continuous and stable data input basis.

[0053] In the implementation manner of the present application, in step S1, the distribution network on-site image data is obtained through the monitoring device, and an initialization background model is constructed based on the continuous frame images. An Internet of Things acquisition device is used to connect to the monitoring device, and a supporting program is used to realize real-time collection of relevant data, and the camera is adjusted according to the scene conditions.

[0054] The implementation of this application addresses the issue of significant grayscale drift in the background area of ​​wide dynamic range cameras in complex lighting environments. It employs a combined approach of "five-frame weighted averaging + binarization processing + foreground area separation." This approach controls image stability by setting a weighted window with a fixed frame length, significantly improving the robustness of the initial background modeling in low-light, backlit, and nighttime monitoring scenarios. This processing approach is more suitable for the practical needs of long-term distribution network monitoring scenarios where the background environment is static but the brightness fluctuates frequently. It addresses the high misjudgment rate and large response delay of existing background modeling methods in such special environments.

[0055] Constructing the initialization background model includes: collecting a continuous sequence of five frames of images, using the weighted average method to perform grayscale processing on the k-5, k-4, k-3, k-2, and k-1 frames to obtain the average image of the background area; then binarizing the grayscale image, using pixel-level operations to generate a background template with a pixel value of 1 and an initialization background image with a pixel value of 0; performing differential processing on the k-th frame image and the initialization background image to separate the foreground target area from the static area, and completing the initialization background modeling.

[0056] Specifically, first, the IoT acquisition terminal is used to collect n frames of video image sequence, and the weighted average method is used to process the k-5, k-4, k-3, k-2, and k-1 frames using formula (1) to obtain grayscale images:

[0057] t a (x,y)=0.289R a (x,y)+0.499G a (x,y)+

[0058] 0.126B a (x,y)

[0059] a=k-5, k-4, k-3, k-2, k-1

[0060] Among them, t a (x, y) is the grayscale value of the a-th frame image at position (x, y), R a (x, y) is the red channel value of the a-th frame image at position (x, y), G a (x, y) is the green channel value of the a-th frame image at position (x, y), B a (x,y) is the blue channel value of the a-th frame image at position (x,y).

[0061] Convert the grayscale image into a binary image and perform the following operation:

[0062] C k (x,y)=|t k-3 (x,y)-tk-4 (x,y)∣∩∣t k-3 (x,y)-t k-5 (x,y)∣∩∣t k-2 (x,y)

[0063] -t k-3 (x,y)∣∩∣t k-1 (x,y)-t k-3 (x,y)|

[0064] Among them, C k (x,y) is a binary image, t k-5 (x,y),t k-4 (x,y),t k-3 (x,y),t k-2 (x,y),t k-1 (x, y) are the grayscale images of the k-5th, k-4th, k-3th, k-2nd, and k-1th frames respectively;

[0065] Thus, a background template with all pixel values ​​of 1 and an initialization background image with pixel values ​​of 0 are created, and the static area and foreground target area of ​​the k-th frame image are separated using the binary image formula; thereby, an initialization background model is constructed.

[0066] In an optional embodiment, the initialization background model is constructed using a method based on the Gaussian mixture model (GMM), and a dynamic distribution model is established for each pixel in the image to determine whether it belongs to the background or foreground, and the model parameter self-update mechanism is used to adapt to lighting changes and dynamic noise interference in different scenes.

[0067] In another optional embodiment, the initialization background model is constructed based on inter-frame median filtering, and the median value of each pixel in multiple image frames is used as the background pixel value, which can effectively suppress random disturbances and is suitable for monitoring environments with a small number of image frames and high scene stability.

[0068] The steps of the present invention can quickly and accurately separate the background and moving target areas, avoiding interference of stationary objects or local occlusions in the environment on the motion detection results, thereby improving the accuracy and real-time performance of subsequent inter-frame difference detection, and enhancing the stability and response speed in actual operation.

[0069] In an embodiment of the present application, in step S2, inter-frame difference processing is performed on continuous image frames, including: selecting the current frame image and the previous frame or adjacent multiple frames to perform pixel-level grayscale difference operation, calculating the absolute value of the grayscale difference for each pixel point, and comparing it with a preset threshold; when the difference of a certain pixel area exceeds the threshold, it is marked as a moving target area; then, image features such as centroid difference and rectangular outline aspect ratio are extracted from the moving target area, and based on these features, it is determined whether it is an abnormal moving target; the extraction result generates an abnormal target image, and is converted into a photo format for subsequent identification and analysis.

[0070] Specifically, the inter-frame difference method is used to automatically identify moving targets by comparing the differences between consecutive frames in the video sequence; differential operations are performed on two or three consecutive frames of images in time, and the pixels corresponding to different frames are subtracted to determine the absolute value of the grayscale difference. When the absolute value exceeds a certain threshold, it can be determined as a moving target, thereby realizing the target detection function.

[0071] Feature extraction: Feature extraction is performed on the detected moving target, including the center of mass difference and rectangle aspect ratio, to describe the dynamic characteristics and shape of the target.

[0072] The center of mass difference distribution, the position of the center of mass can accurately reflect the position of the human body and is not affected by changes in human posture. The center of mass coordinates can be calculated according to the following formula:

[0073]

[0074] Where x is the horizontal coordinate of the pixel point (x, y), y is the vertical coordinate of the pixel point (x, y), h(x, y) is the grayscale value at the pixel point (x, y), ∑ (x,y)∈rect h(x,y) is the grayscale value of all pixels in the region. The absolute value of the grayscale difference of the object is calculated. When the calculated value exceeds the threshold, it is judged that the object has moved.

[0075] Thus, the difference distribution β of the centroid is obtained:

[0076]

[0077] Among them, β is the centroid difference distribution, Y k is the vertical coordinate of the center of mass of the k-th frame, is the average value of the centroid ordinate of the current k frames, i is the i-th frame image, i=1,2,...,k.

[0078] The aspect ratio of a rectangle is calculated by calculating the change in the rectangle to obtain the change ratio of the width W and height H of the rectangle:

[0079]

[0080] in, and are the x-axis and y-axis coordinate values ​​of the rectangular area rect, B is the aspect ratio of the rectangle, W is the width of the rectangle, and H is the height of the rectangle.

[0081] In an optional embodiment, inter-frame difference processing is based on differential fusion processing of three frames of images, and the current frame is temporally differentiated with the two previous and next frames and the intersection area is taken to enhance the continuity of the target boundary; an adaptive threshold update strategy is adopted to automatically adjust the pixel difference threshold according to the scene lighting conditions and image dynamics to improve the accuracy of target extraction.

[0082] In another optional embodiment, inter-frame difference processing calculates the motion vector field between pixel points through the optical flow method, analyzes the motion trajectory of objects in the image over time, and identifies target areas with abnormal displacement or abnormal speed changes. It is suitable for scenes with high image frame rates, small moving targets but obvious motion characteristics.

[0083] The present invention can realize the rapid detection and positioning of dynamic targets in distribution network monitoring images. The recognition process has low computational complexity and fast response speed. It is particularly suitable for the real-time target extraction needs in scenarios such as abnormal behaviors such as approaching towers and temporary gatherings in fixed camera monitoring areas, providing high-confidence image input for subsequent recognition models.

[0084] In the embodiment of the present application, in step S3, as Figure 3 As shown in the figure, an abnormal target image is used as input to a violation recognition model built on YOLOv8. The model consists of an input terminal, a backbone network structure, a neck structure, and a head structure. The backbone feature extraction structure extracts basic semantic features; the neck structure integrates multi-scale target features; and the head structure generates target bounding boxes and corresponding classification labels. Furthermore, a triple attention mechanism is embedded between the backbone feature extraction structure and the neck structure. This mechanism weights the feature maps in the channel, vertical, and horizontal dimensions to improve the model's perception of key areas and enhance the recognition of small targets and complex behaviors.

[0085] The violation identification model described in this application does not simply adopt the existing YOLO series network, but based on the YOLOV8 backbone structure, it innovatively introduces a triple attention mechanism (TripletAttention) and embeds it between the feature extraction stage and the multi-scale fusion stage to construct a fusion attention module of "inter-layer guidance + channel direction + spatial dimension joint constraint". This structure is particularly aimed at the problem that small targets (such as helmets, kite lines, fishing rods, etc.) in distribution network scenarios are difficult to distinguish from behavioral actions. By strengthening high-frequency local feature responses, the model's perception of target semantic boundaries and small posture changes is effectively improved. Compared with the problem of decreased small target detection accuracy in complex environments in the conventional YOLOV8 structure, this solution realizes scene-customized optimization and has substantial differences.

[0086] Specifically, a database was constructed using data augmentation. 1,000 photos of illegal activities, including pond fishing, kite flying near a pylon, and construction violations, were collected. The images were annotated and the data was divided into 75% training, 15% testing, and 15% validation sets. A random sample of all data was used as the training set, 15% testing, and 15% validation sets, avoiding the problem of data acting as both the athlete and the referee.

[0087] Random flipping, rotation, scaling, and color transformation are used to increase the amount of sample data, improve the generalization ability of the model, and reduce overfitting; data enhancement is completed through cropping and patching (RICAP), increasing the original 1,000 violation photos to 3,000.

[0088] YOLOV8's feature recognition framework is as follows Figure 3 As shown in the figure, it consists of four parts: input end, backbone network structure Backbone, neck structure (Neck) and head structure (Head).

[0089] The input end is to segment the violation pictures in the database as the input objects of the network.

[0090] The backbone network structure aggregates and forms a convolutional neural network of image features at different image granularities.

[0091] Neck structure extraction image features are used for multi-scale fusion.

[0092] The head structure is a prediction of the image and generates a bounding box.

[0093] The attention mechanism adopts the triplet attention mechanism (TA), which captures the interaction information between channels and spatial dimensions through three parallel branches; the computational complexity of this mechanism is relatively small because it avoids the introduction of fully connected layers and additional parameters, and thus performs well on devices with limited computing resources.

[0094] The first branch focuses on calculating channel attention, which obtains the correlation between channels through global average pooling and convolution operations; the second branch captures the global information of the feature map in the vertical direction by pooling in the height dimension and then calculating the channel attention; the third branch performs similar operations in the width dimension to obtain the global information of the feature map in the horizontal direction.

[0095] Ultimately, the outputs of these three branches are averaged and aggregated to form the final attention weights. This design allows the TA mechanism to effectively establish interactions between channels and spatial information without reducing feature dimensions, thereby optimizing model training and improving recognition performance.

[0096] In an optional implementation, the violation identification model adopts a dual-channel structure to process the spatial information and temporal change information of the image separately, and realizes the joint identification of the static position features and motion trends of the image target through the linkage analysis of parallel convolutional neural networks and temporal networks (such as LSTM). It is suitable for identifying continuous violations such as overstaying beyond the boundary, slow approach and other scenarios.

[0097] In another optional implementation, the violation identification model is built based on the Transformer architecture, uses a sequence of image blocks as input, and utilizes a global attention mechanism to model the entire image. It is suitable for behavior judgment and interference removal between multiple targets in large-scale scenes, and is suitable for partitioned violation detection scenarios such as when multiple workers are working at the same time.

[0098] The present invention can effectively enhance the ability to express complex targets and behavioral features in distribution network monitoring images, especially in environments where the target size is small, the lighting changes greatly, and the behavioral details are difficult to distinguish, showing good recognition accuracy, providing a reliable judgment basis for the early warning response mechanism.

[0099] In the implementation manner of the present application, in step S4, the violation warning is divided into on-site broadcast and system prompt.

[0100] On-site broadcast: When the violation identification model considers the behavior to be a violation, a warning will be issued through the loudspeaker of the on-site monitoring equipment, prompting the violator to stop the violation and leave.

[0101] The system prompts that in the distribution network violation system, when a violation is identified, a prompt will be given in the system and the corresponding staff will continue to track it, thereby establishing a complete smart distribution network violation identification and early warning system.

[0102] In an optional implementation, the early warning response mechanism performs multi-level response processing on violations based on the behavior level classification rules: for example, violations are divided into three categories: critical violations (such as approaching the edge of a restricted area), general violations (such as not wearing protective equipment), and serious violations (such as entering a high-voltage area); the system controls the voice prompt content and background prompt style of different intensities according to the recognition result level, and determines whether it is necessary to start trajectory tracking or link other system equipment.

[0103] In another optional implementation, the early warning response mechanism performs a local initial judgment by accessing the edge computing node and compares the recognition results with the historical behavior records; if the same target is continuously identified as a violation within a set time period, a "behavior accumulation warning" mechanism can be triggered, and the camera can be linked to track and shoot and force the entire process of the behavior to be recorded, which is suitable for preventing habitual violations or multiple violations without response.

[0104] The present invention can realize dual-channel response of sound intervention and system recording as soon as the violation is identified, effectively avoiding the delay caused by relying solely on manual identification, improving the real-time, closed-loop and traceability of the system in actual operation, and is suitable for the rapid disposal of violations at high-risk sites in distribution networks.

[0105] Example 3, as Figure 4 As shown above, this is a schematic scheme of a method for identifying and warning violations in a smart distribution network. It should be noted that the technical solution of the system for identifying and warning violations in a smart distribution network is the same as the technical solution of the method for identifying and warning violations in a smart distribution network. For details not described in detail in the technical solution of the system for identifying and warning violations in a smart distribution network in this embodiment, please refer to the description of the technical solution of the method for identifying and warning violations in a smart distribution network.

[0106] This embodiment also provides a smart distribution network violation identification and early warning system, including:

[0107] The image acquisition module is used to obtain continuous frame images of distribution network field image data through the monitoring device, and to construct an initialization background model based on the continuous frame images to separate the static area and the foreground target area.

[0108] The image processing module is used to perform inter-frame difference processing in the foreground target area based on the grayscale difference between consecutive frame images, and to determine whether there is an abnormal target image in combination with a preset threshold.

[0109] The violation identification model module is used to perform feature extraction processing on abnormal target images, including extracting the spatial features and motion features of the image, and inputting the extraction results into the violation identification model integrated with the attention mechanism to make violation judgments.

[0110] The violation warning module is used to determine whether there is a violation based on the identification results, and to activate the warning response mechanism when a violation is identified.

[0111] This embodiment also provides an electronic device suitable for a situation where violations of regulations in a smart distribution network are identified and warned, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a method for identifying and warning violations in a smart distribution network as proposed in the above embodiment.

[0112] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, a method for identifying and warning of illegal behaviors in a smart distribution network as proposed in the above embodiment is implemented.

[0113] The storage medium proposed in this embodiment and the method for realizing a violation identification and early warning method for a smart distribution network proposed in the above embodiment belong to the same inventive concept. For technical details not described in detail in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0114] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general hardware, and of course can also be implemented by hardware. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for identifying and warning of illegal behaviors in a smart distribution network, characterized by: include, Continuous frame images of distribution network field image data are obtained through the monitoring device, and an initial background model is constructed based on the continuous frame images to separate the static area and the foreground target area; In the foreground target area, inter-frame difference processing is performed based on the grayscale difference between consecutive frame images, and a preset threshold is combined to determine whether there is an abnormal target image; Perform feature extraction on abnormal target images, including extracting spatial and motion features of the images, and input the extraction results into the violation recognition model integrated with the attention mechanism to determine the violation behavior; Based on the identification results, determine whether there is any violation, and activate the early warning response mechanism when a violation is identified.

2. The method for identifying and warning of illegal behaviors in a smart distribution network according to claim 1, characterized in that: The monitoring device includes a first image acquisition device and a second image acquisition device, which are configured in a distribution network monitoring area; The first image acquisition device is a fixedly deployed image acquisition device with a wide dynamic range and infrared night vision function, and the second image acquisition device is an image acquisition device with maneuverability.

3. The method for identifying and warning of illegal behaviors in a smart distribution network according to claim 2, characterized in that: The initialization background model is constructed based on continuous multi-frame images, and a static area and a foreground target area separation structure are formed through image grayscale and binarization processing.

4. The method for identifying and warning of illegal behaviors in a smart distribution network according to claim 3, wherein: The illegal behavior recognition model includes a classification model trained based on image features and behavior patterns; The early warning response mechanism includes classifying the recognition results into different levels and generating different forms of early warning prompt information accordingly.

5. The method for identifying and warning of illegal behaviors in a smart distribution network according to claim 3, characterized in that: The initialization background model is constructed based on the continuous frame images, including: Collect five consecutive image sequences and perform weighted averaging on the k-5, k-4, k-3, k-2, and k-1 frames to obtain a grayscale image. Convert the grayscale image into a binary image and perform pixel processing to create a background template with all pixel values ​​1 and an initialization background image with pixel values ​​0; Compare the k-th frame image with the initialization background image to separate the foreground target area and the static area, and construct the initialization background model; Based on the constructed initialization background model, the inter-frame difference method is used to perform differential operations on temporally continuous image frames to determine whether the grayscale difference of the corresponding pixel exceeds the preset threshold and identify the moving target; Extracting features from the identified moving target image, wherein the features include centroid difference and rectangle aspect ratio; The abnormal target image is extracted according to the feature result, and a corresponding abnormal target image photo is generated.

6. The method for identifying and warning of illegal behaviors in a smart distribution network according to claim 4, characterized in that: The input is fed into the violation behavior recognition model to perform target recognition and behavior judgment, including: Input the abnormal target image into the violation behavior recognition model based on YOLOV8, including the input end, backbone network structure, neck structure and head structure; Extracting basic features of the image in the backbone network structure, fusing multi-scale image features in the neck structure, and generating a bounding box and corresponding classification results in the head structure; A triple attention mechanism is embedded between the backbone network structure and the neck structure, performing attention calculations in the channel dimension, vertical dimension, and horizontal dimension to enhance the image feature expression capability. Determine whether there are any violations in the image based on the output results.

7. The method for identifying and warning of illegal behaviors in a smart distribution network according to claim 4, characterized in that: The early warning response mechanism includes on-site broadcast and system prompts; On-site broadcast: When the violation recognition model identifies a behavior as a violation, a warning will be issued through the loudspeaker of the on-site monitoring equipment, prompting the violator to stop the violation and leave; The system prompts that in the distribution network violation system, when a violation is identified, a prompt will be given in the system and the corresponding staff will continue to track it.

8. A smart distribution network violation identification and early warning system, applying the smart distribution network violation identification and early warning method according to any one of claims 1 to 7, characterized in that: include: Image acquisition module, image processing module, violation behavior recognition model module and violation warning module; The image acquisition module is used to obtain continuous frame images of the distribution network field image data through the monitoring device, and to construct an initial background model based on the continuous frame images to separate the static area and the foreground target area; The image processing module is used to perform inter-frame difference processing based on the grayscale difference between consecutive frame images in the foreground target area, and determine whether there is an abnormal target image based on the preset threshold. The violation recognition model module is used to extract features from abnormal target images, including extracting spatial and motion features of the images. The extracted results are then fed into the violation recognition model that integrates the attention mechanism to determine the violation. The violation warning module is used to determine whether there is a violation based on the identification results, and to activate the warning response mechanism when a violation is identified.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the smart distribution network violation identification and early warning method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for identifying and warning violations in a smart distribution network according to any one of claims 1 to 7 are implemented.

Images