Safety early warning method and device for transformer substation area intrusion and medium

Abstract

The invention discloses a safety early warning method and device for transformer substation area intrusion and a medium. The method comprises the following steps: constructing background point cloud and a world coordinate system of a monitoring area; scanning whether a suspected target of which the volume is greater than a threshold value exists in the monitoring area or not; analyzing the monitoring image by using an optimized OfficientNetV2 network model so as to identify whether the suspected target is a foreign object and is located in the prevention and control area, and if so, immediately sending out alarm information; and if the suspected target is a foreign object and is outside the edge of the prevention and control area, tracking and predicting the trajectory of the foreign object. According to the invention, multi-dimensional data of the transformer substation can be acquired, an intruding target can be accurately positioned, the target can be tracked at a high speed, and an alarm is given after the target is identified to enter a prevention and control area.

Description

Technical Field

[0001] This application relates to the technical field of intelligent security monitoring, and particularly to a security early warning method, device and medium for intrusion into a substation area. Background Art

[0002] A substation is an important operating hub connecting power plants and users. Its safety and stability are not only related to the reliability of power supply in the power system, but also related to the development of the national economy. However, substations are vulnerable to intrusion by foreign objects such as wild animals, and such intrusion may cause power failures. Therefore, identifying foreign object intrusion in dangerous areas of substations is very important for ensuring stable power supply in the area. In recent years, with the great development of monitoring technology and information technology, substation security monitoring systems based on video monitoring technology have been widely applied. However, most of these monitoring systems still mainly rely on manual checking by security supervisors, have poor automatic analysis functions for video content, and cannot well track the movement trajectories of intruders, and are prone to problems such as tracking lag and loss of high-speed moving objects. Therefore, it is necessary to propose an efficient security early warning method for intrusion into a substation area, which can timely solve potential safety hazards and maintain stable power supply in the area. Summary of the Invention

[0003] In view of the above problems, this application provides a security early warning method for intrusion into a substation area, which is used to solve the problems of poor automatic analysis function of the above-mentioned video monitoring and inability to effectively track the movement trajectories of intruders.

[0004] To achieve the above object, this application provides a security early warning method for intrusion into a substation area, including the following steps:

[0005] Use a lidar to construct a background point cloud of the monitoring area of the substation, and use a camera to construct a world coordinate system for the monitoring area;

[0006] The lidar drives the camera to focus and obtain monitoring information. When a suspected target with a volume greater than a threshold is scanned in the monitoring area, the lidar calculates the distance and position information of the suspected target and sends it to the camera;

[0007] The camera obtains a monitoring image of the suspected target according to the distance and the position information; and uses an optimized EfficientNetV2 network model to analyze the monitoring image to identify whether the suspected target is a foreign object and is located within the prevention and control area. If so, an alarm message is immediately sent; if the suspected target is a foreign object and outside the edge of the prevention and control area, track and predict the trajectory of the foreign object, including:

[0008] Determine the position of the target based on the previous frame in the monitoring image, and delimit an initial detection range centered on the determined position;

[0009] Use trajectory continuity to perform area scanning on the initial detection range; when performing area scanning, use the initial detection range as a template, use the MedianFlow filter to estimate the new position of the target, and then update the bounding box of the target's detection range according to the new position of the target;

[0010] In each frame of the area scanning, use a feature extractor to extract the appearance features of the target, and use the extracted appearance features together with the position of the target for model update to determine whether the target is lost; and when the target is lost, reset the target and perform global scanning.

[0011] To solve the above technical problems, the present application also provides another technical solution: a security early warning device for substation area intrusion, including:

[0012] A sample construction module, configured to use a lidar to construct a background point cloud of the monitoring area of the substation, and use a camera to construct a world coordinate system for the monitoring area;

[0013] A network construction module, configured to construct a basic neural network for identifying foreign objects invading the perimeter of the substation, and use the sample data in the typical intrusion sample library of the substation perimeter to train the basic neural network for foreign objects invading the perimeter of the substation, and obtain a network model for identifying individual types of intrusions on the perimeter of the substation;

[0014] A real-time monitoring module, configured to obtain a monitoring image of a suspected target through the camera according to the distance and the position information; and use the network model to analyze the monitoring image to identify whether the suspected target is a foreign object and is located within the prevention and control area. If so, immediately send an alarm message; if the suspected target is a foreign object and outside the edge of the prevention and control area, track and predict the trajectory of the foreign object, including:

[0015] Determine the position of the target based on the previous frame in the monitoring image, and delimit an initial detection range centered on the determined position;

[0016] Use trajectory continuity to perform area scanning on the initial detection range; when performing area scanning, use the initial detection range as a template, use the MedianFlow filter to estimate the new position of the target, and then update the bounding box of the target's detection range according to the new position of the target;

[0017] In each frame of the area scan, an appearance feature extractor is used to extract the appearance features of the target, and the extracted appearance features and the position of the target are used together for model update to determine whether the target is lost; and when the target is lost, the target is reset and a global scan is performed.

[0018] To solve the above technical problems, the present application also provides another technical solution: a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the substation area intrusion security warning method described in any one of the above technical solutions.

[0019] Different from the prior art, the above technical solution combines the advantages of lidar and high-definition cameras, can obtain multi-dimensional information of foreign objects, and perform precise positioning; and in this technical solution, an optimized EfficientNet network model is used, and a custom cyclic learning rate strategy is introduced during the training process to obtain the optimal recognition model. At the same time, in the target tracking module, the trajectory search is optimized, and local search is added, which improves the speed of tracking and detection, and further improves the timeliness and accuracy of foreign object intrusion detection.

[0020] The above relevant descriptions of the invention content are only an overview of the technical solutions of the present application. In order to enable those of ordinary skill in the art to more clearly understand the technical solutions of the present application, and then can be implemented according to the content recorded in the text of the specification and the drawings, and in order to make the above objects, other objects, features and advantages of the present application more easily understood, the following is described in conjunction with the specific embodiments of the present application and the drawings. Description of the Drawings

[0021] The drawings are only used to illustrate the principles, implementation methods, applications, features and effects of the specific embodiments of the present invention and other related contents, and should not be considered as a limitation to the present application.

[0022] In the drawings of the specification:

[0023] Figure 1 It is a flowchart of the substation area intrusion security warning method described in the specific embodiment;

[0024] Figure 2 It is a block diagram of the modules of the substation area intrusion security warning device described in the specific embodiment;

[0025] Figure 3 It is a block diagram of the modules of the computer-readable storage medium described in the specific embodiment;

[0026] The reference numerals involved in the above drawings are explained as follows:

[0027] 200. Safety warning device for intrusion into substation area; 201. Sample construction module; 202. Network construction module; 203. Real-time monitoring module; 300. Computer-readable storage medium. Detailed implementation manners

[0028] To describe in detail the possible application scenarios, technical principles, specific implementable solutions, achievable purposes and effects of the present application, etc., the following will be described in detail with reference to the listed specific embodiments and in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, so they are only used as examples and cannot be used to limit the protection scope of the present application.

[0029] Referring to "embodiment" in this article means that the specific features, structures or characteristics described in connection with the embodiment may be included in at least one embodiment of the present application. The term "embodiment" appearing in various positions in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or relevance to other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0030] Unless otherwise defined, the meanings of the technical terms used in this article are the same as those generally understood by those skilled in the technical field to which the present application belongs; the use of the relevant terms in this article is only for describing specific embodiments and is not intended to limit the present application.

[0031] In the description of the present application, the phrase "and / or" is an expression used to describe the logical relationship between objects, indicating that there can be three relationships, for example, A and / or B, which means: there is A, there is B, and there is both A and B at the same time. In addition, the character " / " in this article generally represents an "or" logical relationship between the associated objects before and after.

[0032] In the present application, terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, primary or secondary, or order relationship between these entities or operations.

[0033] Without more limitations, in the present application, the open expressions such as "including", "comprising", "having" or other similar expressions used in the statements are intended to cover non-exclusive inclusion. These expressions do not exclude that there may be other elements in the process, method or product including the said elements, so that the process, method or product including a series of elements may not only include those defined elements, but also include other elements not explicitly listed, or also include elements inherent to this process, method or product.

[0034] Similar to the understanding in the "Examination Guidelines", in this application, expressions such as "greater than", "less than", "exceeding", etc. are understood as not including the number itself; expressions such as "above", "below", "within", etc. are understood as including the number itself. In addition, in the description of the embodiments of this application, the meaning of "multiple" is two or more (including two), and similar expressions related to "many" are also understood in this way, such as "multiple groups", "multiple times", etc., unless otherwise clearly and specifically defined.

[0035] In the description of the embodiments of this application, the spatially related expressions used, such as "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "perpendicular", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the specific embodiment or the accompanying drawings, and is only for the convenience of describing the specific embodiments of this application or facilitating the understanding of the reader, rather than indicating or implying that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, it cannot be understood as a limitation to the embodiments of this application.

[0036] Unless otherwise clearly specified or limited, in the description of the embodiments of this application, the terms "installed", "connected", "connected", "fixed", "set", etc. should be understood in a broad sense. For example, the "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be directly connected, or indirectly connected through an intermediate medium; it can be the communication inside two components or the interaction relationship between two components. For those skilled in the art to which this application belongs, the specific meanings of the above terms in the embodiments of this application can be understood according to specific circumstances.

[0037] Please refer to Figure 1 , this embodiment provides a security early warning method for intrusion in a substation area. This security early warning method for intrusion in a substation area can be applied to each substation. This security early warning method for intrusion in a substation area can be applied to the detection and warning of foreign object intrusion at the perimeter of the substation, reducing potential safety hazards and improving the stability of substation operation.

[0038] In this embodiment, the security early warning method for intrusion in the substation area includes the following steps:

[0039] S101. Use a lidar to construct a background point cloud for the monitoring area of the substation, and use a camera to construct a world coordinate system for the monitoring area;

[0040] S102. The lidar drives the camera to focus and obtain monitoring information. When a suspected target with a volume larger than the threshold is scanned in the monitored area, the lidar calculates the distance and position information of the suspected target and sends it to the camera;

[0041] S103. The camera obtains a monitoring image of the suspected target according to the distance and the position information; and analyzes the monitoring image by using an optimized EfficientNetV2 network model to identify whether the suspected target is a foreign object and is located within the prevention and control area. If so, execute step S104 and immediately send out an alarm message; if the suspected target is a foreign object and outside the edge of the prevention and control area, execute step S105 to track and predict the trajectory of the foreign object.

[0042] Among them, tracking and predicting the trajectory of the foreign object includes the following steps:

[0043] Determine the position of the target according to the previous frame in the monitoring image, and delimit an initial detection range with the determined position as the center;

[0044] Use trajectory continuity to perform area scanning on the initial detection range; when performing area scanning, use the initial detection range as a template, use the MedianFlow filter to estimate the new position of the target, and then update the bounding box of the detection range of the target according to the new position of the target.

[0045] In each frame of the area scanning, use a feature extractor to extract the appearance features of the target, and use the extracted appearance features and the position of the target together for model update to determine whether the target is lost; and when the target is lost, reset the target and perform global scanning.

[0046] In this embodiment, the construction of the background point cloud and the world coordinate system includes the following steps:

[0047] Use the lidar to perform multi-angle and all-round scanning on the monitored area of the substation, and store the collected background point cloud information;

[0048] Prepare calibration pictures, and the calibration pictures should contain some known feature points, such as the corner points of the chessboard; use the camera to collect information from the pictures, and pay attention to maintaining the stability of the camera during the collection process to avoid image blurring and shaking;

[0049] Use a feature point extraction algorithm to extract the feature points in the image and perform feature point matching to obtain a matching result;

[0050] Calculate the essential matrix between two images using the feature matching results. The essential matrix describes the rotation and translation relationship between the images. Based on the essential matrix and the feature matching results, and using the principle of triangulation, estimate the internal parameters (focal length, principal point, etc.) of the camera by minimizing the reprojection error.

[0051] Perform distortion correction on the images according to the internal parameters of the camera. Common distortions include radial distortion and tangential distortion. Distortion correction can improve the image quality and provide a better basis for subsequent vision tasks.

[0052] Iteratively optimize the internal parameters of the camera through the Levenberg-Marquardt optimization algorithm to further reduce the reprojection error, improve the calibration accuracy, and evaluate the accuracy of the calibration results until they meet the error criteria.

[0053] Map the position information of the lidar into the world coordinate system constructed by the camera and store it.

[0054] Among them, the Levenberg-Marquardt algorithm is an iterative algorithm for solving nonlinear least squares problems. In this embodiment, the iterative optimization of the internal parameters of the camera by the Levenberg-Marquardt optimization algorithm includes the following specific steps:

[0055] 1. Initialize the parameter vector θ.

[0056] 2. Calculate the residual of the objective function: γ = y - F(θ^k), where y is the observed data and F is the model function. K is the number of iterations.

[0057] 3. Calculate the Jacobian matrix, that is, the derivative of F with respect to the parameter θ.

[0058] 4. Calculate the approximation of the Hessian matrix.

[0059] 5. Update the parameters according to the current parameters and the approximation of the Jacobian matrix.

[0060] 6. Repeat steps 2 - 5 until convergence.

[0061] Furthermore, the steps of the lidar driving the camera are as follows:

[0062] Use the lidar to perform real-time scanning of the substation monitoring area. After processing the lidar point cloud data, compare the current scene point cloud with the background point cloud information to determine whether there are suspected foreign objects.

[0063] There is a suspected foreign object in the current scene. Determine whether its volume is greater than the threshold. If it is greater than the threshold, collect the orientation information, distance information, and volume information (width and height) of the object and report them to the camera (the camera is a dome camera with a pan-tilt head that can rotate);

[0064] Based on the position information of the radar and the uploaded obstacle orientation information and distance information, the camera calculates the actual position of the obstacle and instructs the camera to rotate and focus on the foreign object to obtain a surveillance image;

[0065] Use the constructed foreign object intrusion detection network to discriminate the collected surveillance image.

[0066] In the above embodiment, the network module for constructing the foreign object intrusion detection includes the following steps:

[0067] Use the camera to collect multi-dimensional samples of intrusion objects around the substation, construct a typical intrusion sample library of the substation perimeter as negative samples, classify the intrusion samples into different intrusion levels, and at the same time collect samples within the monitoring area of the substation as positive samples;

[0068] Use the optimized EfficientNetV2 network model to train the collected samples to obtain a basic network model that can discriminate intrusion objects; among them, the optimized EfficientNetV2 network model uses transfer learning and makes the final prediction based on the feature extraction layer of EfficientNetV2 and the fully connected layer and prediction layer constructed by itself.

[0069] According to the surveillance image obtained by the camera, use the trained neural network to discriminate and detect whether it is an intrusion object.

[0070] Furthermore, the training of the optimized EfficientNetV2 includes the following content:

[0071] Select the EfficientNetV2-S model pre-trained with weights based on ImageNet as the basic model for training. The EfficientNetV2 architecture has various sizes (small, medium, and large), providing a scalable solution that can be customized according to the available computing resources and the complexity of the task. This flexibility allows the selection of a model variant that balances computational efficiency and classification performance; at the same time, due to the small amount of collected and labeled data, the pre-trained model can be fine-tuned on a specific dataset, thus significantly improving the learning process. In this embodiment, transfer learning is used during training, and the weight model pre-trained on a large dataset is fine-tuned on its own dataset, and at the same time, a cyclic learning rate is introduced to improve the learning process.

[0072] Load the dataset of the sample library, analyze the quantity of each data category, perform upsampling on the categories with a small quantity to solve the problem of sample imbalance, and split the dataset into a training set, a validation set, and a test set;

[0073] Implement data augmentation techniques, including but not limited to horizontal flipping, rotation, brightness and contrast adjustment, and random cropping, etc., to enhance the diversity of the data and improve the generalization ability of the model;

[0074] Use a custom layer to extend the model for multi-class classification. The model passes through the feature extraction layer of EfficientNetV2 to output a 1280-dimensional feature vector for each image. Subsequently, the Dense layer compresses these features from 1280 dimensions to 512 dimensions, thus allowing a more compact but information-rich representation of the image features. Among them, the Dense layer is a network layer of the EfficientNetV2 model. The Dense layer is a densely connected layer. First, use the EfficientNetV2 feature extraction layer to extract features to obtain a one-dimensional feature vector of length 1280, then connect it with two fully connected layers (Dense), and finally input it into the prediction layer (softmax) to obtain the final prediction result.

[0075] To enhance the training process and prevent overfitting, add a batch normalization layer to normalize the output of the Dense layer through scaling and shifting. This is crucial for stable learning and maintaining the consistency of data batches. Then use a Dropout layer to randomly zero out a part of the input units, which further helps with generalization and prevents the model from relying too much on specific features.

[0076] After configuring the model, start the training. During the training process, use a custom cyclic learning rate. The formula for the cyclic learning rate is as follows:

[0077]

[0078] Where CLR is the current learning rate, LR max and LR min are the maximum and minimum values of the fixed learning rate, β is an empirical constant with a value of 0.5 here, t is the current iteration number, and N is the total number of iterations; during the transfer learning process, using a cyclic learning rate can reduce the instability of the training effect and improve the acquisition rate of the optimal model;

[0079] Furthermore, the tracking of the target outside the dangerous area includes the following steps:

[0080] Initialization. Use the camera to determine that the detected foreign object is outside the dangerous area, and at the same time extract the position of the target object, which is used as the initial bounding box of the target;

[0081] Tracking detection. Use trajectory continuity to perform area scanning. Based on the accurately tracked target positions in previous frames, predict the possible position range of the object in the next frame, and use this range as the detection area of the detector, which can greatly reduce the detection range. When a sudden change in motion occurs, that is, the trajectory is discontinuous, the detector performs a global scan; during the scan, use the initial bounding box of the target as a template, use the MedianFlow filter to estimate the new position of the target, and then update the bounding box of the target according to the new position of the target; in this embodiment, the tracking algorithm is improved based on the TLD algorithm. The TLD detector performs a global scan to determine the target position, resulting in a large number of windows to be detected. In this embodiment, based on the position of the target determined in the previous frame, a detection range is delimited with the position of the target as the center, and a local scan is first performed on the detection range. When the tracked target is not detected, a global scan is performed, thus avoiding resource waste to a certain extent.

[0082] Comprehensive judgment. In each frame, use the current position and appearance information of the target to update the model of the target, that is, use a feature extractor (such as Haar-like features) to extract the appearance features of the target, and use them together with the position information of the target for model update; when the target is lost during the search, use the detection module to reset the target;

[0083] Sample library update. According to the comprehensive judgment results, perform online update on the sample library, and use the sample library to retrain the detection module to further improve its detection ability.

[0084] Judge the distance between the target position of each frame and the boundary line of the dangerous area. If it is less than the threshold, give an alarm. If there is no position change for a long time, exit the tracking and monitoring of the object.

[0085] As Figure 2 shown, in another embodiment, a security early warning device 200 for substation area intrusion is provided. The security early warning device for substation area intrusion includes:

[0086] A sample construction module 201, configured to use a lidar to construct a background point cloud of the monitoring area of the substation, and use a camera to construct a world coordinate system for the monitoring area;

[0087] A network construction module 202, configured to construct a basic neural network for identifying foreign object intrusion objects on the perimeter of the substation, and use the sample data in the typical intrusion sample library on the perimeter of the substation to train the basic neural network for foreign object intrusion objects on the perimeter of the substation, and obtain a network model for identifying individual types of intrusion on the perimeter of the substation;

[0088] The real-time monitoring module 203 is used to obtain a monitoring image of a suspected target according to the distance and the position information through the camera; and analyze the monitoring image by using the network model to identify whether the suspected target is a foreign object and is located within the prevention and control area. If so, an alarm message is immediately sent; if the suspected target is a foreign object and outside the edge of the prevention and control area, the trajectory of the foreign object is tracked and predicted. The tracking and predicting the trajectory of the foreign object includes:

[0089] Determine the position of the target according to the previous frame in the monitoring image, and delimit an initial detection range with the determined position as the center;

[0090] Perform regional scanning on the initial detection range by using trajectory continuity; when performing regional scanning, use the initial detection range as a template, use the MedianFlow filter to estimate the new position of the target, and then update the bounding box of the detection range of the target according to the new position of the target;

[0091] In each frame of the regional scanning, use a feature extractor to extract the appearance features of the target, and use the extracted appearance features together with the position of the target for model update to determine whether the target is lost; and when the target is lost, reset the target and perform global scanning.

[0092] In an embodiment, the security early warning device for substation area intrusion further includes an alarm module, which is used to generate an alarm message when a foreign object intrusion is detected in the monitoring area, wherein the alarm message includes an intrusion detection result and an intrusion level.

[0093] The network model is an optimized EfficientNetV2 network model, and the optimized EfficientNetV2 network model is obtained through the following steps:

[0094] Select the EfficientNetV2-S model pre-trained with weights based on ImageNet as the basic model for training;

[0095] Load the data set of the sample library, analyze the quantity of each data category, use upsampling for the categories with less quantity to solve the problem of sample imbalance, and split the data set into a training set, a validation set and a test set;

[0096] Perform data augmentation on the data set, and the data augmentation includes any one or more of horizontal flipping, rotation, brightness and contrast adjustment, and random cropping;

[0097] The model is extended with custom layers for multi-class classification. The model passes through the feature extraction layer of EfficientNetV2 to output a 1280-dimensional feature vector for each image. Subsequently, the Dense layer compresses the feature vector from 1280 dimensions to 512 dimensions;

[0098] A batch normalization layer is set up to normalize the output of the Dense layer by scaling and shifting;

[0099] After configuring the model, training begins. A custom cyclical learning rate is used during the training process; the formula for defining the cyclical learning rate is as follows:

[0100]

[0101] where CLR is the current learning rate, LR max and LR min are the maximum and minimum values of the fixed learning rate, β is an empirical constant with a value of 0.5 here, t is the current iteration number, and N is the total number of iterations.

[0102] In this embodiment, the present solution combines the advantages of lidar and high-definition cameras. It can not only obtain multi-dimensional information of foreign objects to make up for the problems existing in general solutions in detecting the size, depth, position, and other information of foreign objects, but also utilize the fast processing capabilities of cameras and computer vision algorithms to provide accurate and fast safety warnings. This embodiment uses lidar for precise positioning, does not rely entirely on ambient light, avoids some common problems of traditional lighting equipment deficiencies, and at the same time avoids the problem of excessive data volume and difficulty in processing caused by only using lidar. After precise positioning by lidar, a high-definition camera is used for accurate and fast analysis. This embodiment also optimizes the training process of the EfficientNet network model, introduces a custom cyclical learning rate strategy during the training process to obtain an optimal recognition model, and at the same time optimizes the trajectory search in the target tracking module, adds local search, improves the speed of tracking and detection, and further optimizes the detection accuracy and warning efficiency of the solution.

[0103] As Figure 3 shown, in another embodiment, a computer-readable storage medium 300 is provided. A computer program is stored on the storage medium, and when the computer program is executed by a processor, it implements the safety warning method for substation area intrusion in any one of the above embodiments.

[0104] Among them, the computer-readable storage medium may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a ferromagnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory may be a disk memory or a tape memory.

[0105] The volatile memory may be a random access memory (RAM), which serves as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), sync link dynamic random access memory (SLDRAM), direct rambus random access memory (DRRAM). The computer-readable storage medium described in embodiments of the present invention is intended to include these and any other suitable types of memory.

[0106] In some embodiments, the processor may be implemented by software, hardware, firmware, or a combination thereof, and at least one of a circuit, a single or multiple application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), central processing units (CPUs), controllers, microcontrollers, and microprocessors may be used, so that the processor can execute some steps, all steps, or any combination of the steps in the security warning method for substation area intrusion in various embodiments of the present application.

[0107] Finally, it should be noted that although the above embodiments have been described in the text and drawings of the specification of the present application, the patent protection scope of the present application cannot be limited thereby. Any technical solution obtained by replacing or modifying an equivalent structure or equivalent process based on the substantial concept of the present application and using the content recorded in the text and drawings of the specification of the present application, as well as any technical solution directly or indirectly implementing the above embodiments in other related technical fields, is included in the patent protection scope of the present application.

Claims

1. A security early warning method for substation area intrusion, characterized in that: The following steps are involved: Using a laser radar to construct a background point cloud for the monitoring area of ​​the substation, and using a camera to construct a world coordinate system for the monitoring area; The laser radar drives the camera to focus and obtain monitoring information. When a suspected target with a volume greater than a threshold is scanned in the monitoring area, the laser radar calculates the distance and position information of the suspected target and sends it to the camera; The camera obtains a monitoring image of the suspected target according to the distance and the location information; and uses the optimized EfficientNetV2 network model to analyze the monitoring image to identify whether the suspected target is a foreign object and is located in the control area, and if so, immediately issues an alarm message; If the suspected target is a foreign object and is outside the edge of the control area, the trajectory of the foreign object is tracked and predicted, including: Determine the position of the target according to the previous frame in the monitoring image, and define an initial detection range with the determined position as the center; Performing a regional scan on the initial detection range using trajectory continuity; using the initial detection range as a template during the regional scan, estimating a new position of the target using a MedianFlow filter, and then updating a bounding box of the detection range of the target according to the new position of the target; In each frame of the area scan, a feature extractor is used to extract the appearance features of the target, and the extracted appearance features are used together with the position of the target for model update to determine whether the target is lost; and when the target is lost, the target is reset and a global scan is performed.

2. The security early warning method for substation area intrusion according to claim 1 is characterized in that: The tracking and predicting the trajectory of the foreign object also includes: The distance between the position of the target in each frame and the boundary line of the control area is judged. If the distance is less than the threshold, an alarm is issued. If the position of the target does not change for a long time, the tracking and monitoring of the target is exited.

3. The security early warning method for substation area intrusion according to claim 1 is characterized in that: The optimized EfficientNetV2 network model is obtained by the following steps: Select the EfficientNetV2-S model with weight pre-training based on ImageNet as the basic model for training; Load the dataset of the sample library, analyze the number of each data category, use upsampling to solve the problem of sample imbalance for categories with fewer numbers, and split the dataset into training set, validation set, and test set; Performing data enhancement on the data set, wherein the data enhancement includes any one or more of horizontal flipping, rotation, brightness contrast adjustment, and random cropping; Use custom layers to extend the model for multi-class classification. The model outputs a 1280-dimensional feature vector for each image through the feature extraction layer of EfficientNetV2, and then the Dense layer compresses the feature vector from 1280 dimensions to 512 dimensions. Set up a batch normalization layer to normalize the output of the Dense layer by scaling and shifting; After configuring the model, start training. During the training process, use a custom cyclic learning rate. The definition formula of the cyclic learning rate is as follows: Among them, CLR is the current learning rate, LR max and LR min is the maximum and minimum value of the fixed learning rate, β is an empirical constant, which is 0.5 here, t is the current number of iterations, and N is the total number of iterations.

4. The security early warning method for substation area intrusion according to claim 1 is characterized in that: The background point cloud and world coordinate system construction includes the following steps: Use laser radar to scan the monitoring area of ​​the substation from multiple angles and in all directions, and store the collected background point cloud information; Prepare a calibration image, which should contain some known feature points, and use the camera to collect information from the image; Use feature point extraction algorithm to extract feature points in the image, and perform feature point matching to obtain matching results; Using the feature matching result, the essential matrix between the two images is calculated, where the essential matrix describes the rotation and translation relationship between the images; Based on the triangulation principle, the essential matrix and the feature matching result, estimating the intrinsic parameters of the camera; Performing distortion correction on the image according to the intrinsic parameters of the camera; Iteratively optimizing the internal parameters by using the Levenberg-Marquardt optimization algorithm, and evaluating the accuracy of the calibration results until the error standard is met; The position information of the lidar is mapped to the world coordinate system constructed by the camera and stored.

5. The security early warning method for substation area intrusion according to claim 1 is characterized in that: The laser radar drives the camera to focus and obtain monitoring information, including the following steps: Use laser radar to scan the substation monitoring area in real time. After processing the laser point cloud data, compare the current scene point cloud with the background point cloud information to determine whether there are suspected external objects. If there is a suspected external object in the current scene, determine whether its volume is greater than a threshold. If it is greater than the threshold, collect the object's orientation information, distance information, and volume information and report them to the camera; The camera calculates the actual position of the obstacle based on the radar's position information and the uploaded obstacle's position and distance information, and instructs the camera to rotate and focus on the foreign object to obtain the monitoring image; The constructed foreign body intrusion detection network is used to identify the collected monitoring images.

6. The security early warning method for substation area intrusion according to claim 5 is characterized in that: The construction of the foreign body intrusion detection network includes the following contents: Use cameras to collect multi-dimensional samples of intrusion objects around the substation, build a typical intrusion sample library around the substation as negative samples, and classify the intrusion samples into different levels of intrusion. At the same time, collect samples within the monitoring area of ​​the substation as positive samples. The collected samples are trained using the optimized EfficientNetV2 to obtain a basic network that can identify intrusion objects; Based on the monitoring images obtained by the camera, a trained neural network is used to determine whether it is an intrusion object.

7. A safety warning device for substation area intrusion, characterized in that: include: A sample construction module, for constructing a background point cloud for a monitoring area of ​​a substation using a laser radar, and constructing a world coordinate system for the monitoring area using a camera; A network construction module is used to construct a basic neural network for identifying foreign objects intruding into the substation perimeter, and to train the basic neural network for identifying foreign objects intruding into the substation perimeter using sample data in a typical intrusion sample library of the substation perimeter, so as to obtain a network model for identifying individual types of intrusion into the substation perimeter; A real-time monitoring module is used to obtain a monitoring image of a suspected target through the camera according to the distance and location information; and to analyze the monitoring image using the network model to identify whether the suspected target is a foreign object and is located in the control area, and if so, to immediately issue an alarm message; If the suspected target is a foreign object and is outside the edge of the control area, the trajectory of the foreign object is tracked and predicted, including: Determine the position of the target according to the previous frame in the monitoring image, and define an initial detection range with the determined position as the center; Performing a regional scan on the initial detection range using trajectory continuity; using the initial detection range as a template during the regional scan, estimating a new position of the target using a MedianFlow filter, and then updating a bounding box of the detection range of the target according to the new position of the target; In each frame of the area scan, a feature extractor is used to extract the appearance features of the target, and the extracted appearance features are used together with the position of the target for model update to determine whether the target is lost; and when the target is lost, the target is reset and a global scan is performed.

8. The safety warning device for substation area intrusion according to claim 7 is characterized in that: It also includes an alarm module, which is used to generate alarm information when foreign objects are detected in the monitoring area, wherein the alarm information includes the intrusion detection result and the intrusion level.

9. The safety warning device for substation area intrusion according to claim 7, characterized in that: The network model is an optimized EfficientNetV2 network model, and the optimized EfficientNetV2 network model is obtained by the following steps: Select the EfficientNetV2-S model with weight pre-training based on ImageNet as the basic model for training; Load the dataset of the sample library, analyze the number of each data category, use upsampling to solve the problem of sample imbalance for categories with fewer numbers, and split the dataset into training set, validation set, and test set; Performing data enhancement on the data set, wherein the data enhancement includes any one or more of horizontal flipping, rotation, brightness contrast adjustment, and random cropping; Use custom layers to extend the model for multi-class classification. The model outputs a 1280-dimensional feature vector for each image through the feature extraction layer of EfficientNetV2, and then the Dense layer compresses the feature vector from 1280 dimensions to 512 dimensions. Set up a batch normalization layer to normalize the output of the Dense layer by scaling and shifting; After configuring the model, start training. During the training process, use a custom cyclic learning rate. The definition formula of the cyclic learning rate is as follows: Among them, CLR is the current learning rate, LR max and LR min is the maximum and minimum value of the fixed learning rate, β is an empirical constant, which is 0.5 here, t is the current number of iterations, and N is the total number of iterations.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the security early warning method for substation area intrusion according to any one of claims 1 to 7.

Images