Transformer substation monitoring identification method and device, electronic equipment and storage medium

Abstract

According to the transformer substation monitoring identification method and device, the electronic equipment and the storage medium provided by the invention, the diversity and robustness of training image samples are increased through data enhancement processing, so that a model can better adapt to image features in different environments. And then, the model is continuously optimized by utilizing feedback circulation of initial annotation data and a prediction result, so that the accuracy of target detection is improved. And through correcting the labeled data and training again, identification errors caused by labeling errors are further reduced, and high precision of the target detection model is ensured, so that the stability of transformer substation monitoring identification is improved, the generalization ability of the model is enhanced, the accuracy of environment state identification of a monitoring area is effectively ensured, and the method is suitable for popularization and application. And therefore, the occurrence of false alarm or missing alarm can be effectively avoided.

Description

Technical Field

[0001] The present application relates to the technical field of substation management, and in particular to a substation monitoring and identification method, device, electronic equipment and storage medium. Background Art

[0002] A substation is a place in the power system that transforms voltage and current, receives electric energy and distributes electric energy. As a key component of the power system, its safe operation is directly related to the safety and reliability of the entire power grid. Therefore, it is necessary to monitor and identify the substation in real time and effectively.

[0003] Based on the above situation, the substation monitoring and identification method used in the prior art will be affected by factors such as shooting angle, lighting and weather, and cannot guarantee the accuracy of substation monitoring and identification, which will lead to false alarms or missed alarms. Summary of the invention

[0004] In view of this, the purpose of this application is to propose a substation monitoring and identification method, device, electronic device and storage medium to solve the above-mentioned technical problems.

[0005] Based on the above purpose, the first aspect of the present application provides a substation monitoring and identification method, comprising:

[0006] Acquire training image samples of the monitoring area of ​​the substation and construct an initial model, wherein the training image samples are provided with initial image annotation data;

[0007] Performing data enhancement processing on the training image samples to obtain image-enhanced training image samples;

[0008] Inputting the image samples for training that have undergone image enhancement into the initial model, and outputting a first predicted target detection result through the initial model;

[0009] Determine a first loss function using the initial image annotation data and the first predicted target detection result, and train and adjust the initial model according to a minimization result obtained in the process of minimizing the first loss function to obtain a trained initial model;

[0010] Acquire a test image sample of a monitoring area of ​​a substation, wherein the test image sample is provided with actual annotation data;

[0011] Inputting the test image sample into the trained initial model, and outputting a second predicted target detection result through the trained initial model;

[0012] The initial image annotation data is corrected using the second predicted target detection result to obtain corrected annotation data, and a second loss function is determined using the corrected annotation data and the second predicted target detection result, and the trained initial model is trained and adjusted according to the minimization result obtained in the process of minimizing the second loss function to obtain a target detection model;

[0013] Acquire an image to be identified in a monitoring area of ​​a substation;

[0014] The image to be identified is input into the target detection model, the target detection model outputs a monitoring identification result, and the environmental state of the monitoring area is determined according to the monitoring identification result.

[0015] Optionally, performing data enhancement processing on the training image samples to obtain image-enhanced training image samples includes:

[0016] The training image sample is rotated and transformed using the preset image rotation angle according to the following formula to obtain the image-enhanced training image sample:

[0017]

[0018] Wherein, x′ represents the horizontal coordinate of the training image sample after image enhancement, y′ represents the vertical coordinate of the training image sample after image enhancement, x represents the horizontal coordinate of the training image sample, y represents the vertical coordinate of the training image sample, and θ represents the image rotation angle.

[0019] Optionally, performing data enhancement processing on the training image samples to obtain image-enhanced training image samples includes:

[0020] The saturation and / or brightness of the training image samples are adjusted to obtain image-enhanced training image samples.

[0021] Optionally, performing data enhancement processing on the training image samples to obtain image-enhanced training image samples includes:

[0022] Determine a first color vector of a training image sample in a color space and a feature value of the first color vector;

[0023] Determine a second color vector of the training image sample in the color space and a characteristic value of the second color vector;

[0024] Determine a third color vector of the training image sample in the color space and a eigenvalue of the third color vector;

[0025] Based on the first color vector, the eigenvalue of the first color vector, the second color vector, the eigenvalue of the second color vector, the third color vector, and the eigenvalue of the third color vector, the color transformation is performed on the training image sample by the following formula to obtain the image enhanced training image sample:

[0026] [p 1 , p 2 , p 3 ][α 1 λ 1 , α 2 λ 2 , α 3 λ 3 ] T

[0027] Among them, p 1 represents the first color vector, p 2 represents the second color vector, p 3 represents the third color vector, λ 1 represents the eigenvalue of the first color vector, λ 2 represents the eigenvalue of the second color vector, λ 3 represents the eigenvalue of the third color vector, α 1 , α 2 and α 3 represents a random variable with mean 0 and variance 0.1.

[0028] Optionally, the initial model includes a convolution layer, and the convolution layer is used to perform convolution processing on the input image sample by the following formula;

[0029]

[0030] Among them, Conv x，y Represents the convolution feature response value at the coordinate (x, y) in the input image sample, w i represents the weight at the i-th position in the convolution kernel space size, b i Represents the deviation value at the i-th position in the convolution kernel space size, v i Represents the brightness value at the i-th position in the input image sample, and p×q represents the size of the convolution kernel.

[0031] Optionally, the initial model also includes a pooling layer, which is used to downsample the convolution feature map obtained by the convolution layer according to a window of a preset size.

[0032] Optionally, the initial model also includes an activation function layer, and the activation function layer is used to perform nonlinear transformation processing on the down-sampled feature map obtained by the pooling layer.

[0033] Based on the same inventive concept, the second aspect of the present application provides a substation monitoring and identification device, comprising:

[0034] A first acquisition module is configured to acquire training image samples of a monitoring area of ​​a substation and construct an initial model, wherein the training image samples are provided with initial image annotation data;

[0035] A data enhancement module is configured to perform data enhancement processing on the training image samples to obtain image-enhanced training image samples;

[0036] A first prediction module is configured to input the image samples for training that have undergone image enhancement into the initial model, and output a first predicted target detection result through the initial model;

[0037] A first training module is configured to determine a first loss function using the initial image annotation data and the first predicted target detection result, and to train and adjust the initial model according to a minimization result obtained in a process of minimizing the first loss function to obtain a trained initial model;

[0038] A second acquisition module is configured to acquire a test image sample of a monitoring area of ​​the substation, wherein the test image sample is provided with actual annotation data;

[0039] A second prediction module is configured to input the test image sample into the trained initial model, and output a second predicted target detection result through the trained initial model;

[0040] A second training module is configured to correct the initial image annotation data using the second predicted target detection result to obtain corrected annotation data, and determine a second loss function using the corrected annotation data and the second predicted target detection result, and train and adjust the trained initial model according to the minimization result obtained in the process of minimizing the second loss function to obtain a target detection model;

[0041] A third acquisition module is configured to acquire an image to be identified in a monitoring area of ​​the substation;

[0042] The monitoring and recognition module is configured to input the image to be recognized into the target detection model, output the monitoring and recognition result through the target detection model, and determine the environmental status of the monitoring area according to the monitoring and recognition result.

[0043] Based on the same inventive concept, the third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the method described in the first aspect above when executing the computer program.

[0044] Based on the same inventive concept, the fourth aspect of the present application provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the method described in the first aspect above.

[0045] From the above, it can be seen that the substation monitoring and identification method, device, electronic device and storage medium provided by the present application increase the diversity and robustness of the training image samples through data enhancement processing, so that the model can better adapt to the image features in different environments. Then, the model is continuously optimized by using the feedback loop of the initial annotation data and the prediction results to improve the accuracy of target detection. By correcting the annotation data and training again, the recognition errors caused by the annotation errors are further reduced, ensuring the high accuracy of the target detection model, thereby not only improving the stability of the substation monitoring and identification, but also enhancing the generalization ability of the model, effectively ensuring the accuracy of the environmental status recognition in the monitoring area, and thus effectively avoiding the occurrence of false alarms or omissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the present application or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 A flowchart of a substation monitoring and identification method according to an embodiment of the present application;

[0048] Figure 2 A schematic diagram of a substation monitoring and identification process according to an embodiment of the present application;

[0049] Figure 3 This is a schematic diagram of an image to be identified in an embodiment of the present application being an image of a first substation gate;

[0050] Figure 4 This is a schematic diagram of an abnormal result of identification in an embodiment of the present application, which is an abnormal movement of a substation gate;

[0051] Figure 5 This is a schematic diagram of an image to be identified in an embodiment of the present application being an image of a second substation gate;

[0052] Figure 6This is a schematic diagram of an embodiment of the present application in which the abnormal result of identification is a human intrusion;

[0053] Figure 7 This is a schematic diagram of an image to be identified in an embodiment of the present application being a first perimeter engineering vehicle image;

[0054] Figure 8 This is a schematic diagram of an abnormal identification result of an embodiment of the present application, which is an identification result of an engineering vehicle of a first substation;

[0055] Fig. 9 A schematic diagram of an image to be identified in an embodiment of the present application is a second perimeter engineering vehicle image;

[0056] Fig.10 This is a schematic diagram of an abnormal identification result of an embodiment of the present application, which is an identification result of an engineering vehicle of a second substation;

[0057] Fig.11 This is a structural block diagram of a substation monitoring and identification device according to an embodiment of the present application;

[0058] Fig.12 A schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0060] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present application belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0061] It is understandable that before using the technical solutions of each embodiment of the present application, the type, scope of use, usage scenarios, etc. of the personal information involved will be informed to the user in an appropriate manner, and the user's authorization will be obtained.

[0062] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly remind the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can independently choose whether to provide personal information to the electronic device, application, server, storage medium or other software or hardware that performs the operation of the technical solution of the present application according to the prompt message.

[0063] As an optional but non-limiting implementation, in response to receiving the user's active request, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0064] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation method of this application. Other methods that meet relevant laws and regulations may also be applied to the implementation method of this application.

[0065] A substation is a place in the power system that transforms voltage and current, receives electric energy and distributes electric energy. As a key component of the power system, its safe operation is directly related to the safety and reliability of the entire power grid. Therefore, it is necessary to monitor and identify the substation in real time and effectively.

[0066] Based on the above situation, the substation monitoring and identification method used in the prior art will be affected by factors such as shooting angle, lighting and weather, and cannot guarantee the accuracy of substation monitoring and identification, which will lead to false alarms or missed alarms.

[0067] The inventors of this application discovered that the accuracy of substation monitoring and identification cannot be guaranteed due to factors such as shooting angle, lighting and weather, which leads to false alarms or missed alarms. One of the reasons is that it affects the internal features of the collected substation equipment images, resulting in sample imbalance and poor quality, thereby improving the accuracy of image recognition and the false alarm rate are both low.

[0068] In addition, at present, the perimeter of substations still mainly relies on manual on-site inspections, which have low inspection efficiency, untimely fault detection, and are time-consuming and labor-intensive. With the increase in the number of newly built substations, the structural shortage of staff has become increasingly prominent, making it difficult to achieve all-weather inspections and data monitoring of substations.

[0069] Therefore, the study of substation perimeter image recognition based on artificial intelligence technology has important practical applications and significance. The substation monitoring and recognition method of this application combines data enhancement and convolutional neural network, which can solve the problems of sample imbalance and poor quality caused by factors such as shooting angle, lighting and weather, so that the recognition rate of substation perimeter recognition results is high and the false alarm rate is low, meeting the needs of practical applications.

[0070] The embodiment of the present application provides a substation monitoring and identification method, which increases the diversity and robustness of training image samples through data enhancement processing, so that the model can better adapt to image features in different environments. Then, the model is continuously optimized by using the feedback loop of initial annotation data and prediction results to improve the accuracy of target detection. By correcting the annotation data and retraining, the recognition errors caused by annotation errors are further reduced, ensuring the high accuracy of the target detection model, thereby not only improving the stability of substation monitoring and identification, but also enhancing the generalization ability of the model, effectively ensuring the accuracy of environmental status recognition in the monitoring area, and thus effectively avoiding the occurrence of false alarms or omissions.

[0071] like Figure 1 As shown, the method of this embodiment includes:

[0072] Step 101: obtain training image samples of the monitoring area of ​​the substation and construct an initial model, wherein the training image samples are provided with initial image annotation data.

[0073] In this step, a certain amount of image data is collected from the monitoring area of ​​the substation, which will be used to train the initial model of image recognition. These images should cover different scenes, different times (such as daytime and nighttime), different weather conditions (such as sunny days, rainy days, foggy days, etc.) and various possible equipment states (such as normal operation and fault state, etc.) in the monitoring area of ​​the substation.

[0074] Image samples can be acquired automatically through surveillance cameras installed in the substation, or manually at specific time points and under specific conditions.

[0075] For each collected image sample, it needs to be annotated, that is, the area or object of interest in the image is marked. These annotated data can be used to indicate which parts of the model are important and need to be recognized and distinguished.

[0076] The annotation data may include the type of equipment (such as transformers, circuit breakers) present in the substation environment, the location of the equipment, the status of the equipment (such as normal, abnormal), and the objects that change in the substation environment (such as people or objects, etc.).

[0077] Step 102: Perform data enhancement processing on the training image samples to obtain image-enhanced training image samples.

[0078] In this step, data augmentation is used to increase the diversity of the data set by performing a series of random transformations on the original data, thereby helping the model learn more generalized features and reducing the risk of overfitting.

[0079] Data augmentation can include rotation, random scaling, cropping, flipping, random color changes, adjusting brightness and contrast, adding noise, etc. These transformations are designed to simulate different viewing angles and lighting conditions, allowing the model to better adapt to unseen data.

[0080] After data augmentation, the original training image samples are converted into a series of transformed new images that retain the basic features of the original images (such as object categories) but have different appearances (such as angles, sizes, lighting, etc.).

[0081] In the substation perimeter recognition scenario, there are problems such as sample imbalance and poor quality caused by factors such as shooting angle, lighting and weather. Data enhancement processing can improve the characteristics of large-scale noise and lighting changes, expand the number of data samples and extract model features. These enhanced images are used as input for the model in the training process, which can help improve the generalization ability and robustness of the model.

[0082] Step 103: input the image samples for training that have undergone image enhancement into the initial model, and output a first predicted target detection result through the initial model.

[0083] In this step, the training image samples that have undergone image enhancement are used to train the machine learning model (i.e., the initial model). In the target detection task, these images usually contain some labeled targets (such as people or objects, etc.), and these labeled information (such as the category and location of the target) will be used as supervision information in the training process.

[0084] The initial model is a convolutional neural network (CNN), which has the ability to extract useful features from the input image. The initial model usually has some predefined architecture and parameters, which will be adjusted and optimized during the training process.

[0085] The augmented training image samples are input into the initial model. The model processes these images and extracts features through its internally defined layers (such as convolutional layers, pooling layers, fully connected layers, etc.).

[0086] After the model processes the input image, it will output the prediction result of the target object in the input image based on the learned features and the specific requirements of the object detection task (such as the category and location of the target). The prediction result usually includes the category of the target, the bounding box (that is, the location and size of the target in the image), and the confidence (that is, the probability that the model believes that this prediction is correct).

[0087] The first predicted object detection result means that this is the prediction given by the model in the early stage of training or when it is not fully optimized. As training progresses, these predictions usually become more accurate.

[0088] Step 104: determine a first loss function using the initial image annotation data and the first predicted target detection result, and train and adjust the initial model according to the minimization result obtained in the process of minimizing the first loss function to obtain a trained initial model.

[0089] In this step, the image samples for training after image enhancement are used as the basic data set for training the model, where each image is equipped with corresponding initial image annotation data. These initial image annotation data usually include information such as the location of the target object (for example, marked by a bounding box), category, etc.

[0090] The first predicted target detection result is the prediction result obtained after the initial model (i.e., the model that has not been fully trained) is used to detect the target in the image. These prediction results may include the predicted target location, category, and corresponding confidence (i.e., the probability that the model believes the prediction is correct).

[0091] The first loss function is a function that evaluates the difference between the model prediction results and the actual labeled data.

[0092] In the object detection task, the loss function may include positioning loss (measuring the difference between the predicted bounding box and the actual bounding box), classification loss (measuring the difference between the predicted category and the actual category), etc. The first loss function is calculated based on the initial image annotation data and the first predicted object detection result, and is used to quantify the current performance of the model. For example, the first loss function is determined based on the difference between the initial image annotation data and the first predicted object detection result.

[0093] During the training process, the goal is to find a set of model parameters that minimizes the value of the loss function. This is usually achieved through optimization algorithms such as backpropagation and gradient descent. In each iteration, the model parameters are adjusted according to the gradient of the loss function to gradually reduce the loss value.

[0094] The minimization result refers to the model parameter updates obtained through the above optimization process, which reduce the value of the loss function, that is, the performance of the model is improved.

[0095] Based on the minimization process, the initial model is adjusted using the optimized model parameters. This process may include multiple iterations, each of which updates the model parameters based on the new prediction results and loss function.

[0096] After multiple iterations of training, when the loss function value of the initial model reaches a preset threshold or no longer decreases significantly, the model is considered to have been trained. This trained model is now able to more accurately identify the target object in the image.

[0097] The process of training and tuning an object detection model by minimizing a loss function, with the goal of improving the accuracy and efficiency of the model in identifying target objects in an image.

[0098] Step 105 : obtaining a test image sample of the monitoring area of ​​the substation, wherein the test image sample is provided with actual annotation data.

[0099] In this step, image data is collected or generated from the monitoring area of ​​the substation. The monitoring area of ​​the substation generally refers to the area inside or around the substation that needs to be monitored and managed, which may include key facilities such as transformers, switchgear, and transmission lines.

[0100] Actual labeled data refers to information that has been manually marked or annotated on image samples. These annotations can be about the location of objects in the image (such as bounding boxes), categories (such as transformers, switchgear), attributes (such as color, size), etc.

[0101] Real-world labeled data is essential for training and supervising learning algorithms because it provides the standard answers that the algorithm needs to learn. During the testing phase, the algorithm's output is compared with this labeled data to evaluate its performance.

[0102] Step 106: input the test image sample into the trained initial model, and output a second predicted target detection result through the trained initial model.

[0103] In this step, the trained initial model refers to a model that has been trained with a large amount of training data. During the training process, the model learns how to extract features from the input data and predict the existence, location, category, and other information of the target based on these features.

[0104] Inputting test image samples into the trained initial model means passing the test image data to the model and letting the model process it. This step is a key step in evaluating the performance of the model because the model will make predictions about the input images based on its learned knowledge.

[0105] After the trained initial model processes the input test image, it will output one or more prediction results. These prediction results usually include information such as the existence, location (for example, represented by a bounding box), and category of the target object. Here, the second predicted target detection result means that this is the prediction result output by the model during the testing phase, which corresponds to the first prediction result output during the training initial model phase (used for feedback and adjustment during the training process).

[0106] Step 107: Use the second predicted target detection result to correct the initial image annotation data to obtain corrected annotation data, and use the corrected annotation data and the second predicted target detection result to determine a second loss function, and adjust the trained initial model according to the minimization result obtained in the process of minimizing the second loss function to obtain a target detection model.

[0107] In this step, considering the impact of data annotation quality on image recognition accuracy, a self-iterative method is used to perform multiple corrections on the initial image annotation data (eg, annotation box) based on the second predicted target detection result.

[0108] The second loss function measures the difference between the model prediction result and the actual labeled data. Use the corrected labeled data and the second predicted target detection result to determine a new loss function (i.e., the second loss function). This new loss function reflects the performance of the model in the current training state and is used to guide subsequent training adjustments.

[0109] By minimizing the second loss function, that is, finding the optimal solution for the model parameters, the difference between the model's prediction results and the actual labeled data is minimized. This minimization process may involve optimization algorithms such as gradient descent.

[0110] Based on the results of the minimization process, the trained initial model is further trained and adjusted. These adjustments include updating the model's weights and biases to make the model perform better in the next iteration.

[0111] After the above iterative optimization process, the final result is an object detection model with improved performance. This model can more accurately identify objects in images and provide their location and category information.

[0112] The above method further reduces the recognition errors caused by labeling errors and ensures the high accuracy of the target detection model, which not only improves the stability of substation monitoring and identification, but also enhances the generalization ability of the model, effectively guarantees the accuracy of environmental status identification in the monitoring area, and effectively avoids the occurrence of false alarms or missed alarms.

[0113] Step 108: Acquire the image to be identified in the monitoring area of ​​the substation.

[0114] In this step, specific monitoring areas are usually divided in the substation, which are the focus of monitoring. The selection of monitoring areas is usually based on the importance of the equipment, the possibility of failure, and the impact of the failure on the operation of the power system. Various sensors, cameras and other equipment will be installed in these areas to monitor the operating status of the equipment and changes in the surrounding environment in real time.

[0115] Images to be identified refer to images captured by cameras and other devices in the monitoring area that need to be further identified or analyzed. These images may contain information such as the appearance of the equipment, the status of the operating indicator lights, changes in the surrounding environment, etc. By processing and analyzing these images, remote monitoring and fault diagnosis of substation equipment status can be achieved, improving operation and maintenance efficiency and safety.

[0116] In the specific monitoring area of ​​the substation, images that need to be further identified or analyzed are captured by cameras and other equipment in order to monitor and analyze the operating status of the substation equipment, which is of great significance for ensuring the safe and stable operation of the power system.

[0117] Step 109: input the image to be identified into the target detection model, output a monitoring and identification result through the target detection model, and determine the environmental status of the monitoring area according to the monitoring and identification result.

[0118] In this step, the image to be recognized is fed into a pre-trained object detection model. An object detection model is a machine learning or deep learning model that has been trained with a large amount of data to recognize specific targets in an image (such as people, cars, animals, buildings, or other objects). These models are usually able to locate the location of the target in the image and may provide some additional information about the target (such as size, shape, category, etc.).

[0119] After processing the input image, the object detection model will output one or more recognition results. These results usually include the category of the object detected in the image, the location (possibly given in the form of a bounding box), and possibly a confidence score (indicating the model's confidence in its recognition result).

[0120] The environmental status of the monitoring area is determined based on the output of the target detection model, and the environmental status is used to indicate whether there is an abnormality in the monitoring area. For example, if the monitoring recognition result output by the target detection model is a person recognition frame, it means that there is a person intrusion in the monitoring area, and timely warning can be issued based on this.

[0121] Through the above scheme, the diversity and robustness of the training image samples are increased through data enhancement processing, so that the model can better adapt to the image features in different environments. Then, the model is continuously optimized by using the feedback loop of the initial labeled data and the prediction results, which improves the accuracy of target detection. By correcting the labeled data and retraining, the recognition errors caused by the labeling errors are further reduced, ensuring the high accuracy of the target detection model, which not only improves the stability of substation monitoring and identification, but also enhances the generalization ability of the model, effectively guarantees the accuracy of environmental status identification in the monitoring area, and can effectively avoid the occurrence of false alarms or omissions.

[0122] In some embodiments, step 102 includes:

[0123] The training image sample is rotated and transformed using the preset image rotation angle according to the following formula to obtain the image-enhanced training image sample:

[0124]

[0125] Wherein, x′ represents the horizontal coordinate of the training image sample after image enhancement, y′ represents the vertical coordinate of the training image sample after image enhancement, x represents the horizontal coordinate of the training image sample, y represents the vertical coordinate of the training image sample, and θ represents the image rotation angle.

[0126] In the above scheme, taking into account the different shooting angles on site, the following rotation matrix formula is used, where is the rotation angle, and a random value is given to rotate the image by an angle θ, and the image size obtained by changing the direction of the image does not change;

[0127]

[0128] The image rotation angle refers to one or more specific rotation angles that have been determined before the image rotation transformation. These angles can be any value, but are usually selected within a certain range to ensure that the generated enhanced image can increase the diversity of the data without deviating too much from the distribution of the original image.

[0129] The above rotation matrix formula is used to map each pixel point in the original image (represented by coordinates) to a new position (represented by coordinates) after rotation.

[0130] In specific implementation, the new position of each pixel needs to be calculated based on the rotation angle and the image center (usually the geometric center of the image). Since rotation may cause some pixels to be mapped outside the image boundary, boundary processing (such as cropping, padding, etc.) may need to be considered in actual operation. The implementation process is as follows:

[0131] First, determine the rotation angle to use. Then, for each pixel in the original image, use the above formula to calculate its new position in the rotated image. Finally, construct the rotated image based on the calculated new position. If the new position is not an integer, you may need to use interpolation to determine the pixel value at that position.

[0132] Through this process, one or more enhanced training image samples can be obtained. These enhanced images can be used to train machine learning models to help the models learn more features, thereby improving the performance and generalization ability of the models.

[0133] In some embodiments, step 102 includes:

[0134] The saturation and / or brightness of the training image samples are adjusted to obtain image-enhanced training image samples.

[0135] In the above scheme, saturation refers to the vividness of the colors in the training image samples. High saturation means that the colors are more vivid, while low saturation makes the colors look duller or closer to gray.

[0136] Brightness refers to how light or dark the training image samples are. High brightness means the image looks brighter, while low brightness makes the image look darker.

[0137] By adjusting the saturation and / or brightness of the image while keeping the hue unchanged, the visual effect of the image can be changed, the diversity of training image samples is increased, the model can be better adapted to different image conditions, and training with enhanced images can improve the accuracy or robustness of the model.

[0138] In addition, it can also help the model learn more generalized features instead of just remembering specific patterns in the training data.

[0139] In some embodiments, step 102 includes:

[0140] Step A1, determining a first color vector of a training image sample in a color space and a feature value of the first color vector.

[0141] Step A2: determine a second color vector of the training image sample in the color space and a feature value of the second color vector.

[0142] Step A3: Determine a third color vector of the training image sample in the color space and a characteristic value of the third color vector.

[0143] Step A4, based on the first color vector, the eigenvalue of the first color vector, the second color vector, the eigenvalue of the second color vector, the third color vector and the eigenvalue of the third color vector, color transformation is performed on the training image sample by the following formula to obtain an image enhanced training image sample:

[0144] [p 1 , p 2 , p 3 ][α 1 λ 1 , α 2 λ 2 , α 3 λ 3 ] T

[0145] Among them, p 1 represents the first color vector, p 2 represents the second color vector, p 3 represents the third color vector, λ 1 represents the eigenvalue of the first color vector, λ 2 represents the eigenvalue of the second color vector, λ 3 represents the eigenvalue of the third color vector, α 1 , α 2 and α 3 represents a random variable with mean 0 and variance 0.1.

[0146] In the above scheme, for a training image sample, three different color vectors are first determined in the color space: a first color vector, a second color vector, and a third color vector. These color vectors can represent different color components or channels in the image, such as red, green, and blue channels in the RGB color space, or components in a more complex color space.

[0147] Next, the eigenvalue is calculated for each color vector. The eigenvalue can be a certain statistic of the color vector, such as the mean, median, standard deviation, etc., which is used to describe the performance or distribution of the color vector in the image.

[0148] Use the following formula to transform the color of the training image samples:

[0149] [p 1 , p 2 , p 3 ][α 1 λ 1 , α 2 λ2 , α 3 λ 3 ] T

[0150] Among them, p 1 represents the first color vector (e.g., the red channel), p 2 represents the second color vector (e.g., the green channel), p 3 Represents the third color vector (e.g., the blue channel). 1 , α 2 and α 3 represents a random variable with a mean of 0 and a variance of 0.1. The introduction of these random variables can increase the color diversity of the transformed image, which may improve the generalization ability of the image recognition or classification model.

[0151] By applying the above formula, the color of the training image samples is transformed to obtain the image-enhanced training image samples.

[0152] This augmentation may help improve the performance of image recognition models because it increases the diversity of the training data, allowing the model to better adapt to different lighting conditions, color changes, etc. This process enhances the image by introducing randomness and transformations based on the eigenvalues ​​of the color vector, aiming to improve the accuracy and robustness of the image recognition or classification model.

[0153] For example, using color change: perform color transformation based on the three basic color spaces of RGB in the image to obtain three main direction vectors and three eigenvalues ​​p in the RGB space. 1 、p 2 、p 3 , 1 , 2 , 3 , and then transform each pixel on the image as follows:

[0154] [p 1 , p 2 , p 3 ][α 1 λ 1 , α 2 λ 2 , α 3 λ 3 ] T

[0155] Among them, α i (i=1, 2, 3) is a random variable with mean 0 and variance 0.1.

[0156] In some embodiments, the initial model includes a convolution layer, and the convolution layer is used to perform convolution processing on the input image sample according to the following formula;

[0157]

[0158] Among them, Conv x，y Represents the convolution feature response value at the coordinate (x, y) in the input image sample, w i represents the weight at the i-th position in the convolution kernel space size, b i Represents the deviation value at the i-th position in the convolution kernel space size, v i Represents the brightness value at the i-th position in the input image sample, and p×q represents the size of the convolution kernel.

[0159] In the above scheme, the initial model is a convolutional neural network, which adopts a training method based on a convolutional neural network during the training process, while considering automatically adjusting the confidence according to the target detection results.

[0160] The initial model includes a convolution layer. Taking a two-dimensional (2D) image input as an example, the convolution kernel is translated and scanned on the image according to a certain step size, and the weighted sum of each parameter in the convolution kernel and the pixel value in the corresponding local area of ​​the image is obtained to obtain the convolution feature output. Assuming that the convolution kernel size is p×q, the convolution process can be expressed as:

[0161]

[0162] Conv x，y represents the convolution feature response value at the coordinate (x, y) in the image, w i and b i Respectively represent the weight and bias value at the i-th position in the convolution kernel space size, v i is the brightness value at the corresponding position in the image. A convolutional layer usually contains multiple convolution kernels. Different convolution kernels have different parameter values ​​and are used to extract different feature information in the image. The convolutional layer effectively reduces the number of parameters through two powerful tools: local receptive field and weight sharing, making the network easier to train and reducing the risk of overfitting.

[0163] In some embodiments, the initial model also includes a pooling layer, and the pooling layer is used to downsample the convolution feature map obtained by the convolution layer according to a window of a preset size.

[0164] In the above scheme, the initial model is a convolutional neural network, which adopts a training method based on a convolutional neural network during the training process, while considering automatically adjusting the confidence according to the target detection results.

[0165] The initial model includes a pooling layer, the main purpose of which is to further reduce the amount of data in the network by downsampling the convolutional feature map. The pooling operation specifically refers to sampling the convolutional feature map with a certain step size according to the window size, taking the average value (average pooling) or the maximum value (maximum pooling) of all feature response values ​​in the window as the representative feature of the window area, thereby effectively reducing the size of the convolutional feature map while retaining the significant convolutional feature response part.

[0166] In some embodiments, the initial model also includes an activation function layer, and the activation function layer is used to perform nonlinear transformation processing on the down-sampled feature map obtained by the pooling layer.

[0167] In the above scheme, the initial model is a convolutional neural network, which adopts a training method based on a convolutional neural network during the training process, while considering automatically adjusting the confidence according to the target detection results.

[0168] The initial model also includes an activation function layer, which usually follows the convolutional layer and adds nonlinear transformations to enhance the network's feature expression capabilities, enabling it to handle more complex data features.

[0169] Activation functions include the Sigmoid function, the tanh function, and the ReLU function. Specifically, the expression of the Sigmoid function is as follows:

[0170]

[0171] The Sigmoid function is monotonically continuous and can map input data between (0, 1), but its derivative is large in the middle and small on both sides, which can cause the gradient to disappear during network training.

[0172] The expression of tanh function is:

[0173]

[0174] The tanh function maps the input data to (-1, 1), and this function also has the problem of gradient disappearance.

[0175] The ReLUc function expression is:

[0176] ReLU(x)=max(0,x).

[0177] The ReLU activation function can effectively solve the problem of gradient disappearance in the positive interval, and it only needs to determine whether the input number is greater than 0, so it has a very fast calculation speed, and its convergence speed is also better than the Sigmoid and tanh functions.

[0178] In some embodiments, the initial model is a convolutional neural network, which adopts a training method based on a convolutional neural network during the training process, while considering automatically adjusting the confidence according to the target detection results.

[0179] The initial model also includes a fully connected layer. The convolutional neural network is stacked with multiple convolutional layers and pooling layers, and usually one or two fully connected layers are added at the end. In the fully connected layer, each neuron is fully connected to all neurons in the previous layer. By integrating the different feature information extracted by the network convolutional layer and pooling layer, a discernible global feature expression is obtained, and the result is further output after mapping.

[0180] In some embodiments, the substation monitoring and identification method of the present application integrates data enhancement and convolutional neural networks. First, in order to address the problems of sample imbalance and poor quality caused by factors such as shooting angle, lighting and weather in the substation perimeter defect identification scenario, data enhancement techniques such as random image scaling, cropping, rotation and random color changes are used to improve the characteristics of large-scale noise and lighting changes, and expand the extraction of target data sets and model features; secondly, considering the impact of data annotation quality on image recognition accuracy, a self-iterative method is used to correct the annotation box multiple times; finally, single-stage training is performed based on the convolutional neural network, and the confidence threshold is automatically adjusted according to the target detection results. After actual data testing and verification, effective adaptation of the algorithm model is achieved. Figure 2 As shown, the following steps are included:

[0181] Step S1: Dataset construction. Image annotation is performed after image acquisition. Then, sample expansion is performed by introducing data enhancement techniques such as random scaling, cropping, brightness, rotation, splicing, symmetry, and color operations such as dithering, so as to increase the diversity of training data and enrich the target dataset, thereby improving the generalization ability of the model.

[0182] Step S2: Data pre-training (i.e., network pre-training): A new adaptive model training strategy, namely, adaptive inference threshold, is adopted. This strategy can automatically adjust the confidence threshold according to the target detection results, thereby improving the performance of the model in different scenarios.

[0183] Step S3: Data testing: Use actual field data to test the model, and continuously adjust the data annotation box and model training parameters (i.e., model parameter modification) according to the test results.

[0184] Step S4: Model self-iteration and optimization: Use actual field data to continuously expand the data training set (i.e., increase data samples to optimize the model) and perform self-iterative training. At the same time, use model optimization technology to reduce model parameters and calculation amount, thereby improving the reasoning speed and efficiency of the model.

[0185] Step S5: Determine whether the substation perimeter is abnormal based on the trained model, and output the abnormality identification result (i.e., the environmental status of the monitoring area).

[0186] For example, the image to be identified in the monitoring area of ​​the substation is the first substation gate image, such as Figure 3 The abnormal recognition result corresponding to the first substation gate image (i.e., the substation gate is abnormal) is shown in Figure 4 as shown; or,

[0187] The image to be identified in the monitoring area of ​​the substation is the second substation gate image, such as Figure 5 The abnormal recognition result (i.e., human intrusion) corresponding to the second substation gate image is shown in Figure 6 as shown; or,

[0188] The image to be identified in the monitoring area of ​​the substation is the first perimeter engineering vehicle image, such as Figure 7 The abnormal recognition result corresponding to the first perimeter engineering vehicle image (i.e., the first substation engineering vehicle recognition result) is shown in Figure 8 shown.

[0189] The image to be identified in the monitoring area of ​​the substation is the second perimeter engineering vehicle image, such as Fig. 9 The abnormal recognition result corresponding to the second perimeter engineering vehicle image (i.e., the second substation engineering vehicle recognition result) is shown in Fig.10 shown.

[0190] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.

[0191] It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0192] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a substation monitoring and identification device.

[0193] refer to Fig.11, the substation monitoring and identification device comprises:

[0194] The first acquisition module 1101 is configured to acquire training image samples of the monitoring area of ​​the substation and construct an initial model, wherein the training image samples are provided with initial image annotation data;

[0195] The data enhancement module 1102 is configured to perform data enhancement processing on the training image samples to obtain image-enhanced training image samples;

[0196] The first prediction module 1103 is configured to input the image samples for training that have undergone image enhancement into the initial model, and output a first predicted target detection result through the initial model;

[0197] A first training module 1104 is configured to determine a first loss function using the initial image annotation data and the first predicted target detection result, and to train and adjust the initial model according to a minimization result obtained in a process of minimizing the first loss function to obtain a trained initial model;

[0198] The second acquisition module 1105 is configured to acquire a test image sample of the monitoring area of ​​the substation, wherein the test image sample is provided with actual annotation data;

[0199] The second prediction module 1106 is configured to input the test image sample into the trained initial model, and output a second predicted target detection result through the trained initial model;

[0200] The second training module 1107 is configured to correct the initial image annotation data using the second predicted target detection result to obtain corrected annotation data, and determine a second loss function using the corrected annotation data and the second predicted target detection result, and adjust the trained initial model according to the minimization result obtained in the process of minimizing the second loss function to obtain a target detection model;

[0201] The third acquisition module 1108 is configured to acquire an image to be identified in the monitoring area of ​​the substation;

[0202] The monitoring and recognition module 1109 is configured to input the image to be recognized into the target detection model, output a monitoring and recognition result through the target detection model, and determine the environmental status of the monitoring area according to the monitoring and recognition result.

[0203] In some embodiments, the data enhancement module 1102 is specifically configured to:

[0204] The training image sample is rotated and transformed using the preset image rotation angle according to the following formula to obtain the image-enhanced training image sample:

[0205]

[0206] Wherein, x′ represents the horizontal coordinate of the training image sample after image enhancement, y′ represents the vertical coordinate of the training image sample after image enhancement, x represents the horizontal coordinate of the training image sample, y represents the vertical coordinate of the training image sample, and θ represents the image rotation angle.

[0207] In some embodiments, the data enhancement module 1102 is specifically configured to:

[0208] The saturation and / or brightness of the training image samples are adjusted to obtain image-enhanced training image samples.

[0209] In some embodiments, the data enhancement module 1102 is specifically configured to:

[0210] Determine a first color vector of a training image sample in a color space and a feature value of the first color vector;

[0211] Determine a second color vector of the training image sample in the color space and a characteristic value of the second color vector;

[0212] Determine a third color vector of the training image sample in the color space and a eigenvalue of the third color vector;

[0213] Based on the first color vector, the eigenvalue of the first color vector, the second color vector, the eigenvalue of the second color vector, the third color vector, and the eigenvalue of the third color vector, the color transformation is performed on the training image sample by the following formula to obtain the image enhanced training image sample:

[0214] [p 1 , p 2 , p 3 ][α 1 λ 1 , α 2 λ 2 , α 3 λ 3 ] T

[0215] Among them, p 1 represents the first color vector, p 2 represents the second color vector, p 3 represents the third color vector, λ 1 represents the eigenvalue of the first color vector, λ2 represents the eigenvalue of the second color vector, λ 3 represents the eigenvalue of the third color vector, α 1 , α 2 and α 3 represents a random variable with mean 0 and variance 0.1.

[0216] In some embodiments, the initial model includes a convolution layer, and the convolution layer is used to perform convolution processing on the input image sample according to the following formula;

[0217]

[0218] Among them, Conv x，y Represents the convolution feature response value at the coordinate (x, y) in the input image sample, w i represents the weight at the i-th position in the convolution kernel space size, b i Represents the deviation value at the i-th position in the convolution kernel space size, v i Represents the brightness value at the i-th position in the input image sample, and p×q represents the size of the convolution kernel.

[0219] In some embodiments, the initial model also includes a pooling layer, and the pooling layer is used to downsample the convolution feature map obtained by the convolution layer according to a window of a preset size.

[0220] In some embodiments, the initial model also includes an activation function layer, and the activation function layer is used to perform nonlinear transformation processing on the down-sampled feature map obtained by the pooling layer.

[0221] For the convenience of description, the above device is described in terms of functions divided into various modules. Of course, when implementing the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0222] The device of the above embodiment is used to implement the corresponding substation monitoring and identification method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0223] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the substation monitoring and identification method described in any of the above embodiments is implemented.

[0224] Fig.12A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1201, a memory 1202, an input / output interface 1203, a communication interface 1204, and a bus 1205. The processor 1201, the memory 1202, the input / output interface 1203, and the communication interface 1204 are connected to each other through the bus 1205 in the device.

[0225] The processor 1201 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0226] The memory 1202 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1202 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1202 and are called and executed by the processor 1201.

[0227] The input / output interface 1203 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0228] The communication interface 1204 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0229] The bus 1205 includes a path that transmits information between the various components of the device (eg, the processor 1201 , the memory 1202 , the input / output interface 1203 , and the communication interface 1204 ).

[0230] It should be noted that, although the above device only shows the processor 1201, the memory 1202, the input / output interface 1203, the communication interface 1204 and the bus 1205, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0231] The electronic device of the above embodiment is used to implement the corresponding substation monitoring and identification method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0232] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the substation monitoring and identification method described in any of the above embodiments.

[0233] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0234] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the substation monitoring and identification method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0235] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0236] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.

[0237] Although the present application has been described in conjunction with specific embodiments of the present application, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0238] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A substation monitoring and identification method, characterized in that: include: Acquire training image samples of the monitoring area of ​​the substation and construct an initial model, wherein the training image samples are provided with initial image annotation data; Performing data enhancement processing on the training image samples to obtain image-enhanced training image samples; Inputting the image samples for training that have undergone image enhancement into the initial model, and outputting a first predicted target detection result through the initial model; Determine a first loss function using the initial image annotation data and the first predicted target detection result, and train and adjust the initial model according to a minimization result obtained in the process of minimizing the first loss function to obtain a trained initial model; Acquire a test image sample of a monitoring area of ​​a substation, wherein the test image sample is provided with actual annotation data; Inputting the test image sample into the trained initial model, and outputting a second predicted target detection result through the trained initial model; The initial image annotation data is corrected using the second predicted target detection result to obtain corrected annotation data, and a second loss function is determined using the corrected annotation data and the second predicted target detection result, and the trained initial model is trained and adjusted according to the minimization result obtained in the process of minimizing the second loss function to obtain a target detection model; Acquire an image to be identified in a monitoring area of ​​a substation; The image to be identified is input into the target detection model, the target detection model outputs a monitoring identification result, and the environmental state of the monitoring area is determined according to the monitoring identification result.

2. The method according to claim 1, characterized in that: The performing data enhancement processing on the training image samples to obtain image-enhanced training image samples includes: The training image sample is rotated and transformed using the preset image rotation angle according to the following formula to obtain the image-enhanced training image sample: Among them, x′ represents the horizontal coordinate of the training image sample after image enhancement, y′ represents the vertical coordinate of the training image sample after image enhancement, x represents the horizontal coordinate of the training image sample, y represents the vertical coordinate of the training image sample, and θ represents the image rotation angle.

3. The method according to claim 1, characterized in that The performing data enhancement processing on the training image samples to obtain image-enhanced training image samples includes: The saturation and / or brightness of the training image samples are adjusted to obtain image-enhanced training image samples.

4. The method according to claim 1, characterized in that: The performing data enhancement processing on the training image samples to obtain image-enhanced training image samples includes: Determine a first color vector of a training image sample in a color space and a feature value of the first color vector; Determine a second color vector of the training image sample in the color space and a characteristic value of the second color vector; Determine a third color vector of the training image sample in the color space and a eigenvalue of the third color vector; Based on the first color vector, the eigenvalue of the first color vector, the second color vector, the eigenvalue of the second color vector, the third color vector, and the eigenvalue of the third color vector, the color transformation is performed on the training image sample by the following formula to obtain the image enhanced training image sample: [p1, p2, p3][α1λ1, α2λ2, α3λ3] T Among them, p1 represents the first color vector, p2 represents the second color vector, p3 represents the third color vector, λ1 represents the eigenvalue of the first color vector, λ2 represents the eigenvalue of the second color vector, λ3 represents the eigenvalue of the third color vector, α1, α2 and α3 represent random variables with mean 0 and variance 0.

1.

5. The method according to claim 1, characterized in that The initial model includes a convolution layer, and the convolution layer is used to perform convolution processing on the input image sample according to the following formula; Among them, Conv x，y Represents the convolution feature response value at the coordinate (x, y) in the input image sample, w i represents the weight at the i-th position in the convolution kernel space size, b i Represents the deviation value at the i-th position in the convolution kernel space size, v i Represents the brightness value at the i-th position in the input image sample, and p×q represents the size of the convolution kernel.

6. The method according to claim 5, characterized in that The initial model also includes a pooling layer, which is used to downsample the convolution feature map obtained by the convolution layer according to a window of a preset size.

7. The method according to claim 6, characterized in that The initial model also includes an activation function layer, which is used to perform nonlinear transformation processing on the down-sampled feature map obtained through the pooling layer.

8. A substation monitoring and identification device, characterized in that: include: A first acquisition module is configured to acquire training image samples of a monitoring area of ​​a substation and construct an initial model, wherein the training image samples are provided with initial image annotation data; A data enhancement module is configured to perform data enhancement processing on the training image samples to obtain image-enhanced training image samples; A first prediction module is configured to input the image samples for training that have undergone image enhancement into the initial model, and output a first predicted target detection result through the initial model; A first training module is configured to determine a first loss function using the initial image annotation data and the first predicted target detection result, and to train and adjust the initial model according to a minimization result obtained in a process of minimizing the first loss function to obtain a trained initial model; A second acquisition module is configured to acquire a test image sample of a monitoring area of ​​the substation, wherein the test image sample is provided with actual annotation data; A second prediction module is configured to input the test image sample into the trained initial model, and output a second predicted target detection result through the trained initial model; A second training module is configured to correct the initial image annotation data using the second predicted target detection result to obtain corrected annotation data, and determine a second loss function using the corrected annotation data and the second predicted target detection result, and train and adjust the trained initial model according to the minimization result obtained in the process of minimizing the second loss function to obtain a target detection model; A third acquisition module is configured to acquire an image to be identified in a monitoring area of ​​the substation; The monitoring and recognition module is configured to input the image to be recognized into the target detection model, output the monitoring and recognition result through the target detection model, and determine the environmental status of the monitoring area according to the monitoring and recognition result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images