Method and device for identifying substation equipment defects, electronic device and storage medium

Abstract

Embodiments of the present application disclose a power substation equipment defect identification method and device, electronic equipment and a storage medium, wherein the method comprises: obtaining a target equipment image of a power substation equipment to be detected; inputting the target equipment image into a pre-trained equipment defect identification model to obtain an identification result of the power substation equipment defect; wherein the equipment defect identification model comprises a convolution module and a full connection module in parallel, the convolution module comprises at least one convolution layer and at least one full connection layer connected in series with the convolution layer, and the full connection module comprises a full connection module composed of at least one full connection layer. The technical scheme of the embodiments of the present application can simultaneously identify the location of defects in the target equipment image and locate the defect type by pre-establishing the equipment defect identification model with the convolution module and the full connection module in parallel, thereby reducing the degree of human participation and improving the efficiency and accuracy of the power substation equipment defect identification.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computer image technology, and more particularly to a method, apparatus, electronic device and storage medium for identifying defects in substation equipment. Background Technology

[0002] Substations are an important component of the power system, housing numerous electrical equipment. After prolonged use, the surfaces of these devices are prone to damage, leading to corrosion of the internal metal components. Therefore, it is necessary to conduct regular defect inspections on substation equipment to prevent serious power accidents.

[0003] The current method of detecting defects in substation equipment relies on manual assessment of the degree of internal corrosion based on the extent of damage. This method is prone to misjudgment, resulting in inaccurate test results. Furthermore, manual operation consumes unnecessary time and manpower, which does not meet the actual application requirements. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for identifying defects in substation equipment, so as to achieve intelligent analysis and identification of defects in substation equipment.

[0005] In a first aspect, embodiments of the present invention provide a method for identifying defects in substation equipment, the method comprising:

[0006] Acquire images of the target equipment in the substation to be inspected;

[0007] The target equipment image is input into a pre-trained equipment defect recognition model to obtain the identification results of substation equipment defects;

[0008] The equipment defect identification model includes parallel convolutional modules and fully connected modules. The convolutional module includes at least one convolutional layer and at least one fully connected layer connected in series with the convolutional layer. The fully connected module includes a fully connected module composed of at least one fully connected layer.

[0009] Secondly, embodiments of the present invention also provide a device for identifying defects in substation equipment, the device comprising:

[0010] The equipment image acquisition module is used to acquire the target equipment image of the substation equipment to be inspected;

[0011] The equipment image input module is used to input the target equipment image into the pre-trained equipment defect recognition model to obtain the identification results of substation equipment defects;

[0012] The equipment defect identification model includes parallel convolutional modules and fully connected modules. The convolutional module includes at least one convolutional layer and at least one fully connected layer connected in series with the convolutional layer. The fully connected module includes a fully connected module composed of at least one fully connected layer.

[0013] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0014] One or more processors;

[0015] Storage device for storing one or more programs.

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the substation equipment defect identification method provided in any embodiment of the present invention.

[0017] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for identifying substation equipment defects provided in any embodiment of the present invention.

[0018] This invention proposes a method for identifying defects in substation equipment. This method is based on a neural network model and employs parallel-connected convolutional and fully connected modules, which effectively improves the robustness of the equipment defect identification model compared to a serially connected model structure. Furthermore, by fusing the output results of fully connected and convolutional modules, the accuracy of the output results can be effectively improved, thus achieving accurate and efficient substation defect identification. This solves the problems of high workload, low efficiency, and inaccuracy caused by manual inspection of substation defects in existing technologies. In addition, during the training process of the equipment defect identification model, this invention proposes a method for constructing a training dataset. By augmenting the acquired sample equipment images through image augmentation, the comprehensiveness of the training dataset is improved, and the generalization ability of the neural network model is enhanced. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of exemplary embodiments of the present invention, the accompanying drawings used in describing the embodiments are briefly introduced below. Obviously, the accompanying drawings described are only a portion of the drawings of the embodiments to be described in this invention, and not all of the drawings. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort.

[0020] Figure 1This is a flowchart illustrating a method for identifying defects in substation equipment provided in Embodiment 1 of the present invention.

[0021] Figure 2 This is a schematic diagram of the equipment defect identification model in a substation equipment defect identification method provided in Embodiment 1 of the present invention;

[0022] Figure 3 This is a flowchart illustrating a method for identifying defects in substation equipment provided in Embodiment 2 of the present invention.

[0023] Figure 4 This is a schematic diagram of the structure of a substation equipment defect identification device provided in Embodiment 3 of the present invention;

[0024] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0025] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0026] It should also be noted that, for ease of description, the accompanying drawings show only the parts relevant to the invention and not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the drawings. The process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0027] Example 1

[0028] Figure 1 This is a flowchart illustrating a method for identifying defects in substation equipment according to Embodiment 1 of the present invention. This embodiment is applicable to situations where defects are detected in substation equipment. The method can be executed by a device for identifying defects in substation equipment. This device can be implemented by software and / or hardware and can be configured in a terminal and / or server to implement the method for identifying defects in substation equipment according to the present invention.

[0029] like Figure 1 As shown, the method in this embodiment may specifically include:

[0030] S110. Obtain the target equipment image of the substation equipment to be inspected.

[0031] In this embodiment, the target equipment image can be understood as an image acquired during the inspection of the substation equipment to be inspected. Exemplarily, the target equipment image can be acquired through an image acquisition device or an image database, etc., and this embodiment does not limit this. For example, a fixed-point shooting device or an inspection robot can be used to photograph the substation equipment to be inspected, and the captured image can be used as the target equipment image.

[0032] In practice, image acquisition devices can be used to acquire images of the substation equipment to be inspected, and the acquired images can be used as target equipment images of the substation equipment to be inspected.

[0033] S120. Input the target equipment image into the pre-trained equipment defect recognition model to obtain the identification results of substation equipment defects.

[0034] In this embodiment, the equipment defect identification model can be a neural network model used to identify equipment defects in substation equipment to be detected through target equipment images. Exemplarily, the equipment defect identification model can be composed of at least one of convolutional neural networks, recurrent neural networks, or deep neural networks; this embodiment does not limit this to any particular type.

[0035] Figure 2 This is a schematic diagram of the equipment defect identification model in an embodiment of the present invention. In this embodiment, the specific structure of the equipment defect identification model can be as follows: Figure 2 As shown.

[0036] Optionally, the equipment defect identification model may include parallel convolutional modules and fully connected modules. The convolutional module may include at least one convolutional layer and at least one fully connected layer connected in series with the convolutional layer. The fully connected module may include a fully connected module composed of at least one fully connected layer.

[0037] In this model, the convolutional module can be understood as a deep feedforward neural network module used to perform convolution calculations. The fully connected layer can be understood as a layered structure composed of multiple neurons, with each neuron connected to all inputs by weights. In the equipment defect identification model, the convolutional layer extracts feature information from the target equipment image, while the fully connected layer acts as a "classifier." Therefore, the convolutional module can not only extract feature information from the target equipment image but also perform classification analysis.

[0038] Since the overall substation equipment image occupies a large proportion of the target equipment image, while the defective parts that need to be identified are usually located inside the substation equipment and occupy a small proportion of the target equipment image, the equipment defect identification model adopts parallel convolutional modules and fully connected modules. This can enhance the bias of the equipment defect identification model in feature information extraction and classification analysis. Compared with serially connecting the convolutional modules and fully connected modules, it can effectively improve the sensitivity of the equipment defect identification model to the identification of substation equipment defects in the target equipment image.

[0039] This invention proposes a device defect identification method based on a neural network model. It adopts a parallel connection of convolutional and fully connected modules and fuses the output results of each module. Compared with the serial connection model structure, this method effectively improves the robustness of the device defect identification model. Furthermore, by fusing the output results of fully connected modules with the output results of convolutional modules, the accuracy of the output results can be effectively improved, thereby achieving the technical effect of accurate and efficient substation defect identification.

[0040] Optionally, the equipment defect identification model may also include a feature map extraction module, which may include at least one convolutional layer.

[0041] The feature map extraction module can be understood as a neural network module used to extract feature information from the target device image. For example, the feature map extraction module can be composed of at least one of a convolutional neural network structure, a deep neural network structure, or other forms of neural network structure, such as VGG-Net or RES-Net.

[0042] It should be noted that a convolutional module can be used to extract feature information from the target device image, while a fully connected module can be used to extract spatial relationship information from the target device image. That is, it is possible to determine the location of the defective part in the target device image, as well as the type of defect in the target device image.

[0043] It should also be noted that before inputting the target device image into the pre-trained device defect recognition model, a series of image preprocessing operations can be performed on the target device image to improve the diversity of the image dataset, reduce the risk of overfitting, and improve the generalization of the device defect recognition model. Optionally, the preprocessing operations on the target device image may include, but are not limited to, geometric transformations (e.g., rotation, scaling, cropping, translation, or affine transformations), color space transformations (e.g., contrast changes, brightness changes, saturation changes, histogram enhancement, or grayscale adjustment), or pixel relationship adjustments (e.g., blurring, sharpening, or noise reduction). This embodiment does not limit these operations.

[0044] In this embodiment, the equipment defect identification model can be trained based on sample equipment images and the expected identification results corresponding to the sample equipment images.

[0045] Based on the above technical solution, the method further includes: constructing a training dataset for training the equipment defect identification model, wherein the training dataset includes sample equipment images and expected identification results corresponding to the sample equipment images; inputting the sample equipment images into the model to be trained to obtain the model output results corresponding to the sample equipment images; adjusting the model to be trained according to the model output results corresponding to the sample equipment images, the expected identification results, and the pre-constructed model loss function to obtain the equipment defect identification model.

[0046] The training images can be standard images of the equipment used for model training. The expected recognition results can be pre-labeled data used to identify equipment defects. The expected recognition results can be used as a basis for evaluating subsequent predictions and classification results.

[0047] Optionally, a training dataset for training the equipment defect identification model is constructed, including: acquiring historical equipment images of substation equipment, determining sample equipment images based on the historical equipment images; labeling the sample equipment images to obtain the expected identification results corresponding to the sample equipment images; and constructing a training dataset for training the equipment defect identification model based on the sample equipment images and the expected identification results corresponding to the sample equipment images.

[0048] Among them, historical equipment images can be images of power equipment collected at a certain point in history.

[0049] Specifically, when constructing the training dataset for the model, historical equipment images of substation equipment can be obtained, and sample equipment images can be determined by filtering from the historical equipment images. Furthermore, equipment defects can be labeled on the sample equipment images to obtain the expected recognition results corresponding to the sample equipment images. Finally, the training dataset for the equipment defect equipment model can be constructed based on the sample equipment images and the expected recognition results corresponding to the sample equipment images.

[0050] It should be noted that historical device images can be acquired in real time from image acquisition devices, or from image databases, or other acquisition methods. This embodiment of the invention does not limit the acquisition method of sample device images. The annotation process of the desired recognition results can be achieved by manually annotating the sample device images, or by other methods. This embodiment does not limit this.

[0051] Optionally, determining the sample device image based on the historical device image includes: performing image augmentation processing on the historical device image according to a preset image augmentation method to obtain an augmented device image, and using the augmented device image and the historical device image as the sample device image.

[0052] The preset image augmentation method can be a pre-defined image processing method used to augment the image data. For example, the preset image augmentation method includes at least one of geometric transformation, color space change, and pixel relationship adjustment.

[0053] Optionally, the geometric transformation includes at least one of rotation transformation, scaling transformation, clipping transformation, translation transformation, and affine transformation.

[0054] In geometry, an affine transformation can be described as a linear transformation of a vector space followed by a translation, transforming it into another vector space.

[0055] Optionally, color space variations include at least one of contrast variations, brightness variations, saturation variations, histogram enhancement, and grayscale adjustments.

[0056] Histogram enhancement can be a processing method that enhances an image by adjusting its histogram. For example, histogram enhancement may include histogram equalization and histogram matching.

[0057] Optionally, pixel relationship adjustment includes at least one of blurring, sharpening, and noise reduction.

[0058] Generally, when determining the sample device images used for model training, in order to expand the training dataset, improve its comprehensiveness, and enhance the generalization of the device defect recognition model, historical device images can be augmented according to a preset image augmentation method to obtain augmented device images. Then, the augmented device images and historical device images can be used as sample device images. Furthermore, the sample device images are labeled to obtain the expected recognition results corresponding to the sample device images. Based on the sample device images and the expected recognition results corresponding to the sample device images, a training dataset for model training is constructed.

[0059] This invention proposes a method for constructing a training dataset. By augmenting the acquired sample device images through image augmentation, the comprehensiveness of the training dataset is improved, while the generalization of the neural network model is enhanced.

[0060] In practice, sample device images are input into the device defect recognition model to be trained. The feature map extraction module extracts feature information from the training image dataset to obtain a preliminary feature image dataset. The preliminary feature image dataset is then input into the parallel-connected convolutional module and fully connected module to obtain the convolutional output and the fully connected output, respectively. The convolutional output and the fully connected output are fused to obtain the model output corresponding to the sample device images. Then, the model parameters are adjusted based on the model output corresponding to the sample device images, the expected output, and the model loss function until the training termination condition is met, resulting in a trained device defect recognition model.

[0061] Optionally, the target loss function of the equipment defect identification model can be determined based on the loss values ​​of the fully connected modules and the convolutional modules of the equipment defect identification model.

[0062] Optionally, the target loss function of the equipment defect identification model can be determined by weighted calculation of the loss values ​​of the convolutional module and the fully connected module.

[0063] Specifically, the objective loss function of the equipment defect identification model is determined based on the loss values ​​of the fully connected modules and the convolutional modules of the model, as follows:

[0064]

[0065] Where L represents the objective loss function of the equipment defect identification model, L fc L represents the loss value of the fully connected module. conv This represents the loss value of the convolutional module, where ω represents the weight factor. fc For |ln(1-L) fc )|,ω conv for This indicates the parameter to be corrected.

[0066] The correction parameter can be a value used to correct the target loss value of the equipment defect identification model. It can be a constant set based on experience, and its value is not specifically limited here.

[0067] It should be noted that the training termination condition of the equipment defect identification model can be that the target loss function calculated based on the above formula during the training process tends to converge, or that the identification accuracy of the training results output by the equipment defect identification model reaches the preset accuracy, etc. This embodiment does not limit this.

[0068] In practice, the target equipment image is input into a pre-trained equipment defect recognition model. The feature map extraction module extracts feature information from the target equipment image to obtain a target equipment feature image. The target equipment feature image is then input into a parallel-connected convolutional module and a fully connected module. The convolutional module extracts defect feature information from the target equipment feature image to determine the location information and / or defect type of the defect region in the target equipment feature image. Simultaneously, the fully connected module classifies the defect type of the defect region in the target equipment feature image, and finally outputs the identification result of substation equipment defects.

[0069] The technical solution of this invention acquires the target equipment image of the substation equipment to be inspected and inputs the target equipment image into a pre-trained equipment defect recognition model to obtain the identification result of the substation equipment defect. The equipment defect recognition model adopts parallel-connected convolutional modules and fully connected modules. Compared with the serial connection model structure, it can effectively improve the robustness of the equipment defect recognition model, thereby improving the recognition accuracy. It also solves the problems of large workload, low detection efficiency and inaccuracy caused by manual inspection of substation equipment defects in the prior art. It achieves the technical effect of accurately identifying substation equipment defects while reducing labor costs.

[0070] Example 2

[0071] Figure 2 This is a flowchart illustrating a method for identifying substation equipment defects according to Embodiment 2 of the present invention. Based on the above technical solution, this embodiment further refines the technical solution. Optionally, based on any optional technical solution in the embodiments of the present invention, the step of inputting the target equipment image into a pre-trained equipment defect identification model to obtain the substation equipment defect identification result includes: inputting the target equipment image into the pre-trained equipment defect identification model to obtain the convolution output result output by the convolution module and the fully connected output result output by the fully connected module; and determining the substation equipment defect identification result based on the fully connected output result and the convolution output result.

[0072] Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0073] like Figure 2 As shown, the method in this embodiment specifically includes the following steps:

[0074] S210: Receive the target equipment image of the substation equipment to be inspected, collected by the target inspection equipment.

[0075] The target inspection equipment can be understood as a field inspection device used to conduct on-site inspections of substation equipment and to acquire images of the equipment. For example, the target inspection equipment may include, but is not limited to, image acquisition devices installed at fixed locations or substation inspection robots equipped with cameras; this embodiment does not limit this to a specific type.

[0076] In practice, the target inspection equipment can be used to inspect the substation equipment to be inspected, and during the inspection, the equipment status images of the substation equipment can be collected. In this way, the collected equipment status images can be used as target equipment images of the substation equipment to be inspected, and the defects of the substation equipment can be identified.

[0077] It should be noted that the target equipment images collected by the target inspection equipment during the inspection process can be used to identify and analyze substation equipment defects at the same time as they are collected. Alternatively, the collected target equipment images can be pre-stored in an image database, and then used for substation equipment defect identification and analysis after the inspection process is completed or during the substation equipment defect identification stage. This embodiment does not limit this.

[0078] S220. Input the target device image into the pre-trained device defect recognition model to obtain the convolution output result of the convolution module and the fully connected output result of the fully connected module.

[0079] In this embodiment, optionally, the convolution output result may include the convolution classification result of substation equipment defects, and may also include the convolution recognition result of the defect region of the substation equipment. The convolution classification result can be understood as the classification result obtained after applying the convolution module to classify the substation equipment defect types in the target equipment image. The convolution recognition result can be understood as the recognition result of applying the convolution module to perform feature recognition on the defect region of the substation equipment defects in the target equipment image.

[0080] Optionally, the fully connected output may include the fully connected classification results of substation equipment defects. Here, the fully connected classification results can be understood as the classification results obtained by applying the fully connected module to classify the types of substation equipment defects in the target equipment image.

[0081] In this embodiment, optionally, the target device image is input into a pre-trained device defect recognition model to obtain the convolution output result from the convolution module and the fully connected output result from the fully connected module, respectively. This includes: inputting the target device image into the feature map extraction module of the pre-trained device defect recognition model to obtain a preliminary feature map; inputting the preliminary feature map into the fully connected module to obtain a fully connected output result; and inputting the preliminary feature map into the convolution module to obtain a convolution output result.

[0082] In some embodiments, optionally, the target device image is input into a pre-trained device defect recognition model to obtain the convolutional output result of the convolutional module and the fully connected output result of the fully connected module, respectively. This may further include: setting a first feature map extraction module and a second feature map extraction module before the fully connected module and the convolutional module, respectively; inputting the target device image into the first feature map extraction module to obtain a first preliminary feature map; inputting the first preliminary feature map into the fully connected module to obtain a fully connected output result; inputting the target device image into the second feature map extraction module to obtain a second preliminary feature map; and inputting the second preliminary feature map into the convolutional module to obtain a convolutional output result.

[0083] It should be noted that the terms "first" and "second" mentioned in the above embodiments do not indicate any order, quantity, or importance; they simply refer to different target subjects. For example, "first feature map extraction module" and "second feature map extraction module" can refer to the same feature map extraction module or different feature map extraction modules. Similarly, "first preliminary feature map" and "second preliminary feature map" can refer to the same preliminary feature map or different preliminary feature maps.

[0084] S230. Determine the identification result of substation equipment defects based on the fully connected output result and the convolution output result.

[0085] In this embodiment, optionally, the identification result of substation equipment defects may include the target classification result of the substation equipment defects. The target classification result can be understood as the result obtained by the equipment defect identification model after classifying the defect types in the target equipment image.

[0086] It should be noted that, since the fully connected output includes the fully connected classification result of substation equipment defects, and the convolutional output includes the convolutional classification result of substation equipment defects, in some embodiments, the target classification result of substation equipment defects can be the fully connected classification result, the convolutional classification result, or the classification result obtained by fusing the fully connected classification result and the convolutional classification result, etc.

[0087] In this embodiment, optionally, the identification result of substation equipment defects is determined based on the fully connected output result and the convolutional output result, including: fusing the fully connected classification result and the convolutional classification result to obtain the target classification result of substation equipment defects.

[0088] It should also be noted that during the fusion of the fully connected classification results and the convolutional classification results, since the fully connected module can effectively classify and analyze the types of defects in substation equipment, the convolutional classification results can be corrected based on the fully connected classification results to obtain more accurate target classification results.

[0089] Optionally, the fully connected classification results and the convolutional classification results are fused to obtain the target classification result for the defects of the substation equipment, including: fusing the fully connected classification results and the convolutional classification results based on the following formula to obtain the target classification result for the defects of the substation equipment:

[0090]

[0091] Where, p i This represents the defect classification results of the defect area of ​​substation equipment output by the equipment defect identification model. This represents the defect classification results of the convolutional module in the equipment defect identification model. This represents the defect classification results of the fully connected module of the equipment defect identification model.

[0092] This invention proposes a device defect identification method based on a neural network model. It employs parallel-connected convolutional and fully connected modules, which effectively improves the robustness of the device defect identification model compared to a serially connected model structure. Furthermore, by fusing the output results of the fully connected modules with the output results of the convolutional modules, the accuracy of the output results can be effectively improved. This achieves accurate and efficient substation defect identification, solving the problems of high workload, low efficiency, and inaccuracy caused by manual inspection of substation equipment defects in existing technologies.

[0093] In this embodiment, optionally, the identification result of substation equipment defects may further include the target identification result of the defect area of ​​the substation equipment. The target identification result can be understood as the identification result of the equipment defect identification model of the substation equipment defect area in the target equipment image. The target identification result can characterize the approximate location of the substation equipment defect area in the target equipment image.

[0094] Optionally, the identification result of substation equipment defects is determined based on the fully connected output result and the convolution output result, including: using the convolution identification result of the defect area of ​​the substation equipment as the target identification result of the defect area of ​​the substation equipment.

[0095] In practice, the acquired target equipment image is input into a pre-trained equipment defect recognition model. The fully connected output of the fully connected module and the convolutional output of the convolutional module are obtained respectively. The weighted result of the fully connected output and the convolutional output is then fused to obtain the identification result of the substation equipment defect. Compared with the traditional serial connection of the convolutional module and the fully connected module, the parallel connection of the two neural network modules and the weighted fusion of the two outputs can yield a more accurate recognition result.

[0096] It should be noted that, in this embodiment, the equipment defect identification model can be packaged into a whole container, encapsulated as a dynamic link library, compiled into an executable file, burned into firmware, and integrated into a front-end chip, and can be connected to the back-end system of the target inspection equipment and the camera device in the substation to achieve synchronous analysis of the acquired video and real-time identification of substation equipment defects.

[0097] The technical solution of this invention receives target equipment images of substation equipment to be inspected from target inspection equipment, and inputs the target equipment images into a pre-trained equipment defect recognition model. This yields fully connected output results from a fully connected module and convolutional output results from a convolutional module. Furthermore, the fully connected output results and the convolutional output results are fused to obtain the final identification result of the substation equipment defect. This solves the problem in existing technologies where manual analysis of substation equipment defects leads to untimely defect detection, thus affecting the safe operation of the power grid. Moreover, by fusing the fully connected output results and the convolutional output results, the accuracy of the output results can be effectively improved, thereby achieving accurate and efficient substation defect identification.

[0098] Example 3

[0099] Figure 3 This is a schematic diagram of a substation equipment defect identification device provided in Embodiment 3 of the present invention. The substation equipment defect identification device provided in this embodiment can be implemented by software and / or hardware, and can be configured in a terminal and / or server to implement the substation equipment defect identification method in this embodiment of the present invention. Specifically, the device may include: an equipment image acquisition module 310 and an equipment image input module 320.

[0100] Among them, the equipment image acquisition module 310 is used to acquire the target equipment image of the substation equipment to be inspected;

[0101] The equipment image input module 320 is used to input the target equipment image into a pre-trained equipment defect recognition model to obtain the recognition result of substation equipment defects. The equipment defect recognition model includes a parallel convolutional module and a fully connected module. The convolutional module includes at least one convolutional layer and at least one fully connected layer connected in series with the convolutional layer. The fully connected module includes a fully connected module composed of at least one fully connected layer.

[0102] The technical solution of this invention acquires the target equipment image of the substation equipment to be inspected and inputs the target equipment image into a pre-trained equipment defect recognition model to obtain the identification result of the substation equipment defect. The equipment defect recognition model adopts parallel-connected convolutional modules and fully connected modules. Compared with the serial connection model structure, it can effectively improve the robustness of the equipment defect recognition model, thereby improving the recognition accuracy. It also solves the problems of large workload, low detection efficiency and inaccuracy caused by manual inspection of substation equipment defects in the prior art. It achieves the technical effect of accurately identifying substation equipment defects while reducing labor costs.

[0103] Optionally, the device image input module 320 may further include a device image input unit and a recognition result determination unit.

[0104] The equipment image input unit is used to input the target equipment image into the pre-trained equipment defect recognition model to obtain the convolution output result of the convolution module and the fully connected output result of the fully connected module, respectively; the recognition result determination unit is used to determine the recognition result of the substation equipment defect based on the fully connected output result and the convolution output result.

[0105] Optionally, the equipment defect identification model further includes a feature map extraction module, which includes at least one convolutional layer;

[0106] Correspondingly, the device image input unit is also used to input the target device image into the feature map extraction module of the pre-trained device defect recognition model to obtain a preliminary feature map; input the preliminary feature map into the fully connected module to obtain a fully connected output result; and input the preliminary feature map into the convolution module to obtain a convolution output result.

[0107] Optionally, the fully connected output result includes the fully connected classification result of the defects of the substation equipment, and the convolutional output result includes the convolutional classification result of the defects of the substation equipment; the identification result of the substation equipment defects includes the target classification result of the defects of the substation equipment.

[0108] Correspondingly, the identification result determination unit is also used to fuse the fully connected classification results and the convolutional classification results to obtain the target classification result of the defects of the substation equipment.

[0109] Optionally, the identification result determination unit is also used to fuse the fully connected classification result and the convolutional classification result based on the following formula to obtain the target classification result of the defects of the substation equipment:

[0110]

[0111] Where, p i This represents the defect classification results of the defect area of ​​substation equipment output by the equipment defect identification model. This represents the defect classification results of the convolutional module in the equipment defect identification model. This represents the defect classification results of the fully connected module of the equipment defect identification model.

[0112] Optionally, the convolution output result may also include the convolution recognition result of the defect region of the substation equipment; the defect recognition result of the substation equipment may include the target recognition result of the defect region of the substation equipment.

[0113] Similarly, the identification result determination unit is also used to use the convolutional identification result of the defect area of ​​the substation equipment as the target identification result of the defect area of ​​the substation equipment.

[0114] Optionally, the target loss function of the equipment defect identification model is determined based on the loss value of the fully connected module and the loss value of the convolutional module of the equipment defect identification model.

[0115] Optionally, the target loss function of the equipment defect identification model is determined based on the loss values ​​of the fully connected module and the convolutional module of the equipment defect identification model in the following specific way:

[0116]

[0117] Where L represents the objective loss function of the equipment defect identification model, L fc L represents the loss value of the fully connected module. conv This represents the loss value of the convolutional module, where ω represents the weight factor. fc For |ln(1-L) fc )|,ω conv for This indicates the parameter to be corrected.

[0118] Optionally, the equipment image acquisition module 310 is also used to receive target equipment images of the substation equipment to be inspected collected by the target inspection equipment.

[0119] Optionally, the device further includes a dataset construction module, an image input module, and a model adjustment module.

[0120] The dataset construction module is used to construct a training dataset for training the equipment defect identification model. The training dataset includes sample equipment images and expected identification results corresponding to the sample equipment images. The image input module is used to input the sample equipment images into the model to be trained to obtain the model output results corresponding to the sample equipment images. The model adjustment module is used to adjust the model to be trained based on the model output results corresponding to the sample equipment images, the expected identification results, and a pre-constructed model loss function to obtain the equipment defect identification model.

[0121] Optionally, the dataset construction module includes a device image determination unit, an image annotation unit, and a dataset construction unit.

[0122] The equipment image determination unit is used to acquire historical equipment images of substation equipment and determine sample equipment images based on the historical equipment images; the image annotation unit is used to annotate the sample equipment images to obtain the expected recognition results corresponding to the sample equipment images; and the dataset construction unit is used to construct a training dataset for training the equipment defect recognition model based on the sample equipment images and the expected recognition results corresponding to the sample equipment images.

[0123] Optionally, the device image determination unit is further configured to perform image augmentation processing on the historical device image according to a preset image augmentation method to obtain an augmented device image, and use the augmented device image and the historical device image as sample training images; wherein, the preset image augmentation method includes at least one of geometric transformation, color space change and pixel relationship adjustment.

[0124] Optionally, the geometric transformation includes at least one of rotation transformation, scaling transformation, clipping transformation, translation transformation, and affine transformation.

[0125] Optionally, the color space variation includes at least one of contrast variation, brightness variation, saturation variation, histogram enhancement, and grayscale adjustment.

[0126] Optionally, the pixel relationship adjustment includes at least one of blurring, sharpening, and noise reduction.

[0127] The aforementioned substation equipment defect identification device can execute the substation equipment defect identification method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the substation equipment defect identification method.

[0128] Example 4

[0129] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. Figure 5A block diagram is shown of an exemplary electronic device 40 suitable for implementing embodiments of the present invention. Figure 5 The electronic device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0130] like Figure 5 As shown, electronic device 40 is represented in the form of a general-purpose computing device. The components of electronic device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0131] Bus 403 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0132] Electronic device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 40, including volatile and non-volatile media, removable and non-removable media.

[0133] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 404 and / or cache memory 405. Electronic device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 406 may be used to read and write non-removable, non-volatile magnetic media (… Figure 5 Not shown; usually referred to as a "hard drive"). Although Figure 5 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 403 via one or more data media interfaces. Memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0134] A program / utility 408 having a set (at least one) of program modules 407 may be stored, for example, in memory 402. Such program modules 407 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 407 typically perform the functions and / or methods described in the embodiments of the present invention.

[0135] Electronic device 40 can also communicate with one or more external devices 409 (e.g., keyboard, pointing device, display 410, etc.), and with one or more devices that enable a user to interact with the electronic device 40, and / or with any device that enables the electronic device 40 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 411. Furthermore, electronic device 40 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 412. As shown, network adapter 412 communicates with other modules of electronic device 40 via bus 403. It should be understood that, although... Figure 5 Not shown, other hardware and / or software modules may be used in conjunction with electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0136] The processing unit 401 executes various functional applications and data processing by running programs stored in the system memory 402, such as implementing the substation equipment defect identification method provided in the embodiments of the present invention.

[0137] Example 7

[0138] Embodiment 7 of the present invention also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a method for identifying defects in substation equipment. The method includes:

[0139] Acquire images of the target equipment in the substation to be inspected;

[0140] The target equipment image is input into a pre-trained equipment defect recognition model to obtain the identification results of substation equipment defects;

[0141] The equipment defect identification model includes parallel convolutional modules and fully connected modules. The convolutional module includes at least one convolutional layer and at least one fully connected layer connected in series with the convolutional layer. The fully connected module includes a fully connected module composed of at least one fully connected layer.

[0142] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0143] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0144] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0145] Computer program code for performing the operations of embodiments of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0146] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method of identifying substation equipment defects, characterized by, The method comprises: obtaining a target equipment image of a substation equipment to be detected; inputting the target equipment image into a pre-trained equipment defect recognition model to obtain a recognition result of a substation equipment defect; wherein the recognition result of the substation equipment defect comprises a target classification result of the substation equipment defect and a target recognition result of a defect region of the substation equipment; wherein the equipment defect recognition model comprises a convolution module and a full connection module connected in parallel, the convolution module comprises at least one convolution layer and at least one full connection layer connected in series with the convolution layer, and the full connection module comprises a full connection module composed of at least one full connection layer; inputting the target equipment image into the pre-trained equipment defect recognition model to obtain the recognition result of the substation equipment defect, comprising: inputting the target equipment image into the pre-trained equipment defect recognition model to obtain a convolution output result output by the convolution module and a full connection output result output by the full connection module respectively; determining the recognition result of the substation equipment defect according to the full connection output result and the convolution output result; the full connection output result comprises a full connection classification result of the substation equipment defect, and the convolution output result comprises a convolution classification result of the substation equipment defect; the recognition result of the substation equipment defect comprises a target classification result of the substation equipment defect; determining the recognition result of the substation equipment defect according to the full connection output result and the convolution output result, comprising: performing result fusion on the full connection classification result and the convolution classification result to obtain a target classification result of the substation equipment defect.

2. The method of claim 1, wherein, The equipment defect recognition model further comprises a feature map extraction module, and the feature map extraction module comprises at least one convolution layer; inputting the target equipment image into the pre-trained equipment defect recognition model to obtain the convolution output result output by the convolution module and the full connection output result output by the full connection module respectively, comprising: inputting the target equipment image into the feature map extraction module of the pre-trained equipment defect recognition model to obtain a preliminary feature map; inputting the preliminary feature map into the full connection module to obtain the full connection output result, and inputting the preliminary feature map into the convolution module to obtain the convolution output result.

3. The method of claim 1, wherein, performing result fusion on the full connection classification result and the convolution classification result to obtain a target classification result of the substation equipment defect, comprising: performing result fusion on the full connection classification result and the convolution classification result to obtain a target classification result of the substation equipment defect based on the following formula: ； wherein, represents a defect classification result of a defect region of a substation equipment output by the equipment defect recognition model, represents a defect classification result of a convolution module of the equipment defect recognition model, represents a defect classification result of a full connection module of the equipment defect recognition model.

4. The method of claim 1, wherein, the convolution output result further comprises a convolution recognition result of a defect region of the substation equipment; and the recognition result of the substation equipment defect comprises a target recognition result of the defect region of the substation equipment; determining the recognition result of the substation equipment defect according to the full connection output result and the convolution output result, comprising: taking the convolution recognition result of the defect region of the substation equipment as the target recognition result of the defect region of the substation equipment.

5. The method of claim 1, wherein, The target loss function of the equipment defect identification model is determined according to the loss value of the full connection module of the equipment defect identification model and the loss value of the convolution module.

6. The method of claim 5, wherein, The specific manner in which the target loss function of the equipment defect identification model is determined according to the loss value of the full connection module of the equipment defect identification model and the loss value of the convolution module is as follows: + ； wherein, a target loss function representing a device defect recognition model, a loss value of a full connection module, a loss value of a convolution module, a weight factor, is , is , a correction parameter.

7. The method of claim 1, wherein, The target equipment image of the substation equipment to be detected is obtained, including: Receiving the target equipment image of the substation equipment to be detected collected by the target inspection equipment.

8. The method of claim 1, wherein, Also includes: A training data set for training an equipment defect identification model is constructed, wherein the training data set includes a sample equipment image and an expected identification result corresponding to the sample equipment image; The sample equipment image is input into the to-be-trained model to obtain a model output result corresponding to the sample equipment image; The to-be-trained model is adjusted according to the model output result corresponding to the sample equipment image, the expected identification result, and a pre-constructed model loss function to obtain an equipment defect identification model.

9. The method of claim 8, wherein, The construction of the training data set for training the equipment defect identification model includes: Obtaining a historical equipment image of a substation equipment in a substation, and determining a sample equipment image according to the historical equipment image; Labeling the sample equipment image to obtain an expected identification result corresponding to the sample equipment image; The sample equipment image and the expected identification result corresponding to the sample equipment image are used to train a training data set for an equipment defect identification model.

10. The method of claim 9, wherein, The sample training image is determined according to the historical equipment image, including: The historical equipment image is processed by image expansion according to a preset image expansion method to obtain an expanded equipment image, and the expanded equipment and the historical equipment image are used as sample training images; The preset image expansion method includes at least one of geometric transformation, color space change, and pixel relationship adjustment.

11. The method of claim 10, wherein, The geometric transformation includes at least one of rotation transformation, scaling transformation, cropping transformation, translation transformation, and affine transformation.

12. The method of claim 10, wherein, The color space change includes at least one of contrast change, brightness change, saturation change, histogram enhancement, and gray scale adjustment.

13. The method of claim 10, wherein, The pixel relationship adjustment includes at least one of blur processing, sharpening processing, and noise processing.

14. A device for identifying defects in substation equipment, characterized in that, Including: An equipment image acquisition module is configured to obtain a target equipment image of a substation equipment to be detected; An equipment image input module is configured to input the target equipment image into a pre-trained equipment defect identification model to obtain an identification result of a substation equipment defect, wherein the identification result of the substation equipment defect includes a target classification result of the defect of the substation equipment and / or a target identification result of the defect region of the substation equipment; The equipment defect identification model includes a convolution module and a full connection module in parallel, the convolution module includes at least one convolution layer and at least one full connection layer connected in series with the convolution layer, and the full connection module includes a full connection module composed of at least one full connection layer; The equipment image input module further includes an equipment image input unit and an identification result determination unit; The device image input unit is configured to input a target device image into a pre-trained device defect recognition model to obtain a convolution output result output by a convolution module and a full connection output result output by a full connection module respectively. The full connection output result includes a full connection classification result of the defect of the substation device, and the convolution output result includes a convolution classification result of the defect of the substation device. The recognition result determination unit is further configured to perform result fusion on the full connection classification result and the convolution classification result to obtain a target classification result of the defect of the substation device.

15. An electronic device, comprising: The electronic device includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the substation device defect recognition method as claimed in any one of claims 1-13.

16. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the substation device defect recognition method as claimed in any one of claims 1-13.

Images