L4 level electric power intelligent outdoor inspection unmanned vehicle inspection method and system

Abstract

The present application discloses a patrol method and system based on an L4-level electric power intelligent outdoor patrol unmanned vehicle, which relates to the field of intelligent patrol. It obtains the target meter image collected by the camera of the electric power intelligent outdoor patrol unmanned vehicle, and uses an artificial intelligence-based image analysis and processing algorithm to perform feature analysis on the target meter image, thereby automatically evaluating whether there is defocus based on the spatial shallow feature information of the target meter image, and in response to the presence of defocus, adjusting the shooting angle and shooting distance of the camera until there is no defocus. In this way, automated patrol inspection of target meters can be achieved, reducing manual workload and costs, improving patrol efficiency, reducing misjudgments and missed judgments, and improving the accuracy of evaluation to ensure the reliability of patrol results.

Description

Technical Field

[0001] The present application relates to the field of intelligent inspection, and more specifically, to an inspection method and system for an unmanned vehicle based on L4-level electric power intelligent outdoor inspection. Background Art

[0002] The L4-level electric power intelligent outdoor inspection unmanned vehicle is a highly automated robotic vehicle specially designed to perform automatic inspection tasks in the outdoor environment of power facilities. The application of the L4-level electric power intelligent outdoor inspection unmanned vehicle can significantly improve the efficiency and safety of power facility inspections, reduce the risks and costs of manual inspections, and improve the reliability of power systems through real-time monitoring.

[0003] The inspection unmanned vehicle can obtain the usage data of the power system by taking images of the target meters and monitor the operation of the power equipment, such as power consumption, voltage, current, etc. These data are crucial for billing, energy consumption analysis and management decisions, and provide important data support and security for the power system.

[0004] However, due to environmental factors, lighting conditions, etc., the target meter images captured by traditional methods may be blurred, have insufficient light, poor angles, etc., affecting the clarity and accuracy of the image. In addition, for special cases such as the meter being blocked, damaged, contaminated, the dial being blurred, or the dial being tilted, the images captured by traditional inspection unmanned vehicles may not be accurately recognized and recorded, and manual intervention is required to correct the image angle, adjust the focal length, or handle abnormal conditions in the image, which is time-consuming, labor-intensive, and inefficient.

[0005] Therefore, an optimized inspection method based on L4 level power intelligent outdoor inspection unmanned vehicle is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an inspection method and system based on a L4-level electric power intelligent outdoor inspection unmanned vehicle, which obtains the target meter image collected by the camera of the electric power intelligent outdoor inspection unmanned vehicle, and uses an artificial intelligence-based image analysis and processing algorithm to perform feature analysis on the target meter image, thereby automatically evaluating whether there is defocus based on the spatial shallow feature information of the target meter image, and in response to the existence of defocus, adjusting the shooting angle and shooting distance of the camera until there is no defocus. In this way, automated inspection of target meters can be achieved, reducing manual workload and costs, improving inspection efficiency, while reducing misjudgments and missed judgments, and improving the accuracy of evaluation to ensure the reliability of inspection results.

[0007] According to one aspect of the present application, a method for inspection by an unmanned vehicle based on an L4 level electric power intelligent outdoor inspection is provided, which includes:

[0008] S1: Obtaining the target meter image collected by the camera of the power intelligent outdoor inspection unmanned vehicle;

[0009] S2: Calculate the gradient value of each pixel in the target meter image to obtain a target meter gradient value statistical matrix;

[0010] S3: extracting shallow features of the target meter gradient value statistical matrix to obtain a shallow feature map of the target meter image;

[0011] S4: performing spatial attention enhancement on the shallow feature map of the target meter image to obtain a shallow feature map of the spatially manifested target meter image as a shallow feature of the spatially manifested target meter image;

[0012] S5: obtaining an evaluation result based on the shallow features of the spatially manifested target meter image, and adjusting the shooting angle and shooting distance of the camera based on the evaluation result;

[0013] S6: Repeat S1 to S5 until the evaluation result shows that there is no defocus.

[0014] According to another aspect of the present application, there is provided an unmanned vehicle inspection system based on L4 level electric power intelligent outdoor inspection, which includes:

[0015] The target meter image acquisition module is used to obtain the target meter image collected by the camera of the power intelligent outdoor inspection unmanned vehicle;

[0016] A gradient value calculation module, used to calculate the gradient value of each pixel in the target meter image to obtain a target meter gradient value statistical matrix;

[0017] An image shallow feature extraction module, used for performing image shallow feature extraction on the target meter gradient value statistical matrix to obtain a target meter image shallow feature map;

[0018] A spatial attention enhancement module, used for performing spatial attention enhancement on the shallow feature map of the target meter image to obtain the shallow feature map of the spatially manifested target meter image as the shallow feature of the spatially manifested target meter image;

[0019] The evaluation result generating module is used to obtain the evaluation result based on the shallow features of the spatially manifested target meter image, and adjust the shooting angle and shooting distance of the camera based on the evaluation result.

[0020] Compared with the prior art, the present application provides a method and system for inspection based on an L4-level electric power intelligent outdoor inspection unmanned vehicle, which obtains the target meter image collected by the camera of the electric power intelligent outdoor inspection unmanned vehicle, and uses an artificial intelligence-based image analysis and processing algorithm to perform feature analysis on the target meter image, thereby automatically evaluating whether there is defocus based on the spatial shallow feature information of the target meter image, and in response to the presence of defocus, adjusting the shooting angle and shooting distance of the camera until there is no defocus. In this way, automated inspection of target meters can be achieved, reducing manual workload and costs, improving inspection efficiency, reducing misjudgments and missed judgments, and improving the accuracy of evaluation to ensure the reliability of inspection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 This is a flow chart of the unmanned vehicle inspection method based on L4 level electric power intelligent outdoor inspection according to an embodiment of the present application.

[0023] Figure 2 It is a schematic diagram of the architecture of the unmanned vehicle inspection method based on L4 level electric power intelligent outdoor inspection according to an embodiment of the present application.

[0024] Figure 3 This is a flowchart of obtaining the shallow feature map of the spatially manifested target meter image by passing the shallow feature map of the target meter image through a multi-scale pan-spatial attention module in the L4-level electric power intelligent outdoor inspection unmanned vehicle inspection method according to an embodiment of the present application.

[0025] Figure 4 The present invention is a flowchart for training the shallow feature extractor of the image based on the hole convolutional neural network model, the multi-scale pan-spatial attention module and the classifier-based image quality assessor in the inspection method based on the L4-level electric power intelligent outdoor inspection unmanned vehicle according to an embodiment of the present application.

[0026] Figure 5 It is a block diagram of an unmanned vehicle inspection system based on L4-level electric power intelligent outdoor inspection according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0028] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0029] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0030] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0031] The L4-level electric power intelligent outdoor inspection unmanned vehicle is a highly automated robotic vehicle specially designed and used to perform automatic inspection tasks in the outdoor environment of power facilities. The L4 level refers to the ability of the vehicle to achieve fully automatic driving under certain conditions without driver intervention. The L4-level electric power intelligent outdoor inspection unmanned vehicle is a reflection of the automation and intelligent trend of the power industry. The application of the L4-level electric power intelligent outdoor inspection unmanned vehicle can significantly improve the efficiency and safety of power facility inspections, reduce the risks and costs of manual inspections, and improve the reliability of the power system through real-time monitoring.

[0032] The inspection unmanned vehicle can obtain the usage data of the power system by taking images of the target meters and monitor the operation of the power equipment, such as power consumption, voltage, current, etc. These data are crucial for billing, energy consumption analysis and management decisions, and provide important data support and security for the power system.

[0033] However, due to environmental factors, lighting conditions, etc., the target meter images captured by traditional methods may be blurred, have insufficient light, poor angles, etc., affecting the clarity and accuracy of the image. In addition, for special cases such as the meter being blocked, damaged, contaminated, the dial being blurred, or the dial being tilted, the images captured by traditional inspection unmanned vehicles may not be accurately recognized and recorded, and manual intervention is required to correct the image angle, adjust the focal length, or handle abnormal conditions in the image, which is time-consuming, labor-intensive, and inefficient.

[0034] Therefore, in response to the above technical problems, the technical concept of this application is to obtain the target meter image collected by the camera of the electric power intelligent outdoor inspection unmanned vehicle, and use the image analysis and processing algorithm based on artificial intelligence to perform feature analysis of the target meter image, so as to automatically evaluate whether there is defocus based on the spatial shallow feature information of the target meter image, and in response to the existence of defocus, adjust the shooting angle and shooting distance of the camera until there is no defocus. In this way, the automatic inspection of the target meter can be realized, the manual workload and cost can be reduced, the efficiency of the inspection can be improved, and at the same time, the misjudgment and missed judgment can be reduced, and the accuracy of the evaluation can be improved to ensure the reliability of the inspection results.

[0035] Figure 1 This is a flow chart of the unmanned vehicle inspection method based on L4 level electric power intelligent outdoor inspection according to an embodiment of the present application. Figure 2 FIG. 1 is a schematic diagram of the architecture of an unmanned vehicle inspection method based on an L4-level electric power intelligent outdoor inspection according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the inspection method based on the L4-level electric power intelligent outdoor inspection unmanned vehicle includes: S1: obtaining a target meter image collected by a camera of the electric power intelligent outdoor inspection unmanned vehicle; S2: calculating the gradient value of each pixel in the target meter image to obtain a target meter gradient value statistical matrix; S3: performing image shallow feature extraction on the target meter gradient value statistical matrix to obtain a target meter image shallow feature map; S4: performing spatial attention enhancement on the target meter image shallow feature map to obtain a spatially manifested target meter image shallow feature map as a spatially manifested target meter image shallow feature; S5: obtaining an evaluation result based on the spatially manifested target meter image shallow feature, and adjusting the shooting angle and shooting distance of the camera based on the evaluation result; and, S6: repeating S1 to S5 until the evaluation result is that there is no defocus.

[0036] In step S1, the target meter image captured by the camera of the electric power intelligent outdoor inspection unmanned vehicle is obtained. It should be understood that the target meter image refers to the image of the instrument or equipment used to record and display key parameters such as power usage, voltage, current, etc. in the power system. These instruments usually include electric meters, voltmeters, ammeters, etc., which are crucial for monitoring the operating status, billing, maintenance and fault diagnosis of the power system. Based on this, in the technical solution of the present application, the target meter image captured by the camera of the electric power intelligent outdoor inspection unmanned vehicle is obtained, and it is analyzed and processed. It can accurately determine whether the captured target meter image has a defocus problem, which improves the inspection efficiency of the unmanned inspection vehicle and reduces the cost and time of manual intervention.

[0037] In step S2, the gradient value of each pixel in the target meter image is calculated to obtain a target meter gradient value statistical matrix. Accordingly, considering that the target meter image contains important key features, edge information and texture information in this task, and considering that in image processing and computer vision, the gradient refers to the rate of change or slope of the pixel value in the image, the gradient value represents the degree of grayscale change of each pixel in the image, which can be used to detect edge information, texture information or other significant change areas in the image. Based on this, in order to accurately represent the position and size of the edge in the target meter image, which is helpful for the subsequent defocus assessment, in the technical solution of the present application, the gradient value of each pixel in the target meter image is calculated to obtain a target meter gradient value statistical matrix. It should be understood that the edge is where the grayscale changes rapidly in the image, and the change in the gradient value can usually accurately represent the position and intensity of these edges. By calculating the gradient value of each pixel point in the target meter image, it is possible to identify whether the target is within the available range. At the same time, the gradient value can help extract key instrument feature information and texture information in the target meter image, and help distinguish different areas in the target meter image, so as to obtain a target meter gradient value statistical matrix containing key features of the target meter, thereby providing assistance and support for subsequent out-of-focus evaluation.

[0038] In step S3, the target meter gradient value statistical matrix is ​​subjected to image shallow feature extraction to obtain a target meter image shallow feature map. Specifically, in an embodiment of the present application, the target meter gradient value statistical matrix is ​​subjected to image shallow feature extraction to obtain a target meter image shallow feature map, including: passing the target meter gradient value statistical matrix through an image shallow feature extractor based on a deep neural network model to obtain the target meter image shallow feature map. In particular, the image shallow feature extractor based on a deep neural network model described herein is an image shallow feature extractor based on a hole convolutional neural network model. It should be understood that, considering that the target meter gradient value statistical matrix contains shallow implicit feature information of different scales about the target meter, such as the outline of the meter, the edge of the pointer, the number or the letter; the smoothness of the meter surface, the reflective area or the wear mark, etc., these shallow feature information of the target meter plays a great role in subsequent processing, and the convolutional neural network is a deep learning model specially used for image implicit feature processing and analysis, with powerful feature extraction and capture capabilities, but the receptive field of the traditional convolutional neural network is subject to certain limitations, and will lead to an increase in the number of parameters and the amount of calculation. Therefore, in order to expand the receptive field without increasing the parameters and the amount of calculation, in the technical solution of the present application, the target meter gradient value statistical matrix is ​​passed through the image shallow feature extractor based on the hole convolutional neural network model to obtain the shallow feature map of the target meter image. It is worth mentioning that the hole convolutional neural network can expand the receptive field of the convolution kernel by introducing the hole parameter, and can flexibly adjust the hole parameter to adapt to the feature extraction requirements of different scales, so as to better capture the feature information of different scales in the target meter gradient value statistical matrix.

[0039] In step S4, the shallow feature map of the target meter image is subjected to spatial attention enhancement to obtain a shallow feature map of the spatially explicit target meter image as a shallow feature of the spatially explicit target meter image. Specifically, in an embodiment of the present application, the shallow feature map of the target meter image is subjected to spatial attention enhancement to obtain a shallow feature map of the spatially explicit target meter image, including: passing the shallow feature map of the target meter image through a multi-scale pan-spatial attention module to obtain the shallow feature map of the spatially explicit target meter image. Accordingly, considering that the local spatial dimension in the shallow feature map of the target meter image has feature information of different scales about the shallow features of the target meter image, that is, the shallow feature map of the target meter image contains shallow feature information of different spatial scales about the target meter image. Therefore, in order to mine and extract the shallow semantic information of multi-scale target meter images at different spatial scales from the shallow feature map of the target meter image, and provide a basis and support for the subsequent defocus judgment, in the technical solution of the present application, the shallow feature map of the target meter image is passed through a multi-scale pan-spatial attention module to obtain a shallow feature map of the spatially manifested target meter image. In other words, the shallow feature expression information of the target meter image with different local spatial degrees in the shallow feature map of the target meter image, such as the smooth, rough or texture with a specific direction in the target meter image, or the overall or local shape of an object. The multi-scale pan-spatial attention module introduces average pooling, maximum pooling and random pooling to extract the statistical information, the most significant feature points and the randomly distributed feature points about the shallow layer of the target meter image in the shallow feature map of the target meter image. The features obtained after the pooling process are merged to form a comprehensive pooled feature representation, and then the merged pooled features are convolved and activated to further refine and extract the useful information in the shallow features of the target meter image and output the spatial attention weights. These weights reflect the importance of the target meter image in different spatial regions in the shallow feature map of the target meter image. Finally, the spatial attention weights are weighted and summed with the shallow feature map of the target meter image to obtain the shallow feature map of the spatially explicit target meter image. In this way, the feature information at different spatial scales of the shallow feature map of the target meter image can be considered at the same time, which helps the model to more comprehensively understand the semantic content in the shallow feature map of the target meter image, so as to capture the important features of different spaces in the shallow feature map of the target meter image, thereby improving the richness of the shallow features of the spatially explicit target meter image.

[0040] Figure 3 The present invention is a flowchart of obtaining the shallow feature map of the spatially manifested target meter image by passing the shallow feature map of the target meter image through a multi-scale pan-spatial attention module in the L4-level electric power intelligent outdoor inspection unmanned vehicle inspection method according to the embodiment of the present application. More specifically, in the embodiment of the present application, Figure 3 As shown, the shallow feature map of the target meter image is passed through a multi-scale pan-spatial attention module to obtain the shallow feature map of the spatially manifested target meter image, including: S110, performing global mean pooling processing along the channel dimension on the shallow feature map of the target meter image to obtain a global mean pooling feature matrix of the target meter image; S120, performing maximum value pooling processing along the channel dimension on the shallow feature map of the target meter image to obtain a maximum value pooling feature matrix of the target meter image; S130, performing random pooling processing along the channel dimension on the shallow feature map of the target meter image to obtain a random pooling feature matrix of the target meter image; S140, performing global mean pooling processing on the feature map of the target meter image The matrix, the maximum pooling feature matrix of the target meter image and the random pooling feature matrix of the target meter image are spliced ​​to obtain a spliced ​​target meter image feature matrix; S150, the spliced ​​target meter image feature matrix is ​​convolved to obtain a target meter image implicit correlation feature matrix; S160, the target meter image implicit correlation feature matrix is ​​input into a Sigmoid activation function to obtain a target meter image spatial attention weight matrix; and, S170, each feature matrix along the channel dimension in the shallow feature map of the target meter image is multiplied by the spatial attention weight matrix of the target meter image to obtain the shallow feature map of the spatially manifested target meter image.

[0041] Preferably, in an embodiment of the present application, the shallow feature map of the target meter image is passed through a multi-scale pan-spatial attention module to obtain the shallow feature map of the spatially manifested target meter image, comprising: using the multi-scale pan-spatial attention module to process the shallow feature map of the target meter image with the following spatial attention formula to obtain the shallow feature map of the spatially manifested target meter image; wherein the spatial attention formula is:

[0042] M s (F) = σ(f N×N ([(AvgPool(F));MaxPool(F);StoPool(F)]))

[0043]

[0044] Wherein, F represents the shallow feature map of the target meter image, f N×N represents a convolution kernel of size N×N, AvgPool(·), MaxPool(·) and StoPool(·) represent global mean pooling, maximum pooling and random pooling of the feature map along the channel dimension, respectively, σ(·) represents the Sigmoid activation function, [·; ·; ·] represents the concatenation operation, M s(F) is the spatial attention weight matrix of the shallow feature map of the target meter image, represents the Hadamard product, F s Represents a shallow feature map of the spatially manifested target meter image.

[0045] In step S5, based on the shallow features of the spatially explicit target meter image, an evaluation result is obtained, and based on the evaluation result, the shooting angle and shooting distance of the camera are adjusted. Specifically, in the embodiment of the present application, based on the shallow features of the spatially explicit target meter image, an evaluation result is obtained, and based on the evaluation result, the shooting angle and shooting distance of the camera are adjusted, including: passing the shallow feature map of the spatially explicit target meter image through a classifier-based image quality evaluator to obtain an evaluation result, and the evaluation result is whether there is defocus; and, in response to the evaluation result that there is defocus, adjusting the shooting angle and shooting distance of the camera. That is, the shallow features of the spatially explicit target meter image obtained by spatial attention saliency of the shallow feature map of the target meter image are used for classification processing, so as to automatically evaluate whether there is defocus according to the spatial shallow feature information of the target meter image, and in response to the existence of defocus, adjust the shooting angle and shooting distance of the camera until there is no defocus. In this way, automated inspection of target meters can be achieved, reducing manual workload and costs, improving inspection efficiency, while reducing misjudgments and missed judgments, and improving assessment accuracy to ensure the reliability of inspection results.

[0046] It is worth mentioning that those skilled in the art should be aware that before applying a deep neural network model for inference, the deep neural network model must be trained so that the deep neural network can implement specific functional capabilities.

[0047] Specifically, in an embodiment of the present application, it also includes a method for training the image shallow feature extractor based on the hole convolutional neural network model, the multi-scale pan-spatial attention module and the classifier-based image quality assessor.

[0048] Figure 4 The flowchart is a method for training the shallow feature extractor of the image based on the hole convolutional neural network model, the multi-scale pan-spatial attention module and the image quality assessor based on the classifier in the L4-level electric power intelligent outdoor inspection unmanned vehicle inspection method according to the embodiment of the present application. Figure 4As shown, the training step includes: S210, obtaining training data, the training data including the training target meter image collected by the camera of the electric power intelligent outdoor inspection unmanned vehicle, and the true value of whether there is out-of-focus; S220, calculating the gradient value of each pixel point in the training target meter image to obtain the training target meter gradient value statistical matrix; S230, passing the training target meter gradient value statistical matrix through the image shallow feature extractor based on the void convolutional neural network model to obtain the training target meter image shallow feature map; S240, passing the training target meter image shallow feature map through the multi-scale pan-space attention module to obtain the training space explicit target meter image shallow feature map. Figure; S250, optimizing the shallow feature map of the training space manifested target meter image to obtain an optimized shallow feature map of the training space manifested target meter image; S260, passing the optimized shallow feature map of the training space manifested target meter image through the classifier-based image quality assessor to obtain a classification loss function value; and, S270, based on the classification loss function value and through back propagation of gradient descent, training the shallow feature extractor of the image based on the void convolutional neural network model, the multi-scale pan-spatial attention module and the classifier-based image quality assessor, wherein in each iteration of the model, the shallow feature map of the training space manifested target meter image is optimized.

[0049] Specifically, in step S270, the shallow feature extractor of the image based on the hole convolutional neural network model, the multi-scale pan-space attention module and the image quality assessor based on the classifier are trained based on the classification loss function value and through the back propagation of gradient descent, wherein, in each iteration of the model, the shallow feature map of the training spatial explicit target meter image is optimized. It should be understood that in the technical solution of the present application, the shallow feature map of the training target meter image represents the shallow semantic features of the image based on the convolution kernel filter of the training target meter gradient value statistical matrix of the training target meter image, which generally includes lines, edges, shapes and textures in the image. After the shallow feature map of the training target meter image passes through the multi-scale pan-spatial attention module, a variety of different pooling methods are actually used to capture the different significant expressions of the shallow feature map of the training target meter image in the local spatial dimension to perform multi-scale local feature space distribution enhancement on the shallow feature map of the training target meter image. In this way, the shallow feature map of the training space explicit target meter image obtained has a complex image semantic representation dimension based on the multi-dimensional spatial significant representation of image semantic features. Therefore, it is expected to improve the feature expression effect under the complex image semantic representation dimension during the model iteration process, so as to further improve the accuracy of the evaluation result obtained by passing the shallow feature map of the training space explicit target meter image through the image quality evaluator based on the classifier. Based on this, in the technical solution of the present application, the shallow feature map of the training space explicit target meter image is optimized during each iteration of the model.

[0050] Specifically, in the embodiment of the present application, in each iteration of the model, the shallow feature map of the training space explicit target meter image is optimized, including: in each iteration of the model, the shallow feature map of the training space explicit target meter image is optimized by the following optimization formula to obtain the optimized shallow feature map of the training space explicit target meter image; wherein the correction formula is:

[0051]

[0052] Where B represents the shallow feature map of the target meter image in the training space, μ, σ and f max They are the feature sets f i ∈B, log is the logarithm with base 2, and α is the weight hyperparameter, ⊙ represents positional multiplication, ⊕ represents positional addition, represents subtraction by position, and F′ represents the shallow feature map of the target meter image in the optimized training space.

[0053] Specifically, here, by using an interactive representation containing probability statistical features represented by mean μ and standard value σ and f as a latent variable feature i and f max The short sequence of the distribution interaction representation is used as the sub-manifold latent motifs under the complex manifold network of the shallow feature map F of the training space explicit target meter image, and the latent motif feature information pattern and feature distribution pattern of the shallow feature map F of the training space explicit target meter image are used as its global structure inference unit, so as to reconstruct the complex manifold structure of the shallow feature map F of the training space explicit target meter image based on the connection in the form of the global structure latent motif dictionary, so as to improve the generation and evolution understanding ability of the model for the manifold structure corresponding to the image semantic features of the complex image semantic representation dimension in the iterative process, and improve the feature fusion expression effect under the complex image semantic representation dimension in the model iteration process. In this way, the accuracy of the evaluation result obtained by the shallow feature map of the training space explicit target meter image through the image quality evaluator based on the classifier. In this way, the automatic inspection of the target meter can be realized, the manual workload and cost can be reduced, the efficiency of the inspection can be improved, and at the same time, the misjudgment and missed judgment can be reduced, and the accuracy of the evaluation can be improved to ensure the reliability of the inspection results.

[0054] In summary, the inspection method based on the L4-level electric power intelligent outdoor inspection unmanned vehicle according to the embodiment of the present application is explained, which obtains the target meter image collected by the camera of the electric power intelligent outdoor inspection unmanned vehicle, and uses the image analysis and processing algorithm based on artificial intelligence to perform feature analysis of the target meter image, thereby automatically evaluating whether there is defocus based on the spatial shallow feature information of the target meter image, and in response to the presence of defocus, adjusting the shooting angle and shooting distance of the camera until there is no defocus. In this way, automated inspection of target meters can be achieved, reducing manual workload and costs, improving inspection efficiency, reducing misjudgments and missed judgments, and improving the accuracy of evaluation to ensure the reliability of inspection results.

[0055] Figure 5 FIG. 1 is a block diagram of an unmanned vehicle inspection system based on an L4-level electric power intelligent outdoor inspection according to an embodiment of the present application. Figure 5As shown, according to the embodiment of the present application, the L4-level electric power intelligent outdoor inspection unmanned vehicle inspection system 100 includes: a target meter image acquisition module 110, which is used to obtain the target meter image collected by the camera of the electric power intelligent outdoor inspection unmanned vehicle; a gradient value calculation module 120, which is used to calculate the gradient value of each pixel in the target meter image to obtain the target meter gradient value statistical matrix; an image shallow feature extraction module 130, which is used to perform image shallow feature extraction on the target meter gradient value statistical matrix to obtain a target meter image shallow feature map; a spatial attention enhancement module 140, which is used to perform spatial attention enhancement on the target meter image shallow feature map to obtain a spatially manifested target meter image shallow feature map as a spatially manifested target meter image shallow feature; and an evaluation result generation module 150, which is used to obtain an evaluation result based on the spatially manifested target meter image shallow feature, and adjust the shooting angle and shooting distance of the camera based on the evaluation result.

[0056] It can be understood by those skilled in the art that the specific operations of each step in the above-mentioned L4-level electric power intelligent outdoor inspection unmanned vehicle inspection system have been referred to above. Figures 1 to 4 It has been introduced in detail in the description of the L4-level electric power intelligent outdoor inspection unmanned vehicle inspection method, and therefore, its repeated description will be omitted.

[0057] As described above, the unmanned vehicle inspection system 100 based on L4-level electric power intelligent outdoor inspection according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with an unmanned vehicle inspection algorithm based on L4-level electric power intelligent outdoor inspection. In one possible implementation, the unmanned vehicle inspection system 100 based on L4-level electric power intelligent outdoor inspection according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the unmanned vehicle inspection system 100 based on L4-level electric power intelligent outdoor inspection can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the unmanned vehicle inspection system 100 based on L4-level electric power intelligent outdoor inspection can also be one of the many hardware modules of the wireless terminal.

[0058] Alternatively, in another example, the L4-level electric power intelligent outdoor patrol unmanned vehicle inspection system 100 and the wireless terminal may also be separate devices, and the L4-level electric power intelligent outdoor patrol unmanned vehicle inspection system 100 may be connected to the wireless terminal through a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0059] The foregoing is merely an example of the principles of the present disclosure, and various modifications may be made by those skilled in the art without departing from the scope of the present disclosure. The above embodiments are presented for the purpose of illustration and not limitation. The present disclosure may also take many forms other than those explicitly described herein. Therefore, it is emphasized that the present disclosure is not limited to the methods, systems and devices explicitly disclosed, but is intended to include variations and modifications within the spirit and scope of the appended claims.

Claims

1. A method for inspection based on L4 level electric power intelligent outdoor inspection unmanned vehicle, characterized in that: include: S1: Obtaining the target meter image collected by the camera of the power intelligent outdoor inspection unmanned vehicle; S2: Calculate the gradient value of each pixel in the target meter image to obtain a target meter gradient value statistical matrix; S3: extracting shallow features of the target meter gradient value statistical matrix to obtain a shallow feature map of the target meter image; S4: performing spatial attention enhancement on the shallow feature map of the target meter image to obtain a shallow feature map of the spatially manifested target meter image as a shallow feature of the spatially manifested target meter image; S5: obtaining an evaluation result based on the shallow features of the spatially manifested target meter image, and adjusting the shooting angle and shooting distance of the camera based on the evaluation result; S6: Repeat S1 to S5 until the evaluation result shows that there is no defocus.

2. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 1 is characterized in that: The target meter gradient value statistical matrix is ​​subjected to image shallow feature extraction to obtain a target meter image shallow feature map, including: the target meter gradient value statistical matrix is ​​passed through an image shallow feature extractor based on a deep neural network model to obtain the target meter image shallow feature map.

3. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 2 is characterized in that: The image shallow feature extractor based on the deep neural network model is an image shallow feature extractor based on the hollow convolutional neural network model.

4. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 3 is characterized in that: The shallow feature map of the target meter image is subjected to spatial attention enhancement to obtain the shallow feature map of the spatially manifested target meter image, including: the shallow feature map of the target meter image is passed through a multi-scale pan-spatial attention module to obtain the shallow feature map of the spatially manifested target meter image.

5. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 4 is characterized in that: The shallow feature map of the target meter image is passed through a multi-scale pan-spatial attention module to obtain the shallow feature map of the spatially manifested target meter image, including: Performing global mean pooling processing on the shallow feature map of the target meter image along the channel dimension to obtain a global mean pooling feature matrix of the target meter image; Performing maximum pooling processing on the shallow feature map of the target meter image along the channel dimension to obtain a maximum pooling feature matrix of the target meter image; Performing random pooling processing on the shallow feature map of the target meter image along the channel dimension to obtain a random pooling feature matrix of the target meter image; The target meter image global mean pooling feature matrix, the target meter image maximum pooling feature matrix and the target meter image random pooling feature matrix are spliced ​​to obtain a spliced ​​target meter image feature matrix; Performing convolution processing on the spliced ​​target meter image feature matrix to obtain the target meter image implicit correlation feature matrix; Inputting the target meter image implicit correlation feature matrix into a Sigmoid activation function to obtain a target meter image spatial attention weight matrix; Each feature matrix along the channel dimension in the shallow feature map of the target meter image is multiplied by the spatial attention weight matrix of the target meter image at each position point to obtain the shallow feature map of the spatially manifested target meter image.

6. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 5 is characterized in that: Based on the shallow features of the spatially manifested target meter image, an evaluation result is obtained, and based on the evaluation result, a shooting angle and a shooting distance of the camera are adjusted, including: The shallow feature map of the spatially manifested target meter image is passed through a classifier-based image quality evaluator to obtain an evaluation result, wherein the evaluation result is whether there is out of focus; In response to the evaluation result that out-of-focus exists, the shooting angle and shooting distance of the camera are adjusted.

7. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 6 is characterized in that: It also includes a training step: used to train the image shallow feature extractor based on the void convolutional neural network model, the multi-scale pan-spatial attention module and the classifier-based image quality assessor.

8. The inspection method based on L4 level electric power intelligent outdoor inspection unmanned vehicle according to claim 7 is characterized in that: The training step comprises: Acquire training data, wherein the training data includes training target meter images collected by a camera of an electric power intelligent outdoor inspection unmanned vehicle, and a true value of whether there is a defocused state; Calculating the gradient value of each pixel in the training target meter image to obtain a training target meter gradient value statistical matrix; The training target meter gradient value statistical matrix is ​​passed through the image shallow feature extractor based on the hole convolutional neural network model to obtain a training target meter image shallow feature map; Passing the shallow feature map of the training target meter image through the multi-scale pan-spatial attention module to obtain the shallow feature map of the training spatial explicit target meter image; Optimizing the shallow feature map of the training space manifested target meter image to obtain an optimized shallow feature map of the training space manifested target meter image; Passing the optimized training space manifested target meter image shallow feature map through the classifier-based image quality evaluator to obtain a classification loss function value; The image shallow feature extractor based on the hole convolutional neural network model, the multi-scale pan-spatial attention module and the classifier-based image quality assessor are trained based on the classification loss function value and through back propagation of gradient descent, wherein in each iteration of the model, the shallow feature map of the training space explicit target meter image is optimized.

9. A L4 level electric power intelligent outdoor inspection unmanned vehicle inspection system, characterized in that: include: The target meter image acquisition module is used to obtain the target meter image collected by the camera of the power intelligent outdoor inspection unmanned vehicle; A gradient value calculation module, used to calculate the gradient value of each pixel point in the target meter image to obtain a target meter gradient value statistical matrix; An image shallow feature extraction module, used for performing image shallow feature extraction on the target meter gradient value statistical matrix to obtain a target meter image shallow feature map; A spatial attention enhancement module, used for performing spatial attention enhancement on the shallow feature map of the target meter image to obtain the shallow feature map of the spatially manifested target meter image as the shallow feature of the spatially manifested target meter image; The evaluation result generating module is used to obtain the evaluation result based on the shallow features of the spatially manifested target meter image, and adjust the shooting angle and shooting distance of the camera based on the evaluation result.

10. The L4 level electric power intelligent outdoor inspection unmanned vehicle inspection system according to claim 9 is characterized in that: The image shallow feature extraction module is used to: pass the target meter gradient value statistical matrix through an image shallow feature extractor based on a deep neural network model to obtain the target meter image shallow feature map.

Images