Transmission tower identification method and device, terminal equipment and storage medium

Abstract

The invention discloses a transmission tower identification method and device, terminal equipment and a storage medium, and belongs to the field of power systems, and the method comprises the steps: obtaining remote sensing image data of a to-be-identified region; inputting the remote sensing image data into a trained tower recognition model, so that the tower recognition model predicts the position of a tower in the remote sensing image data, and outputting a plurality of pieces of tower position data; constructing a tower space topological relation graph according to all tower position data; in combination with a preset tower distance relation constraint condition, obtaining misrecognition tower position data and misrecognition tower position data; and in all the pole tower position data, removing mistakenly-identified pole tower position data and adding missed-identified pole tower position data to obtain target pole tower position data. Compared with the prior art, the method has the advantages that industrial experts do not need to manually analyze tower spatial data, data analysis is automatically carried out through the tower identification model, and therefore the problems of low efficiency and high error rate in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a method, device, terminal equipment and storage medium for identifying a transmission tower. Background Art

[0002] my country is one of the countries most severely affected by typhoons. Every summer, typhoons impact power lines in many coastal provinces, leading to widespread power outages and significant losses. To mitigate the impact of typhoons and other severe weather on power facilities, regular inspections are necessary to monitor the operational status of transmission towers along power lines. Furthermore, after a typhoon or other severe weather event, the towers must be inspected for potential hazards as soon as possible. The key to this is accurate location information for transmission towers along power lines.

[0003] A traditional method for identifying power towers involves expert analysis combined with a geographic information system (GIS). This approach combines geospatial data analysis with expert domain expertise. This method leverages expert industry expertise to analyze image data and, with the assistance of GIS software, identifies information such as the location and overall distribution of power towers. Traditional expert analysis relies on subjective judgment, which is prone to errors. Furthermore, manual analysis of massive amounts of data is extremely time-consuming.

[0004] In summary, the traditional tower identification method has the disadvantages of high error rate and low efficiency. Summary of the Invention

[0005] The present invention provides a transmission tower identification method, apparatus, terminal equipment and storage medium, which can solve the problem that the existing tower identification technology relies on subjective analysis by experts, resulting in high error rate and low efficiency.

[0006] An embodiment of the present invention provides a method for identifying a transmission tower, comprising:

[0007] Obtain remote sensing image data of the area to be identified;

[0008] Inputting the remote sensing image data into a trained tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data and outputs a plurality of tower position data;

[0009] Constructing a tower space topology diagram based on all the tower position data;

[0010] According to the tower spatial topology relationship diagram and the preset tower distance relationship constraint conditions, misidentified tower position data and missed tower position data are obtained;

[0011] Among all the tower position data, remove the misidentified tower position data and add the missed tower position data to obtain the target tower position data;

[0012] The training process of the tower recognition model includes:

[0013] Obtain a number of remote sensing image data samples marked with tower location information;

[0014] Inputting each of the remote sensing image data samples into a tower recognition model, so that the tower recognition model predicts the tower position in the remote sensing image data sample to obtain a plurality of prediction results of the position data;

[0015] A loss function value is calculated based on the prediction result and the tower position labeling information, and the tower recognition model is adjusted based on the loss function value to obtain a trained tower recognition model.

[0016] Furthermore, the tower recognition model includes: a tower feature extraction backbone, a multi-level branch tower feature enhancement module and a regression classification module;

[0017] The tower feature extraction backbone is used to extract features layer by layer from the remote sensing image data through a basic feature extraction module, and obtain shallow features, middle features and deep features in sequence;

[0018] The multi-level branch tower feature enhancement module is used to perform fusion enhancement processing on the shallow layer features, the middle layer features and the deep layer features to obtain fusion features;

[0019] The regression classification module is used to generate a plurality of tower target frames according to the fusion features; determine the tower target according to the tower target probability in each of the tower target frames, and output the tower position data of the tower target.

[0020] Furthermore, the remote sensing image data is subjected to layer-by-layer feature extraction by the basic feature extraction module to obtain shallow features, middle features and deep features in sequence, including:

[0021] Extracting features from the remote sensing image data using the basic feature extraction module to obtain shallow features;

[0022] Extracting the shallow features by the basic feature extraction module to obtain the mid-level features;

[0023] The mid-level features are extracted by the basic feature extraction module to obtain the deep-level features.

[0024] Furthermore, the fusion enhancement processing is performed on the shallow features, the middle features and the deep features to obtain fused features, including:

[0025] Performing upsampling processing on the deep layer features and adding the sampling results to the middle layer features to obtain the first features;

[0026] Performing upsampling on the first feature and adding the sampling result to the shallow feature to obtain a second feature;

[0027] Performing convolution downsampling on the second feature, and adding the sampling result to the first feature to obtain a third feature;

[0028] Performing convolution downsampling on the third feature, and adding the sampling result to the deep feature to obtain a fourth feature;

[0029] The second feature, the third feature and the fourth feature are fused to obtain the fused feature.

[0030] Furthermore, the basic feature extraction module includes, in sequence: a 3×3 standard convolution layer, a first batch normalization layer, a first SiLU activation function layer, a 5×5 depth-separable convolution layer, a 1×1 point-by-point convolution layer, a second batch normalization layer, and a second SiLU activation function layer.

[0031] Furthermore, the step of obtaining a plurality of remote sensing image data samples marked with tower position information includes:

[0032] Acquire a number of original remote sensing sample data obtained through manual labeling; wherein the original remote sensing sample data is labeled with rough labeling information of tower positions;

[0033] For each of the original remote sensing sample data, a straight line fitting is performed on the tower point according to the rough marking information of the tower position to obtain a fitted transmission line;

[0034] Generating a transmission channel range with a preset width along the fitted transmission line;

[0035] Within the transmission channel, search for unmarked towers and mark them.

[0036] Furthermore, the tower space topology diagram is constructed based on all the tower position data, including:

[0037] Generating a Delaunay triangulated mesh graph with the towers as nodes based on all the tower position data;

[0038] The tower spatial topology relationship diagram is generated according to the Delaunay triangulated mesh diagram using a minimum spanning tree algorithm.

[0039] Another embodiment of the present invention further provides a transmission tower identification device, comprising: a data acquisition module, a tower identification module and an identification result optimization module;

[0040] The data acquisition module is used to acquire remote sensing image data of the area to be identified;

[0041] The tower recognition module is used to input the remote sensing image data into a trained tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data and outputs a plurality of tower position data; wherein the training process of the tower recognition model includes: obtaining a plurality of remote sensing image data samples marked with tower position information; inputting each of the remote sensing image data samples into the tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data sample and obtains a plurality of prediction results of position data; calculating a loss function value based on the prediction result and the tower position labeling information, and adjusting the tower recognition model based on the loss function value to obtain a trained tower recognition model;

[0042] The recognition result optimization module is used to construct a tower space topology relationship diagram based on all the tower position data; obtain misidentified tower position data and missed tower position data based on the tower space topology relationship diagram and preset tower distance relationship constraints; and remove the misidentified tower position data and add the missed tower position data from all the tower position data to obtain the target tower position data.

[0043] Another embodiment of the present invention further provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the transmission tower identification method of the present invention are implemented.

[0044] Another embodiment of the present invention further provides a computer-readable storage medium item, comprising: a stored computer program, which controls the device where the computer-readable storage medium is located to execute the steps of the transmission tower identification method of the present invention when the computer program is executed.

[0045] The following beneficial effects are achieved by implementing the present invention:

[0046] The present invention obtains remote sensing image data of the area to be identified, makes predictions based on the remote sensing image data through a trained pole tower recognition model, and outputs the predicted pole tower position data. The pole tower position data is then further optimized, and a pole tower spatial topology relationship diagram is constructed based on the pole tower position data. According to the pole tower relationship constraint conditions, the misidentified pole tower position data and the missed pole tower position data are determined, the misidentified data are removed, and the missed data are added to obtain the final target pole tower position data. The present invention does not require industry experts to manually analyze the pole tower spatial data, but automatically performs data analysis through the pole tower recognition model, greatly improving the efficiency of data analysis, avoiding errors caused by subjective judgment of experts, and reducing the error rate of the pole tower recognition results. In addition, the present invention further optimizes the model recognition results, further improving the accuracy of the pole tower recognition results. In summary, the present invention has the advantages of high efficiency and low error rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0048] Figure 1 This is a flow chart of a method for identifying a transmission tower provided by one embodiment of the present invention;

[0049] Figure 2 It is a schematic diagram of the tower marking method;

[0050] Figure 3 It is a structural diagram of the tower identification model;

[0051] Figure 4 It is a structural diagram of the basic feature extraction module;

[0052] Figure 5 This is a schematic diagram of the optimization process of the recognition results based on the tower spatial topology relationship diagram;

[0053] Figure 6 The figure is a schematic structural diagram of a transmission tower identification device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0056] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0057] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0058] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0059] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0060] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0061] In order to solve the problem that the existing tower identification technology relies on subjective analysis by experts, resulting in high error rate and low efficiency, an embodiment of the present invention provides a transmission tower identification method, see Figure 1 , which is a flow chart of the transmission tower identification method, including:

[0062] S1. Obtain remote sensing image data of the area to be identified.

[0063] It should be noted that remote sensing image data includes optical image data and SAR image data. Optical remote sensing images not only have higher resolution but also provide characteristic information such as the spectral texture and shape of the tower, which is consistent with the general understanding of ground objects by the human eye and is helpful for tower interpretation. SAR images can obtain the electromagnetic reflection characteristics of the tower target and have the ability to work in extreme weather and all day and night.

[0064] S2. Inputting the remote sensing image data into a trained tower recognition model, so that the tower recognition model predicts the tower position in the remote sensing image data and outputs a plurality of tower position data;

[0065] The training process of the tower recognition model includes:

[0066] Obtain a number of remote sensing image data samples marked with tower location information;

[0067] Inputting each of the remote sensing image data samples into a tower recognition model, so that the tower recognition model predicts the tower position in the remote sensing image data sample to obtain a plurality of prediction results of the position data;

[0068] A loss function value is calculated based on the prediction result and the tower position labeling information, and the tower recognition model is adjusted based on the loss function value to obtain a trained tower recognition model.

[0069] It should be noted that the present invention is aimed at power grid emergency disaster survey scenarios under extreme weather conditions. During the sample data collection process, in order to ensure that the tower targets are at the minimum identifiable level, remote sensing image data samples are obtained as much as possible under various resolution conditions and various terrain scenes (such as mountainous areas, plains, cities, hills, etc.) to simulate the characteristics of tower targets under different imaging conditions (cloudy, rainy, etc.) and enhance the richness of sample data.

[0070] Since remote sensing images are often affected by factors such as solar altitude, atmospheric absorption and scattering, sensor, and terrain during the acquisition process, and these factors often change over time, this can lead to differences in the radiation values ​​of unchanged ground objects in images at different times. Therefore, standardized preprocessing is required to construct a high-quality, standardized, and unified deep learning sample library for training and optimizing the subsequent tower recognition model. In view of the characteristics of remote sensing images in tower recognition tasks, the preprocessing methods included in this disclosure include georegistration, geometric correction, radiometric correction, pan-sharpening, and image sliding window segmentation.

[0071] Original remote sensing imagery generally lacks geographic coordinates, only pixel coordinates. Georeferencing is required to add a geographic reference coordinate system to the remote sensing image. This invention, after georeferencing, provides geographic reference information to the remote sensing image. This allows for the extraction of meaningful geographic coordinates (latitude and longitude) from subsequently identified tower targets, helping inspection personnel quickly locate the true spatial position of the identified towers.

[0072] Differences between remote sensing satellite images and sensors can lead to a certain degree of geometric distortion of objects in remote sensing images. Furthermore, images acquired by different sensors exhibit systematic differences in pixel values. This can cause geometric distortion of the shapes of tower targets in the images, making them more difficult to identify. Geometric and radiometric correction preprocessing aims to correct for geometric distortion and background radiation differences in the original remote sensing images. After geometric and radiometric correction, tower targets appear more consistent in their geometric shape and pixel brightness in remote sensing images. This is crucial for building a high-quality, standardized sample library.

[0073] Optical remote sensing images typically contain both multispectral and panchromatic data. Multispectral data has higher spectral resolution, while panchromatic data has higher spatial resolution. To obtain optical remote sensing images with both good spatial and spectral resolution, sharpening and fusion processing of the panchromatic and multispectral data is required. The fused remote sensing images have higher spatial and spectral resolutions, providing richer characteristic information about tower targets.

[0074] The large-scale remote sensing images and tower annotation results are synchronously divided into blocks through sliding window cropping, and data enhancement methods such as rotation, flipping, and brightness changes are used to simulate the differences in tower targets in images under different lighting conditions and satellite imaging perspectives, fully ensuring the diversity requirements of the constructed sample library.

[0075] The model is trained using the back propagation algorithm through sample data, and the model parameters are tuned to obtain the optimized tower recognition model.

[0076] The present invention uses a tower recognition model to achieve automated, high-precision recognition of tower targets in remote sensing images, solving the problem of high error rates and low efficiency caused by existing tower recognition technologies relying on subjective analysis by experts.

[0077] In a preferred embodiment, the step of obtaining a plurality of remote sensing image data samples marked with tower position information includes:

[0078] Acquire a number of original remote sensing sample data obtained through manual labeling; wherein the original remote sensing sample data is labeled with rough labeling information of tower positions;

[0079] For each of the original remote sensing sample data, a straight line fitting is performed on the tower point according to the rough marking information of the tower position to obtain a fitted transmission line;

[0080] Generating a transmission channel range with a preset width along the fitted transmission line;

[0081] Within the transmission channel, search for unmarked towers and mark them.

[0082] It should be noted that the present invention constructs a sample library based on preprocessed remote sensing image data samples and designs a rapid data sample annotation method based on the distribution characteristics of tower targets in remote sensing images. The present invention notes that in large-scale satellite remote sensing images, transmission towers are typically distributed in a linear pattern, densely packed in a specific direction, while in most other areas, they are sparsely distributed or almost non-existent. During the annotation process, searching along a specific linear direction greatly improves the annotation speed of tower samples.

[0083] like Figure 1 FIG. 1 is a schematic diagram of a tower marking method designed in the present invention. The present invention notes that the distribution of transmission towers is generally arranged linearly along a certain direction, and there is often a large tower density in this direction, while other types of towers, such as telecommunication base stations or other types of towers, are scattered throughout the area.

[0084] According to the prior distribution rules of transmission towers, a method for quickly labeling data samples is designed in the present invention. The steps of the method are as follows:

[0085] 1. First, perform the initial manual identification and rough labeling of transmission towers on the remote sensing image data sample. During the initial identification, there is no need to fully label the transmission towers. Only sparse and uniform general labeling of transmission towers in various areas of the entire image is required, which greatly reduces the labeling time. The black triangle in the figure above shows the initial labeling results of the towers.

[0086] 2. Then, using the sparse transmission tower results initially marked in step 1, a straight line fitting is performed through the sparse transmission pole identification points to obtain the fitting result of the transmission line, such as Figure 2 As shown by the dotted line in the middle. And set a certain size of buffer radius as Figure 2 As shown in d, the transmission channel range is constructed as above. Figure 2 As shown in the solid channel;

[0087] 3. Finally, using the transmission channel range constructed in step 2, focus on searching for unlabeled transmission towers within the constructed transmission channel range and accurately label the transmission towers.

[0088] The proposed labeling method first establishes a transmission line corridor by labeling sparse tower data. This established transmission line corridor guides the subsequent dense labeling of transmission towers. Compared to a global search and visual interpretation approach for labeling towers, this approach reduces workload and mitigates the risk of misidentification of scattered non-transmission tower objects outside the transmission corridor.

[0089] Finally, after the transmission tower marking is completed, the quality of the sample library is checked manually to check whether the marked transmission towers have misidentified non-tower targets, and whether there are any missed identifications of towers in the transmission channel, so as to ensure the high quality requirements of the transmission tower sample library.

[0090] In a preferred embodiment, the tower recognition model includes: a tower feature extraction backbone, a multi-level branch tower feature enhancement module and a regression classification module;

[0091] The tower feature extraction backbone is used to extract features layer by layer from the remote sensing image data through a basic feature extraction module, and obtain shallow features, middle features and deep features in sequence;

[0092] The multi-level branch tower feature enhancement module is used to perform fusion enhancement processing on the shallow layer features, the middle layer features and the deep layer features to obtain fusion features;

[0093] The regression classification module is used to generate a plurality of tower target frames according to the fusion features; determine the tower target according to the tower target probability in each of the tower target frames, and output the tower position data of the tower target.

[0094] like Figure 3 Figure 2 shows the schematic diagram of the tower recognition model designed by the present invention. The model structure is generally divided into three parts: a tower feature extraction backbone, a multi-level branch tower feature enhancement module, and a regression classification module. The tower feature extraction backbone is responsible for processing input remote sensing image data and extracting target features of transmission towers. The multi-level branch tower feature enhancement module fuses and enhances the tower recognition features extracted by the backbone, generating more robust tower recognition features for subsequent recognition. Finally, a regression classification module processes the output features of the previous module and predicts the tower location data of the output tower target.

[0095] The designed regression classification module consists of a regression branch and a classification branch. The regression branch is responsible for the prediction of the tower target frame, and the classification module complexly judges the probability of the tower target in the target frame, and finally obtains the prediction result of the tower target after processing.

[0096] In a preferred embodiment, the remote sensing image data is subjected to layer-by-layer feature extraction by a basic feature extraction module to obtain shallow features, mid-layer features, and deep features in sequence, including:

[0097] Extracting features from the remote sensing image data using the basic feature extraction module to obtain shallow features;

[0098] Extracting the shallow features by the basic feature extraction module to obtain the mid-level features;

[0099] The mid-level features are extracted by the basic feature extraction module to obtain the deep-level features.

[0100] It should be noted that the model design takes into account the significant differences in scale between remote sensing images of varying resolutions and different types of transmission tower targets. To enhance the model's ability to recognize transmission towers of various sizes and types, the model's feature extraction backbone uses a Basic Feature Extraction Module (BFEM) to repeatedly perform feature processing to obtain shallow, mid-level, and deep-level features of the tower targets.

[0101] In a preferred embodiment, the basic feature extraction module includes, in sequence: a 3×3 standard convolution layer, a first batch normalization layer, a first SiLU activation function layer, a 5×5 depth-separable convolution layer, a 1×1 point-by-point convolution layer, a second batch normalization layer, and a second SiLU activation function layer.

[0102] The designed basic feature extraction module structure is as follows Figure 4 As shown:

[0103] The basic feature extraction module consists of functional layers such as 3×3 standard convolution, 5×5 depth-separable convolution, 1×1 point-by-point convolution, batch normalization layer, and SiLU activation function.

[0104] First, the input features are processed using a 3×3 convolutional layer, and then normalized and nonlinearly activated using a batch normalization layer and an activation function layer. To enhance the global expressiveness of the features, a 5×5 depthwise separable convolution is used for global feature extraction, and the number of output channels is adjusted using a 1×1 point-by-point convolution. Finally, the features are normalized and nonlinearly activated using a batch normalization layer and a SiLU activation function.

[0105] Finally, a backbone network is constructed using the basic feature extraction module to extract features layer by layer from the input data. During the extraction process, the global expressive power of the features obtained becomes increasingly stronger, but the spatial resolution of the features gradually decreases. Shallow features have a higher spatial resolution for tower targets, while deep features have a relatively stronger global expressive power.

[0106] In a preferred embodiment, the fusing and enhancing the shallow features, the middle features, and the deep features to obtain the fused features includes:

[0107] Performing upsampling processing on the deep layer features and adding the sampling results to the middle layer features to obtain the first features;

[0108] Performing upsampling on the first feature and adding the sampling result to the shallow feature to obtain a second feature;

[0109] Performing convolution downsampling on the second feature, and adding the sampling result to the first feature to obtain a third feature;

[0110] Performing convolution downsampling on the third feature, and adding the sampling result to the deep feature to obtain a fourth feature;

[0111] The second feature, the third feature and the fourth feature are fused to obtain the fused feature.

[0112] It should be noted that in the tower target recognition task, features with strong global expression capabilities are needed to improve recognition accuracy, and it is also necessary to ensure that the features have high spatial resolution to achieve accurate positioning of the transmission tower. Therefore, a multi-level branch tower feature enhancement module is designed in the present invention to further fuse and enhance the acquired multi-level tower features. The multi-level branch tower feature enhancement module obtains the first feature by upsampling the deep features and using them to supplement the global expression capabilities of the middle-level features. The first feature is then upsampled as a supplement to the shallow features and fused to obtain the second feature. The second feature is convoluted downsampled to supplement the spatial resolution information lacking in the first feature to obtain the third feature. The third feature is then convoluted downsampled to enhance the resolution of the deep feature to obtain the fourth feature. Finally, the second, third and fourth features are fused to obtain the fused feature, which is used for the final tower recognition and positioning.

[0113] The fused features contain both the global feature information and spatial resolution information of the tower. The fused features are further processed through the regression classification module to obtain the spatial position information of the target tower.

[0114] S3. Construct a tower space topology relationship diagram based on all the tower position data.

[0115] In a preferred embodiment, constructing a tower space topology diagram based on all the tower position data includes:

[0116] Generating a Delaunay triangulated mesh graph with the towers as nodes based on all the tower position data;

[0117] The tower spatial topology relationship diagram is generated according to the Delaunay triangulated mesh diagram using a minimum spanning tree algorithm.

[0118] S4. Obtaining misidentified tower position data and missed tower position data based on the tower spatial topology relationship diagram and preset tower distance relationship constraints.

[0119] S5. Remove misidentified tower position data from all the tower position data and add missed tower position data to obtain target tower position data.

[0120] It should be noted that when using models to identify tower targets in remote sensing images, there are often cases where targets partially or highly obscured by complex terrain are missed, as well as misidentification of scattered suspected tower targets in the periphery. For example, tower targets are obscured by clouds, trees, tall buildings, and can be confused with other types of towers, such as telecommunications towers, leading to missed and misidentified targets. Therefore, the present invention further adjusts the recognition results by constructing spatial topological relationships between towers and pre-defined prior knowledge.

[0121] like Figure 5 As shown in FIG, a schematic diagram of the optimization process of the recognition result based on the pole tower spatial topological relationship diagram is shown, wherein FIG (1) is a schematic diagram of the pole tower recognition result obtained based on the model, FIG (2) is a pole tower spatial topological relationship diagram, FIG (3) is a schematic diagram of misidentification and missed recognition, and FIG (4) is a schematic diagram of the recognition result after removing misidentification and adding missed recognition.

[0122] This invention further optimizes and refines the original identification results based on prior knowledge of the distribution of transmission towers to improve the accuracy of transmission tower identification. This prior knowledge includes the fact that tower targets are often arranged linearly and are generally evenly spaced along the line. The process for constructing the spatial topology of towers and optimizing tower identification is as follows:

[0123] 1. First, rely on Figure 5 -(1) The tower recognition results are used to construct the spatial topological connection relationship of the identified tower targets, and the following is obtained: Figure 5 -(2) The spatial topological relationship diagram of the tower shown;

[0124] 2. Secondly, considering the prior knowledge of the target distribution of power grid towers (such as the distribution distance between towers, etc.), the spatial adjacency relationship of tower points is constructed in sequence to generate the following Figure 5 -The Delaunay triangulated mesh shown in the dashed line in (2);

[0125] 3. Then, the minimum spanning tree algorithm is used to generate the initial connected network, ensuring that all points are connected with the minimum total length, and finally the following is obtained: Figure 5 -(2) The tower topology network shown by the solid line;

[0126] 4. Finally, add constraints to the tower topology network (such as the distance threshold between towers, etc.) and perform topology optimization. Based on the constraints and the constructed topology network, optimize the tower targets identified in the figure. For example, according to the preset rules (equal spacing), Figure 5 -(3) The boxed area is suspected to be missed, and the circled area is suspected to be misidentified (isolated points outside the topological relationship). Subsequently, the transmission towers are checked for omissions and the optimization results are output, and finally the following is obtained: Figure 5 -(4) shows the optimized tower identification results.

[0127] In the above optimization process, the present invention introduces the prior knowledge that transmission tower targets are generally evenly spaced and there are almost no isolated towers outside the transmission line. The tower identification results are optimized using the constructed topological relationship diagram, such as Figure 5 As shown in (3), the tower target in the circle is outside the constructed topological association relationship and is suspected to be a misidentified target of a non-tower target. The box-marked area is above the topological association relationship. According to the prior knowledge that the transmission towers are evenly spaced, there is a suspected missed tower target in the box-marked target. Based on the above situation, further analysis and judgment are made on the tower target areas suspected of missed identification and misidentification. Finally, the optimized tower recognition result is obtained as follows: Figure 5 -(4) shown.

[0128] The present invention enhances and improves the original tower identification results through the above-mentioned tower spatial topological relationship establishment process and the tower identification result optimization process based on the topological relationship, especially for non-power tower targets that are highly shielded on the transmission line and outside the transmission line, and has a strong ability to "check for omissions and fill in gaps" and "eliminate the false and retain the true" .

[0129] like Figure 6 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0130] An embodiment of the present invention provides a transmission tower identification device, comprising: a data acquisition module, a tower identification module, and an identification result optimization module;

[0131] The data acquisition module is used to acquire remote sensing image data of the area to be identified;

[0132] The tower recognition module is used to input the remote sensing image data into a trained tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data and outputs a plurality of tower position data; wherein the training process of the tower recognition model includes: obtaining a plurality of remote sensing image data samples marked with tower position information; inputting each of the remote sensing image data samples into the tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data sample and obtains a plurality of prediction results of position data; calculating a loss function value based on the prediction result and the tower position labeling information, and adjusting the tower recognition model based on the loss function value to obtain a trained tower recognition model;

[0133] The recognition result optimization module is used to construct a tower space topology relationship diagram based on all the tower position data; obtain misidentified tower position data and missed tower position data based on the tower space topology relationship diagram and preset tower distance relationship constraints; and remove the misidentified tower position data and add the missed tower position data from all the tower position data to obtain the target tower position data.

[0134] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, and can implement the transmission tower identification method provided by any of the above-mentioned method embodiments of the present invention.

[0135] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Furthermore, in the drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which may be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement the present invention without inventive effort.

[0136] Based on the embodiments of the above-mentioned method items, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the transmission tower identification method of any embodiment of the present invention.

[0137] For example, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device.

[0138] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0139] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0140] Based on the above method embodiments, another embodiment of the present invention provides a computer-readable storage medium, including a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the transmission tower identification method described in any one of the above method embodiments of the present invention.

[0141] Wherein, the module / unit integrated in the device / terminal equipment, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0142] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying a transmission tower, characterized in that: include: Obtain remote sensing image data of the area to be identified; Inputting the remote sensing image data into a trained tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data and outputs a plurality of tower position data; Constructing a tower space topology diagram based on all the tower position data; According to the tower spatial topology relationship diagram and the preset tower distance relationship constraint conditions, misidentified tower position data and missed tower position data are obtained; Among all the tower position data, remove the misidentified tower position data and add the missed tower position data to obtain the target tower position data; The training process of the tower recognition model includes: Obtain a number of remote sensing image data samples marked with tower location information; Inputting each of the remote sensing image data samples into a tower recognition model, so that the tower recognition model predicts the tower position in the remote sensing image data sample to obtain a plurality of prediction results of the position data; A loss function value is calculated based on the prediction result and the tower position labeling information, and the tower recognition model is adjusted based on the loss function value to obtain a trained tower recognition model.

2. The transmission tower identification method according to claim 1, wherein: The tower recognition model includes: a tower feature extraction backbone, a multi-level branch tower feature enhancement module and a regression classification module; The tower feature extraction backbone is used to extract features layer by layer from the remote sensing image data through a basic feature extraction module, and obtain shallow features, middle features and deep features in sequence; The multi-level branch tower feature enhancement module is used to perform fusion enhancement processing on the shallow layer features, the middle layer features and the deep layer features to obtain fusion features; The regression classification module is used to generate a plurality of tower target frames according to the fusion features; determine the tower target according to the tower target probability in each of the tower target frames, and output the tower position data of the tower target.

3. The transmission tower identification method according to claim 2, wherein: The remote sensing image data is subjected to layer-by-layer feature extraction by the basic feature extraction module to obtain shallow features, middle features, and deep features in sequence, including: Extracting features from the remote sensing image data using the basic feature extraction module to obtain shallow features; Extracting the shallow features by the basic feature extraction module to obtain the mid-level features; The mid-level features are extracted by the basic feature extraction module to obtain the deep-level features.

4. The method for identifying a transmission tower according to claim 2, wherein: The fusing and enhancing the shallow features, the middle features, and the deep features to obtain fused features includes: Performing upsampling processing on the deep layer features and adding the sampling results to the middle layer features to obtain the first features; Performing upsampling on the first feature and adding the sampling result to the shallow feature to obtain a second feature; Performing convolution downsampling on the second feature, and adding the sampling result to the first feature to obtain a third feature; Performing convolution downsampling on the third feature, and adding the sampling result to the deep feature to obtain a fourth feature; The second feature, the third feature and the fourth feature are fused to obtain the fused feature.

5. The transmission tower identification method according to claim 2, wherein: The basic feature extraction module includes, in sequence: a 3×3 standard convolution layer, a first batch normalization layer, a first SiLU activation function layer, a 5×5 depthwise separable convolution layer, a 1×1 point-by-point convolution layer, a second batch normalization layer, and a second SiLU activation function layer.

6. The method for identifying a transmission tower according to claim 1, wherein: The step of obtaining a plurality of remote sensing image data samples marked with tower position information includes: Acquire a number of original remote sensing sample data obtained through manual labeling; wherein the original remote sensing sample data is labeled with rough labeling information of tower positions; For each of the original remote sensing sample data, a straight line fitting is performed on the tower point according to the rough marking information of the tower position to obtain a fitted transmission line; Generating a transmission channel range with a preset width along the fitted transmission line; Within the transmission channel, search for unmarked towers and mark them.

7. The method for identifying a transmission tower according to claim 1, wherein: The step of constructing a tower space topology diagram based on all the tower position data includes: Generating a Delaunay triangulated mesh graph with the towers as nodes based on all the tower position data; The tower spatial topology relationship diagram is generated according to the Delaunay triangulated mesh diagram using a minimum spanning tree algorithm.

8. A transmission tower identification device, characterized in that: include: Data acquisition module, tower identification module and identification result optimization module; The data acquisition module is used to acquire remote sensing image data of the area to be identified; The tower recognition module is used to input the remote sensing image data into a trained tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data and outputs a plurality of tower position data; wherein the training process of the tower recognition model includes: obtaining a plurality of remote sensing image data samples marked with tower position information; inputting each of the remote sensing image data samples into the tower recognition model so that the tower recognition model predicts the tower position in the remote sensing image data sample and obtains a plurality of prediction results of position data; calculating a loss function value based on the prediction result and the tower position labeling information, and adjusting the tower recognition model based on the loss function value to obtain a trained tower recognition model; The recognition result optimization module is used to construct a tower space topology relationship diagram based on all the tower position data; obtain misidentified tower position data and missed tower position data based on the tower space topology relationship diagram and preset tower distance relationship constraints; and remove the misidentified tower position data and add the missed tower position data from all the tower position data to obtain the target tower position data.

9. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for identifying a transmission tower according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that include: A stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the transmission tower identification method according to any one of claims 1 to 7.

Images