Electric power tower identification method and system based on satellite remote sensing image

Abstract

The invention provides an electric power tower identification method and system based on a satellite remote sensing image. The method comprises the following steps: acquiring image data of a target area; preprocessing the image data of the target area to obtain preprocessed image data; substituting the preprocessed image data into a pre-trained deep learning model to obtain power tower information of the target area; wherein the pre-trained deep learning model comprises a front denoising network, a backbone detection network and a separation and attention module. According to the method, the remote sensing image is processed by using the pre-trained deep learning model, and the problems that in the prior art, satellite remote sensing image coordinates and actual geographic coordinates cannot be accurately matched, and the recognition success rate is low under the high-dip-angle shooting condition are solved.

Description

Technical Field

[0001] The present invention relates to the field of satellite remote sensing, and in particular to a method and system for identifying electric poles and towers based on satellite remote sensing images. Background Art

[0002] In the operation and management of modern power systems, power towers are the key supporting structures of transmission lines, and their accurate identification and positioning are of great significance. With the continuous expansion of the scale and increasing complexity of power networks, traditional manual inspection methods can no longer meet the needs of efficient, accurate and large-scale power tower identification. Satellite remote sensing images can cover large geographical areas, have macroscopic, periodic and multi-spectral characteristics, and can quickly obtain rich surface information.

[0003] However, satellite remote sensing images face many challenges when they are actually used for power pole tower identification. On the one hand, atmospheric factors interfere with the radiation signals received by satellite sensors, causing radiation errors in the images, affecting the image quality and the accurate expression of the features of the objects. On the other hand, satellite remote sensing images will produce geometric distortions during the imaging process due to factors such as sensor attitude and earth curvature, making it impossible to accurately match the image coordinates with the actual geographic coordinates, which brings great inconvenience to the subsequent identification and analysis of power pole towers based on location information. In addition, satellite remote sensing images will inevitably introduce noise during data collection, transmission and storage, which will mask the true characteristic information of power pole towers and reduce the accuracy and reliability of their identification.

[0004] The Chinese patent application with application number 202310632940.1 discloses a method for monitoring the operating status of transmission towers using satellite remote sensing based on improved Yolov5, including: obtaining high-resolution satellite remote sensing images of each transmission tower; obtaining the operating status monitoring results of each transmission tower based on the high-resolution satellite remote sensing images combined with the improved Yolov5 model. However, there are still problems in that the coordinates of the satellite remote sensing images cannot be accurately matched with the actual geographic coordinates and the recognition success rate is not high under high-angle shooting conditions. Summary of the invention

[0005] In order to solve the problems in the prior art that the coordinates of satellite remote sensing images cannot be accurately matched with the actual geographic coordinates and the noise reduction effect of satellite remote sensing images is poor, the present invention provides a method for identifying power towers based on satellite remote sensing images, comprising:

[0006] Acquire image data of the target area;

[0007] Preprocessing the image data of the target area to obtain preprocessed image data;

[0008] Substituting the pre-processed image data into a pre-trained deep learning model to obtain power tower information in the target area;

[0009] Among them, the pre-trained deep learning model is obtained by training the deep learning model using regional image data corresponding to known power pole tower location points.

[0010] Preferably, the deep learning model includes: a front denoising network, a backbone detection network, and a separation and attention module;

[0011] The front denoising network is used to remove noise from the image data to obtain denoised image data;

[0012] The backbone detection network is used to perform real-time detection of the tower information in the denoised image data using a detection model to obtain the tower information;

[0013] The separation and attention module is used to dynamically adjust the focus point in the feature map by learning the relationship between occlusion and unocclusion, thereby improving the detection effect of the detection model.

[0014] Preferably, the training of the deep learning model includes:

[0015] Obtain image data of known power tower locations;

[0016] The image data is denoised using a pre-denoising network to obtain denoised image data;

[0017] The denoised image data and the corresponding power pole tower location points constitute a data pair, the data pair constitutes a sample set, and the sample set is divided into a training set and a validation set according to a set ratio;

[0018] The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model;

[0019] The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that passes the validation is used as a trained deep learning model.

[0020] Preferably, the denoising process of the image data using a pre-denoising network to obtain denoised image data includes:

[0021] The noise in the image data is removed by using dilated convolution and regular convolution through the sparse block in the front denoising network;

[0022] The feature enhancement block in the front denoising network integrates the global and local feature information of the image data after noise removal to improve the denoising effect;

[0023] The noise information hidden in the background of the image data is extracted through the attention block in the front denoising network;

[0024] The denoised image data is constructed according to the noise information and the image data through a reconstruction block in the front denoising network.

[0025] Preferably, a dynamic snake-shaped convolutional layer is used to replace the convolutional layer in the backbone detection network.

[0026] Preferably, the denoised image data in the data pair in the training set is used as input data, the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model, including:

[0027] Substituting the denoised image data and the corresponding power pole tower position points in the data pair into the backbone detection network, adjusting the shape, size and weight of the convolution kernel according to the characteristics and structure of the input denoised image data through the dynamic serpentine convolution layer in the backbone detection network, predicting the offset through the offset convolution layer in the dynamic serpentine convolution layer, adjusting the actual position of the convolution kernel, and interpolating the input denoised image data to obtain a feature map;

[0028] The relationship between occluded and unoccluded feature maps is learned through separation and attention modules, and the focus points in the feature maps are dynamically adjusted to obtain a preliminary trained model.

[0029] Preferably, the method of using a validation set to validate the preliminarily trained model and using the preliminarily trained model that has passed the validation as a trained deep learning model includes:

[0030] Substitute the denoised image data in the validation set into the preliminarily trained model to obtain the predicted power tower location points corresponding to the denoised image data;

[0031] The power pole tower location points corresponding to the denoised image data in the validation set are compared with the power pole tower location points corresponding to the predicted denoised image data. If the error is less than the set threshold, the validation is passed and the preliminarily trained model is used as the trained deep learning model. Otherwise, the validation fails and the preliminarily trained model continues to be trained based on the training set.

[0032] Preferably, the preprocessing of the image data of the target area to obtain preprocessed image data includes:

[0033] The image data of the target area is corrected and enhanced to obtain preprocessed image data.

[0034] On the other hand, based on the same concept, the present application also provides a power pole tower identification system based on satellite remote sensing images, including:

[0035] An acquisition module, used for acquiring image data of a target area;

[0036] A preprocessing module, used for preprocessing the image data of the target area to obtain preprocessed image data;

[0037] A recognition module, used to substitute the pre-processed image data into a pre-trained deep learning model to obtain the power tower information of the target area;

[0038] Among them, the pre-trained deep learning model is obtained by training the deep learning model using regional image data corresponding to known power pole tower location points.

[0039] Preferably, the deep learning model includes: a front denoising network, a backbone detection network, and a separation and attention module;

[0040] The front denoising network is used to remove noise from the image data to obtain denoised image data;

[0041] The backbone detection network is used to perform real-time detection of the tower information in the denoised image data using a detection model to obtain the tower information;

[0042] The separation and attention module is used to dynamically adjust the focus point in the feature map by learning the relationship between occlusion and unocclusion, thereby improving the detection effect of the detection model.

[0043] Preferably, the training module is used to:

[0044] Obtain image data of known power tower locations;

[0045] The image data is denoised using a pre-denoising network to obtain denoised image data;

[0046] The denoised image data and the corresponding power pole tower location points constitute a data pair, the data pair constitutes a sample set, and the sample set is divided into a training set and a validation set according to a set ratio;

[0047] The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model;

[0048] The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that passes the validation is used as a trained deep learning model.

[0049] Preferably, the training module uses a pre-denoising network to perform denoising on the image data to obtain denoised image data, and the specific implementation steps include:

[0050] The noise in the image data is removed by using dilated convolution and regular convolution through the sparse block in the front denoising network;

[0051] The feature enhancement block in the front denoising network integrates the global and local feature information of the image data after noise removal to improve the denoising effect;

[0052] The noise information hidden in the background of the image data is extracted through the attention block in the front denoising network;

[0053] The denoised image data is constructed according to the noise information and the image data through a reconstruction block in the front denoising network.

[0054] Preferably, the training module uses the denoised image data in the data pair in the training set as input data, and uses the corresponding power pole tower position points in the data pair as output data, and substitutes them into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model. The specific implementation steps include:

[0055] Substituting the denoised image data and the corresponding power pole tower position points in the data pair into the backbone detection network, adjusting the shape, size and weight of the convolution kernel according to the characteristics and structure of the input denoised image data through the dynamic serpentine convolution layer in the backbone detection network, predicting the offset through the offset convolution layer in the dynamic serpentine convolution layer, adjusting the actual position of the convolution kernel, and interpolating the input denoised image data to obtain a feature map;

[0056] The relationship between occluded and unoccluded feature maps is learned through separation and attention modules, and the focus points in the feature maps are dynamically adjusted to obtain a preliminary trained model.

[0057] Preferably, the validation set is used to validate the preliminarily trained model, and the preliminarily trained model that has passed the validation is used as a trained deep learning model. The specific implementation steps include:

[0058] Substitute the denoised image data in the validation set into the preliminarily trained model to obtain the predicted power tower location points corresponding to the denoised image data;

[0059] The power pole tower location points corresponding to the denoised image data in the validation set are compared with the power pole tower location points corresponding to the predicted denoised image data. If the error is less than the set threshold, the validation is passed and the preliminarily trained model is used as the trained deep learning model. Otherwise, the validation fails and the preliminarily trained model continues to be trained based on the training set.

[0060] In another aspect, the present application further provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0061] The memory is used to store one or more programs;

[0062] When the one or more programs are executed by the at least one processor, the above-mentioned method for identifying power poles and towers based on satellite remote sensing images is implemented.

[0063] On the other hand, the present application also provides a readable storage medium having an execution program stored thereon, and when the execution program is executed, the above-mentioned method for identifying power poles and towers based on satellite remote sensing images is implemented.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The present invention provides a method for identifying electric poles and towers based on satellite remote sensing images, comprising: acquiring image data of a target area; preprocessing the image data of the target area to obtain preprocessed image data; substituting the preprocessed image data into a pre-trained deep learning model to obtain information about electric poles and towers in the target area. The method uses a pre-denoising network, a backbone detection network, and a separation and attention module to perform noise reduction and feature enhancement processing on remote sensing images, thereby solving the problems in the prior art that the coordinates of satellite remote sensing images cannot be accurately matched with the actual geographic coordinates and the recognition success rate is poor under high-angle shooting conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a flow chart of the power tower identification method of the present invention;

[0067] Figure 2 This is a manual recognition effect diagram;

[0068] Figure 3 To identify the effect diagram using the present invention;

[0069] Figure 4 This is a structural diagram of the electronic device described in the present invention. DETAILED DESCRIPTION

[0070] In order to better understand the present invention, the content of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0071] Embodiment 1:

[0072] The present invention proposes a method for identifying power towers based on satellite remote sensing images. Figure 1 As shown, including:

[0073] Step S1: Acquire image data of the target area;

[0074] Step S2: preprocessing the image data of the target area to obtain preprocessed image data;

[0075] Step S3: Substituting the pre-processed image data into a pre-trained deep learning model to obtain the power tower information of the target area;

[0076] Among them, the pre-trained deep learning model is obtained by training the deep learning model using regional image data corresponding to known power pole tower location points.

[0077] In this embodiment, the deep learning model specifically includes: a front denoising network, a backbone detection network, and a separation and attention module;

[0078] The front denoising network is used to remove noise from the image data to obtain denoised image data;

[0079] The backbone detection network is used to perform real-time detection of the tower information in the denoised image data using a detection model to obtain the tower information;

[0080] The separation and attention module is used to dynamically adjust the focus point in the feature map by learning the relationship between occlusion and unocclusion, thereby improving the detection effect of the detection model.

[0081] In this embodiment, the training of the deep learning model includes:

[0082] Obtain image data of known power tower locations;

[0083] The image data is denoised using a pre-denoising network to obtain denoised image data;

[0084] The denoised image data and the corresponding power pole tower location points constitute a data pair, the data pair constitutes a sample set, and the sample set is divided into a training set and a validation set according to a set ratio;

[0085] The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model;

[0086] The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that passes the validation is used as a trained deep learning model.

[0087] In this embodiment, the denoising process of the image data using the pre-denoising network to obtain the denoised image data includes:

[0088] The noise in the image data is removed by using dilated convolution and regular convolution through the sparse block in the front denoising network;

[0089] The feature enhancement block in the front denoising network integrates the global and local feature information of the image data after noise removal to improve the denoising effect;

[0090] The noise information hidden in the background of the image data is extracted through the attention block in the front denoising network;

[0091] The denoised image data is constructed according to the noise information and the image data through a reconstruction block in the front denoising network.

[0092] In this embodiment, a dynamic snake-shaped convolutional layer is used to replace the convolutional layer in the backbone detection network.

[0093] In this embodiment, the denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model, including:

[0094] Substituting the denoised image data and the corresponding power pole tower position points in the data pair into the backbone detection network, adjusting the shape, size and weight of the convolution kernel according to the characteristics and structure of the input denoised image data through the dynamic serpentine convolution layer in the backbone detection network, predicting the offset through the offset convolution layer in the dynamic serpentine convolution layer, adjusting the actual position of the convolution kernel, and interpolating the input denoised image data to obtain a feature map;

[0095] The relationship between occluded and unoccluded feature maps is learned through separation and attention modules, and the focus points in the feature maps are dynamically adjusted to obtain a preliminary trained model.

[0096] In this embodiment, the validation set is used to validate the preliminarily trained model, and the preliminarily trained model that has passed the validation is used as a trained deep learning model, including:

[0097] Substitute the denoised image data in the validation set into the preliminarily trained model to obtain the predicted power tower location points corresponding to the denoised image data;

[0098] The power pole tower location points corresponding to the denoised image data in the validation set are compared with the power pole tower location points corresponding to the predicted denoised image data. If the error is less than the set threshold, the validation is passed and the preliminarily trained model is used as the trained deep learning model. Otherwise, the validation fails and the preliminarily trained model continues to be trained based on the training set.

[0099] In this embodiment, step S2: preprocessing the image data of the target area to obtain preprocessed image data includes:

[0100] The image data of the target area is corrected and enhanced to obtain preprocessed image data.

[0101] In this embodiment, step S3: substituting the pre-processed image data into a pre-trained deep learning model to obtain the power tower information of the target area.

[0102] The present invention proposes a method for identifying power poles and towers based on satellite remote sensing images, which is applied in pre-disaster preparation or daily inspection in the power field to identify power poles and towers in a large range, determine the location of the poles and towers, and facilitate power grid companies to quickly grasp the status of the poles and towers, including the following beneficial effects:

[0103] A front-end remote sensing image denoising network is used to reduce the noise in the remote sensing image and improve the robustness of the model. A post-separation and attention module is used to improve the detection effect of the model in the case of occlusion / partial cutting of the pole tower. The general detection model YOLOv10 is used as the backbone detection network. The dynamic snake convolution layer is used to replace the convolution layer in the backbone detection network to improve the detection effect of the model on the pole tower target.

[0104] Example 2

[0105] The present invention provides a method for identifying power poles and towers based on satellite remote sensing images, comprising:

[0106] Step S1: Obtain image data of the target area. The image data here includes image data.

[0107] Step S2: preprocessing the image data of the target area to obtain preprocessed image data, specifically including:

[0108] In power tower recognition, high-resolution optical images are usually used to obtain image data of the target area. Image processing refers to the preprocessing of the acquired optical images, including image correction and enhancement, to improve the image quality and recognition effect.

[0109] Before step S3, the deep learning model is also introduced and trained. The deep learning model is further introduced below.

[0110] According to the initial position data set, the image data of the corresponding area in the optical image is intercepted as the input of the subsequent deep learning model.

[0111] Deep learning algorithms include model training, model verification, and iterative optimization. Model training refers to training deep learning models using regional image data corresponding to known power pole location points. Deep learning models are YOLO models or Faster R-CNN models based on convolutional neural networks (CNN). Through model training, the model can learn the characteristics and patterns of power poles and towers, so that it has recognition capabilities. Model verification refers to inputting pre-processed SAR image data into the trained deep learning model. Based on the input data, the model extracts the characteristics of the power poles and towers, and performs matching and recognition. The recognition results are output, including the location, shape and other information of the power poles. Iterative optimization refers to iteratively updating and optimizing the deep learning model based on the recognition results and actual conditions. The recognition accuracy and generalization ability of the model can be improved by adjusting model parameters and increasing training data.

[0112] Step 1: Use the front-end remote sensing image denoising network to denoise the remote sensing image, including:

[0113] The front-end network is constructed based on the attention mechanism and includes four modules: Sparse Block (SB), which removes noise by adopting dilated convolution and conventional convolution to achieve a balance between performance and efficiency; Feature Enhancement Block (FEB), which integrates global and local feature information through long paths to enhance the expression ability of the denoising model; Attention Block (AB), which is used to finely extract noise information hidden in complex backgrounds. The FEB module is integrated with the AB module to improve the effectiveness of the denoising model and reduce the computational complexity; Reconstruction Block (RB), which constructs the denoised image through the obtained noise map and the given noise image. The front-end denoising network has significant denoising performance for different types of noise (such as synthetic noise, real noise and blind noise), and can retain the original details and texture of the image, thereby avoiding problems such as over-smoothing or distortion.

[0114] Step 2: Use the post-separation and attention modules to process the remote sensing image to improve the detection effect under object occlusion, including:

[0115] This module is mainly used in the Neck layer of the detection network to enhance the network's responsiveness to occluded targets. Its core goal is to compensate for the response loss of the occluded area by enhancing the feature response of the unoccluded area. By learning the relationship between occlusion and unocclusion, the focus in the feature map is dynamically adjusted, so that the model can process the occluded information more effectively. This module combines spatial attention and feature enhancement mechanisms. It improves the detection effect under object occlusion by focusing on the importance of the unoccluded area and improving the overall feature representation. Specifically, this module is implemented by a combination of depthwise separable convolution and residual connection. Depthwise separable convolution operates on a channel-by-channel basis, which can learn the importance of different channels and reduce the amount of parameters, but ignores the information relationship between channels. To compensate for this loss, the outputs of different depthwise convolutions are combined through point-to-point (1x1) convolutions. Then a two-layer fully connected network is used to fuse the information of each channel to enhance the connection between all channels.

[0116] Step 3: The present invention uses YOLOv10 as the backbone detection network to detect remote sensing images.

[0117] YOLOv10 is the latest version of the YOLO (You Only Look Once) series and is a real-time target detection algorithm. Compared with the previous generation of YOLO algorithms, YOLOv10 solves the problem of relying on non-maximum suppression (NMS) for post-processing in the YOLO series of algorithms, achieves end-to-end deployment, and reduces inference latency; it adopts a model design strategy driven by overall efficiency and accuracy, and comprehensively optimizes all parts of the model from the perspectives of efficiency and accuracy. YOLOv10 can detect targets of different sizes at the same time, improves the detection coverage, and adopts a fully convolutional network design, so that the original resolution can be maintained when processing images, reducing the loss of feature maps, and significantly improving the inference speed and mAP (mean average precision).

[0118] Step 4: Dynamic snake convolution layer, which performs convolution processing on the remote sensing image;

[0119] Dynamic Snake Convolution Layer (DSCL) is a convolutional neural network (CNN) technology in deep learning, which is designed to deal with certain limitations in convolution operations to improve the network's ability to process irregular data. Its core features include: Dynamic adjustment of convolution kernels. The core idea of ​​the dynamic snake convolution layer is to dynamically adjust the shape and size of the convolution kernel according to the specific features and structure of the input data.

[0120] This adjustment is obtained through learning, so that the convolution kernel can better capture local features in irregular data; serpentine path. The dynamic snake convolution layer introduces the concept of serpentine path, so that the convolution kernel can follow the serpentine path when calculating the sliding window. The choice of this path can better adapt to the feature distribution of different image regions, thereby improving the expressiveness and performance of the model; offset prediction and feature interpolation. The dynamic snake convolution layer usually contains an offset convolution layer to predict the offsets, which are used to adjust the actual position of the convolution kernel. By using these offsets, the input feature map can be interpolated to obtain the adjusted feature map. This process involves calculating offsets, generating coordinate mappings, and interpolating input features; adaptive weight adjustment. The dynamic snake convolution layer can also adaptively adjust the weights of the convolution kernel, so that the network can pay more attention to the local features of the target structure. This adaptive adjustment helps to enhance the model's perception of the target structure, thereby improving the accuracy and robustness of detection.

[0121] Step S3: Substituting the pre-processed image data into a pre-trained deep learning model to obtain the power tower information of the target area, specifically including:

[0122] The image data pre-processed in step S2 is substituted into the pre-trained deep learning model to obtain the power tower locations in the target area.

[0123] The effectiveness of the method of the present invention:

[0124] In order to verify the actual effect of the present invention, the verification was carried out on a self-built data set, which includes 337 actual remote sensing images of power towers. After data expansion by random rotation, cropping, contrast adjustment and other methods, the remote sensing image data set was expanded to 4044 images. The resolution of the satellite remote sensing image is 0.3 meters, and the data is collected from Google Earth. For the convenience of processing, the original satellite images are uniformly scaled to 800*800 pixels during training and verification.

[0125] The experimental results are as follows Figure 2 and Figure 3 As shown in the figure, the final experimental results show that the tower recognition accuracy reaches 95.4%.

[0126] Embodiment 3:

[0127] The present invention based on the same inventive concept also provides a power pole tower identification system based on satellite remote sensing images, comprising:

[0128] An acquisition module, used for acquiring image data of a target area;

[0129] A preprocessing module, used for preprocessing the image data of the target area to obtain preprocessed image data;

[0130] A recognition module, used to substitute the pre-processed image data into a pre-trained deep learning model to obtain the power tower information of the target area;

[0131] Among them, the pre-trained deep learning model is obtained by training the deep learning model using regional image data corresponding to known power pole tower location points.

[0132] In this embodiment, the deep learning model includes: a front denoising network, a backbone detection network, and a separation and attention module;

[0133] The front denoising network is used to remove noise from the image data to obtain denoised image data;

[0134] The backbone detection network is used to perform real-time detection of the tower information in the denoised image data using a detection model to obtain the tower information;

[0135] The separation and attention module is used to dynamically adjust the focus point in the feature map by learning the relationship between occlusion and unocclusion, thereby improving the detection effect of the detection model.

[0136] In this embodiment, the training module is used to:

[0137] Obtain image data of known power tower locations;

[0138] The image data is denoised using a pre-denoising network to obtain denoised image data;

[0139] The denoised image data and the corresponding power pole tower location points constitute a data pair, the data pair constitutes a sample set, and the sample set is divided into a training set and a validation set according to a set ratio;

[0140] The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model;

[0141] The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that passes the validation is used as a trained deep learning model.

[0142] In this embodiment, the training module uses a pre-denoising network to perform denoising on the image data to obtain denoised image data. The specific implementation steps include:

[0143] The noise in the image data is removed by using dilated convolution and regular convolution through the sparse block in the front denoising network;

[0144] The feature enhancement block in the front denoising network integrates the global and local feature information of the image data after noise removal to improve the denoising effect;

[0145] The noise information hidden in the background of the image data is extracted through the attention block in the front denoising network;

[0146] The denoised image data is constructed according to the noise information and the image data through a reconstruction block in the front denoising network.

[0147] In this embodiment, the training module uses the denoised image data in the data pair in the training set as input data, and uses the corresponding power pole tower position point in the data pair as output data, and substitutes it into the backbone detection network and the separation and attention module for training to obtain a preliminarily trained model. The specific implementation steps include:

[0148] Substituting the denoised image data and the corresponding power pole tower position points in the data pair into the backbone detection network, adjusting the shape, size and weight of the convolution kernel according to the characteristics and structure of the input denoised image data through the dynamic serpentine convolution layer in the backbone detection network, predicting the offset through the offset convolution layer in the dynamic serpentine convolution layer, adjusting the actual position of the convolution kernel, and interpolating the input denoised image data to obtain a feature map;

[0149] The relationship between occluded and unoccluded feature maps is learned through separation and attention modules, and the focus points in the feature maps are dynamically adjusted to obtain a preliminary trained model.

[0150] In this embodiment, the validation set is used to validate the preliminarily trained model, and the preliminarily trained model that has passed the validation is used as a trained deep learning model. The specific implementation steps include:

[0151] Substitute the denoised image data in the validation set into the preliminarily trained model to obtain the predicted power tower location points corresponding to the denoised image data;

[0152] The power pole tower location points corresponding to the denoised image data in the validation set are compared with the power pole tower location points corresponding to the predicted denoised image data. If the error is less than the set threshold, the validation is passed and the preliminarily trained model is used as the trained deep learning model. Otherwise, the validation fails and the preliminarily trained model continues to be trained based on the training set.

[0153] Example 4

[0154] like Figure 4As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0155] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding functions, so as to implement the steps of a method for identifying power poles and towers based on satellite remote sensing images in the above-mentioned embodiment.

[0156] Example 5

[0157] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in the electronic device for storing programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device and the extended storage medium supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of a method for identifying power poles and towers based on satellite remote sensing images in the above embodiment.

[0158] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0160] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0161] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0162] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A method for identifying power towers based on satellite remote sensing images, characterized in that: include: Acquire image data of the target area; Preprocessing the image data of the target area to obtain preprocessed image data; Substituting the pre-processed image data into a pre-trained deep learning model to obtain power tower information in the target area; Among them, the pre-trained deep learning model is obtained by training the deep learning model using regional image data corresponding to known power pole tower location points.

2. The method according to claim 1, characterized in that: The deep learning model includes: a front denoising network, a backbone detection network, and a separation and attention module; The front denoising network is used to remove noise from the image data to obtain denoised image data; The backbone detection network is used to perform real-time detection of the tower information in the denoised image data using a detection model to obtain the tower information; The separation and attention module is used to dynamically adjust the focus point in the feature map by learning the relationship between occlusion and unocclusion, thereby improving the detection effect of the detection model.

3. The method according to claim 2, characterized in that: The training of the deep learning model includes: Obtain image data of known power tower locations; The image data is denoised using a pre-denoising network to obtain denoised image data; The denoised image data and the corresponding power pole tower location points constitute a data pair, the data pair constitutes a sample set, and the sample set is divided into a training set and a validation set according to a set ratio; The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model; The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that passes the validation is used as a trained deep learning model.

4. The method according to claim 3, characterized in that The method of using a pre-denoising network to denoise the image data to obtain denoised image data includes: The noise in the image data is removed by using dilated convolution and regular convolution through the sparse block in the front denoising network; The feature enhancement block in the front denoising network integrates the global and local feature information of the image data after noise removal to improve the denoising effect; The noise information hidden in the background of the image data is extracted through the attention block in the front denoising network; The denoised image data is constructed according to the noise information and the image data through a reconstruction block in the front denoising network.

5. The method according to claim 3, characterized in that The dynamic snake convolutional layer is used to replace the convolutional layer in the backbone detection network.

6. The method according to claim 5, characterized in that The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model, including: Substituting the denoised image data and the corresponding power pole tower position points in the data pair into the backbone detection network, adjusting the shape, size and weight of the convolution kernel according to the characteristics and structure of the input denoised image data through the dynamic serpentine convolution layer in the backbone detection network, predicting the offset through the offset convolution layer in the dynamic serpentine convolution layer, adjusting the actual position of the convolution kernel, and interpolating the input denoised image data to obtain a feature map; The relationship between occluded and unoccluded feature maps is learned through separation and attention modules, and the focus points in the feature maps are dynamically adjusted to obtain a preliminary trained model.

7. The method according to claim 3, characterized in that The method of using a validation set to validate the preliminarily trained model and using the preliminarily trained model that has passed the validation as a trained deep learning model includes: Substitute the denoised image data in the validation set into the preliminarily trained model to obtain the predicted power tower location points corresponding to the denoised image data; The power pole tower location points corresponding to the denoised image data in the validation set are compared with the power pole tower location points corresponding to the predicted denoised image data. If the error is less than the set threshold, the validation is passed and the preliminarily trained model is used as the trained deep learning model. Otherwise, the validation fails and the preliminarily trained model continues to be trained based on the training set.

8. The method according to claim 1, characterized in that: The preprocessing of the image data of the target area to obtain preprocessed image data includes: The image data of the target area is corrected and enhanced to obtain preprocessed image data.

9. A power tower identification system based on satellite remote sensing images, characterized in that: include: An acquisition module, used for acquiring image data of a target area; A preprocessing module, used for preprocessing the image data of the target area to obtain preprocessed image data; A recognition module, used to substitute the pre-processed image data into a pre-trained deep learning model to obtain the power tower information of the target area; Among them, the pre-trained deep learning model is obtained by training the deep learning model using regional image data corresponding to known power pole tower location points.

10. The system according to claim 9, characterized in that The deep learning model includes: a front denoising network, a backbone detection network, and a separation and attention module; The front denoising network is used to remove noise from the image data to obtain denoised image data; The backbone detection network is used to perform real-time detection of the tower information in the denoised image data using a detection model to obtain the tower information; The separation and attention module is used to dynamically adjust the focus point in the feature map by learning the relationship between occlusion and unocclusion, thereby improving the detection effect of the detection model.

11. The system according to claim 9, characterized in that Also included are training modules for: Obtain image data of known power tower locations; The image data is denoised using a pre-denoising network to obtain denoised image data; The denoised image data and the corresponding power pole tower location points constitute a data pair, the data pair constitutes a sample set, and the sample set is divided into a training set and a validation set according to a set ratio; The denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model; The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that passes the validation is used as a trained deep learning model.

12. The system according to claim 11, characterized in that The training module uses a pre-denoising network to denoise the image data to obtain denoised image data. The specific implementation steps include: The noise in the image data is removed by using dilated convolution and regular convolution through the sparse block in the front denoising network; The feature enhancement block in the front denoising network integrates the global and local feature information of the image data after noise removal to improve the denoising effect; The noise information hidden in the background of the image data is extracted through the attention block in the front denoising network; The denoised image data is constructed according to the noise information and the image data through a reconstruction block in the front denoising network.

13. The system according to claim 11, characterized in that In the training module, the denoised image data in the data pair in the training set is used as input data, and the corresponding power pole tower position point in the data pair is used as output data, and is substituted into the backbone detection network and the separation and attention modules for training to obtain a preliminarily trained model. The specific implementation steps include: Substituting the denoised image data and the corresponding power pole tower position points in the data pair into the backbone detection network, adjusting the shape, size and weight of the convolution kernel according to the characteristics and structure of the input denoised image data through the dynamic serpentine convolution layer in the backbone detection network, predicting the offset through the offset convolution layer in the dynamic serpentine convolution layer, adjusting the actual position of the convolution kernel, and interpolating the input denoised image data to obtain a feature map; The relationship between occluded and unoccluded feature maps is learned through separation and attention modules, and the focus points in the feature maps are dynamically adjusted to obtain a preliminary trained model.

14. The system according to claim 11, characterized in that The validation set is used to validate the preliminarily trained model, and the preliminarily trained model that has passed the validation is used as a trained deep learning model. The specific implementation steps include: Substitute the denoised image data in the validation set into the preliminarily trained model to obtain the predicted power tower location points corresponding to the denoised image data; The power pole tower location points corresponding to the denoised image data in the validation set are compared with the power pole tower location points corresponding to the predicted denoised image data. If the error is less than the set threshold, the validation is passed and the preliminarily trained model is used as the trained deep learning model. Otherwise, the validation fails and the preliminarily trained model continues to be trained based on the training set.

15. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a method for identifying power poles and towers based on satellite remote sensing images as described in any one of claims 1 to 8 is implemented.

16. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, a method for identifying power poles and towers based on satellite remote sensing images as described in any one of claims 1 to 8 is implemented.

Images