Power transmission channel along the line of easy floating object detection model training method and detection method

Abstract

The application relates to a power transmission channel along-line easy-floating object detection model training method and a detection method. The method comprises the following steps: inputting a historical power transmission channel along-line remote sensing image obtained to an initial easy-floating object detection model constructed, so that an easy-floating object deep feature extraction module extracts easy-floating object features of different scales of the historical power transmission channel along-line remote sensing image, and generates a power transmission channel easy-floating object feature map based on the easy-floating object features of different scales; the power transmission channel easy-floating object feature map is subjected to a pooling and channel weighting operation through an easy-floating object feature attention module to obtain a power transmission channel easy-floating object feature enhancement map; the class, position information and confidence of a prediction box are determined based on the easy-floating object features of different scales and the power transmission channel easy-floating object feature enhancement map; and the parameters of the initial easy-floating object detection model are iteratively updated based on the error of a real box and a prediction box to train an easy-floating object detection model after training. The method can support high-precision easy-floating object detection.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for training a model of easily floating objects along a power transmission channel. Background Technology

[0002] Easily floating objects along power transmission lines, such as plastic greenhouses, mulch films, and dust-proof green nets, are easily blown up and attached to power transmission lines under severe weather conditions such as strong winds and heavy rain. This can cause short circuits, tripping, and other faults in the power lines, seriously affecting the stable operation of the power system and creating significant safety hazards to the power transmission lines.

[0003] With the rapid development of remote sensing technology, existing technologies utilize deep learning algorithms to analyze and process remote sensing images of power transmission channels, enabling the automatic identification and detection of easily floating objects.

[0004] However, since easily floating objects such as plastic greenhouses and mulch films may present complex shapes and textures in remote sensing images, existing technologies for detecting easily floating objects in remote sensing images of power transmission channels have difficulty accurately matching target boundaries, thus resulting in low accuracy in the detection results of easily floating objects. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for training a model of easily floating objects along power transmission channels, which can improve the accuracy of easily floating object detection, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for training a model for detecting easily floating objects along a power transmission channel, including:

[0007] Acquire remote sensing images along historical power transmission channels, wherein the remote sensing images along historical power transmission channels include category labels and location information of the true bounding boxes of easily floating objects;

[0008] The remote sensing images along the historical power transmission channel are input into the constructed initial floating object detection model. The initial floating object detection model extracts floating object features at different scales from the remote sensing images along the historical power transmission channel through the floating object deep feature extraction module, and generates a floating object feature map of the power transmission channel based on the floating object features at different scales.

[0009] The easy-floating object feature attention module performs pooling and channel weighting operations on the easy-floating object feature map of the power transmission channel to capture the easy-floating object features and enhance the easy-floating object feature representation in the easy-floating object feature map of the power transmission channel, thereby obtaining the easy-floating object feature enhancement map of the power transmission channel.

[0010] Based on the features of easily floating objects at different scales and the enhanced feature map of easily floating objects in the power transmission channel, the category, location information and confidence of the prediction box are determined. The prediction box is used to identify easily floating objects in the remote sensing images along the historical power transmission channel. The initial easily floating object detection model includes an easily floating object deep feature extraction module and an easily floating object feature attention module.

[0011] The error between the ground truth bounding box and the predicted bounding box is determined based on the category label and location information of the ground truth bounding box, as well as the category, location information, and confidence level of the predicted bounding box.

[0012] Based on the error, the parameters of the initial floating object detection model are iteratively updated until the preset training termination condition is met, thus obtaining the trained floating object detection model.

[0013] Secondly, this application also provides a method for detecting easily floating objects along a power transmission line, including:

[0014] Acquire remote sensing images along the power transmission corridor;

[0015] Using the remote sensing image along the power transmission channel as input, the trained floating object detection model is called to obtain the floating object detection result. The floating object detection model is trained based on the above-mentioned floating object detection model training method along the power transmission channel.

[0016] An early warning will be issued based on the detection results of the easily floating objects.

[0017] Thirdly, this application also provides a training device for a detection model of easily floating objects along a power transmission channel, comprising:

[0018] The image acquisition module is used to acquire remote sensing images along historical power transmission channels, which include category labels and location information of the true bounding boxes of easily floating objects;

[0019] The model training module is used to input the remote sensing images along the historical power transmission channel into the constructed initial floating object detection model. The initial floating object detection model extracts floating object features at different scales from the remote sensing images along the historical power transmission channel through a floating object deep feature extraction module, and generates a floating object feature map of the power transmission channel based on the floating object features at different scales. The floating object feature attention module performs pooling and channel weighting operations on the floating object feature map of the power transmission channel to capture the floating object features and enhance the floating object feature representation in the floating object feature map of the power transmission channel, resulting in a floating object feature enhancement map of the power transmission channel. Based on the floating object features at different scales and the floating object feature enhancement map of the power transmission channel, the category, location information and confidence of the prediction box are determined. The prediction box is used to identify floating objects in the remote sensing images along the historical power transmission channel. The initial floating object detection model includes a floating object deep feature extraction module and a floating object feature attention module.

[0020] An error determination module is used to determine the error between the real bounding box and the predicted bounding box based on the category label and location information of the real bounding box, as well as the category, location information and confidence level of the predicted bounding box.

[0021] The parameter update module is used to iteratively update the parameters of the initial floating object detection model until the preset training termination condition is met, thereby obtaining the trained floating object detection model.

[0022] Fourthly, this application also provides a device for detecting easily floating objects along a power transmission line, comprising:

[0023] The remote sensing image acquisition module is used to acquire remote sensing images along the power transmission corridor.

[0024] The floating object detection module is used to take the remote sensing image along the power transmission channel as input, call the trained floating object detection model, and obtain the floating object detection result. The floating object detection model is trained based on the above-mentioned floating object detection model training method along the power transmission channel.

[0025] The early warning module is used to issue early warnings based on the detection results of the easily floating objects.

[0026] Fifthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in any of the above embodiments of the method for training a model for detecting easily floating objects along a power transmission channel, and the steps in an embodiment of a method for detecting easily floating objects along a power transmission channel.

[0027] Sixthly, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in any of the above embodiments of the method for training a model for detecting easily floating objects along a power transmission channel, and the steps in an embodiment of a method for detecting easily floating objects along a power transmission channel.

[0028] Seventhly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the method for training a model for detecting easily floating objects along a power transmission channel, and the steps in an embodiment of a method for detecting easily floating objects along a power transmission channel.

[0029] The aforementioned training method, apparatus, computer equipment, computer-readable storage medium, and computer program product for detecting easily floating objects along power transmission channels consider the impact of the complex morphology and texture of easily floating objects in remote sensing images on detection accuracy. A deep feature extraction module for easily floating objects is designed to extract deep features of easily floating objects at different scales in remote sensing images. Based on these features, a feature map of easily floating objects along the power transmission channel is generated, which enhances the robustness and accuracy of the detection model. Simultaneously, considering that easily floating objects may appear as small or low-contrast targets in remote sensing images, such as thin materials like mulch film and dust-proof green netting, and are easily obscured by trees and buildings or confused with the background, an easily floating object feature attention module is designed. This module performs pooling and channel weighting operations on the feature map of easily floating objects along the power transmission channel, capturing and enhancing the easily floating object features and representations within the map. This results in an enhanced feature map of easily floating objects along the power transmission channel, which improves the model's detection accuracy and reduces the possibility of missed detections. Based on the features of easily floating objects at different scales and the enhanced feature map of easily floating objects along the power transmission channel, the category, location information, and confidence level of the predicted bounding box are determined. Then, based on the category label and location information of the ground truth bounding box, and the category, location information, and confidence level of the predicted bounding box, the error between the ground truth bounding box and the predicted bounding box is determined. The model parameters are iteratively updated based on the error until a preset training termination condition is met, resulting in a trained easily floating object detection model. Using this trained easily floating object detection model improves the detection accuracy of easily floating objects in different scenarios along power transmission channels, and enhances the model's adaptability to high-precision easily floating object detection in various complex scenarios.

[0030] The aforementioned method, device, computer equipment, computer scale storage medium, and computer program product for detecting easily floating objects along power transmission channels acquire remote sensing images along the transmission channel and input them into a trained easily floating object detection model obtained through the aforementioned easily floating object detection model training method. This yields easily floating object detection results, which are then used to issue early warnings. On the one hand, the model-based detection method improves the accuracy and efficiency of easily floating object detection; on the other hand, the rapid and accurate detection results of easily floating objects along the transmission channel based on the model facilitate the timely discovery of potential hazards in the power system, enabling corresponding measures to be taken to reduce the possibility of accidents and improve the stability and safety of power system operation. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a diagram illustrating the application environment of a training method for a detection model of easily floating objects along a power transmission channel in one embodiment.

[0033] Figure 2 This is a flowchart illustrating a method for training a model to detect easily floating objects along a power transmission channel in one embodiment.

[0034] Figure 3 This is a structural block diagram of a model for detecting easily floating objects along a power transmission channel in one embodiment;

[0035] Figure 4 This is a structural block diagram of a deep feature extraction module for easily floating objects in one embodiment;

[0036] Figure 5 This is a schematic diagram of the structure of the attention module for easily floating objects in one embodiment;

[0037] Figure 6 This is a flowchart illustrating a method for detecting easily floating objects along a medium-sized power transmission line.

[0038] Figure 7 This is a structural block diagram of a training device for a detection model of easily floating objects along a power transmission channel in one embodiment.

[0039] Figure 8 This is a structural block diagram of a device for detecting easily floating objects along a power transmission channel in one embodiment;

[0040] Figure 9This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0042] The method for training a detection model for easily floating objects along power transmission lines provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0043] Specifically, an operator can upload historical remote sensing images along the power transmission channel to server 104 via terminal 102, and then send model training messages to server 104 via terminal 102. Server 104 acquires the remote sensing images along the power transmission channel and inputs them into a pre-constructed initial floating object detection model. The initial floating object detection model extracts floating object features at different scales from the historical power transmission channel remote sensing images using a deep floating object feature extraction module. Based on these features, a floating object feature map of the power transmission channel is generated. Then, a floating object feature attention module performs pooling and channel weighting operations on the floating object feature map to capture the floating object features in the map. The system generates a feature enhancement map of floating objects along a power transmission channel by using features of floating objects at different scales and the enhanced map. Then, based on these features and the enhanced map, the category, location information, and confidence level of predicted bounding boxes are determined. These predicted bounding boxes are used to identify floating objects in remote sensing images along the historical power transmission channel. The initial floating object detection model includes a deep feature extraction module and a feature attention module. Finally, based on the category label and location information of the ground truth bounding boxes, and the category, location information, and confidence level of the predicted bounding boxes, the error between the ground truth bounding boxes and the predicted bounding boxes is determined. Based on this error, the parameters of the initial floating object detection model are iteratively updated until a preset training termination condition is met, resulting in a trained floating object detection model.

[0044] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0045] In one exemplary embodiment, such as Figure 2 As shown, a method for training a detection model for easily floating objects along a power transmission channel is provided, and this method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S100 to S600. Among them:

[0046] S100, acquire remote sensing images along the historical power transmission channel, the remote sensing images along the historical power transmission channel containing category labels and location information of the true bounding boxes of easily floating objects.

[0047] The historical remote sensing images along the power transmission corridor can be images acquired during a historical period of time along the corridor. Easily floating objects along the transmission corridor refer to objects that are easily moved or floated by wind or other natural forces and may pose a threat to the transmission lines. Easily floating objects include, but are not limited to, plastic greenhouses, mulch films, and dust-proof green netting. Contact with power lines by easily floating objects may cause short circuits, tripping, and other power outages, affecting the stable operation and safety of the power system. The category label for the ground truth bounding box can include easily floating objects and the background. Location information can include the coordinates of the top-left and bottom-right corners of the ground truth bounding box.

[0048] In practical applications, images of the area along the power transmission corridor can be acquired using UAV remote sensing. Obtaining historical remote sensing images of the power transmission corridor can involve acquiring multiple historical images of the corridor containing easily floating objects, collected by UAVs within a historical timeframe. To facilitate model processing, the dimensions of these remote sensing images can be standardized. Specifically, the historical remote sensing images along the power transmission corridor are cropped and resized to ensure uniform image size. For these historical remote sensing images, the locations of easily floating objects are selected using a labeling tool, i.e., the coordinates of the top-left and bottom-right vertices of the box are determined, and category labels are used for annotation.

[0049] In other implementations, to enhance model performance, data augmentation processing is performed on remote sensing images along historical power transmission corridors. Data augmentation includes, but is not limited to, rotation, flipping, scaling and cropping, brightness adjustment, and noise addition. After data augmentation, the augmented remote sensing images along historical power transmission corridors can be divided into training, validation, and test sets in an 8:1:1 ratio. The initial buoyancy detection model is trained using the remote sensing images along historical power transmission corridors from the training set.

[0050] S200, the remote sensing images along the historical power transmission channel are input into the constructed initial floating object detection model. The initial floating object detection model extracts floating object features at different scales from the remote sensing images along the historical power transmission channel through the floating object deep feature extraction module, and generates a floating object feature map of the power transmission channel based on the floating object features at different scales.

[0051] The initial floating object detection model is an improvement upon the pre-trained model. The deep feature extraction module for floating objects includes multiple feature processing branches for extracting deep features of floating objects.

[0052] In practical applications, to improve the accuracy of capturing the features of easily floating objects in remote sensing images, this application designs a deep feature extraction module for easily floating objects, comprising multiple feature processing branches, to extract features of easily floating objects in remote sensing images across different dimensions. Through this designed deep feature extraction module, the architecture of a pre-trained model (such as the YOLO model) is improved, resulting in an initial easily floating object detection model. Using historical remote sensing images along power transmission lines as input, multi-dimensional feature extraction and feature fusion operations are performed through multiple feature processing branches to extract easily floating object features at different scales from the historical power transmission line remote sensing images. Feature enhancement, further feature extraction, and feature fusion processing are then performed on these features at different scales to generate a feature map of easily floating objects along the power transmission line.

[0053] S300, the easy-floating object feature attention module performs pooling and channel weighting operations on the easy-floating object feature map of the power transmission channel to capture the easy-floating object features and enhance the easy-floating object feature representation in the easy-floating object feature map of the power transmission channel, and obtains the easy-floating object feature enhancement map of the power transmission channel.

[0054] Among them, the easy-to-float object feature attention module is used to enhance the feature representation of easy-to-float objects.

[0055] In practical applications, considering that easily floating objects may appear as small or low-contrast targets in remote sensing images, such as thin materials like mulch film and dust-proof green netting, and are easily obscured by trees and buildings or confused with the background, an easily floating object feature attention module is designed. This module improves the architecture of the pre-trained model (such as the YOLO model) through a deep feature extraction module and the easily floating object feature attention module, resulting in an initial easily floating object detection model. For example... Figure 3 As shown. Initialize all neural network parameters of the initial floating object detection model, and set relevant hyperparameters such as training epochs, batch size, optimizer selection, and learning rate. Use a floating object feature attention module to perform pooling operations (such as global pooling and max pooling) on ​​the floating object feature map of the transmission channel to capture the floating object features of the input feature map. Enhance the floating object feature representation through channel weighting operations to obtain the enhanced floating object feature map of the transmission channel.

[0056] S400, based on the features of easily floating objects at different scales and the enhanced feature map of easily floating objects in the power transmission channel, determine the category, location information and confidence of the prediction box. The prediction box is used to identify easily floating objects in the remote sensing images along the historical power transmission channel. The initial easily floating object detection model includes an easily floating object deep feature extraction module and an easily floating object feature attention module.

[0057] In practical applications, feature extraction and feature fusion are performed on features of easily floating objects at different scales and enhanced maps of easily floating objects in power transmission channels to generate prediction boxes that identify the location of easily floating objects in remote sensing images. The prediction boxes are then output with their category, location information, and confidence level.

[0058] S500, based on the category label and location information of the ground truth bounding box, and the category, location information and confidence level of the predicted bounding box, determine the error between the ground truth bounding box and the predicted bounding box.

[0059] In practical applications, the cross-union ratio (CUP) between the predicted and ground truth bounding boxes can be determined based on their location information. The error between the predicted and ground truth bounding boxes can then be determined by calculating the difference between their locations, the difference between their categories, the confidence level, and the CUP using a loss function. Loss functions include, but are not limited to, mean squared error loss, cross-entropy loss, and smoothed L1 loss.

[0060] S600, based on the error, iteratively update the parameters of the initial floating object detection model until the preset training termination condition is met, and obtain the trained floating object detection model.

[0061] In practical applications, after obtaining the error between the ground truth bounding box and the predicted bounding box using the loss function, backpropagation is performed starting from the output layer of the initial floating object detection model based on the loss function value (error). The parameter gradient values ​​of each layer are calculated, and the parameters of the initial floating object detection model are updated to minimize the loss function value. A preset training termination condition can be set such that the loss function value is continuously less than a preset loss threshold within a preset number of iterations, at which point training of the initial floating object detection model is stopped, resulting in the trained floating object detection model.

[0062] In other implementations, after completing one round of training on the images in the training set, the model performance is verified using historical remote sensing images of power transmission channels in the validation set. The training strategy is then adjusted based on the validation results, such as terminating training early or adjusting the learning rate.

[0063] In the training method for the above-mentioned floating object detection model along the power transmission channel, considering the impact of the complex morphology and texture of floating objects in remote sensing images on detection accuracy, a deep feature extraction module for floating objects is designed. This module extracts deep features of floating objects at different scales from the remote sensing images and generates a feature map of floating objects along the power transmission channel based on these features. This enhances the robustness and accuracy of the floating object detection model. Simultaneously, considering that floating objects may appear as small or low-contrast targets in remote sensing images, such as thin materials like mulch film and dust-proof green netting, and are easily obscured by trees and buildings or confused with the background, a floating object feature attention module is designed. This module performs pooling and channel weighting operations on the feature map of floating objects along the power transmission channel, capturing and enhancing the floating object features and representations within the map. This results in an enhanced feature map of floating objects along the power transmission channel, which improves the model's detection accuracy and reduces the possibility of missed detections. Based on the features of easily floating objects at different scales and the enhanced feature map of easily floating objects along the power transmission channel, the category, location information, and confidence level of the predicted bounding box are determined. Then, based on the category label and location information of the ground truth bounding box, and the category, location information, and confidence level of the predicted bounding box, the error between the ground truth bounding box and the predicted bounding box is determined. The model parameters are iteratively updated based on the error until a preset training termination condition is met, resulting in a trained easily floating object detection model. Using this trained easily floating object detection model improves the detection accuracy of easily floating objects in different scenarios along power transmission channels, and enhances the model's adaptability to high-precision easily floating object detection in various complex scenarios.

[0064] In one exemplary embodiment, the initial floating object detection model includes a feature extraction module, such as... Figure 3 As shown, S200 includes S220 to S260. Wherein:

[0065] S220: The feature extraction module extracts features from remote sensing images along the historical power transmission channel to obtain the first feature map.

[0066] The first feature map is the feature map obtained by the feature extraction module through feature extraction operations on remote sensing images along the historical power transmission channel.

[0067] In this embodiment, the feature extraction module is the Conv_BN_ReLu module (i.e. Figure 3 The Conv_BN_ReLu feature extraction module (in the context of the project). In practical applications, the Conv_BN_ReLu module extracts features from the input historical remote sensing images along the power transmission channel using a sliding convolution kernel, generating a first feature map.

[0068] S240, through the feature dimensionality reduction branch, multi-scale feature fusion branch and feature pooling branch in the deep feature extraction module for easily floating objects, convolution operation, channel weighting operation and pooling operation are performed on the first feature map respectively to obtain the second feature map.

[0069] The feature dimensionality reduction branch can include multiple convolutional modules to reduce the dimensionality of the input feature map. The multi-scale feature fusion branch can include pooling layers and feature fusion modules to fuse features at multiple scales. The feature pooling branch can include different pooling layers (such as max pooling and average pooling). The second feature map is the feature map obtained by the deep feature extraction module for easily floating objects from the first feature map.

[0070] In practical applications, the first feature map can be input into modules in the feature dimensionality reduction branch, the multi-scale feature fusion branch, and the feature pooling branch respectively: the first feature map is convolved by multiple convolution modules in the feature dimensionality reduction branch; the first feature map is pooled by pooling layers in the multi-scale feature fusion branch; the feature map obtained after pooling is fused with the feature map obtained by convolution in the feature dimensionality reduction branch; the first feature map is pooled by different pooling layers in the feature pooling branch; and the feature maps obtained from different feature processing branches are fused to obtain the second feature map.

[0071] S260, based on the feature extraction module and the deep feature extraction module for easily floating objects, extracts features of the second feature map at different scales, and obtains the feature map of easily floating objects in the power transmission channel based on the features at different scales.

[0072] In practical applications, the feature extraction module (Conv_BN_ReLu module) and the deep feature extraction module for easily floating objects can be stacked multiple times in sequence. Through multiple convolution operations of the stacked feature extraction module and the deep feature extraction module for easily floating objects, features of the second feature map at different scales are extracted. Convolution operations and feature fusion operations are then performed on the features at different scales to obtain the feature map of easily floating objects in the power transmission channel.

[0073] In this embodiment, the input feature map is processed by feature dimensionality reduction branch, multi-scale feature fusion branch and feature pooling branch, which helps to enhance the robustness of the model and the ability to represent features of easily floating objects, thereby improving the detection accuracy of easily floating objects.

[0074] In an exemplary embodiment, S240 includes S241 to S245. Wherein:

[0075] S241, the first feature map is obtained by dilated convolution of the first feature map through the dilated convolution layer in the deep feature extraction module for easily floating objects.

[0076] Among them, the first deep feature map is the feature map obtained by performing a dilated convolution operation on the first deep feature map by the dilated convolution layer in the deep feature extraction module for easily floating objects.

[0077] In practical applications, this application designs an Easily Floating Objects Deep-feature Extraction module (EFODE). The structure of the EFODE module is as follows: Figure 4 As shown, the first feature map Q1 is input into a dilated convolutional layer with a kernel size of 3×3 and a dilation rate of 7. The dilated convolutional layer performs dilated convolution on the first feature map to expand the receptive field and capture a wider range of contextual information, thus obtaining the first deep feature map Q2.

[0078] S242, by performing convolution, batch normalization, nonlinear processing and convolution operations sequentially on the first deep feature map through the feature dimensionality reduction branch, the second deep feature map is obtained.

[0079] The feature dimensionality reduction branch includes concatenated convolutions and 1×1 convolutional layers, and batch normalization layers. Figure 4 The convolutional layers consist of BN batch normalization, ReLU activation function, and a kernel size of 1×1.

[0080] In practical applications, the first deep feature map Q2 is subjected to convolutional operations, batch normalization operations, nonlinear processing, and convolutional operations in sequence through the concatenated convolutional layers, batch normalization layers, ReLU activation functions, and convolutional layers in the feature dimensionality reduction branch, so as to achieve feature dimensionality reduction and nonlinear enhancement, and obtain the second deep feature map Q3.

[0081] S243, the first feature map is pooled through a multi-scale feature fusion branch to obtain the third deep feature map. The third and second deep feature maps are then channel-weighted through a convolutional layer to obtain the fourth deep feature map. Finally, the fourth deep feature map is convolved to obtain the fifth deep feature map.

[0082] The multi-scale feature fusion branch includes a max pooling layer, a channel weighting module, and a convolutional layer with a kernel size of 1×1.

[0083] In practical applications, the input first feature map Q2 is processed through a multi-scale feature fusion branch. First, the first feature map Q2 is pooled through a max pooling layer to obtain the third deep feature map Q4. Channel weighting is performed on the second deep feature map Q3 and the third feature map Q4 to obtain the fourth deep feature map Q5. Then, a convolution operation with a kernel size of 1×1 is performed on the fourth deep feature map to obtain the fifth deep feature map Q6.

[0084] S244, the first feature map is subjected to max pooling and average pooling operations through the feature pooling branch. The results of the max pooling and average pooling operations are fused to obtain the sixth deep feature map. The sixth deep feature map is then subjected to channel weighting to obtain the seventh deep feature map.

[0085] The feature pooling branch includes a cascaded max pooling layer and a ReLU activation function, a cascaded average pooling layer and a ReLU activation function, a feature fusion module, and a channel weighting module.

[0086] In practical applications, the first feature map Q2 is pooled and nonlinearly processed by a series of max pooling and ReLU activation functions in the feature pooling branch, and then average pooling and nonlinearly processed by a series of average pooling layers and ReLU activation functions. The results of the two processes are then fused along the channel dimension to obtain the sixth deep feature map Q7. In this way, the feature representation is enriched by multiple pooling methods. The fifth deep feature map Q6 and the sixth deep feature map Q7 are then subjected to channel weighting to obtain the seventh deep feature map Q8.

[0087] S245, perform feature fusion and convolution operations on the second deep feature map and the seventh deep feature map to obtain the second feature map.

[0088] In practical applications, the second and seventh deep feature maps are input into a cascaded feature fusion module and a 1×1 convolutional module for feature fusion and convolution operations, resulting in the second feature map Q9 (i.e., ...). Figure 3 Feature map X4 in the middle.

[0089] In this embodiment, a feature fusion and weighting method is designed and applied to the deep feature extraction module for easily floating objects. Dilated convolution effectively expands the receptive field, capturing a wide range of contextual information. The feature dimensionality reduction branch utilizes convolution for feature dimensionality reduction and nonlinear enhancement; the multi-scale feature fusion branch fuses feature maps of easily floating objects at different scales to enhance model robustness; and the feature pooling branch introduces various pooling methods to enrich feature representation. Furthermore, channel weighting and fusion operations in the deep feature extraction module for easily floating objects further improve the feature extraction capability.

[0090] In an exemplary embodiment, there are multiple deep feature extraction modules and feature extraction modules for easily floating objects, and S260 includes S261 to S265. Wherein:

[0091] S261 connects the output of each feature extraction module to the input of each easy-to-float deep feature extraction module to obtain the first feature extraction combination module, the second feature extraction combination module and the third feature extraction combination module.

[0092] S262, the first feature extraction combination module extracts features from the third feature map to obtain a first-scale feature map, the second feature extraction combination module extracts features from the first-scale feature map to obtain a second-scale feature map, and the third feature extraction combination module extracts features from the second-scale feature map to obtain a third-scale feature map.

[0093] S263 uses the deep feature extraction module for easily floating objects to extract features from the third-scale feature map, and performs pooling operations on the extracted features to obtain the fourth-scale feature map.

[0094] S264, by upsampling the fourth-scale feature map, the upsampled feature map is fused with the second-scale feature map to obtain a feature fusion map, and the feature is extracted by the easy-to-float deep feature extraction module to obtain the fifth-scale feature map.

[0095] S265, perform an upsampling operation on the fifth-scale feature map, and fuse the upsampled feature map with the first-scale feature map to obtain the feature map of easily floating objects in the power transmission channel.

[0096] In practical applications, such as Figure 3As shown, in the initial floating object detection model, the output of the feature extraction module is connected to the input of the floating object feature extraction module, and the feature extraction module and the floating object feature extraction module are stacked three times to obtain the first feature extraction combination module, the second feature extraction combination module and the third feature extraction combination module respectively.

[0097] The input is the second feature map ( Figure 3 The first feature extraction module extracts features from feature map X4 (in the first feature extraction module), outputting feature map X5. Features of feature map X5 are then extracted using the deep feature extraction module for easily floating objects within the first feature extraction module, resulting in a first-scale feature map X6. Features of the first-scale feature map X6 are extracted using the feature extraction module in the second feature extraction module, outputting feature map X7. Feature map X7 is then input to the deep feature extraction module for easily floating objects to reduce spatial dimensions, resulting in a second-scale feature map X8. Features of the second-scale feature map X8 are extracted using the feature extraction module in the third feature extraction module, outputting feature map X9. Features of feature map X9 are then extracted using the deep feature extraction module for easily floating objects within the third feature extraction module to reduce spatial dimensions, resulting in a third-scale feature map X10. In this embodiment, the steps for feature extraction from the feature map using the deep feature extraction module for easily floating objects are the same as those in the previous embodiment, used to extract features from the first feature map to obtain the second feature map in advance; therefore, they will not be repeated here.

[0098] The third-scale feature map X10 is input into the deep feature extraction module for easily floating objects to extract features, resulting in feature map X11. Feature map X11 is then input into the SPPF pooling layer (Spatial Pyramid Pooling - Fast) for fast pooling operation, resulting in the fourth-scale feature map X12.

[0099] Upsampling is performed on the fourth-scale feature map X12 to obtain feature map X13. Feature map X13 is then fused with the second-scale feature map X8 to obtain feature fusion map X14. Feature fusion map X14 is then input into the deep feature extraction module for easily floating objects to extract features, resulting in the fifth-scale feature map X15.

[0100] Upsampling is performed on the fifth-scale feature map X15 to obtain feature map X16. Feature map X16 is then fused with the first-scale feature map X6 at the channel dimension to obtain the transmission channel feature map. Figure 3 Feature map X17 in the middle.

[0101] In this embodiment, the extraction of features from the second feature map at different scales by the feature extraction combination module is beneficial to improving the detection accuracy and robustness of the module.

[0102] In an exemplary embodiment, the easily floating object feature attention module includes a max pooling layer and an average pooling layer, and S300 includes S310 to S350. Wherein:

[0103] S310, the feature map of easily floating objects in the transmission channel is pooled through the maximum pooling layer and the average pooling layer respectively to obtain the first attention feature map and the second attention feature map.

[0104] S320 performs dilated convolution operations of different scales on the feature maps of easily floating objects in the power transmission channel through the easily floating object feature attention module to obtain the third attention feature map, the fourth attention feature map, and the fifth attention feature map.

[0105] S330, perform channel weighting on the first attention feature map and the third attention feature map to obtain the sixth attention feature map.

[0106] S340, perform channel weighting on the second and fifth attention feature maps to obtain the seventh attention feature map.

[0107] S350 performs feature fusion and feature extraction on the fourth, sixth, and seventh attention feature maps to obtain the eighth attention feature map.

[0108] To reduce background interference and enhance the features of easily floating objects, this application designs an EFO (Easily Floating Objects Feature Attention) module. Its structure is as follows: Figure 5 As shown. In Figure 5 In the middle, the input is the feature map S1 of easily floating objects in the power transmission channel ( Figure 3 Feature map X17 in the middle.

[0109] First, the easy-to-float feature map S1 of the transmission channel is pooled using a global max pooling layer to obtain the first attention feature map S2. Then, the easy-to-float feature map S1 of the transmission channel is pooled using a global average pooling layer to obtain the second attention feature map S3.

[0110] Secondly, the feature map S1 of easily floating objects in the power transmission channel is subjected to dilated convolution with a kernel size of 3×3 and a dilation rate of 3, followed by batch normalization and ReLU activation to obtain the third attention feature map S4; the feature map S1 of easily floating objects in the power transmission channel is subjected to dilated convolution with a kernel size of 5×5 and a dilation rate of 5, followed by batch normalization and ReLU activation to obtain the third attention feature map S6; the feature map S1 of easily floating objects in the power transmission channel is subjected to dilated convolution with a kernel size of 9×9 and a dilation rate of 9, followed by batch normalization and ReLU activation to obtain the third attention feature map S7.

[0111] Next, channel weighting is performed on the first attention feature map S2 and the third attention feature map S4 to obtain the sixth attention feature map S5.

[0112] Then, channel weighting is performed on the second attention feature map S3 and the fifth attention feature map S7 to obtain the seventh attention feature map S8.

[0113] Finally, the fourth attention feature map S6, the sixth attention feature map S5, and the seventh attention feature map S8 are fused along the channel dimension, and then convolutional operations with a 1×1 kernel are performed to obtain the enhanced feature map S9 of the power transmission channel's easily floating objects.

[0114] In this embodiment, the most salient and average features of the input feature map are captured by global max pooling and global average pooling; multi-scale dilated convolution is used, combining convolution kernels of different sizes and dilation rates to expand the receptive field and fuse contextual information of easily floating objects; through channel weighting operations, combined with global pooling features and dilated convolution features, a deep feature map is generated to enhance the model's sensitivity to key features of easily floating objects; finally, the multi-scale features are fused in the channel dimension and integrated through convolution to obtain the final enhanced feature map of easily floating objects in the power transmission channel, which helps to reduce background interference, highlight target features, and thus reduce the possibility of missed detections and false detections by the model.

[0115] In an exemplary embodiment, the determination of the prediction box category, location information, and confidence level based on the different scales of the floating object features and the enhanced map of the floating object features of the transmission channel includes steps S420 to S460. Wherein:

[0116] S420: The features of the enhanced feature map of the easily floating object in the power transmission channel are extracted by the feature extraction module in the initial easily floating object detection model to obtain the first target feature map. The features of the first target feature map and the fifth size feature map are fused to obtain the second target feature map. The feature representation of the second target feature map is enhanced by the easily floating object feature attention module to obtain the first scale target feature fusion map.

[0117] In practical applications, such as Figure 3As shown, the enhanced feature map of easily floating objects in the transmission channel is input to the feature extraction module for feature extraction, outputting a first target feature map X19. The first target feature map X19 and the fifth-size feature map X15 are fused to obtain a second target feature map X20. The second target feature map X20 is input to the easily floating object feature attention module for feature enhancement processing to obtain a first-scale target feature fusion map X21. In this embodiment, the step of performing feature enhancement processing on the input feature map through the easily floating object feature attention module refers to the above-described step of performing pooling and channel weighting operations on the easily floating object feature map of the transmission channel through the easily floating object feature attention module to capture the easily floating object features and enhance the easily floating object feature representation in the easily floating object feature map of the transmission channel, and obtain the enhanced feature map of easily floating objects in the transmission channel. This will not be repeated here.

[0118] S440, the features of the first-scale target feature fusion map are extracted by the feature fusion module in the initial floating object detection model to obtain the third target feature map. The features of the third target feature map and the fourth-scale feature map are fused to obtain the fourth target feature map. The feature representation of the fourth target feature map is enhanced by the floating object feature attention module to obtain the second-scale target feature fusion map.

[0119] In practical applications, the first-scale target feature fusion map X21 is input into the feature fusion module for feature extraction to obtain the third target feature map X22. The third target feature map X22 and the fourth-scale feature map X12 are fused along the channel dimension to obtain the fourth target feature map X23. The enhanced fourth target feature map X23 is input into the easily floating object feature attention module for feature enhancement processing to obtain the second-scale target feature fusion map X24.

[0120] S460: Input the first-scale target feature fusion map, the second-scale target feature fusion map, and the enhanced feature map of easily floating objects in the transmission channel into the target detection head of the initial easily floating object detection model to generate multiple prediction boxes and output the category, location information, and confidence of each prediction box.

[0121] In practical applications, the first-scale target feature fusion map, the second-scale target feature fusion map, and the enhanced feature map of easily floating objects in the power transmission channel are respectively input into the target prediction head of the initial easily floating object detection model. Multiple prediction boxes for identifying easily floating objects are generated through the target prediction head. The category of each prediction box is predicted, and the category, location information, and confidence level of the prediction box are output.

[0122] In this embodiment, the first-scale target feature fusion map, the second-scale target feature fusion map, and the enhanced map of easily floating objects in the power transmission channel are extracted by the easily floating object feature attention module and the feature extraction module. In this way, the detection head outputs the category, location information and confidence of the prediction box, thereby improving the detection accuracy.

[0123] To measure the quality of model predictions, in an exemplary embodiment, S500 includes S510 to S550. Wherein:

[0124] S510: Based on the location information of the ground truth bounding boxes and the location information of the predicted bounding boxes, determine the intersection-union ratio (IUU) between the ground truth bounding boxes and the predicted bounding boxes.

[0125] In practical applications, the intersection-union ratio (IUR) of the ground truth bounding boxes and the predicted bounding boxes is calculated based on their location information to measure the degree of overlap between them. The IUR is the ratio of the intersection area to the union area of ​​the ground truth and predicted bounding boxes; a higher IUR indicates a greater degree of overlap.

[0126] S520 determines the predicted box loss value between the ground truth box and the predicted box based on the cross-union ratio.

[0127] Among them, the predicted box loss value represents the positional difference between the predicted box and the ground truth box.

[0128] In practical applications, the predicted bounding box loss value between the ground truth bounding box and the predicted bounding box can be obtained based on a preset loss function and cross-union ratio (CUI). Preset loss functions include, but are not limited to, mean squared error loss function, cross-entropy loss function, and smoothed L1 loss function.

[0129] S530 determines the class loss value between the ground truth bounding box and the predicted bounding box based on the class label of the ground truth bounding box, the class of the predicted bounding box, and the preset loss function.

[0130] Among them, the category loss value is used to measure the difference between the predicted box category distribution and the actual category.

[0131] In practical applications, obtaining the class labels of the ground truth boxes and the class probability distribution of the predicted boxes can be achieved by using a sigmoid function combined with a binary cross-entropy loss function to determine the class loss value between the ground truth boxes and the predicted boxes.

[0132] S540, determine the confidence loss value between the ground truth box and the predicted box based on the intersection-union ratio of the ground truth box and the predicted box, and the confidence of the predicted box.

[0133] The confidence loss value is used to measure the model's prediction error regarding the presence of easily floating objects in the prediction box.

[0134] In practical applications, the intersection-union ratio of the ground truth bounding box and the predicted bounding box can be used as the true confidence level. The confidence loss value between the ground truth bounding box and the predicted bounding box can be determined based on the cross-entropy loss function, the true confidence level, and the confidence level of the predicted bounding box.

[0135] S550 determines the error between the predicted box and the true box based on the predicted box loss value, the category loss value, and the confidence loss value.

[0136] In practical applications, weights can be assigned to the predicted box loss value, the category loss value, and the confidence loss value respectively, and the loss value (error) between the predicted box and the ground truth box can be obtained by weighted summation.

[0137] In this embodiment, the error between the model's prediction result and the true label is determined by the predicted box loss value, class loss value, and confidence loss value between the ground truth box and the predicted box. This helps to guide the adjustment of model parameters based on the error to minimize the error.

[0138] In an exemplary embodiment, S520 includes S522 to S524. Wherein:

[0139] S522, determine the minimum bounding box between the ground truth box and the predicted box.

[0140] In practical applications, the smallest bounding rectangle containing both the ground truth bounding box and the predicted bounding box is defined as the minimum bounding box.

[0141] S524: Based on the size of the minimum bounding box, the size of the ground truth box, and the size of the predicted box, determine the weighting factor, and determine the predicted box loss value between the ground truth box and the predicted box according to the cross-union ratio and the weighting factor.

[0142] Considering that easily floating objects account for a small proportion of the acquired remote sensing images, while background pixels cause significant interference, this application specifically designs an improved cross-union loss function. The loss function introduces weighting factors that are closely related to the size of the ground truth and predicted bounding boxes. These factors take into account the minimum bounding box as well as the width and height of the ground truth and predicted bounding boxes, thereby achieving a significant improvement in the detection accuracy of easily floating targets.

[0143] The loss function is defined as follows:

[0144]

[0145] In the formula, This represents the intersection-union ratio (IoU) between the predicted bounding box and the ground truth bounding box. , and These represent the predicted bounding box and the ground truth bounding box, respectively. and These represent the width and height of the actual bounding box, respectively. and These represent the width and height of the prediction box, respectively; and These represent the width and height of the smallest bounding box composed of the predicted box and the ground truth box, respectively.

[0146] In this embodiment, the calculation method of the cross-union ratio loss function is improved by introducing a weight factor that is closely related to the size of the ground truth box and the predicted box, thereby improving the accuracy of calculating the predicted box loss value and thus improving the detection accuracy of the model.

[0147] To provide a clearer explanation of the training method for the detection model of easily floating objects along the power transmission channel provided in this application, a specific embodiment and appendix are described below. Figure 6 The specific embodiment includes the following steps:

[0148] S1. Acquire remote sensing images along the historical power transmission corridor. These images include category labels and location information for the true bounding boxes of easily floating objects.

[0149] S2, input the remote sensing images along the historical power transmission channel into the constructed initial floating object detection model, and extract the features of the remote sensing images along the historical power transmission channel through the feature extraction module to obtain the first feature map.

[0150] S3, through the dilated convolutional layer in the deep feature extraction module for easily floating objects, dilated convolution is performed on the first feature map to obtain the first deep feature map. Through the feature dimensionality reduction branch, convolution, batch normalization, nonlinear processing, and convolution are performed sequentially on the first deep feature map to obtain the second deep feature map. Through the multi-scale feature fusion branch, pooling is performed on the first feature map to obtain the third deep feature map. Through the convolutional layer, channel weighting is performed on the third and second deep feature maps to obtain the fourth deep feature map. Through the feature pooling branch, max pooling and average pooling are performed on the first feature map respectively. The results of max pooling and average pooling are fused to obtain the sixth deep feature map. Channel weighting is performed on the sixth deep feature map to obtain the seventh deep feature map. Feature fusion and convolution are performed on the second and seventh deep feature maps to obtain the second feature map.

[0151] S4 connects the output of each feature extraction module to the input of each easily floating object deep feature extraction module to obtain a first feature extraction combination module, a second feature extraction combination module, and a third feature extraction combination module. The first feature extraction combination module extracts features from the third feature map to obtain a first-scale feature map. The second feature extraction combination module extracts features from the first-scale feature map to obtain a second-scale feature map. The third feature extraction combination module extracts features from the second-scale feature map to obtain a third-scale feature map. The easily floating object deep feature extraction module extracts features from the third-scale feature map and performs pooling operations on the extracted features to obtain a fourth-scale feature map. The fourth-scale feature map is upsampled and then fused with the second-scale feature map to obtain a feature fusion map. The easily floating object deep feature extraction module extracts features from the fifth-scale feature map and then upsamples the fifth-scale feature map and fuses it with the first-scale feature map to obtain the easily floating object feature map of the power transmission channel.

[0152] S5, pooling operations are performed on the easily floating object feature map of the transmission channel through max pooling layer and average pooling layer respectively to obtain the first attention feature map and the second attention feature map. The easily floating object feature attention module is used to perform dilated convolution operations of different scales on the easily floating object feature map of the transmission channel to obtain the third attention feature map, the fourth attention feature map and the fifth attention feature map. Channel weighting operation is performed on the first attention feature map and the third attention feature map to obtain the sixth attention feature map. Channel weighting operation is performed on the second attention feature map and the fifth attention feature map to obtain the seventh attention feature map. Feature fusion and feature extraction are performed on the sixth attention feature map and the seventh attention feature map to obtain the eighth attention feature map.

[0153] S6. The features of the enhanced feature map of easily floating objects in the transmission channel are extracted by the feature extraction module in the initial easily floating object detection model to obtain the first target feature map. The features of the first target feature map and the fifth-size feature map are fused to obtain the second target feature map. The feature representation of the second target feature map is enhanced by the easily floating object feature attention module to obtain the first-scale target feature fusion map. The features of the first-scale target feature fusion map are extracted by the feature fusion module in the initial easily floating object detection model to obtain the third target feature map. The features of the third target feature map and the fourth-scale feature map are fused to obtain the fourth target feature map. The feature representation of the fourth target feature map is enhanced by the easily floating object feature attention module to obtain the second-scale target feature fusion map. The first-scale target feature fusion map, the second-scale target feature fusion map and the enhanced feature map of easily floating objects in the transmission channel are input into the target detection head of the initial easily floating object detection model to generate multiple prediction boxes. The category, location information and confidence of each prediction box are output.

[0154] S7. Based on the location information of the ground truth boxes and the location information of the predicted boxes, determine the intersection-union ratio (IU / R) between the ground truth boxes and the predicted boxes, determine the minimum bounding box between the ground truth boxes and the predicted boxes, determine the weighting factor based on the size of the minimum bounding box, the size of the ground truth boxes, and the size of the predicted boxes, determine the predicted box loss value between the ground truth boxes and the predicted boxes based on the IU / R and the weighting factor, determine the class loss value between the ground truth boxes and the predicted boxes based on the class label of the ground truth boxes, the class of the predicted boxes, and the preset loss function, determine the confidence loss value between the ground truth boxes and the predicted boxes based on the IU / R and the confidence of the predicted boxes, and determine the error between the predicted boxes and the ground truth boxes based on the predicted box loss value, the class loss value, and the confidence loss value.

[0155] In practical applications, the feature map size is set to H×W×C, where H is the height of the feature map, W is the width of the feature map, and C is the number of channels of the feature map.

[0156] like Figure 3 As shown, the remote sensing image X1 of easily floating objects along the power transmission channel is used as the input to the remote sensing detection model for easily floating objects along the power transmission channel. The size of X1 is 512×512×3. First, the Conv_BN_ReLu feature extraction module is applied to process X1 to obtain feature map X2, which has a size of 512×512×64. Then, the Conv_BN_ReLu feature extraction module is applied to X2 for convolution, batch normalization, and activation operations to obtain the first feature map X3, which has a size of 128×128×128. Next, the EFODE module is applied to process the first feature map X3 to obtain the second feature map X4, which has a size of 128×128×256. Then, the Conv_BN_ReLu extraction module is applied to process the second feature map X4 to obtain feature map X5, which has a size of 64×64×512. X5 is then input into the EFODE module. The first feature map X6, with a size of 64×64×512, is obtained by processing the blocks. Then, the Conv_BN_ReLu feature extraction module is used to perform convolution, batch normalization, and activation operations on X6 to obtain feature map X7, with a size of 32×32×512. X7 is then input into the EFODE module for processing to obtain the second feature map X8, with a size of 32×32×512. Finally, the Conv_BN_ReLu module is used to perform convolution, batch normalization, and activation operations on X8 to obtain feature map X9, with a size of 16×16×512. X9 is then input into the EFODE module for processing to obtain the third feature map X10, with a size of 16×16×512.

[0157] The EFODE module is applied to process the third-scale feature map X10, resulting in feature map X11, which has a size of 16×16×512. Next, SPPF is applied to X11 to obtain the fourth-scale feature map X12, which also has a size of 16×16×512. After upsampling, feature map X13 is obtained, with a size of 32×32×512. Finally, the second-scale feature map X8 and X13 are fused by channel dimension to obtain the thirteenth easily floating object feature map X14, which has a size of [missing information]. The first-scale feature map X14 is 32×32×1024. Then, the sixth EFODE module is used to process the feature fusion map X14 to obtain the fifth-scale feature map X15, which is 32×32×1024 in size. X15 is then upsampled to obtain the feature map X16, which is 64×64×512 in size. Finally, the first-scale feature map X6 and the fifth-scale feature map X15 are fused along the channel dimension to obtain the power transmission channel floating object feature map X17, which is 64×64×1024 in size.

[0158] The feature map X17 of the power transmission channel prone to floating objects is input into the EFO Attention module to obtain the enhanced feature map X18 of the power transmission channel prone to floating objects, with a size of 64×64×512. Next, the Conv_BN_ReLu module is applied to X18 for convolution, batch normalization, and activation operations to obtain the first target feature map X19, with a size of 32×32×512. X19 is then fused with the fifth-size feature map X15 along the channel dimension to obtain the second target feature map X20, with a size of 32×32×1536. Subsequently, EFO is applied... The Attention module processes X20 to obtain the first-scale target feature fusion map X21, which has a size of 32×32×512. The Conv_BN_ReLu module then performs convolution, batch normalization, and activation operations on X21 to obtain the third-scale target feature map X22, which has a size of 16×16×512. Subsequently, the fourth-scale feature map X12 is fused with the third-scale feature map X22 along the channel dimension to obtain the fourth-scale target feature map X23, which has a size of 16×16×1024. Finally, the EFO Attention module processes X23 to obtain the second-scale target feature fusion map X24, which has a size of 16×16×512. Finally, X18, X21, and X24 are input into the target prediction head, outputting tensors containing prediction information. Each row of the tensor corresponds to a detection result, including the predicted bounding box coordinates, defect category label, and confidence score. This structure achieves accurate detection and classification of easily floating objects along power transmission channels in remote sensing images through layer-by-layer extraction, splicing, and fusion.

[0159] S8. Based on the error, iteratively update the parameters of the initial floating object detection model until the preset training termination condition is met, and obtain the trained floating object detection model.

[0160] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0161] The method for detecting tree obstruction hazards in power transmission channels provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0162] Specifically, the operator can send a message about the detection of easily floating objects in the power transmission channel to the server 104 via a terminal. The server receives the message, then acquires remote sensing images along the power transmission channel. Using these images as input, the trained easily floating object detection model is called to obtain the detection results. Based on these results, an early warning is issued.

[0163] In one example embodiment, this application also provides a method for detecting easily floating objects along a power transmission channel, such as... Figure 6 As shown, it includes the following S700 to S900. Wherein:

[0164] S700 acquires remote sensing images along the power transmission corridor.

[0165] Among them, the remote sensing images along the power transmission channel can be images obtained by image acquisition along the power transmission channel.

[0166] In practical applications, this can be done by periodically inspecting the power transmission channels during power grid operations, and obtaining remote sensing images of the power transmission channels through drone remote sensing and satellite remote sensing.

[0167] S800 takes remote sensing images along the power transmission channel as input, calls the trained floating object detection model, and obtains the floating object detection results. The floating object detection model is trained based on the above-mentioned training method for the floating object detection model along the power transmission channel.

[0168] The detection results for easily floating objects can include the detection box and category label of the area where the easily floating object is located.

[0169] In practical applications, during the inspection and acquisition of remote sensing images along power transmission corridors, the acquired images are input into a trained model obtained through the aforementioned method for training a detection model for easily floating objects along power transmission corridors. This model detects the areas where easily floating objects are located within the remote sensing images, yielding the detection results. The easily floating object detection model is trained using the steps outlined in any of the above-described examples of the training method for easily floating object detection models along power transmission corridors. The specific model training process will not be elaborated upon here.

[0170] The S900 issues warnings based on the detection results of easily floating objects.

[0171] In practical applications, when easily floating objects are detected, the method can be as follows: The location information of the easily floating object in the remote sensing image along the power transmission channel where it was detected can be obtained. Based on the source of the remote sensing image, the specific location information of the easily floating object within the power transmission channel can be obtained. Integrating the specific location information of the easily floating object and the time of detection, a warning message can be obtained and pushed out. Methods for pushing warning messages can include pushing alarm messages in the notification bar, displaying alarm messages in a full-screen or half-screen pop-up window, providing alarm prompts through specific sound or vibration patterns, or providing visual alarm prompts through flashing indicator lights. It is understood that the method for pushing alarm messages can be any one of the aforementioned methods or any combination of methods, and is not limited here.

[0172] In this embodiment, remote sensing images along the power transmission channel are acquired and input into a trained floating object detection model obtained through the aforementioned training method for the floating object detection model along the power transmission channel. Floating object detection results are then obtained, and early warnings are issued based on these results. On the one hand, model-based detection improves the accuracy and efficiency of floating object detection; on the other hand, early warnings based on the rapid and accurate detection results of floating objects along the power transmission channel facilitate the timely discovery of potential hazards in the power system, enabling corresponding measures to be taken to reduce the likelihood of accidents and improve the stability and safety of power system operation.

[0173] In one exemplary embodiment, such as Figure 7As shown, a model training device 600 for detecting easily floating objects along a power transmission channel is provided, including: an image acquisition module 610, a model training module 620, an error determination module 630, and a parameter update module 640, wherein:

[0174] The image acquisition module 610 is used to acquire remote sensing images along historical power transmission channels. The remote sensing images along historical power transmission channels contain category labels and location information of the true bounding boxes of easily floating objects.

[0175] The model training module 620 is used to input remote sensing images along the historical power transmission channel into the constructed initial floating object detection model. The initial floating object detection model extracts floating object features at different scales from the remote sensing images along the historical power transmission channel through a floating object deep feature extraction module, and generates a floating object feature map of the power transmission channel based on the floating object features at different scales. The floating object feature attention module performs pooling and channel weighting operations on the floating object feature map of the power transmission channel to capture the floating object features and enhance the floating object feature representation in the floating object feature map of the power transmission channel, resulting in a floating object feature enhancement map of the power transmission channel. Based on the floating object features at different scales and the floating object feature enhancement map of the power transmission channel, the category, location information and confidence of the prediction box are determined. The prediction box is used to identify floating objects in the remote sensing images along the historical power transmission channel. The initial floating object detection model includes a floating object deep feature extraction module and a floating object feature attention module.

[0176] The error determination module 630 is used to determine the error between the ground truth bounding box and the predicted bounding box based on the category label and location information of the ground truth bounding box, as well as the category, location information and confidence level of the predicted bounding box.

[0177] The parameter update module 640 is used to iteratively update the parameters of the initial floating object detection model until the preset training termination condition is met, thus obtaining the trained floating object detection model.

[0178] In an exemplary embodiment, the model training module 620 is further configured to extract features from remote sensing images along the historical power transmission channel using the feature extraction module to obtain a first feature map; and to perform convolution, channel weighting, and pooling operations on the first feature map using the feature dimensionality reduction branch, multi-scale feature fusion branch, and feature pooling branch in the deep feature extraction module for easily floating objects, respectively, to obtain a second feature map; and to extract features of the second feature map at different scales based on the feature extraction module and the deep feature extraction module for easily floating objects, and to obtain a feature map of easily floating objects in the power transmission channel based on the features at different scales.

[0179] In an exemplary embodiment, the model training module 620 is further configured to perform dilated convolution on the first feature map through the dilated convolutional layer in the easy-to-float deep feature extraction module to obtain a first deep feature map; perform convolution, batch normalization, nonlinear processing, and convolution on the first deep feature map sequentially through the feature dimensionality reduction branch to obtain a second deep feature map; perform pooling on the first feature map through the multi-scale feature fusion branch to obtain a third deep feature map; perform channel weighting on the third and second deep feature maps through the convolutional layer to obtain a fourth deep feature map; perform convolution on the fourth deep feature map to obtain a fifth deep feature map; perform max pooling and average pooling on the first feature map through the feature pooling branch; perform feature fusion on the results of max pooling and average pooling to obtain a sixth deep feature map; perform channel weighting on the fifth and sixth deep feature maps to obtain a seventh deep feature map; and perform feature fusion and convolution on the second and seventh deep feature maps to obtain a second feature map.

[0180] In an exemplary embodiment, the model training module 620 is further configured to connect the output end of each feature extraction module to the input end of each easily floating object deep feature extraction module, thereby obtaining a first feature extraction combination module, a second feature extraction combination module, and a third feature extraction combination module.

[0181] The first feature extraction module extracts features from the second feature map to obtain a first-scale feature map. The second feature extraction module then extracts features from the first-scale feature map to obtain a second-scale feature map. The third feature extraction module then extracts features from the second-scale feature map to obtain a third-scale feature map. The third-scale feature map is then processed by a deep feature extraction module for easily floating objects. The extracted features are then pooled to obtain a fourth-scale feature map. The fourth-scale feature map is then upsampled and fused with the second-scale feature map to obtain a fused feature map. The deep feature extraction module for easily floating objects then extracts features from the fourth-scale feature map to obtain a fifth-scale feature map. The fifth-scale feature map is then upsampled and fused with the first-scale feature map to obtain a feature map of easily floating objects in the power transmission channel.

[0182] In an exemplary embodiment, the model training module 620 is further configured to perform pooling operations on the easily floating object feature map of the power transmission channel through max pooling layers and average pooling layers respectively to obtain a first attention feature map and a second attention feature map; perform dilated convolution operations of different scales on the easily floating object feature map of the power transmission channel through the easily floating object feature attention module to obtain a third attention feature map, a fourth attention feature map, and a fifth attention feature map; perform channel weighting operations on the first attention feature map and the third attention feature map to obtain a sixth attention feature map; perform channel weighting operations on the second attention feature map and the fifth attention feature map to obtain a seventh attention feature map; and perform feature fusion and feature extraction on the fourth attention feature map, the sixth attention feature map, and the seventh attention feature map to obtain an enhanced feature map of easily floating objects in the power transmission channel.

[0183] In an exemplary embodiment, the model training module 620 is further configured to extract features from the enhanced feature map of easily floating objects in the transmission channel through the feature extraction module in the initial easily floating object detection model to obtain a first target feature map; fuse the features of the first target feature map and the fifth-scale feature map to obtain a second target feature map; enhance the feature representation of the second target feature map through the easily floating object feature attention module to obtain a first-scale target feature fusion map; extract features from the first-scale target feature fusion map through the feature fusion module in the initial easily floating object detection model to obtain a third target feature map; fuse the features of the third target feature map and the fourth-scale feature map to obtain a fourth target feature map; enhance the feature representation of the fourth target feature map through the easily floating object feature attention module to obtain a second-scale target feature fusion map; input the first-scale target feature fusion map, the second-scale target feature fusion map, and the enhanced feature map of easily floating objects in the transmission channel into the target detection head of the initial easily floating object detection model to generate multiple prediction boxes, and output the category, location information, and confidence score of each prediction box.

[0184] In an exemplary embodiment, the error determination module 630 is further configured to: determine the intersection-union ratio (IU) between the ground truth bounding box and the predicted bounding box based on the location information of the ground truth bounding box and the location information of the predicted bounding box; determine the predicted bounding box loss value between the ground truth bounding box and the predicted bounding box based on the IU of the ground truth bounding box, the category of the predicted bounding box, and a preset loss function; determine the confidence loss value between the ground truth bounding box and the predicted bounding box based on the IU of the ground truth bounding box and the confidence level of the predicted bounding box; and determine the error between the predicted bounding box and the ground truth bounding box based on the predicted bounding box loss value, the category loss value, and the confidence loss value.

[0185] In an exemplary embodiment, the error determination module 640 is further configured to determine the minimum bounding box between the ground truth box and the predicted box; determine a weighting factor based on the size of the minimum bounding box, the size of the ground truth box, and the size of the predicted box; and determine the predicted box loss value between the ground truth box and the predicted box according to the cross-union ratio and the weighting factor.

[0186] In one exemplary embodiment, such as Figure 8 As shown, a device 700 for detecting easily floating objects along a power transmission line is provided, comprising: a remote sensing image acquisition module 710, an easily floating object detection module 720, and an early warning module 730, wherein:

[0187] The remote sensing image acquisition module 710 is used to acquire remote sensing images along the power transmission channel;

[0188] The tree obstacle hazard detection module 720 is used to take the remote sensing image along the power transmission channel as input, call the trained floating object detection model, and obtain the floating object detection result. The floating object detection model is trained based on the above-mentioned floating object detection model training method along the power transmission channel.

[0189] The early warning module 730 is used to issue an early warning based on the detection results of the easily floating objects.

[0190] The modules in the aforementioned power transmission corridor floating object detection model training device 600 and power transmission corridor floating object detection device 700 can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0191] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a training method for a detection model of easily floating objects along a power transmission channel.

[0192] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0193] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in any of the above embodiments of the method for training a model for detecting easily floating objects along a power transmission channel.

[0194] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps in any of the above embodiments of the method for training a model for detecting easily floating objects along a power transmission channel.

[0195] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the method for training a model for detecting easily floating objects along a power transmission channel.

[0196] It should be noted that the data involved in this application (including but not limited to data used for analysis, data stored, data displayed, etc.) are all data that have been fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0197] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0198] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0199] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for training a detection model for easily floating objects along a power transmission line, characterized in that, The method includes: Acquire remote sensing images along historical power transmission channels, wherein the remote sensing images along historical power transmission channels include category labels and location information of the true bounding boxes of easily floating objects; The remote sensing images along the historical power transmission channel are input into the constructed initial floating object detection model. The features of the remote sensing images along the historical power transmission channel are extracted by the feature extraction module of the initial floating object detection model to obtain the first feature map. The first feature map is obtained by dilated convolution of the first feature map through the dilated convolution layer in the deep feature extraction module of the initial floating object detection model. The first deep feature map is subjected to convolution, batch normalization, nonlinear processing and convolution operations in the feature dimensionality reduction branch of the easy-to-float deep feature extraction module to obtain the second deep feature map. The first feature map is pooled by the multi-scale feature fusion branch in the deep feature extraction module for easily floating objects to obtain a third deep feature map. The third deep feature map and the second deep feature map are then channel-weighted by a convolutional layer to obtain a fourth deep feature map. Finally, the fourth deep feature map is convolved to obtain a fifth deep feature map. The first feature map is subjected to max pooling and average pooling operations by the feature pooling branch in the deep feature extraction module for easily floating objects. The results of the max pooling and average pooling operations are fused to obtain the sixth deep feature map. The fifth deep feature map and the sixth deep feature map are subjected to channel weighting to obtain the seventh deep feature map. The second deep feature map and the seventh deep feature map are fused and convolved to obtain the second feature map. Based on the feature extraction module and the deep feature extraction module for easily floating objects, features of the second feature map at different scales are extracted, and based on the features at different scales, a feature map of easily floating objects in the power transmission channel is obtained. The easy-floating object feature attention module performs pooling and channel weighting operations on the easy-floating object feature map of the power transmission channel to capture the easy-floating object features and enhance the easy-floating object feature representation in the easy-floating object feature map of the power transmission channel, thereby obtaining the easy-floating object feature enhancement map of the power transmission channel. Based on the features of easily floating objects at different scales and the enhanced feature map of easily floating objects in the power transmission channel, the category, location information and confidence of the prediction box are determined. The prediction box is used to identify easily floating objects in the remote sensing images along the historical power transmission channel. The initial easily floating object detection model includes an easily floating object deep feature extraction module and an easily floating object feature attention module. The error between the ground truth bounding box and the predicted bounding box is determined based on the category label and location information of the ground truth bounding box, as well as the category, location information, and confidence level of the predicted bounding box. Based on the error, the parameters of the initial floating object detection model are iteratively updated until the preset training termination condition is met, thus obtaining the trained floating object detection model.

2. The method according to claim 1, characterized in that, The number of the easily floating object deep feature extraction module and the feature extraction module is multiple; the step of extracting features of the second feature map at different scales based on the feature extraction module and the easily floating object deep feature extraction module to obtain the easily floating object feature map of the power transmission channel includes: The output terminals of each feature extraction module are connected to the input terminals of each easily floating object deep feature extraction module to obtain a first feature extraction combination module, a second feature extraction combination module, and a third feature extraction combination module; The first feature extraction combination module extracts features from the second feature map to obtain a first-scale feature map; the second feature extraction combination module extracts features from the first-scale feature map to obtain a second-scale feature map; and the third feature extraction combination module extracts features from the second-scale feature map to obtain a third-scale feature map. The deep feature extraction module for easily floating objects extracts features from the third-scale feature map, and performs pooling operation on the extracted features to obtain the fourth-scale feature map. By upsampling the fourth-scale feature map, the upsampled feature map is fused with the second-scale feature map to obtain a feature fusion map. The feature map is then extracted by the deep feature extraction module for easily floating objects to obtain a fifth-scale feature map. An upsampling operation is performed on the fifth-size feature map, and the upsampled feature map is fused with the first-size feature map to obtain a feature map of easily floating objects in the power transmission channel.

3. The method according to claim 2, characterized in that, The easily floating object feature attention module includes a max pooling layer and an average pooling layer; the process of performing pooling and channel weighting operations on the easily floating object feature map of the transmission channel through the easily floating object feature attention module to capture easily floating object features and enhance easily floating object feature representations in the easily floating object feature map of the transmission channel, and obtaining an enhanced easily floating object feature map of the transmission channel, includes: The feature map of easily floating objects in the power transmission channel is pooled using the maximum pooling layer and the average pooling layer, respectively, to obtain a first attention feature map and a second attention feature map. The floating object feature attention module performs dilated convolution operations of different scales on the floating object feature map of the power transmission channel to obtain the third attention feature map, the fourth attention feature map and the fifth attention feature map; A channel-weighted operation is performed on the first attention feature map and the third attention feature map to obtain the sixth attention feature map; The second and fifth attention feature maps are subjected to channel weighting to obtain the seventh attention feature map. Feature fusion and feature extraction are performed on the fourth attention feature map, the sixth attention feature map, and the seventh attention feature map to obtain an enhanced feature map of easily floating objects in the power transmission channel.

4. The method according to claim 2, characterized in that, The determination of the prediction box category, location information, and confidence level based on the features at different scales and the enhanced feature map of easily floating objects in the power transmission channel includes: The features of the power transmission channel floating object feature enhancement map are extracted by the feature extraction module in the initial floating object detection model to obtain a first target feature map. The first target feature map and the fifth size feature map are fused to obtain a second target feature map. The feature representation of the second target feature map is enhanced by the floating object feature attention module to obtain a first scale target feature fusion map. The features of the first-scale target feature fusion map are extracted by the feature fusion module in the initial floating object detection model to obtain the third target feature map. The features of the third target feature map and the fourth-scale feature map are fused to obtain the fourth target feature map. The feature representation of the fourth target feature map is enhanced by the floating object feature attention module to obtain the second-scale target feature fusion map. The first-scale target feature fusion map, the second-scale target feature fusion map, and the enhanced feature map of easily floating objects in the power transmission channel are input into the target detection head of the initial easily floating object detection model to generate multiple prediction boxes. The category, location information, and confidence level of each prediction box are output.

5. The method according to claim 4, characterized in that, The step of determining the error between the ground truth bounding box and the predicted bounding box based on the category label and location information of the ground truth bounding box, and the category, location information, and confidence level of the predicted bounding box, includes: Based on the location information of the ground truth bounding box and the location information of the predicted bounding box, the intersection-union ratio (IUU) between the ground truth bounding box and the predicted bounding box is determined; Based on the intersection-union ratio, determine the predicted box loss value between the ground truth box and the predicted box; Based on the category label of the ground truth bounding box, the category of the predicted bounding box, and a preset loss function, determine the category loss value between the ground truth bounding box and the predicted bounding box; Based on the intersection-union ratio of the ground truth bounding box and the predicted bounding box, and the confidence level of the predicted bounding box, determine the confidence loss value between the ground truth bounding box and the predicted bounding box; The error between the predicted box and the ground truth box is determined based on the predicted box loss value, the category loss value, and the confidence loss value.

6. The method according to claim 5, characterized in that, Determining the predicted bounding box loss value between the ground truth bounding box and the predicted bounding box based on the intersection-union ratio includes: Determine the minimum bounding box between the ground truth bounding box and the predicted bounding box; Based on the size of the minimum bounding box, the size of the ground truth box, and the size of the predicted box, a weighting factor is determined, and the predicted box loss value between the ground truth box and the predicted box is determined according to the intersection-union ratio and the weighting factor.

7. A method for detecting easily floating objects along a power transmission line, characterized in that, The method includes: Acquire remote sensing images along the power transmission corridor; Using the remote sensing image along the power transmission channel as input, the trained floating object detection model is called to obtain the floating object detection result. The floating object detection model is trained based on the training method for the floating object detection model along the power transmission channel as described in any one of claims 1 to 6. An early warning will be issued based on the detection results of the easily floating objects.

8. A training device for detecting easily floating objects along a power transmission line, characterized in that, The device includes: The image acquisition module is used to acquire remote sensing images along historical power transmission channels, which include category labels and location information of the true bounding boxes of easily floating objects; The model training module is used to input the remote sensing images along the historical power transmission channel into the pre-constructed initial floating object detection model. The feature extraction module of the initial floating object detection model extracts features from the remote sensing images along the historical power transmission channel to obtain a first feature map. The first feature map is then subjected to dilated convolution through the dilated convolution layer in the floating object deep feature extraction module of the initial floating object detection model to obtain a first deep feature map. Finally, the first deep feature map is subjected to convolution operations and batch reduction sequentially through the feature dimensionality reduction branch in the floating object deep feature extraction module. The first feature map is obtained by performing a pooling operation, nonlinear processing, and convolution operation. A second deep feature map is then obtained by performing pooling on the first feature map through the multi-scale feature fusion branch of the easy-to-float deep feature extraction module. A third deep feature map is obtained by performing channel-weighted operations on the third and second deep feature maps through a convolutional layer. A fourth deep feature map is then obtained by performing convolution on the fourth deep feature map. A fifth deep feature map is obtained by performing max pooling and average pooling operations on the first feature map through the feature pooling branch of the easy-to-float deep feature extraction module. The process involves fusing the results of max pooling and average pooling operations to obtain a sixth deep feature map. Channel weighting is then applied to the fifth and sixth deep feature maps to obtain a seventh deep feature map. Feature fusion and convolution operations are then performed on the second and seventh deep feature maps to obtain a second feature map. Based on the feature extraction module and the easily floating object deep feature extraction module, features at different scales are extracted from the second feature map, and based on these features, an easily floating object feature map of the transmission channel is obtained. Finally, an easily floating object feature attention module is used to... The feature map of easily floating objects along the power transmission channel is subjected to pooling and channel weighting operations to capture easily floating object features and enhance their representations, resulting in an enhanced easily floating object feature map. Based on the easily floating object features at different scales and the enhanced easily floating object feature map, the category, location information, and confidence level of the prediction box are determined. The prediction box is used to identify easily floating objects in the remote sensing images along the historical power transmission channel. The initial easily floating object detection model includes an easily floating object deep feature extraction module and an easily floating object feature attention module. An error determination module is used to determine the error between the real bounding box and the predicted bounding box based on the category label and location information of the real bounding box, as well as the category, location information and confidence level of the predicted bounding box. The parameter update module is used to iteratively update the parameters of the initial floating object detection model until the preset training termination condition is met, thereby obtaining the trained floating object detection model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images