An intelligent verification method for clearing houses in power transmission line corridors based on high-resolution images

Abstract

The present invention relates to the technical field of change detection, and in particular to an intelligent verification method for houses in power transmission line channel clearance based on high-resolution images. The method collects high-resolution images of power transmission channels in design and construction stages by unmanned aerial vehicles, and pre-processes the collected images, including geometric correction, radiation correction, geographic registration, etc., so as to reduce the influence of the external environment. The processed dual-phase images are cropped into non-repetitive blocks with a resolution of 512×512, and labelme is used to annotate the changed areas to construct a data set for high-resolution house change detection in power transmission channels. The house change detection in power transmission channels is realized based on a ChangeFormer model. The method automatically extracts vector maps of newly added houses and house demolition in power transmission channels by inputting high-resolution remote sensing images of two phases of unmanned aerial vehicles based on intelligent remote sensing interpretation change detection technology, so as to quickly provide comprehensive and accurate verification results for management personnel.

Description

Technical Field

[0001] The present invention relates to the technical field of change detection, and in particular to an intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images. Background Art

[0002] UAV remote sensing technology has developed rapidly in recent years. Due to its advantages of high efficiency, flexibility and low cost, it has been widely used in many fields such as topographic mapping, agricultural monitoring, and environmental monitoring. The acquisition of high-resolution images provides a basis for the extraction of refined ground information, especially in the management of transmission lines that require frequent monitoring and large-scale coverage; with the rapid development of my country's economy, the scale of transmission line construction has continued to expand. In order to protect the lives and property of people in the channel and ensure safe and reliable power supply of the power grid, the power supply company has strengthened the special governance of power line channels, including the demolition of houses that do not meet the safety distance in the transmission line channel, and preventing the phenomenon of "rushing to plant and build" to reduce social disputes. Traditional channel corridor house inspections rely on manual inspections, which have high labor input costs, low efficiency, and lack of effective intelligent management and control methods. With the rapid development of artificial intelligence and deep learning technology, image-based change detection has become a research hotspot. By comparing remote sensing images of different phases, the increase and decrease of surface objects can be automatically identified. Change detection technology has important application value in urban planning, disaster assessment and other fields.

[0003] In the change detection task, traditional image processing methods are limited by the complexity of feature extraction and the diversity of change types, making it difficult to achieve good results. Deep learning models, especially those based on convolutional neural networks, can autonomously learn complex features in images and show higher detection accuracy and robustness. The ChangeFormer model is a change detection model based on the Transformer architecture, which has powerful feature extraction and change recognition capabilities.

[0004] In order to improve the efficiency of transmission corridor channel cleaning and inspection by construction management departments, the present invention provides a method and system for intelligent inspection of transmission line channel cleaning houses based on high-resolution images, which can effectively reduce labor costs and error rates, improve the management level of construction site cleaning work, confirm that there are no omissions or wrong cleaning objects, support the smooth handover and safe operation of engineering construction, and are of great significance to the standardized management of construction sites.

[0005] The existing Chinese patent document CN202310426990.4, by introducing a visual transformer structure, spatial and channel attention, a U-shaped residual module, an enhanced feature extraction module, and a self-attention feature fusion module, can achieve full integration of different building information. It can better distinguish buildings of different regular shapes and sizes to prevent false detection and missed detection, while improving the feature extraction capabilities of buildings of different shapes and their edge details. The prior art has little research on house change detection in power transmission scenarios, and most of them use public data sets for model testing. The present invention aims to provide a full-process control solution for intelligent inspection of power transmission line houses from data set construction to model building and result output, realizing the transformation from manual self-inspection to system automatic inspection, so as to achieve the purpose of reducing costs and increasing efficiency. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for intelligent house inspection for power transmission line channel clearance based on high-resolution images. By inputting two phases of high-resolution remote sensing images from unmanned aerial vehicles and based on intelligent remote sensing interpretation change detection technology, vector maps of newly added houses and house demolition in power transmission channels are automatically extracted, so as to quickly provide comprehensive and accurate inspection results for management personnel.

[0007] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0008] A method for intelligently checking houses for clearing transmission line passages based on high-resolution images comprises the following steps:

[0009] S1: Use drones to collect high-resolution images of the transmission channel during the design and construction phases, and pre-process the collected images, including geometric correction, radiation correction, and geographic registration, in order to reduce the impact of the external environment;

[0010] S2: The processed dual-phase images are cropped into non-repetitive blocks with a resolution of 512×512, and labelme is used to annotate the changed areas to construct a high-resolution house change detection dataset for power transmission channels;

[0011] S3: Model construction and model training: Based on the ChangeFormer model, the transmission channel house change detection is realized, and the data set is divided into training set, validation set and test set in a ratio of 8:1:1 to obtain the training model;

[0012] S4: The two real image data are cropped into image pairs with a resolution of 512×512 without duplication, and the ChangeFormer model is input to calculate the change detection results of each image pair. Finally, the image blocks are merged and the detection frames are restored to the image to obtain the house change information before and after the transmission line is installed.

[0013] Furthermore, the step S1 specifically includes: using Envi and ArcMap software to perform geometric correction, radiation correction, and geographic registration on the image, eliminating deformation caused by factors such as camera angle and earth curvature, and reducing the impact of atmospheric conditions and lighting changes on the image to ensure the accuracy and consistency of the image.

[0014] Furthermore, the ChangeFormer model in step S3 includes two modules, Hierarchical Transformer Encoder and MLP Decoder. The change detection image pair is input, and the twin network extracts multi-level features of the before and after phases respectively; the Difference Module unit in the Hierarchical Transformer Encoder module receives a total of 4 groups of features of different scales generated by different layers of the twin network, and performs splicing and fusion; the multi-level features generated in the previous step are input into the MLP decoder for aggregation to obtain the change detection mask output.

[0015] Further, the Hierarchical Transformer Encoder module includes a DownsamplingBlock unit, a TransformerBlock unit, and a Difference Module unit;

[0016] The Downsampling Block unit includes: performing a downsampling operation on the feature map of each layer, except for the first feature layer which uses a convolutional layer with a convolutional kernel size of 7×7, a stride of 4, and a padding of 3, and the remaining layers use a convolutional layer with a convolutional kernel size of 3×3, a stride of 2, and a padding of 1.

[0017] The TransformerBlock unit includes: using a scaling factor R to reshape the (Q, K, V) dimension from (HW, C) to (HW / R, C·R) for accelerated calculation, and then reducing the dimension CR to C through a linear layer Linear, and finally obtaining a new (Q, K, V) dimension, namely (HW / R, C).

[0018]

[0019] Where S represents the sequence to be reduced, and R is the set scaling factor.

[0020] To add position information, two MLP layers and a 3×3 depthwise separable convolution are used as shown below:

[0021] F out=MLP(GELU(Conv2D 3×3 (MLP(F in ))))+F in

[0022] Where F in Represents the feature map of Self-Attention, and GELU represents the Gaussian Error Linear Units activation function.

[0023] The Difference Module unit includes: receiving features of four different sizes from the two branches of the twin network, performing feature splicing, convolution calculation, and outputting differences. The results calculated by each layer of Difference Module will be passed to the next layer for further calculation.

[0024] Furthermore, the MLP Decoder module specifically includes: processing each multi-scale difference feature through the MLP layer, unifying the channel dimension, upsampling to the size of H / 4×W / 4, combining the difference feature maps at different levels for feature fusion, and then upsampling to H×W using transposed convolution, maintaining the features through a series of residual blocks, and finally generating a change map through a classification convolution.

[0025] Furthermore, the intersection over union (IoU), recall rate (Rec), precision (Pre) and F1 are used as evaluation indicators:

[0026]

[0027]

[0028] Among them, TP is the positive sample that is detected correctly; FN is the negative sample that is detected incorrectly; FP is the false positive sample; TN is the negative sample that is detected correctly.

[0029] A transmission line channel clearing house intelligent verification system based on high-resolution images, the system runs the above method, and the system construction includes the following steps:

[0030] S1: Basic environment preparation: QGIS and Pycharm installation, configuration of PyQGIS environment, running code to debug whether QtDesigner runs normally;

[0031] S2: Interface design: Based on the system functions of the present invention and with reference to ArcMap software, the system interface design is carried out, including: layer tree, map canvas, menu bar, status bar, etc.; the map canvas will render the remote sensing image according to the order of the layer tree and display it in the map canvas. The menu bar will store all the functions of the software; the status bar will display some real-time status such as the XY coordinates (latitude and longitude) of the mouse position;

[0032] S3: Create layer tree and map canvas: Create MainWindow interface through Main Window template, the interface content includes centralwidget (central component), menubar (menu bar) and statusbar (status bar); place map canvas through frame in DockWidget, and place layer tree in DockWidgetContents;

[0033] S4: Image and vector data loading: Use QgsRasterLayer and QgsVectorLayer functions to build image and vector data loading functions and encapsulate them into buttons;

[0034] S5: Algorithm integration: Encapsulate the ChangeFormer change detection model into the detect.py file, add a recognition button, and start the built-in algorithm of the button when data is input to extract house rush construction, house expansion and house demolition, and obtain the change detection result picture;

[0035] S6: Result display: Load the detection result image of the previous step and display it on the map canvas, overlay the vector data of the house demolition distribution map, and intuitively display the house inspection situation of the transmission channel clearance to form a "one picture" of the channel clearance.

[0036] Beneficial effects of the present invention:

[0037] The present invention collects high-resolution images through drones, and automatically identifies and extracts house change information based on intelligent remote sensing interpretation and change detection technology. Traditional transmission line channel house inspections mainly rely on manual inspections, which are time-consuming and labor-intensive, and there is a risk of omissions. UAV high-resolution image acquisition ensures the accuracy and consistency of the image through preprocessing steps, including geometric correction, radiation correction, and geo-registration. Subsequently, the processed images are cropped into 512×512 tiles to ensure higher detection accuracy. ChangeFormer unifies the hierarchical Transformer encoder with the multi-layer perception decoder in the twin network architecture to effectively render the detailed information required for multi-scale remote and accurate change detection. Through the combination of these technical means, the verification efficiency is significantly improved, the labor cost is reduced, and the work is made more efficient and economical.

[0038] Existing SOTA change detection methods are mainly based on deep convolutional networks (ConvNets) because they can extract powerful discriminative features. Since the contextual information must be captured in both space and time to identify relevant changes in multi-temporal images, the latest change detection research has been focused on increasing the receptive field of the change detection model. Based on this, many change detection models with stacked convolutional layers, dilated convolutions, and attention mechanisms (channel and spatial attention) have been proposed. Although attention-based methods are effective in capturing global details, they have difficulty in linking contextual details in space and time because they use attention to reweight the bi-temporal features obtained by ConvNets in channel and spatial dimensions. The present invention uses the ChangeFormer model for house change detection, which effectively addresses the limitations of traditional image processing methods in feature extraction complexity and the diversity of change types. The ChangeFormer model achieves multi-level feature extraction and feature fusion through the cooperation of Hierarchical Transformer Encoder and MLP Decoder modules. Specifically, the Downsampling Block unit downsamples the feature map, and the TransformerBlock unit extracts features using the scaling factor R and the linear layer Linear. The Difference Module unit further receives features of multiple different sizes, performs concatenation, convolution and difference output. This process ensures efficient and accurate identification of changes in house additions and subtractions, avoids missing or mistakenly cleaning objects, and provides comprehensive and accurate verification results.

[0039] The present invention provides a scientific basis for the house change information before and after the installation of power transmission lines through precise change detection technology. The high-resolution images collected by drones ensure the high accuracy of the images after geometric correction, radiation correction and geo-registration processing. The change detection technology based on the ChangeFormer model extracts subtle differences in the images through model training and verification. These change information are converted into actionable verification results through image block splicing and restoration steps. Managers can quickly understand the changes at the construction site, ensure the cleaning of houses that do not meet the safety distance, and prevent the phenomenon of "rushing to plant and build". In this way, social disputes are effectively reduced, the smooth progress and safe operation of power line construction are ensured, and the standardized management level of the construction site is improved.

[0040] The present invention uses intersection over union (IoU), recall (Rec), precision (Pre) and F1 score as evaluation indicators to comprehensively measure the performance of the model. These evaluation indicators not only take into account the accuracy of detection, but also measure the breadth and depth of the detection results. Specifically, IoU is used to evaluate the degree of overlap between the detection results and the actual change area, Rec is used to evaluate the proportion of correctly detected change areas, Pre is used to evaluate the proportion of actual changes in the detection results, and the F1 score comprehensively considers recall and precision. Through these scientific evaluation indicators, the detection results are comprehensively evaluated to ensure their reliability and effectiveness, thereby providing a solid data foundation for subsequent management decisions.

[0041] In summary, the present invention significantly improves the efficiency and accuracy of transmission line channel verification through efficient image processing and intelligent change detection methods, reduces labor costs, and ensures the safe operation and standardized management of power line construction. Scientific evaluation indicators further ensure the reliability and effectiveness of the detection results, which is an important technical innovation in the transmission line channel verification work.

[0042] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0044] Figure 1 A schematic diagram of the technical route described in an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of UAV image data of a power transmission channel according to an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of a data set for detecting house changes in a power transmission channel according to an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of the ChangeFormer model framework described in an embodiment of the present invention;

[0048] Figure 5 This is a schematic diagram of the housing change detection result of the power transmission channel according to the embodiment of the present invention;

[0049] Figure 6 A schematic diagram of a system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] Example 1

[0052] The intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images described in this embodiment includes the following steps:

[0053] S1: Use drones to collect high-resolution images of the transmission channel during the design and construction phases, and pre-process the collected images, including geometric correction, radiation correction, and geographic registration, in order to reduce the impact of the external environment;

[0054] S2: The processed dual-phase images are cropped into non-repetitive blocks with a resolution of 512×512, and labelme is used to annotate the changed areas to construct a high-resolution house change detection dataset for power transmission channels;

[0055] S3: Model construction and model training: Based on the ChangeFormer model, the transmission channel house change detection is realized, and the data set is divided into training set, validation set and test set in a ratio of 8:1:1 to obtain the training model;

[0056] S4: The two real image data are cropped into non-repetitive image pairs with a resolution of 512×512, and the ChangeFormer model is input to calculate the change detection results of each image pair. Finally, the image blocks are merged and the detection frames are restored to the image to obtain the house change information before and after the transmission line is installed.

[0057] In this embodiment, step S1 specifically includes: using Envi and ArcMap software to perform geometric correction, radiation correction, and geographic registration on the image, eliminating deformation caused by factors such as camera angle and earth curvature, and reducing the impact of atmospheric conditions and lighting changes on the image to ensure the accuracy and consistency of the image. Figure 2 As shown in the figure, the background is complex and diverse, including ponds, farmlands, bare land, vegetation areas, etc. The sample types are rich and representative, which increases the challenge of the detection task, thereby improving the generalization and robustness of the model in practical applications.

[0058] In this embodiment, step S2 specifically includes: since the original size of the image is large, the original image is divided into image blocks of 512×512 for model training, and there is no overlapping between the image blocks. Labelme is used to mark the changed area, and a high-resolution house change detection dataset for the power transmission channel is constructed. The dataset example is as follows: Figure 3 As shown, from left to right are the design stage images, construction stage images, and binary change labels.

[0059] In this embodiment, the ChangeFormer model in step S3 includes two modules, Hierarchical Transformer Encoder and MLP Decoder. A change detection image pair is input, and the twin network extracts multi-level features of the before and after phases respectively; the Difference Module unit in the Hierarchical Transformer Encoder module receives a total of 4 groups of features of different scales generated by different layers of the twin network, and performs splicing and fusion; the multi-level features generated in the previous step are input into the MLP decoder for aggregation to obtain a change detection mask output.

[0060] In this embodiment, the Hierarchical Transformer Encoder module includes a DownsamplingBlock unit, a TransformerBlock unit, and a Difference Module unit;

[0061] (1) Downsampling Block:

[0062] The feature map of each layer is downsampled. Except for the first feature layer, which uses a convolutional kernel size of 7×7, a stride of 4, and a padding of 3, the remaining layers use a convolutional kernel size of 3×3, a stride of 2, and a padding of 1.

[0063] (2) Transformer Block:

[0064] The main building block of Transformer Encoder is self-attention. The traditional self-attention is:

[0065]

[0066] Where Q, K, and V represent Query, Key, and Value, respectively, and have the same dimension HW×C.

[0067] The Transformer Block module uses the scaling factor R to reshape the (Q, K, V) dimension from (HW, C) to (HW / R, C·R) for accelerated calculation, and then reduces the dimension CR to C through a linear layer Linear, finally obtaining a new (Q, K, V) dimension, namely (HW / R, C).

[0068]

[0069]

[0070] Where S represents the sequence to be reduced, and R is the set scaling factor.

[0071] To add position information, two MLP layers and a 3×3 depthwise separable convolution are used as shown below:

[0072] F out =MLP(GELU(Conv2D 3×3 (MLP(F in ))))+F in

[0073] Where F in Represents the feature map of Self-Attention, and GELU represents the Gaussian Error Linear Units activation function.

[0074] (3)Difference Module

[0075] It receives features of four different sizes from the two branches of the twin network, performs feature splicing, convolution calculation, and output difference. The results calculated by each layer of Difference Module will be passed to the next layer for further calculation.

[0076] In this embodiment, the MLP Decoder module specifically includes: processing each multi-scale difference feature through the MLP layer, unifying the channel dimension, upsampling to a size of H / 4×W / 4, combining the difference feature maps at different levels for feature fusion, and then upsampling to H×W using transposed convolution, maintaining the features through a series of residual blocks, and finally generating a change map through a classification convolution.

[0077] In this embodiment, the intersection over union (IoU), recall rate (Rec), precision (Pre) and F1 are used as evaluation indicators:

[0078]

[0079] Among them, TP is the positive sample that is detected correctly; FN is the negative sample that is detected incorrectly; FP is the false positive sample; TN is the negative sample that is detected correctly.

[0080] In order to verify the effectiveness of the model, the data set is divided into training set, validation set and test set in a ratio of 8:1:1. The experimental environment and configuration are as follows: python3.8 and pytorch are selected as the development language and deep learning library, the CPU model is Xeon Gold 6330, the graphics card model is NVIDIAA100, and the video memory is 80GB.

[0081] Example 2

[0082] As described in this embodiment, a transmission line channel clearing house intelligent verification system based on high-resolution images is built in combination with PyQGIS and pyqt. The system integrates map, data preprocessing, algorithm reasoning, data post-processing and other modules. The system supports common map functions such as map zooming in and out, vector and raster data loading, etc. By inputting two phases of drone images, based on the system's built-in ChangeFormer change detection model, it automatically extracts house rush construction, house expansion and house demolition, superimposes the house demolition distribution map, and visualizes the detection results on the map, intuitively displays the transmission channel clearing house verification situation, and forms a "one picture" for channel clearing.

[0083] like Figure 5 As shown in the figure, the results of house change detection in the power transmission channel obtained by using the change detection technology are shown. It can be seen that the model has a good extraction effect on the rush-built and demolished houses, and the house boundaries are clear and complete. Based on experimental comparison and analysis, it can be concluded that the method proposed in the present invention can accurately extract the house change information of the power transmission channel, and can meet the intelligent verification needs of the houses in the power transmission line channel clearance in actual projects.

[0084] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for intelligent inspection of houses for clearing transmission line channels based on high-resolution images, characterized in that: The following steps are involved: S1: Use drones to collect high-resolution images of the transmission channel during the design and construction stages, and pre-process the collected images, including geometric correction, radiation correction, and geographic registration, to reduce the impact of the external environment; S2: The processed dual-phase images are cropped into non-repetitive blocks with a resolution of 512×512, and labelme is used to annotate the changed areas to construct a high-resolution house change detection dataset for power transmission channels; S3: Model construction and model training: Based on the ChangeFormer model, the transmission channel house change detection is realized, and the data set is divided into training set, validation set and test set in a ratio of 8:1:1 to obtain the training model; S4: The two real image data are cropped into non-repetitive image pairs with a resolution of 512×512, and the ChangeFormer model is input to calculate the change detection results of each image pair. Finally, the image blocks are merged and the detection frames are restored to the image to obtain the house change information before and after the transmission line is installed.

2. The intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images as claimed in claim 1, characterized in that: The step S1 specifically includes: using Envi and ArcMap software to perform geometric correction, radiation correction, and geographic registration on the image, eliminating deformation caused by camera angle and earth curvature factors, and reducing the impact of atmospheric conditions and lighting changes on the image to ensure the accuracy and consistency of the image.

3. The intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images as claimed in claim 1, characterized in that: The ChangeFormer model in step S3 includes two modules, Hierarchical Transformer Encoder and MLP Decoder. The change detection image pair is input, and the twin network extracts multi-level features of the before and after phases respectively; the Difference Module unit in the Hierarchical Transformer Encoder module receives a total of 4 groups of features of different scales generated by different layers of the twin network, and performs splicing and fusion; the multi-level features generated in the previous step are input into the MLP decoder for aggregation to obtain the change detection mask output.

4. The intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images as claimed in claim 3 is characterized by: The Hierarchical Transformer Encoder module includes a Downsampling Block unit, a Transformer Block unit, and a Difference Module unit; The Downsampling Block unit includes: performing a downsampling operation on the feature map of each layer, except for the first feature layer using a convolution layer with a convolution kernel size of 7×7, a step size of 4, and a padding of 3, and the remaining layers use a convolution layer with a convolution kernel size of 3×3, a step size of 2, and a padding of 1; The TransformerBlock unit includes: using a scaling factor R to reshape the (Q, K, V) dimension from (HW, C) to (HW / R, C·R) for accelerated calculation, and then reducing the dimension CR to C through a linear layer Linear, and finally obtaining a new (Q, K, V) dimension, namely (HW / R, C); Where S represents the sequence to be reduced, and R is the set scaling factor; To add position information, two MLP layers and a 3×3 depthwise separable convolution are used as shown below: F out =MLP(GELU(Conv2D 3×3 (MLP(F in ))))+F in Where F in Represents the feature map of Self-Attention, GELU represents the Gaussian Error LinearUnits activation function; The Difference Module unit includes: receiving features of four different sizes from the two branches of the twin network, performing feature splicing, convolution calculation, and outputting differences. The results calculated by each layer of Difference Module will be passed to the next layer for further calculation.

5. The intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images as claimed in claim 4 is characterized by: The MLP Decoder module specifically includes: processing each multi-scale difference feature through the MLP layer, unifying the channel dimension, upsampling to the size of H / 4×W / 4, combining the difference feature maps at different levels for feature fusion, and then upsampling to H×W using transposed convolution, maintaining the features through a series of residual blocks, and finally generating a change map through a classification convolution.

6. The intelligent inspection method for clearing houses in power transmission line passages based on high-resolution images as claimed in claim 1, characterized in that: The intersection over union (IoU), recall rate (Rec), precision (Pre) and F1 are used as evaluation indicators: Among them, TP is the positive sample that is detected correctly; FN is the negative sample that is detected incorrectly; FP is the false positive sample; TN is the negative sample that is detected correctly.

7. An intelligent inspection system for clearing houses in power transmission line passages based on high-resolution images, characterized in that: The system runs a smart inspection method for clearing houses in power transmission line passages based on high-resolution images, and the system construction includes the following steps: S1: Basic environment preparation: QGIS and Pycharm installation, configuration of PyQGIS environment, running code to debug whether Qt Designer runs normally; S2: Interface design: Based on the system functions of the present invention and with reference to ArcMap software, the system interface design is carried out, including but not limited to: layer tree, map canvas, menu bar, status bar; S3: Create layer tree and map canvas: Create the MainWindow interface through the Main Window template. The interface content includes the central component centralwidget, menu bar menubar and status bar statusbar; place the map canvas through the frame of the Dock Widget, and place the layer tree in dockWidgetContents; S4: Image and vector data loading: Use QgsRasterLayer and QgsVectorLayer functions to build image and vector data loading functions and encapsulate them into buttons; S5: Algorithm integration: Encapsulate the ChangeFormer change detection model into the detect.py file, add a recognition button, and start the built-in algorithm of the button when data is input to extract house rush construction, house expansion and house demolition, and obtain the change detection result picture; S6: Result display: The detection result image obtained in step S5 is loaded and displayed on the map canvas, and the vector data of the house demolition distribution map is superimposed to intuitively display the house inspection situation of the transmission channel clearance, forming a "one picture" of the channel clearance.

Images