A method and related device for identifying land use types in the base area of ​​power transmission towers

Abstract

This invention pertains to a land use type identification method. To address the technical problem that existing mainstream models for automatically identifying land use types in power transmission tower base areas suffer from insufficient accuracy and speed, this invention provides a method and related apparatus for identifying land use types in power transmission tower base areas. The method utilizes a YOLO v7 object detection network model to filter images containing the tower base from acquired images of the power transmission towers to be identified, thus obtaining the base area. The images containing the tower base are then input into an improved DeepLabV3+ semantic segmentation network model to obtain base pixels and pixels representing various land use types. The improved DeepLabV3+ semantic segmentation network model includes an encoder and a decoder, with the encoder incorporating a dual-attention CBAM module. Based on the base pixels, the effective range of the base is determined. Within the effective range of the base, the proportion of each land use type is calculated based on the percentage of pixels representing each land use type.

Description

Technical Field

[0001] This invention relates to a land use type identification method, specifically a method and related apparatus for identifying land use types in the base area of ​​power transmission towers. Background Technology

[0002] The construction of power grid projects inevitably damages the original ecological environment to some extent, causing ground erosion. Transmission towers, as one of the most crucial components of the entire power transmission system, have foundations whose load-bearing capacity and stability directly affect the normal operation of power transmission. On the one hand, without regular inspection and maintenance, factors such as earthquakes and landslides can lead to soil erosion and tower collapse over the years, causing unnecessary losses. On the other hand, transmission towers are mostly located in remote, sparsely populated areas, making manual inspections time-consuming and labor-intensive.

[0003] Automatically identifying land use types in power transmission tower foundation areas using artificial intelligence technology can not only improve the efficiency of disaster identification and risk prevention, but also help ensure the construction and operation safety of power grid projects. Early image segmentation algorithms were represented by grayscale segmentation and conditional random fields. With the rapid development of hardware resources such as GPUs, semantic segmentation based on deep learning technology has achieved great success. Currently, the most effective solution is to use semantic segmentation methods from the field of computer vision to understand images at the pixel level and assign different types of pixels. Mainstream models include symmetric models represented by U-net, dilated convolutional models represented by DeepLab, and residual network models represented by PSPnet. However, in practice, applying existing models to land use type identification in power transmission tower foundation areas fails to meet the required accuracy and speed. Summary of the Invention

[0004] To address the technical problem that the accuracy and speed of automatically identifying land use types in the base area of ​​power transmission towers using existing mainstream models cannot meet the identification requirements, this invention provides a method and related apparatus for identifying land use types in the base area of ​​power transmission towers.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] In a first aspect, the present invention proposes a method for identifying land use types in the base area of ​​power transmission towers, comprising the following steps:

[0007] S1. Using the YOLO v7 object detection network model, images containing the base of the transmission tower are selected from the collected images of the transmission towers to be identified, and the base area is obtained.

[0008] S2, input the image containing the base of the power transmission tower into the improved DeepLabV3+ semantic segmentation network model to obtain the base pixel and the pixel of various land use types;

[0009] The improved DeepLabV3+ semantic segmentation network model includes an encoder and a decoder;

[0010] The encoder processes the image specifically as follows:

[0011] S2-1, the image is downsampled multiple times to obtain the first preliminary effective feature layer; and dilated convolution with different dilation ratios in parallel is used to extract features from the image to obtain the first feature layer;

[0012] S2-2, the first preliminary effective feature layer and the first feature layer are concatenated, and then the dual attention CBAM module is used to enhance different categories of features and improve the classification accuracy between local features;

[0013] S2-3, features are compressed through convolutional layers to obtain the encoded feature map;

[0014] S3, based on the base pixels, obtain the effective range of the base;

[0015] S4. Within the effective range of the tower base, the proportion of each land use type is obtained based on the proportion of pixels of each land use type.

[0016] Furthermore, step S3 specifically includes:

[0017] S3.1, Based on the base pixel points, count the positions corresponding to the base pixels in the image, and obtain the base center point by averaging;

[0018] S3.2, Calculate the length from the center point of the tower base to the edge pixel of the tower base, and average the lengths to obtain the average length of the tower base;

[0019] S3.3, draw a circle with the center of the tower base as the center and a radius of at least one times the average length of the tower base, and the resulting circle is the effective range of the tower base.

[0020] Furthermore, step S4 specifically includes:

[0021] S4.1, Create a mask image of the same size as the image containing the base of the power transmission tower, and set all pixels of the mask image to zero;

[0022] S4.2, Set the pixel value of the effective range area of ​​the tower base in the mask image to 255;

[0023] S4.3, Perform a bitwise AND operation on the mask image obtained in step S4-2 and the image containing the base of the power transmission tower to obtain the final result image;

[0024] S4.4 Calculate the proportion of pixels of various land use types in the final result image to obtain the proportion of various land use types.

[0025] Furthermore, the decoder processes the image specifically as follows:

[0026] S2-4, the encoded feature map is downsampled multiple times to obtain the second preliminary effective feature layer;

[0027] S2-5, adjust the number of channels in the second preliminary effective feature layer using a convolutional layer to obtain the second feature layer;

[0028] S2-6, the second feature layer and the encoded feature map are stacked, and then multiple depthwise separable convolutions are performed to obtain the decoded effective feature layer, and the base pixel points and various land use type pixel points are output.

[0029] Furthermore, in steps S2-6, after obtaining the decoded effective feature layer, the following steps are also included:

[0030] The number of effective feature layer channels after decoding is adjusted to the number of categories by convolution, and then upsampling is performed by resizing so that the height and width of the image output by the decoder are consistent with the image of the power transmission tower base of the input encoder.

[0031] Further, in step S1, the YOLO v7 object detection network model is trained using the following method:

[0032] S1-1, Use the LabelImg tool to mark the base of the transmission tower on the acquired images of the transmission tower;

[0033] S1-2, Input multiple images of transmission towers with labeled tower bases into the YOLO v7 target detection network model to train the YOLO v7 target detection network model;

[0034] In step S2, the improved DeepLabV3+ semantic segmentation network model is trained using the following method:

[0035] S1-3, On multiple images containing power transmission tower bases selected by the YOLO v7 object detection network model, various land use types are labeled using the Labelme tool to generate a labeling file;

[0036] S1-4, Convert the annotation file into VOC dataset format;

[0037] S1-5, according to a preset ratio, the VOC dataset is divided into a training set, a test set, and a validation set, and the improved DeepLabV3+ semantic segmentation network model is trained, tested, and validated.

[0038] Furthermore, the land use types mentioned in step S2 include vegetation, cultivated land, bare land, and water areas.

[0039] Secondly, the present invention proposes a land use type identification system for the base area of ​​power transmission towers, including a base screening module, a pixel acquisition module, a base range determination module, and an identification module;

[0040] The tower base screening module is used to filter out images containing the tower bases of power transmission towers from the collected images of power transmission towers to be identified using the YOLO v7 target detection network model.

[0041] The pixel acquisition module is used to input an image containing the base of the power transmission tower into the improved DeepLabV3+ semantic segmentation network model to obtain the base pixel and the pixel of various land use types.

[0042] The improved DeepLabV3+ semantic segmentation network model includes an encoder and a decoder;

[0043] The encoder processes the image specifically as follows:

[0044] S2-1, the image is downsampled multiple times to obtain the first preliminary effective feature layer; and dilated convolution with different dilation ratios in parallel is used to extract features from the image to obtain the first feature layer;

[0045] S2-2, the first preliminary effective feature layer and the first feature layer are concatenated, and then the dual attention CBAM module is used to enhance different categories of features and improve the classification accuracy between local features;

[0046] S2-3, features are compressed through convolutional layers to obtain the encoded feature map;

[0047] The pixel acquisition module is used to obtain the effective range of the tower base based on the tower base pixels;

[0048] The identification module is used to obtain the proportion of various land use types based on the proportion of pixels of various land use types within the effective range of the tower base.

[0049] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0050] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] 1. This invention proposes a method for identifying land use types in the base area of ​​power transmission towers. The method utilizes a YOLO v7 object detection network model to screen tower bases, and an improved DeepLabV3+ semantic segmentation network model to distinguish pixels of the tower base and various land use types. The encoder structure of the DeepLabV3+ semantic segmentation network model is adjusted, and a dual-attention CBAM module is added after feature layer concatenation. This not only strengthens features of different categories but also improves the classification accuracy between local features. After identifying the corresponding pixels, the effective range of the tower base is first determined, and then the proportion of various land use types is calculated to complete the identification of various land use types. By combining machine learning methods and improving the corresponding model, and combining the pixels identified by the model, land use type identification can be completed accurately and quickly through clever calculations. This method utilizes and amplifies the advantages of the corresponding model, and the improvements also make the model more suitable for the application scenario of this application.

[0053] 2. When determining the effective range of the tower base, this invention determines the center point of the tower base by using the tower base pixels obtained from the model, and then combines the length between the tower base edge pixels and the tower base center point to obtain the average length of the tower base. Finally, the effective range of the tower base is determined by drawing a circle. By cleverly combining pixel points with geometric calculation methods, the effective range of the tower base can be obtained quickly and accurately.

[0054] 3. This invention converts land use type identification into pixel ratio calculation, making identification easier to achieve, avoiding the complexity of land use types, and thus making the identification results more accurate.

[0055] 4. In this invention, the improved DeepLabV3+ semantic segmentation network model, after completing feature extraction, also adjusts the number of channels in the effective feature layer and upsamples the output image of the decoder to be the same size as the input image, maximizing the preservation of information in the input image and facilitating subsequent calculations, thereby indirectly ensuring the reliability of the recognition results.

[0056] 5. This invention also proposes a land use type identification system for the base area of ​​power transmission towers. This system implements the aforementioned identification method through a modular structure, making it easier to promote and apply the method of this invention.

[0057] 6. The present invention also proposes a computer device and a computer-readable storage medium for implementing the above method, thereby implementing the above identification method with more hardware carriers. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating Embodiment 1 of the present invention;

[0060] Figure 2 This is a schematic diagram of the dual attention CBAM module in Embodiment 2 of the present invention;

[0061] Figure 3 This is a flowchart illustrating Embodiment 2 of the present invention;

[0062] Figure 4 This is a schematic diagram of an example of identification using the identification method of Embodiment 2 of the present invention; wherein, (a) is a diagram of the final result obtained when calculating the proportion of each land use type, and (b) is a schematic diagram of the calculation result of the proportion of each land use type;

[0063] Figure 5 This is a schematic diagram of Example 2 of the identification method according to Embodiment 2 of the present invention; wherein, (a) is the final result diagram obtained when calculating the proportion of each land use type, and (b) is a schematic diagram of the calculation result of the proportion of each land use type;

[0064] Figure 6 This is a schematic diagram of an example of a comparative experiment conducted using the embodiments of the present invention; wherein, (a) is the feature image output by the existing DeepLabV3+ semantic segmentation network model, and (b) is the feature image output by the existing DeepLabV3+ semantic segmentation network model.

[0065] Figure 7 This is a schematic diagram of Example 2, which uses the comparative experiment in the embodiments of the present invention; wherein, (a) is the feature image output by the existing DeepLabV3+ semantic segmentation network model, and (b) is the feature image output by the existing DeepLabV3+ semantic segmentation network model. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0067] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0068] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0069] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0070] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0071] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0072] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments:

[0073] Example 1

[0074] like Figure 1 As shown, a method for identifying land use types in the base area of ​​power transmission towers includes the following steps:

[0075] S101, using the YOLO v7 object detection network model, filters out images containing the base of the power transmission tower from the collected images of the power transmission tower to be identified, and obtains the base area.

[0076] The YOLO v7 object detection network model, once trained, can recognize images of power transmission towers and identify the tower base area. By training the YOLO v7 object detection network model with a sufficient number of power transmission tower images whose base areas are already labeled, the trained model can quickly and accurately identify the tower base area.

[0077] S102, input the image containing the base of the power transmission tower into the improved DeepLabV3+ semantic segmentation network model to obtain the base pixel points and the pixel points of various land use types;

[0078] The improved DeepLabV3+ semantic segmentation network model includes an encoder and a decoder. The improvement in the encoding region mainly lies in the dual-attention CBAM module. Specifically, the encoder processes the image as follows:

[0079] (1) The image is downsampled multiple times to obtain the first preliminary effective feature layer; and the image is extracted by dilated convolution with different dilation ratios in parallel to obtain the first feature layer.

[0080] (2) The first preliminary effective feature layer and the first feature layer are spliced ​​together, and then the dual attention CBAM module is used to enhance different categories of features and improve the classification accuracy between local features.

[0081] Dual-attention CBAM modules generally include a channel attention module and a spatial attention module. The channel attention module enhances features of different categories, while the spatial attention module improves the classification accuracy between different local features.

[0082] (3) The features are compressed through convolutional layers to obtain the encoded feature map.

[0083] S103, Based on the base pixels, obtain the effective range of the base;

[0084] S104. Within the effective range of the tower base, the proportion of each land use type is obtained based on the proportion of pixels of each land use type.

[0085] Example 2

[0086] like Figure 3 Implementation two is a preferred embodiment of the method for identifying land use types in the base area of ​​power transmission towers according to the present invention, specifically including the following steps:

[0087] S201, Acquire images of power transmission towers and perform preprocessing.

[0088] High-resolution images of power transmission towers can be captured using drones. For such high-resolution images, to facilitate subsequent selection of power transmission tower bases and training of network models, the original images can be compressed to a resolution of 1915*1276.

[0089] Existing methods can be used, such as using the LabelImg tool to label the base of the power transmission towers on the acquired images, for subsequent training of the YOLO v7 object detection network model.

[0090] A trained YOLO v7 object detection network model can be used to obtain the base of power transmission towers from images. Then, the LabelImg tool can be used to annotate various land use types on the images labeled with the tower bases. Example 2 uses four land use types—vegetation, cultivated land, bare land, and water area—as examples. After annotation, a labeling file is generated and converted to VOC dataset format. When converting the labeling file to VOC dataset format, a SegmentationClassPNG file is often generated. The SegmentationClassPNG file contains the mask labels of the annotated data. As an example, RGB(128, 0, 0) is defined as the tower base, RGB(0, 128, 0) as vegetation, RGB(128, 0, 128) as cultivated land, RGB(128, 0, 0) as the tower base, RGB(0, 0, 128) as water area, and RGB(128, 128, 0) as bare land. The VOC dataset is converted to grayscale using a weighted average. Pixel values ​​of 75 represent vegetation, 53 represent cultivated land, 38 represent the base of a tower, 14 represent water bodies, and 114 represent bare land. Different pixels represent different areas or land use types. The number of images in the converted VOC dataset can be adjusted according to actual needs. In Example 2, 450 images were labeled in the VOC dataset. As a further preferred method, the VOC dataset can be expanded to 600 images through data augmentation. The data in the VOC dataset is then randomly divided into training, validation, and test sets in an 8:1:1 ratio for training the subsequently improved DeepLabV3+ semantic segmentation network model. The resulting filename list can be stored in the project's directory, such as train.txt, val.txt, and test.txt.

[0091] S202, Building an improved DeepLabV3+ semantic segmentation network model

[0092] The DeepLabV3+ semantic segmentation network model borrows from the U-net symmetric model to design an encoder and a decoder. The encoder extracts features to obtain high-level semantic information, while the decoder gradually reconstructs spatial information.

[0093] The encoder is a trimmed version of the Deeplabv3 network model, primarily used for ASPP (Atrous Spatial Pyramid Pooling) enhanced feature extraction. Specifically, it first obtains an initial effective feature layer through four downsampling passes, then uses parallel dilated convolutions with different dilation ratios for feature extraction, and finally concatenates the features and passes them through a dual-attention CBAM module. Figure 2 As shown, the input features are multiplied once by the channel attention module, and the result of this first multiplication is multiplied twice by the spatial attention module to obtain the refined features. In the dual-attention CBAM module, the channel attention module enhances features of different categories, while the spatial attention module improves the classification accuracy between different local features. Then, a 1*1 convolution is used to compress the features, resulting in the final encoded feature map. At the decoder, a preliminary effective feature layer is obtained through two downsampling operations. After adjusting the number of channels using a 1*1 convolution, the result is stacked with the upsampled result from the encoder. Finally, two depthwise separable convolutions are performed to obtain the decoded effective feature layer. Another 1*1 convolution is used to adjust the number of channels in the decoded effective feature layer to match the number of categories. Finally, resizing is used for upsampling to make the width and height of the final output layer the same as the input image.

[0094] S203, Training the YOLO v7 object detection network model and the improved DeepLabV3+ semantic segmentation network model.

[0095] The YOLO v7 object detection network model is trained by using a certain number of images of power transmission tower bases labeled in step S201, so that the trained YOLO v7 object detection network model can output the power transmission tower base area after inputting power transmission tower images.

[0096] The improved DeepLabV3+ semantic segmentation network model is trained using the VOC dataset obtained in step S201, and tested and validated using the validation and test sets to obtain the trained improved DeepLabV3+ semantic segmentation network model. This model can output the pixels of the power transmission tower base and the pixels of various land use types after inputting the image of the power transmission tower base.

[0097] S204. The image of the power transmission tower to be identified is sequentially input into the YOLO v7 object detection network model and the improved DeepLabV3+ semantic segmentation network model to obtain the pixels of the power transmission tower base and the pixels of various land use types.

[0098] S205: First, determine the effective range of the tower base, then obtain the proportion of each land use type based on the proportion of pixels of each land use type, and identify the land use type by obtaining the proportion.

[0099] (1) Count the positions of the base pixels in the image, and obtain the base center point (x, y) by averaging. Count the length from the base center point to the base edge pixels, and obtain the base mean length R by averaging.

[0100] (2) Taking the center point (x, y) of the tower base as the center, select a multiple of the average length R of the tower base as the radius according to actual needs to obtain the effective range U of the tower base.

[0101] (3) Create a mask image with the same size as the original image, initialize all pixels to 0, and then set all pixel values ​​in the corresponding U area of ​​the mask image to 255.

[0102] (4) Perform an AND operation between the mask image and the original image to obtain the final result image. Then calculate the proportion of each pixel in the result image, which represents the proportion of each land use type. Accordingly, complete the identification of each land use type in the legend.

[0103] The method described in Example 2 was used to perform two actual identifications for verification, such as... Figure 4 This is a schematic diagram of the result of Example 1, as shown below. Figure 5 This is a schematic diagram of the results of Example 2. Figure 4 and Figure 5 It can be seen that the identification results of each land use type are clear, and the identification results are accurate after comparison with the actual situation.

[0104] Furthermore, to demonstrate that the improved DeepLabV3+ semantic segmentation network model outperforms the existing DeepLabV3+ semantic segmentation network model in achieving better recognition results for the application scenarios described in this invention, practical tests were conducted. Under the same experimental environment, datasets were loaded into both the improved DeepLabV3+ semantic segmentation network model and the existing DeepLabV3+ semantic segmentation network model for experiments. The experiment iterated for a total of 4400 times, saving the weights every 400 iterations, using the cross-entropy loss function. Specific environment and parameter configurations are shown in Table 1.

[0105] Table 1 Comparative Experimental Environment Configuration and Model Parameter Settings

[0106]

[0107] The experiment selected cPA, mCPA, and mIOU as evaluation metrics. PA is pixel accuracy, which refers to the proportion of pixels correctly predicted to the total number of pixels; cPA is class pixel accuracy, which refers to the accuracy of pixels that actually belong to class i among the predicted values ​​of class i; mPA is class average pixel accuracy, which is the ratio of the intersection to the union of the model's prediction results and the true values ​​for a certain class; IoU is the intersection-union ratio, which is the ratio of the intersection to the union of the model's prediction results and the true values ​​for a certain class; mIoU is the average intersection-union ratio, which is the sum and average of the intersection to the union of the model's prediction results and the true values ​​for each class. Experimental results show that the improved network model improves the pixel accuracy of each class, as shown in Tables 2 and 3 below:

[0108] Table 2. Experimental results of existing DeepLabV3+ semantic segmentation network models.

[0109]

[0110] Table 3. Experimental results of the improved DeepLabV3+ semantic segmentation network model.

[0111]

[0112] The results in Tables 2 and 3 show that mIOU and mCPA were improved by 1.41% and 1.88% respectively compared with the original model.

[0113] When conducting comparative experiments, we obtained Figure 6 and Figure 7 The comparison chart shown is as follows. Figure 6 and Figure 7 The image shows a comparison of two examples. Figure 6 This is a comparison chart for Example 1. Figure 7 This is a comparison chart for Example 2. Figure 6 and Figure 7 In the image, (a) is the feature image output by the existing DeepLabV3+ semantic segmentation network model, and (b) is the feature image output by the existing DeepLabV3+ semantic segmentation network model.

[0114] A computer device is provided according to an embodiment of the present invention. This computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0115] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0116] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor and memory.

[0117] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0118] The memory can be used to store the computer program and / or module, and the processor implements various functions of the computer device by running or executing the computer program and / or module stored in the memory, and by calling the data stored in the memory.

[0119] If the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0120] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying the land use type of a tower foundation area of a power transmission tower, characterized in that The method comprises the following steps: S1, filtering out an image containing a tower base of a power transmission tower from a collected image to be recognized by using a YOLO v7 target detection network model, and obtaining a tower base area; S2, inputting the image containing the tower base of the power transmission tower into an improved DeepLabV3+ semantic segmentation network model to obtain pixel points of the tower base and pixel points of various land use types; The improved DeepLabV3+ semantic segmentation network model comprises an encoder and a decoder; The processing of the image by the encoder is specifically as follows: S2-1, performing multiple times of down-sampling on the image to obtain a first primary effective feature layer; and performing feature extraction on the image by using different dilated convolution ratios and parallel hole convolution to obtain a first feature layer; S2-2, splicing the first primary effective feature layer and the first feature layer, and then performing different category feature strengthening and improving classification accuracy between local features by using a double-attention CBAM module; S2-3, compressing the features by using a convolution layer to obtain an encoded feature map; The processing of the image by the decoder is specifically as follows: S2-4, performing multiple times of down-sampling on the encoded feature map to obtain a second primary effective feature layer; S2-5, adjusting the channel number of the second primary effective feature layer by using a convolution layer to obtain a second feature layer; S2-6, stacking the second feature layer and the encoded feature map, and then performing multiple times of depth separable convolution to obtain a decoded effective feature layer, and outputting pixel points of the tower base and pixel points of various land use types; S3, obtaining an effective range of the tower base according to the pixel points of the tower base; S4, obtaining proportions of various land use types according to proportions of pixel points of various land use types in the effective range of the tower base: S4.1, creating a mask image with the same size as the image containing the tower base of the power transmission tower, and setting all pixel values of the mask image to zero; S4.2, setting pixel values of a region corresponding to the effective range of the tower base in the mask image to 255; S4.3, performing AND operation on the mask image obtained in step S4-2 and the image containing the tower base of the power transmission tower to obtain a final result image; S4.4, calculating proportions of pixel points of various land use types in the final result image to obtain proportions of various land use types.

2. The method according to claim 1, wherein the method is characterized by, Step S3 is specifically as follows: S3.1, according to the pixel points of the tower base, counting positions of the pixel points of the tower base in the image, and averaging to obtain a tower base center point; S3.2, counting lengths of the tower base center point to tower edge pixel points in the pixel points of the tower base, and averaging to obtain a tower base average length; S3.3, taking the tower base center as a center and at least one times of the tower base average length as a radius to draw a circle, and obtaining an effective range of the tower base.

3. The method according to claim 2, wherein the method is characterized by: In step S2-6, after obtaining the decoded effective feature layer, the following steps are further included: adjusting the channel number of the decoded effective feature layer to a classification number by using convolution, and then performing up-sampling by using resize to make the height and width of the image output by the decoder consistent with the image containing the tower base of the power transmission tower input into the encoder.

4. The method according to claim 3, wherein the method is characterized by: In step S1, the YOLO v7 target detection network model is trained by the following method: S1-1, label the tower base of the transmission tower on the collected transmission tower image through the LabelImg tool; S1-2, input multiple transmission tower images labeled with the tower base of the transmission tower into the YOLO v7 target detection network model to train the YOLO v7 target detection network model; In step S2, the improved DeepLabV3+ semantic segmentation network model is trained by the following method: S1-3, label various land use types on the images containing the tower base of the transmission tower screened by the YOLO v7 target detection network model using the Labelme tool to generate a label file; S1-4, convert the label file into a VOC data set format; S1-5, divide the VOC data set into a training set, a test set, and a validation set according to a preset proportion, and train, test, and validate the improved DeepLabV3+ semantic segmentation network model.

5. The method according to claim 4, wherein the method is characterized by: The land use types in step S2 include vegetation, cultivated land, bare land, and water area.

6. A power transmission tower foundation area land use type identification system, characterized in that: The tower base screening module, the pixel point acquisition module, the tower base range determination module, and the identification module are included. The tower base screening module is configured to screen images containing the tower base of the transmission tower from the collected transmission tower image to be identified through the YOLO v7 target detection network model. The pixel point acquisition module is configured to input the images containing the tower base of the transmission tower into the improved DeepLabV3+ semantic segmentation network model to obtain the pixel points of the tower base and the pixel points of various land use types. The improved DeepLabV3+ semantic segmentation network model includes an encoder and a decoder. The processing of the image by the encoder is specifically as follows: S2-1, perform multiple downsampling on the image to obtain a first preliminary effective feature layer; and perform feature extraction on the image by using different dilated convolutions and parallel convolutions to obtain a first feature layer; S2-2, splice the first preliminary effective feature layer and the first feature layer, and then perform different class feature enhancement and improve the classification accuracy between local features by a double attention CBAM module; S2-3, compress the features by a convolution layer to obtain an encoded feature map; The processing of the image by the decoder is specifically as follows: S2-4, perform multiple downsampling on the encoded feature map to obtain a second preliminary effective feature layer; S2-5, adjust the channel number of the second preliminary effective feature layer by a convolution layer to obtain a second feature layer; S2-6, stack the second feature layer and the encoded feature map, and then perform multiple depth separable convolutions to obtain a decoded effective feature layer, and output the pixel points of the tower base and the pixel points of various land use types; The pixel point acquisition module is configured to obtain the effective range of the tower base according to the pixel points of the tower base. The identification module is configured to obtain the proportion of various land use types according to the proportion of the pixel points of various land use types in the effective range of the tower base. S4.1, create a mask image with the same size as the image containing the tower base of the power transmission tower, and set all pixels of the mask image to zero; S4.2, set the pixel values of the effective range area corresponding to the tower base in the mask image to 255; S4.3, perform an AND operation on the mask image obtained in step S4-2 and the image containing the tower base of the power transmission tower to obtain a final result image; S4.4, calculate the proportion of pixel points of various land use types in the final result image to obtain the proportion of various land use types.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, the computer-readable storage medium comprising: The computer program is executed by the processor to implement the steps of the method of any one of claims 1-5.

Images