A method and system for monitoring the range of disturbance of a substation ring water conservation land

Abstract

The application belongs to the field of substation water and soil conservation, and particularly relates to a substation water and soil conservation land disturbance range monitoring method, which comprises the following steps: in the encoder, a ResNet-50 feature extraction network fused with a lightweight channel attention module is used to perform feature extraction processing on a high-resolution image, each feature layer is subtracted and an absolute value is taken to obtain a difference feature map; the difference feature map is fused into the decoder part through a channel splicing and up-sampling manner; in the decoder, an MLP structure containing only one hidden layer is used to decode the feature map enhanced by the feature difference, generate a binary classification change prediction map, and output a final change result image. The application further provides a substation water and soil conservation land disturbance range monitoring system. Through image information extraction, whether the engineering construction land disturbance range is reasonable is judged to achieve the purpose of rapid monitoring of water and soil conservation in the construction process.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of substation environmental protection, and particularly relates to a substation environmental protection land disturbance range monitoring method and system. BACKGROUND

[0002] Substations are critical components of the power system, responsible for converting high-voltage electricity generated by power plants into voltage levels suitable for long-distance transmission and distribution. Substations also convert high-voltage electricity into lower voltage suitable for household and industrial use.

[0003] The construction and operation of substations are crucial for the stability and reliability of power supply.

[0004] With the rapid development of industrialization and urbanization, environmental protection and water resource protection have become increasingly important. The construction and operation of substations may have an impact on the surrounding environment, especially in terms of land disturbance,

[0005] Land disturbance can lead to a variety of environmental problems, such as soil erosion, water and soil loss, ecological system destruction, and water pollution. These problems not only affect the quality of life of local residents, but also have long-term impacts on the entire ecosystem.

[0006] During the construction process of a substation project, construction personnel excavate the ground, causing differences between the construction area surface and the surrounding environment. These differences mainly manifest in the differences between new soil and original soil, and between bare ground and vegetation-covered ground. The extraction of the boundaries of these differences results in the disturbance area of the project construction, thereby measuring the land disturbance range.

[0007] Traditional remote sensing image change detection mainly uses image difference, ratio, and change vector analysis (CVA) methods. Neighboring information, structural features, and contrast difference maps, mean difference maps, and principal component analysis are applied to change detection, which improves the detection accuracy to some extent. However, the above methods rely partly on manually constructed feature representations, and have limited modeling capabilities for complex high-level change information. Moreover, when the features of change classes and non-change classes overlap or their statistical distribution modeling is inaccurate, the detection results will have large errors, and have certain limitations. SUMMARY

[0008] The purpose of the embodiments of the present application is to provide a substation environmental protection land disturbance range monitoring method and system, which aims to solve the technical problems raised in the background.

[0009] To achieve the above-mentioned purpose, the present application provides the following technical solutions.

[0010] In a first aspect of the present application, a substation environmental protection land disturbance range monitoring method is provided, which comprises the following steps:

[0011] Step S101: obtaining a pre-phase image and a post-phase image of a substation engineering area by using a UAV;

[0012] Step S102: the pre-phase image and the post-phase image are decomposed into high-frequency and low-frequency components through wavelet transform, the high-frequency component is processed by using a bilinear interpolation method, the low-frequency component is processed by using a bicubic interpolation method, then inverse wavelet transform is performed, and a high-resolution image is output;

[0013] Step S103: in an encoder, a ResNet-50 feature extraction network fused with a lightweight channel attention module is used to perform feature extraction processing on the high-resolution image, differences between extracted feature layers are taken, and an absolute value is obtained, thereby obtaining a difference feature map;

[0014] Step S104: the difference feature map is fused into a decoder part through channel splicing and upsampling, in the decoder, a change prediction map is generated by decoding the feature map enhanced through feature difference, and a final change result image is output.

[0015] In further limitation of the embodiment scheme of the present application, before the high-frequency component is processed by using the bilinear interpolation method, the high-frequency component is further processed by using a Canny algorithm to extract edges, thereby obtaining a new high-frequency component, and then the new high-frequency component is processed by using the bilinear interpolation method, which can effectively retain edge information and improve interpolation speed.

[0016] In further limitation of the embodiment scheme of the present application, the ResNet-50 feature extraction network fused with the lightweight channel attention module comprises:

[0017] An input layer is used to receive an input image, in the embodiment of the present application, the input image is a high-resolution pre-phase image and a post-phase image;

[0018] A residual convolution block comprises a convolution, a batch normalization and a ReLU activation function;

[0019] A lightweight channel attention module is added after the residual block;

[0020] An average pooling layer is used to perform average pooling after the residual block and the CAM;

[0021] A fully connected layer is used to map the features after pooling.

[0022] In further limitation of the embodiment of the present application, in the step of performing feature extraction processing on the high-resolution image by using the ResNet-50 feature extraction network fused with the lightweight channel attention module, the pre-phase image and the post-phase image are subjected to feature extraction by two feature extraction modules sharing parameters;

[0023] After the image is subjected to convolution and pooling processing, the residual convolution block is used to extract multi-scale semantic features;

[0024] The residual convolution block is composed of a ResNeXt unit with 2 times down-sampling and a 1x1 convolution, and then unified by an up-sampling module to obtain the same feature map size as the first layer residual block, so that the output features fused with high-level semantic information and low-level spatial information of different scales can be obtained;

[0025] The 4-layer transformed feature maps are concatenated in the channel, and a 3x3 convolution layer is input to obtain the final feature map.

[0026] In further limitation of the embodiment of the present application, when the information of different scales is fused, the data of the same scale is required to be fused, and the image data is kept at the same resolution by using the up-sampling and down-sampling methods, and specifically, the scale calculation method is as follows:

[0027]

[0028] In the formula, padding represents the padding of feature data; kernel_size represents the size of the convolution kernel; stride represents the step size of each convolution movement; H out represents the height of the feature map; W out represents the width of the feature map.

[0029] In further limitation of the embodiment of the present application, the output of the cth convolution layer in the nth residual convolution block is represented as: F n，c = σ (W n，c [F n-1 , F n，1 ,..., F n，c-1 ]) ;

[0030] In the formula, sigma represents the ReLU activation function, and W n，c represents the weight value of the cth convolution layer.

[0031] The output features of each layer of each residual convolution block are mapped and spliced, and the output features of each layer of the previous residual convolution block and the current residual convolution block are connected to the subsequent layer, and based on this form, the local features can be well preserved.

[0032] In further limitation of the embodiment of the present application, three residual convolution blocks are adopted as the backbone of the feature extraction network, after obtaining the local information features of each residual convolution block, a global feature fusion and global residual learning are performed; the output features of each residual convolution block are fused by the full connection layer, and a global residual learning is performed with the initial input features, which can effectively reduce the information loss;

[0033] The fused features are sent into a 3x3 convolution, and the obtained features are fused with themselves again to obtain the final feature result, which is represented as F':

[0034] F' = F cat + Conv(F cat );

[0035] In the formula, Conv represents a 3x3 convolution operation, F cat represents the result of the global fusion and residual learning of the features, which can effectively preserve the low-level features.

[0036] In further limitation of the embodiment of the present application, the step of fusing the change feature map into the decoder part by channel splicing and upsampling of the difference feature map includes: in the f1, f2, f3, f4 and f5 in the difference feature map, f3, f4 and f5 are respectively doubled and upsampled, and then fused with f2 to obtain p2, and then p2 is doubled and upsampled and fused with f1 to obtain p1.

[0037] In further limitation of the embodiment of the present application, in the step of decoding the feature map enhanced by the feature difference to generate a binary change prediction map and outputting the final change result image, the MLP decodes the upsampled feature map p1 to generate a change prediction map, which is a two-dimensional image, wherein the value of each pixel represents the possibility of change at that position, and the change prediction map is converted into a binary image by a threshold, wherein the pixels higher than the threshold are marked as 1, and the pixels not higher than the threshold are marked as 0, 0 and 1 represent the labels of the unchanged area and the changed area respectively, and correspondingly, the changed area is the land disturbance range required to be monitored in the substation project.

[0038] In the second aspect of the present application, a substation ring water protection land disturbance range monitoring system is provided, which comprises:

[0039] An image acquisition module is configured to acquire the pre-phase image and the post-phase image of the substation project area by using a UAV;

[0040] The image processing module is used for decomposing the pre-phase image and the post-phase image into high-frequency and low-frequency components through wavelet transform, processing the high-frequency components by using a bilinear interpolation method, processing the low-frequency components by using a bicubic interpolation method, then performing inverse wavelet transform, and outputting a high-resolution image;

[0041] The encoder module is used for performing feature extraction processing on the high-resolution image by using a ResNet-50 feature extraction network fused with a lightweight channel attention module, performing difference on each feature layer extracted and taking an absolute value to obtain a difference feature map;

[0042] The decoder module is used for fusing the difference feature map into a decoder part by means of channel splicing and subsequent upsampling, and generating a binary classification change prediction map by decoding the feature map enhanced by the feature difference in the decoder, and outputting a final change result image.

[0043] Compared with the prior art, the substation ring water and land disturbance range monitoring method and system has the following beneficial effects:

[0044] Firstly, the pre-phase image and the post-phase image are decomposed into high-frequency and low-frequency components through wavelet transform, the high-frequency components are processed by using a bilinear interpolation method, the low-frequency components are processed by using a bicubic interpolation method, then inverse wavelet transform is performed, and a high-resolution image can be outputted, so that more reliable image data is provided for subsequent model recognition of the image change area;

[0045] Secondly, in the encoder, a ResNet-50 feature extraction network fused with a lightweight channel attention module is used to perform feature extraction processing on the high-resolution image, difference is performed on each feature layer extracted and an absolute value is taken to obtain a difference feature map;

[0046] Thirdly, in the decoder, the difference feature map is fused into a decoder part by means of channel splicing and subsequent upsampling, the feature map enhanced by the feature difference is decoded by means of an MLP structure containing only one hidden layer, a binary classification change prediction map is generated, and a final change result image is outputted.

[0047] In summary, the feature extraction network based on deep learning takes the unmanned aerial vehicle image in the process of substation engineering construction as a data source, establishes a substation engineering construction land disturbance monitoring system, obtains land disturbance information in the substation engineering construction area through image information extraction, judges whether the engineering construction land disturbance range is reasonable, and achieves the purpose of rapid monitoring of water and soil conservation in the construction process. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to make the technical solutions in the embodiments of the present application clearer, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application.

[0049] Figure 1 The figure is a framework of a substation surrounding water and soil conservation land disturbance range monitoring method of the present application.

[0050] Figure 2 The figure is an implementation flowchart of a substation surrounding water and soil conservation land disturbance range monitoring method of the present application.

[0051] Figure 3 The figure is a principle block diagram of image high resolution processing provided by the embodiments of the present application.

[0052] Figure 4 The figure is a structure block diagram of a substation surrounding water and soil conservation land disturbance range monitoring system provided by the present application.

[0053] Figure 5 The figure is a structure block diagram of a computer device provided by the present application. DETAILED DESCRIPTION

[0054] In order to make the technical solutions in the embodiments of the present application clearer, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application.

[0055] The specific implementation of the present application will be described in detail below in combination with specific embodiments.

[0056] As shown in Figure 1 and Figure 2 In one embodiment provided by the present application, a substation surrounding water and soil conservation land disturbance range monitoring method comprises the following steps:

[0057] Step S101: obtaining a pre-phase image and a post-phase image of a substation engineering area by using a UAV;

[0058] The pre-phase image and the post-phase image refer to two images obtained at different times, and the pre-phase image is obtained earlier than the post-phase image. The purpose of the embodiments of the present application is to obtain the change range of the post-phase image relative to the pre-phase image, i.e. the land disturbance range caused by the substation engineering. The disturbance range causes differences between the ground surface of the construction area and the surrounding environment, and the boundary extraction result of the differences is the substation construction disturbance area, so as to monitor the land disturbance range.

[0059] In the selection and preparation of the UAV, the size and terrain complexity of the substation engineering area need to be considered to select a suitable UAV, considering its flight height, endurance, and load capacity; the UAV needs to be equipped with high-resolution photographic equipment, and the UAV automatically takes high-resolution ground images at the predetermined position according to the set parameters.

[0060] The shooting of the post-phase image is selected at an appropriate time point after the implementation of the project, so as to observe the changes after the implementation of the project; specifically, the shooting is performed as much as possible under the same lighting and climate conditions as the pre-phase image, so as to reduce the influence of external factors on the image comparison.

[0061] Specifically, the key of the substation ring water conservation land disturbance range monitoring task is to distinguish the region related to the change and the region irrelevant to the change in the image, wherein the region related to the change is the part that needs to be concerned in the monitoring task.

[0062] Further, please refer to Figure 1 and Figure 2 The monitoring method of the present application further comprises:

[0063] Step S102: The pre-phase image and the post-phase image are decomposed into high-frequency and low-frequency components through wavelet transform, the high-frequency component is processed by using a bilinear interpolation method, the low-frequency component is processed by using a bicubic interpolation method, then inverse wavelet transform is performed, and a high-resolution image is output;

[0064] Most of the image information is stored in the low-frequency component, such as the contour information of the image; and some edge and detail information is stored in the high-frequency component. In order to achieve the best effect, different interpolation methods need to be adopted for the two components.

[0065] Further, please refer to Figure 1 and Figure 2 The monitoring method of the present application further comprises:

[0066] Step S103: In the encoder, a ResNet-50 feature extraction network fused with a lightweight channel attention module is used to perform feature extraction processing on the high-resolution image, the extracted feature layers are subtracted and the absolute value is taken to obtain a difference feature map;

[0067] Step S104: The difference feature map is fused into the decoder part by means of channel splicing and upsampling, in the decoder, the feature map enhanced by the feature difference is decoded through an MLP structure containing only one hidden layer to generate a two-class change prediction map, and an output final change result image is output.

[0068] Wherein, the binary classification change prediction is mapped to 0 or 1, 0 and 1 represent the labels of non-change area and change area respectively, and the change area is the land disturbance range required to be monitored in the substation project.

[0069] Further, before the high-frequency component is processed by the bilinear interpolation method, the high-frequency component is further processed by an edge extraction processing using a Canny algorithm to obtain a new high-frequency component, and then the new high-frequency component is processed by the bilinear interpolation method, which can effectively retain edge information and improve the interpolation speed.

[0070] Further, in the preferred embodiment of the present application, the ResNet-50 feature extraction network fused with the lightweight channel attention module comprises:

[0071] The input layer is used to receive an input image, and in the embodiment of the present application, the input image is a high-resolution pre-phase image and a post-phase image;

[0072] The residual convolution block includes convolution, batch normalization and ReLU activation function;

[0073] The lightweight channel attention module (Channel Attention Module, CAM) is added after the residual block;

[0074] The average pooling layer is used to perform average pooling after the residual block and the CAM;

[0075] The fully connected layer is used to map the features after pooling.

[0076] Specifically, in the step of using the ResNet-50 feature extraction network fused with the lightweight channel attention module to perform feature extraction processing on the high-resolution image, the pre-phase image and the post-phase image are extracted by two feature extraction modules sharing parameters;

[0077] After the image is processed by convolution and pooling, the residual convolution block is used to extract multi-scale semantic features;

[0078] The residual convolution block is composed of a ResNeXt unit with 2 times down-sampling and a 1x1 convolution, and then unified by an up-sampling module to obtain the same feature map size as the first layer residual block, so that the output features fused with high-level semantic information and low-level spatial information of different scales can be obtained;

[0079] Wherein, when the information of different scales is fused, the data of the same scale is required to be fused, and the image data is kept at the same resolution by using the up-sampling and down-sampling method, and specifically, the scale calculation method is as follows:

[0080]

[0081] In the formula, padding represents padding of feature data; kernel_size represents size of a convolution kernel; stride represents stride of each convolution movement; H out represents height of a feature map; W out represents width of a feature map.

[0082] Further, 4 layers of transformed feature maps are concatenated in a channel, and a 3x3 convolution layer is input to obtain a final feature map.

[0083] Further, in the embodiment of the present application, the output of the cth convolution layer in the nth residual convolution block is represented as:

[0084] F n，c = σ(W n，c [F n-1 , F n，1 ,..., F n，c-1 ]);

[0085] In the formula, σ represents a ReLU activation function, W n，c represents weight value of the cth convolution layer;

[0086] The output features of each layer of each residual convolution block are mapped and spliced, and the output features of each layer of the previous residual convolution block and the current residual convolution block are connected to the subsequent layer, and based on this form, local features can be well preserved;

[0087] As a preferred, the embodiment of the present application adopts three residual convolution blocks as the backbone of the feature extraction network, after obtaining local information features of each residual convolution block, a global feature fusion and global residual learning are performed; the output features of each residual convolution block are fused by a full connection layer, and a global residual learning is performed with the initial input features, which can effectively reduce information loss.

[0088] Further, the embodiment of the present application sends the fused features into a 3x3 convolution, and the obtained features are fused with themselves again to obtain a final feature result, and the final feature result F' is represented as:

[0089] F' = F cat + Conv(F cat );

[0090] In the formula, Conv represents a 3x3 convolution operation, F cat represents a result of global fusion and residual learning of features, which can effectively preserve low-level features.

[0091] Further, the step of fusing the change feature map into the decoder part by means of channel concatenation and up-sampling of the difference feature map comprises: in the difference feature map f1, f2, f3, f4 and f5, f3, f4 and f5 are respectively twice up-sampled and fused with f2 after concatenation to obtain p2, and p2 is twice up-sampled and fused with f1 to obtain p1.

[0092] Further, in the step of decoding the feature map enhanced by the feature difference to generate a binary change prediction map and outputting the final change result image, the MLP decodes the up-sampled feature map p1 to generate a change prediction map, which is a two-dimensional image, wherein the value of each pixel represents the possibility of change at that position, and the change prediction map is converted into a binary image by a threshold, wherein the pixels higher than the threshold are marked as 1, and the pixels not higher than the threshold are marked as 0, 0 and 1 represent the labels of the unchanged area and the changed area respectively, and correspondingly, the changed area is the land disturbance range required to be monitored in the substation project.

[0093] It should be noted that the above figures are only schematic representations of the processes included in the method according to the exemplary embodiments of the present application, and are not intended to limit the purpose. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be executed synchronously or asynchronously, for example, in multiple modules.

[0094] It should be understood that although the above steps are described in a certain order, these steps are not necessarily executed in the above order. Unless explicitly stated herein, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, part of the steps of the present embodiment can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0095] In a second aspect of the embodiment of the present application, the present application further provides a substation ring water conservation land disturbance range monitoring system, comprising:

[0096] The image acquisition module 201 is configured to acquire the pre-time phase image and the post-time phase image of the substation project area by using the unmanned aerial vehicle.

[0097] The image processing module 202 is used for decomposing the pre-phase image and the post-phase image into high-frequency and low-frequency components through wavelet transform, processing the high-frequency components through a bilinear interpolation method, processing the low-frequency components through a bicubic interpolation method, then performing inverse wavelet transform, and outputting a high-resolution image;

[0098] The encoder module 203 is used for performing feature extraction processing on the high-resolution image through a ResNet-50 feature extraction network fused with a lightweight channel attention module, performing difference processing on each feature layer extracted, and obtaining a difference feature map through absolute value processing.

[0099] The decoder module 204 is used for fusing the difference feature map into a decoder part through channel splicing and up-sampling, performing decoding on the feature map enhanced through feature difference through an MLP structure containing only one hidden layer in the decoder, generating a binary classification change prediction map, and outputting a final change result image.

[0100] The substation ring water conservation land disturbance range monitoring system has wide application prospects in the field of substation engineering management, can significantly improve the efficiency and reliability of engineering safety construction management, reduce production interruption and loss, and provides higher level support for substation engineering.

[0101] A third aspect of the embodiment of the present application also provides a computer device 300 including a memory and a processor, the memory storing a computer program, and the computer program being executed by the processor to implement the substation ring water conservation land disturbance range monitoring method of any one of the above embodiments.

[0102] The computer device includes a processor and a memory, and can further include an input system and an output system. The processor, the memory, the input system and the output system can be connected through a bus or other means. The input system can receive input digital or character information and generate signal input related to the migration of the substation ring water conservation land disturbance range monitoring. The output system can include a display device such as a display screen.

[0103] The memory, as a non-volatile computer readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the substation ring water conservation land disturbance range monitoring method in the embodiments of the present application. The memory can include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required by a function; the data storage area can store data created by use of the substation ring water conservation land disturbance range monitoring method, and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged with respect to the processor, and these remote memories can be connected to the local module through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0104] The processor can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is generally used to control the overall operation of the computer device. In the present embodiment, the processor is used to run the program code stored in the memory or process data. The processors of the plurality of computer devices of the computer device in the present embodiment execute various function applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory, that is, implement the steps of the substation ring water conservation land disturbance range monitoring method of the above method embodiments.

[0105] Those skilled in the art will also appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the particular application and design constraints imposed on the overall system. Skilled persons can implement the functions in various ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the embodiments disclosed herein.

[0106] Finally, it is noted that the computer-readable storage media of the present disclosure (e.g., the memory) can be volatile or nonvolatile storage. By way of example, and not limitation, nonvolatile memory can include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), which can act as external cache memory. By way of example, and not limitation, RAM is available in many forms such as Static RAM (DRAM), Dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed aspects of the disclosure intend to include all such forms of memory.

[0107] The various illustrative logical blocks, modules, and circuits described in connection with the disclosure herein can be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field

[0108] The foregoing is a summary of an exemplary embodiment of the disclosure, but it should be noted that various changes and modifications can be made without departing from the scope of the disclosure as defined by the claims. The functions, steps and / or actions of the method claims in accordance with the embodiments of the disclosure described herein need not be performed in any particular order. Furthermore, although elements of the disclosure embodiment disclosed can be described or claimed in individual forms, other embodiments can include a plurality of those elements in order to carry out the intended functions of the disclosure.

[0109] It should be understood that, as used herein, "and / or," includes any and all possible combinations of one or more of the associated listed items. The foregoing disclosure of the disclosure embodiment numbers is merely intended to describe the disclosure and is not intended to be limiting.

[0110] Those skilled in the art should understand that the above discussion of any embodiment is only intended to be illustrative and is not intended to be limiting on the scope of the present embodiments, including the claims, which are intended to be limited only by the literal and equivalent language of the claims. The above embodiments, or technical features among different embodiments, can be combined, and there are many other variations of different aspects of the present embodiments as described above, which are not provided in details for the sake of brevity. Therefore, any omission, modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiments should be included in the protection scope of the present embodiments.

Claims

1. A method for monitoring the disturbance range of water and soil conservation in substations, characterized in that, The method includes the following steps: Step S101: Use a drone to acquire the previous and next phase images of the substation engineering area; Step S102: The previous and subsequent time-phase images are decomposed into high-frequency and low-frequency components by wavelet transform. The high-frequency components are processed by edge extraction using the Canny algorithm and then processed by bilinear interpolation. The low-frequency components are processed by bicubic interpolation. Then, inverse wavelet transform is performed to output a high-resolution image. Step S103: In the encoder, a ResNet-50 feature extraction network with a lightweight channel attention module is used to perform feature extraction processing on the high-resolution image. The lightweight channel attention module is added after the residual block. The difference between each extracted feature layer is calculated and the absolute value is taken to obtain the difference feature map. Step S104: The difference feature map is fused into the decoder by upsampling after concatenation by channel. In the decoder, an MLP structure with only one hidden layer decodes the feature map enhanced by feature difference, generating a binary classification change prediction map and outputting the final change result image, including: the features in the difference feature map. , , , and In the middle, , and After double upsampling and After splicing and fusion, the result is Then After doubling upsampling and splicing and fusion to obtain In the step of decoding the feature map after feature difference enhancement to generate a binary classification change prediction map and outputting the final change result image, the MLP upsamples the feature map. Decoding is performed to generate a change prediction map, which is a two-dimensional image. The value of each pixel in the two-dimensional image represents the probability of a change at that location. The change prediction map is converted into a binary image by a threshold. Pixels above the threshold are marked as 1, and pixels below the threshold are marked as 0.

2. The method for monitoring the environmental and soil disturbance range of a substation according to claim 1, characterized in that, The ResNet-50 feature extraction network, which incorporates a lightweight channel attention module, includes: The input layer receives the input image, which is a high-resolution front-phase image and a back-phase image. Residual convolutional blocks: include convolution, batch normalization, and ReLU activation function; Average pooling layer: Average pooling is performed after the residual block and CAM; Fully connected layer: maps the pooled features.

3. The method for monitoring the environmental and soil disturbance range of a substation according to claim 2, characterized in that, In the step of using a ResNet-50 feature extraction network with a lightweight channel attention module to perform feature extraction on high-resolution images, the features of the previous and subsequent phase images are extracted by two feature extraction modules with shared parameters. After the image undergoes convolution and pooling processing, residual convolutional blocks are used to extract multi-scale semantic features. The residual convolutional block consists of a ResNeXt unit with 2x downsampling and a 1×1 convolution. Then, the upsampling module obtains the same feature map size as the first layer residual block, resulting in output features that fuse high-level semantic information and low-level spatial information at different scales. The transformed feature maps of the four layers are concatenated on the channels and input into a 3×3 convolutional layer to obtain the final feature map.

4. The method for monitoring the environmental and soil disturbance range of a substation according to claim 3, characterized in that, When fusing information at different scales, data at the same scale must be fused. Upsampling and downsampling methods are used to maintain the same resolution for the image data. Specifically, the scale calculation method is as follows: ； ； In the formula, padding represents the padding of the feature data; kernel_size represents the size of the convolution kernel; stride represents the stride of each convolution movement; Indicates the height of the feature map; This indicates the width of the feature map.

5. The method for monitoring the environmental and soil disturbance range of a substation according to claim 4, characterized in that, The output of the c-th convolutional layer in the n-th residual convolutional block is represented as: ； In the formula, Represents the ReLU activation function. This represents the weight values ​​of the c-th convolutional layer; The output features of each layer of each residual convolutional block are mapped and concatenated, and the output features of the previous residual convolutional block and the output features of the current residual convolutional block are connected to each subsequent layer.

6. The method for monitoring the environmental and soil disturbance range of a substation according to claim 5, characterized in that, Three residual convolutional blocks are used as the backbone of the feature extraction network. After obtaining the local information features of each residual convolutional block, a global feature fusion and global residual learning are performed. The fully connected layer fuses the output features of each residual convolutional block and performs a global residual learning with the initial input features. The fused features are fed into a 3×3 convolution, and the resulting features are then fused with the original features to obtain the final feature result. Represented as: ； In the formula, Conv represents a 3×3 convolution operation. This represents the result of global fusion and residual learning of the features.

7. A monitoring system for implementing the substation environmental water conservation and land disturbance range monitoring method as described in any one of claims 1-6, characterized in that, The monitoring system includes: The image acquisition module is used to acquire pre- and post-phase images of the substation engineering area using a drone; The image processing module is used to decompose the previous and subsequent time-phase images into high-frequency and low-frequency components through wavelet transform. The high-frequency components are processed using bilinear interpolation, and the low-frequency components are processed using bicubic interpolation. Then, inverse wavelet transform is performed to output a high-resolution image. The encoder module is used to perform feature extraction processing on high-resolution images using a ResNet-50 feature extraction network that incorporates a lightweight channel attention module. It calculates the difference between the extracted feature layers and takes the absolute value to obtain a difference feature map. The decoder module is used to fuse the change feature map into the decoder part by upsampling the difference feature map through channel concatenation. In the decoder, the feature map enhanced by feature difference is decoded through an MLP structure containing only one hidden layer to generate a binary change prediction map and output the final change result image.

Images