Remote sensing monitoring method for power transmission and transformation project construction disturbance

Abstract

The invention discloses a power transmission and transformation project construction disturbance remote sensing monitoring method, which comprises the following steps: acquiring a satellite remote sensing image without cloud or with less cloud coverage in a research area, and selecting remote sensing observation images of two time phases before and after construction for each power transmission and transformation project; s2, establishing a power transmission and transformation project construction disturbance sample database on the basis of the multi-temporal remote sensing image data obtained in the step S1, and providing training samples for subsequent power transmission and transformation project construction disturbance monitoring based on deep learning; extracting image factor data related to construction disturbance from spectral bands, transformation indexes and texture information based on the very-high-resolution remote sensing image; and establishing a convolutional neural network model for monitoring construction disturbance. According to the method, the dual-time-phase remote sensing image is taken as a data source, target detection is carried out on construction disturbance by utilizing a deep learning framework through remote sensing image quality optimization processing, sample library establishment and optimization and remote sensing image variable selection and extraction, and state information of power transmission and transformation project construction disturbance can be effectively monitored.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing monitoring of construction disturbances of power transmission and transformation projects, and specifically to a remote sensing monitoring method for construction disturbances of power transmission and transformation projects, which uses dual-temporal remote sensing images as a data source, optimizes remote sensing image quality, establishes and optimizes a sample library, selects and extracts remote sensing image variables, and uses a deep learning framework to perform target detection on construction disturbances. Background Art

[0002] Power transmission and transformation projects are important infrastructure projects to ensure the safe and stable supply of electricity. In recent years, the number and scale of their construction have continued to grow. During the construction of power transmission and transformation projects, site leveling, foundation pit excavation, road construction and other construction projects will inevitably disturb the surface, damage vegetation, and accelerate the rate of soil erosion. They also have the characteristics of long construction period, high technical requirements, low overall soil erosion intensity but strong local point soil erosion. The large number of scattered tower base stations and line-laying construction access roads have caused certain damage to the surrounding ecological environment. Therefore, it is of great significance to do a good job in monitoring the disturbance of power transmission and transformation project construction and timely grasp the location, quantity, area and other information of the disturbance during the construction process, in order to ensure the conscientious implementation of the project soil and water conservation plan and guide the prevention and control of soil erosion.

[0003] In the prior art, the methods for monitoring disturbances in power transmission and transformation project construction mainly include: field distance measurement method and RTK measurement method. Among them, the field distance measurement method mainly uses a tape measure or a distance meter as a measuring tool, takes the longest and widest part of the construction range as the measuring point, and uses the side length * side length to directly obtain the disturbance area of ​​the tower base and road construction. This method is not only inefficient, but also difficult to verify on-site in rugged mountainous areas, and has low accuracy, which cannot meet the requirements of environmental protection and water conservation monitoring, and is likely to cause misjudgment of whether the disturbance area exceeds the standard. The RTK measurement method uses a mobile station to move along the border, save the coordinates at the obvious corner point, and after returning to the starting point, enter the area calculation interface, select all the points that enclose the area, and directly extract the area calculation result. Compared with the field distance measurement method, this method has improved measurement accuracy, but the accuracy of the RTK measurement area depends on the number of points in the field measurement. The more points are marked, the higher the accuracy is, and the fewer points are marked, the lower the accuracy is, so the accuracy of the RTK measurement area is not absolutely accurate. RTK instruments are expensive, and they require manpower and material resources, and have low work efficiency. Summary of the invention

[0004] In order to solve the defects of the above-mentioned technology, the present invention provides a remote sensing monitoring method for construction disturbance of power transmission and transformation engineering, which uses dual-phase remote sensing images as data source, optimizes remote sensing image quality processing, establishes and optimizes sample libraries, selects and extracts remote sensing image variables, and utilizes a deep learning framework to perform target detection on construction disturbances.

[0005] The technical solution adopted by the present invention to achieve the above technical effects is:

[0006] A method for remote sensing monitoring of disturbances in power transmission and transformation project construction, comprising the steps of:

[0007] S1. Satellite image acquisition and preprocessing: obtain satellite remote sensing images of the study area with no or little cloud coverage. For each power transmission and transformation project, remote sensing observation images of two time phases before and after construction are selected;

[0008] S2. Establish and optimize the sample library. On the basis of the multi-temporal remote sensing image data acquired in step S1, establish a sample database of power transmission and transformation project construction disturbances to provide training samples for subsequent power transmission and transformation project construction disturbance monitoring based on deep learning;

[0009] S3, remote sensing image variable selection and extraction, based on very high resolution remote sensing images, extract image factor data related to construction disturbance from spectral bands, transformation index, and texture information;

[0010] S4. Establish a convolutional neural network model for monitoring construction disturbances.

[0011] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbance of power transmission and transformation projects, the preprocessing includes radiation correction, geometric correction, and image fusion; wherein the radiation correction includes radiation calibration and atmospheric correction; the geometric correction is to use a geometric correction digital model to correct the errors caused by the combined influence of multiple factors in the remote sensing imaging process, such as the geometric position, shape, size, dimension, orientation and other characteristics of the objects on the original image and the corresponding ground objects; the image fusion is to combine two or more high-resolution single-band images with low-resolution multi-band images into a new multi-band image with high spatial resolution.

[0012] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbance of power transmission and transformation project, the sample library establishment includes two steps of sample preparation and sample expansion;

[0013] The sample preparation includes: preparing the construction disturbance samples of power transmission and transformation projects based on field surveys and combined with satellite images. The construction disturbance samples of power transmission and transformation projects include two parts: station projects and line projects. The station projects include four types of samples: station site area, access road, temporary production and living area, and temporary earth pile area. The line projects include four types of samples: tower base construction area, temporary construction road, crossing construction area, and tensioning field.

[0014] The sample expansion includes: using the Gamma transformation method to expand the sample, wherein the transformation formula is: y = (x + esp) a ; where y is the output image, x is the input image, a is the coefficient, and esp is a small quantity.

[0015] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbance of power transmission and transformation projects, the spectral bands include a blue light band, a green light band, a red light band, a red edge band, and a near infrared band.

[0016] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbance of power transmission and transformation project, the transformation index includes three normalized difference vegetation index NDVI, normalized difference greenness index NDGI and normalized soil brightness index NSBI which can effectively reflect vegetation and soil feature information;

[0017] in,

[0018] NDVI = (NIR-R) / (NIR+R);

[0019] NDGI = (GR) / (G+R);

[0020] NSBI = (R-NIR) / (R+NIR);

[0021] Among them, NIR is the reflection value of the near infrared band, R is the reflection value of the red light band, and G is the reflection value of the green light band.

[0022] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbances of power transmission and transformation projects, the texture information includes texture feature remote sensing variables including mean, variance, contrast, information entropy, consistency, second-order moment, correlation and dissimilarity extracted from the first principal component image of the remote sensing image.

[0023] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbance of power transmission and transformation project, the extraction of texture feature remote sensing variables of the first principal component image comprises the steps of:

[0024] The first principal component image of the remote sensing image is extracted using the principal component analysis method;

[0025] The gray level co-occurrence matrix method is used to extract the texture feature remote sensing variables including mean, variance, contrast, information entropy, consistency, second-order moment, correlation and dissimilarity of the first principal component image from different window sizes of 3*3, 5*5 and 7*7.

[0026] Preferably, in the above-mentioned remote sensing monitoring method for construction disturbance of power transmission and transformation project, the key variables are also screened, including the steps of:

[0027] Correlation analysis was used to eliminate variables with high correlation;

[0028] The random forest algorithm is used to screen and optimize variables to obtain key variables that have a greater impact on construction disturbance monitoring.

[0029] The beneficial effects of the present invention are as follows: the monitoring method of the present invention uses dual-phase remote sensing images as data sources, optimizes remote sensing image quality processing, establishes and optimizes sample libraries, selects and extracts remote sensing image variables, and uses a deep learning framework to perform target detection on construction disturbances. It can effectively monitor the status information of construction disturbances in power transmission and transformation projects, and the monitoring data is accurate and reliable. Compared with the existing field distance measurement method and RTK measurement method, the difficulty and cost of monitoring work can be greatly reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0031] In order to further understand the present invention, the present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0032] In the description of the present invention, it should be noted that the terms "vertical", "upper", "lower", "horizontal", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, "first", "second", "third", and "fourth" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0033] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection, it can be a mechanical connection, it can be an electrical connection, it can be a direct connection, it can be connected through an intermediate medium, and it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0034] See also Figure 1 As shown in the figure, an embodiment of the present invention proposes a remote sensing monitoring method for construction disturbance of a power transmission and transformation project, and the monitoring method comprises the steps of:

[0035] S1. Satellite image acquisition and preprocessing: obtain satellite remote sensing images of the study area with no or little cloud coverage. For each power transmission and transformation project, select remote sensing observation images of two time phases before and after construction.

[0036] Specifically, in an embodiment of the present invention, the preprocessing of remote sensing images includes radiation correction, geometric correction, and image fusion; wherein, the radiation correction of remote sensing images includes radiation calibration and atmospheric correction. The purpose of radiation calibration is to transform the measured value obtained by the sensor into a relative value related to physical quantities such as absolute brightness or surface reflectivity. Atmospheric correction is the process of eliminating the radiation error caused by the influence of atmospheric absorption and scattering, and inverting the real surface reflectivity of the object. The geometric correction of remote sensing images is to use the geometric correction digital model to correct the errors caused by the comprehensive influence of various factors in the remote sensing imaging process, such as the geometric position, shape, size, dimension, orientation and other characteristics of the object on the original image and the corresponding ground object. Image fusion of remote sensing images is to combine two or more high-resolution single-band images with low-resolution multi-band images into a new multi-band image with high spatial resolution. Through the above preprocessing, the quality of remote sensing images can be optimized to improve the accuracy of subsequent samples.

[0037] Specifically, in the embodiment of the present invention, the remote sensing image used is the Beijing-3 satellite image, including a panchromatic band with a spatial resolution of 0.3m and a deep blue band, blue band, green band, red band, red edge band, and near infrared band with a spatial resolution of 1.2m. The wavelength range of each band is: panchromatic band 450-800nm, deep blue band 400-450nm, blue band 450-520nm, green band 530-590nm, red band 620-690nm, red edge band 700-750nm, near infrared band 770-880nm. The imaging time of the image used is March 15 and May 28, 2023. First, the FLAASH (Fast line-of-sight atmospheric analysis of spectral hypercubes) correction model is used to perform atmospheric correction on remote sensing image data, effectively eliminating the influence of factors such as atmosphere and illumination on the reflection of objects, and obtaining more accurate real physical model parameters such as the radiation rate of objects. Then, the image of March 15 is selected as the benchmark, and control points are selected on the image of May 28 by image-to-image method. Finally, the cubic polynomial model is used to perform geometric correction on all images in turn, and finally the Pansharp algorithm is used to complete the batch fusion processing of panchromatic and multispectral orthophotos.

[0038] S2. Establish and optimize the sample library. Based on the multi-temporal remote sensing image data obtained in step S1, establish a power transmission and transformation project construction disturbance sample database to provide training samples for subsequent power transmission and transformation project construction disturbance monitoring based on deep learning.

[0039] Specifically, in an embodiment of the present invention, the establishment of a sample library includes two steps: sample preparation and sample expansion; wherein, sample preparation includes: preparing a sample of construction disturbance of a power transmission and transformation project based on field investigation and combined with satellite images; the sample of construction disturbance of a power transmission and transformation project includes two parts: station engineering and line engineering; the station engineering includes four types of samples: station site area, access road, temporary production and living area for construction, and temporary earth-pile area; the line engineering includes four types of samples: tower base construction area, temporary construction road, crossing construction area, and tensioning field. Sample expansion includes: using the Gamma transformation method to expand the sample, wherein the transformation formula is: y = (x + esp) a ; Where y is the output image, x is the input image, a is the coefficient, and esp is the small quantity. The construction disturbance samples of power transmission and transformation projects are prepared using ArcGIS software, and the samples are expanded by Gamma transformation. Through sample preparation and expansion, the total number of samples is guaranteed to be no less than 1,000 to ensure the accuracy of subsequent training model prediction.

[0040] S3. Remote sensing image variable selection and extraction: based on very high resolution remote sensing images, image factor data related to construction disturbance are extracted from spectral bands, transformation indexes, and texture information.

[0041] Specifically, in an embodiment of the present invention, the spectral bands include a blue light band, a green light band, a red light band, a red edge band, and a near infrared band. The transformation index includes three normalized difference vegetation indexes NDVI, normalized difference greenness index NDGI, and normalized soil brightness index NSBI that can effectively reflect vegetation and soil feature information;

[0042] in,

[0043] NDVI = (NIR-R) / (NIR+R);

[0044] NDGI = (GR) / (G+R);

[0045] NSBI = (R-NIR) / (R+NIR);

[0046] Among them, NIR is the reflection value of the near infrared band, R is the reflection value of the red light band, and G is the reflection value of the green light band.

[0047] The texture information includes texture feature remote sensing variables including mean, variance, contrast, information entropy, consistency, second-order moment, correlation and dissimilarity extracted from the first principal component image of the remote sensing image.

[0048] The extraction of the texture feature remote sensing variable of the first principal component image includes the following steps:

[0049] The first principal component image of the remote sensing image is extracted using the principal component analysis method;

[0050] The gray level co-occurrence matrix method is used to extract the texture feature remote sensing variables including mean, variance, contrast, information entropy, consistency, second-order moment, correlation and dissimilarity of the first principal component image from different window sizes of 3*3, 5*5 and 7*7.

[0051] Among them, principal component analysis (PCA) is a dimensionality reduction algorithm that can convert multiple indicators into a few principal components. These principal components are linear combinations of original variables and are independent of each other. They can reflect most of the information of the original data. The remote sensing image data used in this example has 5 bands (images). Through principal component analysis, most of the information in the 5 images is concentrated into a new image, namely the first principal component image. As a prior art, the specific principle process of the principal component analysis method will not be repeated here.

[0052] Specifically, in an embodiment of the present invention, after the remote sensing image variable extraction is completed, the key variables are screened, and the screening includes the steps of:

[0053] Correlation analysis was used to eliminate variables with high correlation;

[0054] The random forest algorithm is used to screen and optimize variables to obtain key variables that have a greater impact on construction disturbance monitoring.

[0055] Among them, in the embodiment of the present invention, the mean variables calculated by the 3*3 and 5*5 windows, the contrast variables calculated by the 5*5 and 7*7 windows, etc., have high correlation (≥0.9 or ≤-0.9). For such variables, only one of them is retained and the other is eliminated. Otherwise, the variables with high correlation will cause high data redundancy and cannot truly reflect the monitoring or evaluation results.

[0056] The random forest algorithm uses the bootstrap sampling method to randomly extract multiple samples with replacement, and performs decision tree modeling on the bootstrap sampling data obtained each time. The random forest algorithm has three parameters that need to be optimized, including ntree (the number of regression trees, the default value is 500), mtry (the number of segmentation variables for each node, for classification algorithms, the default value is one-third of the total number of variables used) and nodesize (the default value is 1). When establishing a random forest model, the embodiment of the present invention selects the ntree and mtry parameters based on the minimum RMSE of the modeling sample, and the nodesize selects the default value. The specific process is as follows:

[0057] 1. The bootstrap sample data set is randomly sampled from the training sample with replacement, and ntree sampling is performed. The ntree bootstrap sample data sets are used for regression modeling to build ntree regression trees. The unsampled data constitute ntree out-of-bag data sets (OOB).

[0058] 2. At each tree node, randomly select mtry explanatory variables for node segmentation;

[0059] 3. For each decision tree, an OOB error estimate can be obtained. Taking the average of the OOB error estimates of all decision trees in the forest, the generalization error estimate of the random forest can be obtained;

[0060] 4. Evaluate the importance value of each variable by predicting the error growth rate of out-of-band data, and sort the variables according to the importance value.

[0061] The variable screening and optimization is completed through the random forest algorithm to obtain the key variables that have a greater impact on the construction disturbance monitoring. In an embodiment of the present invention, the red light band, NDVI index, NDGI index, NSBI index, mean variable calculated by 5*5 window, variance variable calculated by 3*3 window, information entropy variable calculated by 7*7 window, etc., are used as key variables to participate in the construction disturbance range monitoring.

[0062] S4. Establish a convolutional neural network model for monitoring construction disturbances. Select the convolutional neural network model based on the sample data and variable data extracted from remote sensing images to complete the construction disturbance monitoring of power transmission and transformation projects. Specifically, add accuracy evaluation to the model test, and evaluate it through correctness and completeness. Determine the model by the influence of K value selection on accuracy, comparison of optimal image block size, comparison of weight distribution, number of convolutional layers, initialization method and activation function complexity, and finally complete the location, quantity and area information monitoring of construction disturbances of power transmission and transformation projects. The convolutional neural network model (CNN) is generally composed of three different types of layers: convolutional layer (C), pooling layer (pooling layer), and full-connected layer (FC). After multiple debugging parameter attempts, 4 layers of Conv1D are selected to extract feature values. After every 2 layers of Conv1D, 1 layer of MaxPooling1D is added to retain the main features and reduce the amount of calculation. Each convolutional layer uses the rectified linear unit (ReLU) function as the activation function to improve the neural network’s ability to express the model. The pooling layer is fixed to the maximum pooling layer with a window size of 2, followed by a fully connected layer for Softmax classification.

[0063] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions only describe the principles of the present invention. The present invention may have various changes and improvements without departing from the spirit and scope of the present invention, and these changes and improvements fall within the scope of the present invention to be protected, and the scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for remote sensing monitoring of disturbances in power transmission and transformation engineering construction, characterized in that: Includes steps: S1. Satellite image acquisition and preprocessing: obtain satellite remote sensing images of the study area with no or little cloud coverage. For each power transmission and transformation project, remote sensing observation images of two time phases before and after construction are selected; S2. Establish and optimize the sample library. On the basis of the multi-temporal remote sensing image data acquired in step S1, establish a sample database of power transmission and transformation project construction disturbances to provide training samples for subsequent power transmission and transformation project construction disturbance monitoring based on deep learning; S3, remote sensing image variable selection and extraction, based on very high resolution remote sensing images, extract image factor data related to construction disturbance from spectral bands, transformation index, and texture information; S4. Establish a convolutional neural network model for monitoring construction disturbances.

2. The method for remote sensing monitoring of power transmission and transformation engineering construction disturbance according to claim 1 is characterized in that: The preprocessing includes radiation correction, geometric correction, and image fusion; wherein the radiation correction includes radiation calibration and atmospheric correction; the geometric correction is to use a geometric correction digital model to correct the errors caused by the combined influence of multiple factors in the remote sensing imaging process, such as the geometric position, shape, size, dimension, orientation and other characteristics of the objects on the original image and the corresponding ground objects; the image fusion is to combine two or more high-resolution single-band images with low-resolution multi-band images into a new multi-band image with high spatial resolution.

3. The method for remote sensing monitoring of power transmission and transformation engineering construction disturbance according to claim 1 is characterized in that: The sample library establishment includes two steps: sample preparation and sample expansion; The sample preparation includes: preparing the construction disturbance samples of power transmission and transformation projects based on field surveys and combined with satellite images. The construction disturbance samples of power transmission and transformation projects include two parts: station projects and line projects. The station projects include four types of samples: station site area, access road, temporary production and living area, and temporary earth pile area. The line projects include four types of samples: tower base construction area, temporary construction road, crossing construction area, and tensioning field. The sample expansion includes: using the Gamma transformation method to expand the sample, wherein the transformation formula is: y = (x + esp) a ; where y is the output image, x is the input image, a is the coefficient, and esp is a small quantity.

4. The method for remote sensing monitoring of power transmission and transformation engineering construction disturbance according to claim 1 is characterized in that: The spectral bands include a blue light band, a green light band, a red light band, a red edge band, and a near infrared band.

5. The method for remote sensing monitoring of power transmission and transformation engineering construction disturbance according to claim 1, characterized in that: The transformation index includes three normalized difference vegetation index NDVI, normalized difference greenness index NDGI and normalized soil brightness index NSBI which can effectively reflect vegetation and soil information. in, NDVI = (NIR-R) / (NIR+R); NDGI = (GR) / (G+R); NSBI = (R-NIR) / (R+NIR); Among them, NIR is the reflection value of the near infrared band, R is the reflection value of the red light band, and G is the reflection value of the green light band.

6. The method for remote sensing monitoring of power transmission and transformation engineering construction disturbance according to claim 1, characterized in that: The texture information includes texture feature remote sensing variables including mean, variance, contrast, information entropy, consistency, second-order moment, correlation and dissimilarity extracted from the first principal component image of the remote sensing image.

7. The method for remote sensing monitoring of power transmission and transformation engineering construction disturbance according to claim 6 is characterized in that: The extraction of the texture feature remote sensing variable of the first principal component image comprises the steps of: The first principal component image of the remote sensing image is extracted using the principal component analysis method; The gray level co-occurrence matrix method is used to extract the texture feature remote sensing variables including mean, variance, contrast, information entropy, consistency, second-order moment, correlation and dissimilarity of the first principal component image from different window sizes of 3*3, 5*5 and 7*7.

8. The method for remote sensing monitoring of construction disturbance of power transmission and transformation project according to claim 7, characterized in that: It also includes the screening of key variables, including the steps: Correlation analysis was used to eliminate variables with high correlation; The random forest algorithm is used to screen and optimize variables to obtain key variables that have a greater impact on construction disturbance monitoring.

Images