Geological disaster monitoring method based on synthetic aperture radar and point sensor

Abstract

The present invention belongs to the technical field of pipeline geological disaster monitoring, and specifically relates to a geological disaster monitoring method based on synthetic aperture radar and point sensors. The method comprises the following steps: arranging synthetic aperture radar and point sensors in disaster-prone areas; collecting historical data and establishing an original database; preprocessing the original data to obtain a synthetic aperture radar image data set; offline training a pre-designed network and saving the trained network model; collecting real-time data, preprocessing the real-time data, and using the trained network model to perform online monitoring of geological disasters. The present invention has the characteristics of extracting features from an image angle, having low sensitivity to data quality, and being able to provide accurate early warning of geological disasters in oil and gas pipelines.

Description

Technical Field

[0001] The invention belongs to the technical field of pipeline geological disaster monitoring, and in particular relates to a geological disaster monitoring method based on synthetic aperture radar and point sensors. Background Art

[0002] At present, the geological structure of oil and gas pipelines in most areas is complex, the soil is highly erodible, soil erosion is serious, the terrain changes dramatically due to short-term heavy rainfall in summer, and the types of geological disasters are diverse. Oil and gas pipelines are inevitably threatened by cross-river floods, landslides and collapses, loess collapse and other disaster-causing factors, which pose a huge risk to pipeline operation. In addition, due to the long distance and large span of oil and gas pipelines, the geological disaster risks of oil and gas pipelines are highly uncertain and unpredictable. It is difficult to prevent and control them in advance with existing technical means, and the technical methods and management means for issuing forecasts and warnings are still lacking. Once an accident occurs, the losses are huge.

[0003] In the prior art, when monitoring geological disasters in oil and gas pipelines, the methods used are: analyzing the laws of geological disasters, establishing mathematical models and setting thresholds for early warning; or analyzing and early warning the stress of pipelines through finite element methods; or early warning through risks. The methods described lack experience in applying artificial intelligence to the field of oil and gas pipeline address disaster monitoring, and the accuracy and adaptability of the threshold early warning method are poor. When using image information for early warning, due to various reasons such as lighting, climate conditions, and surface vegetation cover, the analysis of images taken by cameras often cannot meet the accuracy requirements. In addition, the existing monitoring technology lacks the combination of multi-source data to analyze the early warning of geological disasters, and the accuracy is poor, which cannot meet the early warning requirements for geological disasters in oil and gas pipelines.

[0004] For example, the pipeline geological disaster monitoring and early warning method and device described in the Chinese invention patent with application number CN201810470350.2 includes: obtaining multiple geological environmental factors along the pipeline, the geological environmental factors along the pipeline include any one selected from the following: earthquake intensity, landform type, altitude, average annual rainfall, vegetation type, rock and soil type and hydrogeological type; based on the multiple geological environmental factors along the pipeline obtained, determining the geological environmental index along the pipeline; obtaining historical geological disaster data and weather information along the pipeline; according to the historical geological disaster data, geological environmental index and weather information along the pipeline, calculating the probability of pipeline geological disasters, and judging whether to perform early warning operations. Although it can monitor and warn of various geological disasters along the pipeline, and quantify the probability of pipeline geological disasters, and improve the reliability of pipeline geological disaster early warning results, its disadvantage is that its monitoring method is only based on the multiple geological environmental factors along the pipeline obtained and the corresponding established data model, the means are single, lack of multi-dimensional data fusion for monitoring and early warning, low accuracy, and has limitations. Summary of the invention

[0005] The present invention aims to overcome the problems in the prior art that the existing oil and gas pipeline geological disaster monitoring method lacks the combination of multi-source data, has poor accuracy, and cannot meet the early warning requirements for oil and gas pipeline geological disasters. The present invention provides a geological disaster monitoring method based on synthetic aperture radar and point sensors that extracts features from image angles, has low sensitivity to data quality, and can provide accurate early warning of oil and gas pipeline geological disasters.

[0006] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0007] The method for monitoring geological disasters based on synthetic aperture radar and point sensors includes the following steps:

[0008] S1, based on historical disaster data, deploy synthetic aperture radars and point sensors in disaster-prone areas;

[0009] S2, collect synthetic aperture radar image data and point sensor geological data to establish the original database;

[0010] S3, importing the synthetic aperture radar image data in the original database, extracting the region of interest ROI by using the three-frame difference method, and obtaining the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method;

[0011] S4, establishing the point sensor coordinates corresponding to the pixel points of the synthetic aperture radar image data, using a bilinear interpolation algorithm to interpolate the data collected by the point sensor, and then integrating it into the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method, to obtain a synthetic aperture radar image data set fused with the point sensor data;

[0012] S5, using the synthetic aperture radar image data set fused with point sensor data, offline training the pre-designed network and saving the trained network model;

[0013] S6, converting the synthetic aperture radar image data and point sensor data obtained in real time into a synthetic aperture radar image data set fused with the point sensor data through steps S3 and S4, and using the trained network model to perform online monitoring of geological disasters.

[0014] Preferably, step S3 comprises the following steps:

[0015] The images of the n+1th frame, the nth frame, and the n-1th frame are f n+1 、f n and f n-1 , the features of the corresponding pixels in the three frames are recorded as f n+1 (xij′), f n (xij′) and f n-1 (xij′), and the difference image D is obtained according to the following formula: n+1 and D n :

[0016] D n (xij′)=f n (xij′)-f n-1 (xij′)

[0017] Where xij′ is the coordinate of synthetic aperture radar image data, f n (xij′) is the complex value of the pixel of the nth frame image;

[0018] The obtained difference image D n+1 and D n According to the following formula, we can get the image D′ n :

[0019] D′ n (xij′)=(f n+1 (xij′)-f n (xij′))∩(f n (xij′)-f n-1 (xij′));

[0020] Then the obtained image D′n Perform threshold processing and connectivity analysis:

[0021] Right now,

[0022] Where T is the threshold, R′ n (xij′) represents the image D′ n Pixel value after threshold processing and connectivity analysis.

[0023] Preferably, step S4 comprises the following steps:

[0024] set up,

[0025] Among them, X means that each frame has n rows and n columns of data, and xnn means the data of the nth row and nth column. |xij ps |=m,xij(1:2)=R′ n (xij′),θ=tan -1 (b / a);

[0026] The collected synthetic aperture radar image data is a complex number, a represents the real part of the collected synthetic aperture radar image data, and b represents the imaginary part of the collected synthetic aperture radar image data; xij is the synthetic aperture radar image data fused with point sensor data; xij(1:2) represents the first two items of the xij data, specifically refers to the data features of the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method, that is, xij(1:2)=R′ n (xij′); xij ps is the point sensor data; |xij ps |=m means that the point sensor collects multi-dimensional data with a dimension of m.

[0027] Preferably, the network model described in step S5 is as follows:

[0028] The input is X matrix, m+2 channels, the sliding window is set to 10, the output is the probability of geological disasters, and the loss function adopts cross entropy. The corresponding formula is as follows:

[0029]

[0030] Among them, y i represents the label of sample i, the positive class is 1 and the negative class is 0; p i represents the probability that sample i is predicted to be a positive class, N represents the number of samples; the sample i is a synthetic aperture radar image data set fused with point sensor data; the positive class represents the occurrence of a geological disaster, and the negative class represents the absence of a geological disaster.

[0031] Preferably, the network offline training in step S5 adopts a back propagation algorithm, the principle of which is to optimize network parameters by a gradient descent method with the goal of minimizing the training loss function, wherein the formula for gradient descent is:

[0032]

[0033] Among them, θ i represents the parameters to be optimized in the network, θ i+1 Represents θ i The next iteration value, α represents the learning rate in gradient descent.

[0034] Preferably, each pixel in the synthetic aperture radar image data contains a phase value reflecting the surface microwave reflection intensity and related to the synthetic aperture radar slant range.

[0035] Compared with the prior art, the present invention has the following beneficial effects: (1) the data obtained by synthetic aperture radar has a higher resolution, and synthetic aperture radar is not restricted by light and climate conditions, and can realize all-day and all-weather ground observation. The quality of the collected data is stable, and the three-frame difference method is used to extract the area of ​​interest, which can also reduce the interference of the background; (2) combined with the data of point sensors, multi-dimensional data is used for early warning, which improves the accuracy of early warning; (3) the network model in the present invention adopts CNN to propose features, inputs them into the LSTM network, and combines with the attention mechanism, that is, extracts features from the perspective of the image, and considers the sequence information of the data, so that the monitoring and early warning of geological disasters in oil and gas pipelines are more accurate; (4) the present invention has good practicality, economic benefits and social benefits, and has broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of the method of the present invention;

[0037] Figure 2 A flow chart of the three-frame difference method in the present invention;

[0038] Figure 3 A position distribution diagram of a point sensor under a bilinear interpolation algorithm in the present invention;

[0039] Figure 4 A structural diagram of the network model in the present invention. DETAILED DESCRIPTION

[0040] In order to more clearly illustrate the embodiments of the present invention, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other accompanying drawings and other implementation methods can be obtained based on these accompanying drawings without creative work.

[0041] Embodiment 1:

[0042] The present invention provides a method for monitoring geological disasters based on synthetic aperture radar and point sensors. Based on the synthetic aperture radar technology and the data obtained by the point sensors, a synthetic aperture radar image data set integrating the point sensor data is established. By using the frame difference method and artificial intelligence method, accurate early warning of geological disasters in oil and gas pipelines can be carried out through learning.

[0043] like Figure 1 As shown, the specific method of the present invention is carried out according to the following steps:

[0044] S1, based on historical disaster data, deploy synthetic aperture radar and point sensors in disaster-prone areas;

[0045] S2, collect synthetic aperture radar image data and point sensor geological data to establish the original database;

[0046] Each pixel in the synthetic aperture radar image data contains a phase value reflecting the surface microwave reflection intensity and related to the synthetic aperture radar slant range (generally sampled at the slant range perpendicular to the platform flight direction). The above two information components can be represented by a complex number (a+b·i):

[0047]

[0048] S3, importing the synthetic aperture radar image data in the original database, extracting the region of interest ROI by using the three-frame difference method, and obtaining the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method;

[0049] The three-frame difference method in step S3 has the following specific process: Figure 2 As shown, the images of the n+1th frame, the nth frame, and the n-1th frame are respectively f n+1 、f n and f n-1 , the features of the corresponding pixels in the three frames are recorded as f n+1 (xij′), f n (xij′) and f n-1 (xij′), and the difference image D is obtained according to the following formula: n+1 and D n :

[0050] Dn (xij′)=f n (xij′)-f n-1 (xij′)

[0051] Where xij′ is the coordinate of synthetic aperture radar image data, f n (xij′) is the complex value of the pixel of the nth frame image;

[0052] The obtained difference image D n+1 and D n According to the following formula, we can get the image D′ n :

[0053] D′ n (xij′)=(f n+1 (xij′)-f n (xij′))∩(f n (xij′)-f n-1 (xij′));

[0054] Then the obtained image D′ n Perform threshold processing and connectivity analysis:

[0055] Right now,

[0056] Where T is the threshold, R′ n (xij′) represents the image D′ n Pixel value after threshold processing and connectivity analysis.

[0057] S4, establishing the point sensor coordinates corresponding to the pixel points of the synthetic aperture radar image data, using a bilinear interpolation algorithm to interpolate the data collected by the point sensor, and then integrating it into the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method, to obtain a synthetic aperture radar image data set fused with the point sensor data;

[0058] set up,

[0059] Among them, X means that each frame has n rows and n columns of data, and xnn means the data of the nth row and nth column. |xij ps |=m,xij(1:2)=R′ n (xij′),θ=tan -1 (b / a), θ is the complex-valued argument;

[0060] The collected synthetic aperture radar image data is a complex number, a represents the real part of the collected synthetic aperture radar image data, and b represents the imaginary part of the collected synthetic aperture radar image data; xij is the synthetic aperture radar image data fused with point sensor data; xij(1:2) represents the first two items of the xij data, specifically refers to the data features of the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method, that is, xij(1:2)=R′ n (xij′); xij ps is the point sensor data; |xij ps |=m means that the point sensor collects multi-dimensional data with a dimension of m.

[0061] The coordinate position of the point sensor is denoted by xij x ,xij y express.

[0062] Specifically, Figure 3 As shown, click Q 12 , Q 22 , Q 11 , Q 21 is the arrangement point of the point sensor, which can be directly obtained by collecting data from the point sensor. Since the density of the point sensor arrangement is much smaller than the pixel density of the synthetic aperture radar image data, if the data of point P in the figure is to be obtained, it can be achieved through bilinear interpolation. The specific steps are:

[0063] S41, perform linear interpolation in the X direction and insert blue points R1 and R2;

[0064]

[0065] where R1=(x,y1)

[0066]

[0067] where R2=(x,y2)

[0068] Among them, Q 11 =(x1,y1),Q 12 =(x1,y2),Q 21 =(x2,y1),Q 22 =(x2,y2), and we have:

[0069] f(Q 11 )=x11 ps ,x1=x11 x ,y1=x11 y , and similarly we can get the relationship of other points:

[0070] f(Q ij )=xijps ,x=xij x ,y=xij y ;

[0071] S42, after completing the interpolation in the X direction, do the interpolation in the Y direction, and calculate point P from points R1 and R2:

[0072]

[0073] Here, f(·) is a vector function with a dimension of m, that is, each dimension of data measured by the point sensor can be expanded by bilinear interpolation.

[0074] S5, using the synthetic aperture radar image data set fused with point sensor data, offline training the pre-designed network and saving the trained network model;

[0075] The network model structure is as follows: Figure 4 As shown, since the sensor collects m-dimensional data, plus the original two dimensions, the channel is set to m+2:

[0076] The input is X matrix, m+2 channels, the sliding window is set to 10, the output is the probability of geological disasters, and the loss function adopts cross entropy. The corresponding formula is as follows:

[0077]

[0078] Among them, y i represents the label of sample i, the positive class is 1 and the negative class is 0; p i represents the probability that sample i is predicted to be a positive class, N represents the number of samples; the sample i is a synthetic aperture radar image data set fused with point sensor data; the positive class indicates that a geological disaster has occurred, and the negative class indicates that no geological disaster has occurred.

[0079] Figure 4 The operation flow represented by the network structure model shown is as follows:

[0080] The collected data is input into the CNN (convolutional neural network) network to extract features, and an LSTM layer is added to extract sequence information. Combined with the attention mechanism, the output layer is connected to a softmax layer for multi-classification, and finally the warning information is output.

[0081] In step S5, the network offline training adopts the back propagation algorithm. The principle of the back propagation algorithm is to optimize the network parameters by the gradient descent method with the goal of minimizing the training loss function, wherein the formula of gradient descent is:

[0082]

[0083] In the formula, θ i represents the parameters to be optimized in the network, θ i+1 Represents θ i The next iteration value, α represents the learning rate in gradient descent.

[0084] S6, converting the synthetic aperture radar image data and point sensor data obtained in real time into a synthetic aperture radar image data set fused with the point sensor data through steps S3 and S4, and using the trained network model to perform online monitoring of geological disasters.

[0085] The process from step S1 to step S5 in the present invention is equivalent to Figure 1 The offline training process shown in;

[0086] The process described in step S6 of the present invention is equivalent to Figure 1 The online prediction process shown in .

[0087] The synthetic aperture radar (SAR) technology on which the present invention is based uses a small antenna to move at a constant speed along the trajectory of a long linear array and radiate coherent signals, and performs coherent processing on the echoes received at different positions, thereby obtaining a high-resolution imaging radar. Synthetic aperture radar has the characteristics of realizing all-day and all-weather earth observation without being restricted by light and climate conditions, and can even obtain information covered by the surface or vegetation. These characteristics make it have broad application prospects in civil fields such as agriculture, forestry, water or geology, and natural disasters.

[0088] The present invention fuses the data obtained by the point sensor with the data of the synthetic aperture radar to expand the angle of the image channel, and converts it into multi-channel, and uses multi-dimensional data for early warning, thereby improving the accuracy of the early warning; the network model in the present invention adopts CNN to propose features, inputs them into the LSTM network, and combines the attention mechanism, that is, extracts features from the perspective of the image, and considers the sequence information of the data, so that the monitoring and early warning of geological disasters in oil and gas pipelines are more accurate.

[0089] The above description is only a detailed description of the preferred embodiments and principles of the present invention. For ordinary technicians in this field, according to the ideas provided by the present invention, there will be changes in the specific implementation methods, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. A method for monitoring geological disasters based on synthetic aperture radar and point sensors, characterized in that: The steps include: S1, based on historical disaster data, deploy synthetic aperture radar and point sensors in disaster-prone areas; S2, collect synthetic aperture radar image data and point sensor geological data to establish the original database; S3, importing the synthetic aperture radar image data in the original database, extracting the region of interest ROI by using the three-frame difference method, and obtaining the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method; S4, establishing the point sensor coordinates corresponding to the pixel points of the synthetic aperture radar image, using a bilinear interpolation algorithm to interpolate the data collected by the point sensor, and then integrating it into the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method, to obtain a synthetic aperture radar image data set that integrates the point sensor data; S5, using the synthetic aperture radar image data set fused with point sensor data, offline training the pre-designed network and saving the trained network model; S6, converting the synthetic aperture radar image data and point sensor data collected in real time into a synthetic aperture radar image data set fused with the point sensor data through steps S3 and S4, and using the trained network model to perform online monitoring of geological disasters; Step S4 includes the following steps: set up, Among them, X means that each frame has n rows and n columns of data, and xnn means the data of the nth row and nth column. |xij ps |=m,xij(1:2)=R n ′(xij′),θ=tan -1 (b / a), θ is the complex-valued argument, R n ′(xij′) represents the image D n ′ is the pixel value after threshold processing and connectivity analysis, and xij′ is the coordinate of the synthetic aperture radar image data; The collected synthetic aperture radar image data is a complex number, a represents the real part of the collected synthetic aperture radar image data, and b represents the imaginary part of the collected synthetic aperture radar image data; xij is the synthetic aperture radar image data fused with point sensor data; xij(1:2) represents the first two items of the xij data, specifically refers to the data features of the synthetic aperture radar image data after the ROI is extracted by the three-frame difference method, that is, xij(1:2)=R n ′(xij′);xij ps is the point sensor data; |xij ps |=m means that the point sensor collects multi-dimensional data with a dimension of m; The network model described in step S5 is as follows: The input is X matrix, m+2 channels, the sliding window is set to 10, the output is the probability of geological disasters, and the loss function adopts cross entropy. The corresponding formula is as follows: Among them, y i represents the label of sample i, the positive class is 1 and the negative class is 0; p i represents the probability that sample i is predicted to be a positive class, N represents the number of samples; the sample i is a synthetic aperture radar image data set fused with point sensor data; the positive class represents the occurrence of a geological disaster, and the negative class represents the absence of a geological disaster.

2. The method for monitoring geological disasters based on synthetic aperture radar and point sensors according to claim 1, characterized in that: Step S3 includes the following steps: The images of the n+1th frame, the nth frame, and the n-1th frame are f n+1 、f n and f n-1 , the features of the corresponding pixels in the three frames are recorded as f n+1 (xij′), f n (xij′) and f n-1 (xij′), and the difference image D is obtained according to the following formula: n+1 and D n : D n (xij′)=f n (xij′)-f n-1 (xij′) Where xij′ is the coordinate of synthetic aperture radar image data, f n (xij′) is the complex value of the pixel of the nth frame image; The obtained difference image D n+1 and D n According to the following formula, we can get image D n ′: D n ′(xij′)=(f n+1 (xij′)-f n (xij′))∩(f n (xij′)-f n-1 (xij′)); Then the obtained image D n 'Perform threshold processing and connectivity analysis: Right now, Where T is the threshold, R n ′(xij′) represents the image D n ′Pixel value after threshold processing and connectivity analysis.

3. The method for monitoring geological disasters based on synthetic aperture radar and point sensors according to claim 1, characterized in that: The network offline training described in step S5 adopts the back propagation algorithm. The principle of the back propagation algorithm is to optimize the network parameters by the gradient descent method with the goal of minimizing the training loss function, wherein the formula of gradient descent is: Among them, θ i represents the parameters to be optimized in the network, θ i+1 Represents θ i The next iteration value, α represents the learning rate in gradient descent.

4. The method for monitoring geological disasters based on synthetic aperture radar and point sensors according to claim 1 or 3, characterized in that: Each pixel in the synthetic aperture radar image data contains a phase value reflecting the microwave reflection intensity of the ground surface and related to the slant range of the synthetic aperture radar.

Images