Batch deployment method for power station pipeline displacement on-line measurement devices

Abstract

The invention relates to a batch deployment method for power station pipeline displacement on-line measurement devices, and specifically, the method comprises the steps: generating an incremental data set based on a camera parameter range covering a plurality of expected working conditions; the camera parameters at least comprise camera internal parameters and camera external parameters; constructing and training a modular displacement reconstruction network; the training uses the incremental data set and the corresponding camera parameters; the modular displacement reconstruction network comprises at least one internal reference module used for processing information related to camera internal reference; the method comprises the following steps: acquiring actual internal reference of a camera actually used on site and a shot on-site target image, and identifying angular points in the on-site target image; and determining the weight of the internal reference module of the modular displacement reconstruction network by using the angular points and the actual internal reference, and completing migration deployment. Repeated training due to internal reference change is avoided, and the method is suitable for batch deployment.

Description

Technical Field

[0001] The present invention relates to the fields of computer vision and artificial intelligence, and specifically to a method for batch deployment of an on-line measurement device for pipeline displacement in a power station. Background Art

[0002] An on-line displacement monitoring device is a device that uses the target images captured by left and right cameras to identify corner points to establish a mapping relationship, and then obtains the three-dimensional coordinates of an object. The vision-based displacement measurement device can realize the three-dimensional displacement measurement of the object to be measured. Its basic process is as follows: calibrate the target to obtain the internal and external parameters of the stereo vision camera; collect the target images and extract the corner points of the target images; use a three-dimensional displacement reconstruction algorithm or a deep learning method to realize displacement calculation. Chinese Patent with the publication number CN110415300B discloses a dynamic displacement measurement method for a stereo vision structure based on three target planes, which introduces pasting concentric circle targets at the measuring points on the surface of the structure, placing a binocular stereo vision system in front of the side where the target is pasted on the structure, and then calibrating the two cameras using the checkerboard calibration method to obtain the internal and external parameters of the binocular stereo vision system, and the subsequent process of obtaining the three-dimensional displacement of the target on the surface of the structure in the structure coordinate system. Chinese Patent with the publication number CN115100284A discloses a binocular feature matching displacement measurement method and device, which emphasizes the process of calculating the scale in the x direction and the scale in the y direction corresponding to the feature point according to the current frame distance corresponding to the feature point and the parameters of the binocular cameras, and finally obtaining the displacement. It can be seen that obtaining the internal and external parameters of the stereo vision camera is one of the cores of the stereo vision measurement method. However, when using the stereo vision algorithm for deployment, it is necessary to obtain the internal parameters of each camera. When deploying each device, it is necessary to consider that the internal parameters of the cameras of different devices are different and train them separately. This not only takes a long training time and affects the on-line measurement effect of the displacement, but also causes huge economic costs and time costs when deploying a large number of devices separately. In large-scale deployment under complex working conditions, it is difficult to achieve rapid displacement measurement, which affects the monitoring efficiency of the equipment. Summary of the Invention

[0003] Aiming at the problems of high time cost and large economic cost caused by different internal parameters of cameras in the displacement measurement device based on the stereo vision principle and the need for separate training of different devices when batch-deploying the on-line measurement device for pipeline displacement in a power station, the present invention provides a method for batch deployment of an on-line measurement device for pipeline displacement in a power station, and the method includes:

[0004] Generating an incremental data set based on a camera parameter range covering a variety of expected working conditions; the camera parameters at least include camera internal parameters and camera external parameters.

[0005] Construct and train a modular displacement reconstruction network; the training uses the incremental dataset and the corresponding camera parameters; the modular displacement reconstruction network includes at least one internal parameter module for processing information related to the camera internal parameters.

[0006] Obtain the actual internal parameters of the camera actually used on site and the on-site target images taken, and identify the corner points in the on-site target images; use the corner points and the actual internal parameters to determine the weights of the internal parameter module of the modular displacement reconstruction network, and complete the migration and deployment.

[0007] Optionally, the step of generating the incremental dataset includes: using the preset three-dimensional target structure parameters, the selected camera internal parameters and the camera external parameters, simulating the camera imaging process through the stereo vision algorithm, and generating simulated image data including the target pixel coordinates and their corresponding gray values.

[0008] Optionally, the step of training the modular displacement reconstruction network includes: taking the target pixel coordinates of the simulated images in the incremental dataset and the camera internal parameters used to generate the simulated images as network inputs, and taking the target three-dimensional spatial positions obtained from the camera external parameters as training targets.

[0009] Optionally, the modular displacement reconstruction network further includes a spatial module and a feature reconstruction module; the spatial module is used to extract spatial features according to the input target pixel coordinate information; the feature reconstruction module is used to combine the internal parameter features and the spatial features and output displacement or coordinate information.

[0010] Optionally, the internal parameter module is composed of at least one fully connected neural network layer, and its input is the camera internal parameter, and the camera internal parameter includes focal length, principal point coordinates, radial distortion parameters and tangential distortion parameters.

[0011] Optionally, the spatial module is composed of at least one fully connected neural network layer and includes a tensor expansion operation, and its input is the pixel coordinates of the corner points identified in the on-site target image.

[0012] Optionally, the feature reconstruction module performs tensor operations on the internal parameter features output by the internal parameter module and the spatial features output by the spatial module, and processes them through at least one fully connected neural network layer to output the three-dimensional displacement or coordinate information of the power station pipeline.

[0013] Optionally, the use of the corner points and the actual internal parameters to determine the weights of the internal parameter module of the modular displacement reconstruction network is specifically:

[0014] Input the actual internal parameters into the internal parameter module for a forward calculation once, extract and fix the output internal parameter feature tensor as the effective weight state of the internal parameter module after migration and deployment; in subsequent measurement inferences, directly use the fixed internal parameter feature tensor.

[0015] Optionally, before identifying the corner points in the on-site target image, the method further includes: applying an image restoration algorithm to the on-site target image for processing, so that the characteristics of the processed on-site target image are consistent with the characteristics of the simulated images in the incremental dataset.

[0016] Optionally, the image restoration algorithm includes: preliminarily identifying the corner points in the on-site target image; calculating the point spread function (PSF) of the camera actually used on-site based on the preliminarily identified corner points and the inclined edge method; performing image restoration processing on the on-site target image using the point spread function (PSF).

[0017] The present invention pre-generates an incremental dataset covering a range of camera parameters for various expected working conditions, and uses this dataset to train a modular displacement reconstruction network, realizing the centralization and generalization of model training. During actual deployment, only the actual internal parameters of the on-site camera and the captured target image need to be obtained, and by identifying the corner points and using the actual internal parameters to determine the weights of the internal parameter module, the migration and deployment can be completed. This method greatly shortens the deployment time of a single device, thereby significantly reducing the overall time cost. Description of the Drawings

[0018] Figure 1 It is a flowchart of Embodiment 1;

[0019] Figure 2 It is a logical flowchart of the present invention in the image classification task. Detailed Embodiments

[0020] In the embodiments of the present invention, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific way for easy understanding.

[0021] It will be understood that the "embodiments" referred to throughout the specification mean that specific features, structures, or characteristics related to the embodiments are included in at least one embodiment of the present application. Thus, the various embodiments mentioned throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. It will be understood that in the various embodiments of the present application, the magnitude of the serial numbers of the various processes does not mean the order of execution, and the order of execution of the various processes should be determined by their functions and internal logics, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0022] In the present invention, unless otherwise specified, the same or similar parts between the various embodiments may be referred to each other. In the various embodiments of the present invention, as well as in each implementation manner / implementation method / realization method in each embodiment, if there is no special specification and logical conflict, the terms and / or descriptions between different embodiments, as well as between each implementation manner / implementation method / realization method in each embodiment, are consistent and can be mutually referred to. The technical features in different embodiments, as well as in each implementation manner / implementation method / realization method in each embodiment, can be combined to form new embodiments, implementation manners, implementation methods, or realization methods according to their internal logical relationships. The implementation manners of the present application described below do not constitute a limitation to the protection scope of the present application.

[0023] Figure 1 The flowchart of the method for batch deployment of the on-line measurement device for the displacement of power station pipelines provided by the present invention is shown, as Figure 1 shown, the method includes:

[0024] Generating an incremental data set based on a range of camera parameters covering a variety of expected working conditions; the camera parameters at least include camera internal parameters and camera external parameters.

[0025] The internal parameters of different cameras are different, and the camera models may also be different. There are differences in their inherent internal parameters (internal parameters) and external positions and postures (external parameters) relative to the monitoring target. Determining the reasonable variation range of these parameters gives the range of camera parameters. For example, the focal length may vary within a certain interval, the principal point coordinates may deviate near the center of the image, the distortion parameters may be set within a range according to the characteristics of different lenses, and the camera external parameters need to consider different installation positions and angles. Then, using these parameter ranges, a large number of camera parameter combinations are generated by random sampling or systematic combination. For each set of parameters, using the three-dimensional target model technique, the imaging process of the camera under these parameters is simulated to generate the corresponding target image, that is, the incremental data set.

[0026] Construct and train a modular displacement reconstruction network; the training uses the incremental dataset and the corresponding camera parameters; the modular displacement reconstruction network includes at least one internal parameter module for processing information related to camera internal parameters.

[0027] Construct a modular displacement reconstruction network, which includes an input layer, a feature extraction layer, an internal parameter module, a spatial module, a feature reconstruction module, and a final output layer. Among them, the input layer receives the target image or corner coordinates and camera internal parameters, and the feature extraction layer uses, for example, a convolutional neural network to process the image. Train using a large amount of generated simulation data and the corresponding camera parameters to enable the network to learn the mapping relationship from input to output.

[0028] Obtain the actual internal parameters of the camera actually used on-site and the on-site target image taken, and identify the corners in the on-site target image; use the corners and the actual internal parameters to determine the weights of the internal parameter module of the modular displacement reconstruction network to complete the migration and deployment.

[0029] During actual deployment, calibrate the camera used on-site to obtain the camera internal parameters. Preferably, use traditional camera calibration methods such as Zhang Zhengyou calibration method or use the taken target image and calibration tool. Take a target image on-site and use a corner detection algorithm, such as Harris corner detector or Shi-Tomasi corner detector, etc., to identify the corners in the image. Input the obtained actual internal parameters into the internal parameter module of the pre-trained network, obtain the output features of this module through a single forward calculation, and use this feature as one of the conditions for subsequent displacement calculation, or directly fix the weights of this module to achieve migration and deployment.

[0030] In an optional embodiment, the step of generating the incremental dataset includes: using the preset three-dimensional target structure parameters, the selected camera internal parameters and external parameters, and simulating the camera imaging process through a stereo vision algorithm to generate simulation image data including the target pixel coordinates and their corresponding gray values.

[0031] Define the point coordinates of the three-dimensional target, and then for the selected camera internal parameters (focal length, principal point, distortion parameters) and external parameters (rotation vector and translation vector), use the cv2.projectPoints() function to project the three-dimensional target points onto the two-dimensional image plane to obtain the corresponding pixel coordinates. In an optional embodiment, in order to generate a simulation image including gray values, use a rendering engine to render the scene into a two-dimensional image to obtain simulation image data including the target pixel coordinates and their corresponding gray values.

[0032] In an alternative embodiment, the step of training the modular displacement reconstruction network includes: using the target pixel coordinates of the simulated images in the incremental dataset and the camera internal parameters used to generate the simulated images as network inputs, and using the target three-dimensional spatial positions obtained from the camera external parameters as the training targets.

[0033] When training the network, it is preferred to use deep learning frameworks such as TensorFlow or PyTorch. For each piece of training data, the pixel coordinates of the target in the simulated image and the camera internal parameters used to generate the image are used as the network inputs. The output layer of the network needs to predict the position of the target in three-dimensional space. The three-dimensional position is calculated from the camera external parameters used to generate the simulated image, for example, by inverse transformation to convert the coordinates of the target in the camera coordinate system to the world coordinate system. A loss function is used to measure the difference between the three-dimensional position predicted by the network and the true three-dimensional position, and the backpropagation algorithm is used to optimize the parameters of the network.

[0034] In an alternative embodiment, the modular displacement reconstruction network further includes a spatial module and a feature reconstruction module; the spatial module is used to extract spatial features according to the input target pixel coordinate information; the feature reconstruction module is used to combine the internal parameter features and the spatial features and output displacement or coordinate information.

[0035] Preferably, the spatial module is a multi-layer perceptron, and the input is the flattened coordinate vector of the target corner points. The feature reconstruction module concatenates the outputs of the internal parameter module and the spatial module, and then further processes them through a series of fully connected layers, and finally outputs the predicted displacement or coordinates. Of course, more complex network structures can also be used. For example, the spatial module can use a convolutional neural network (CNN) if the input is a complete target image, and the feature reconstruction module can use an attention mechanism to better fuse the internal parameter features and the spatial features.

[0036] In an alternative embodiment, the internal parameter module is composed of at least one fully connected neural network layer, and its input is the camera internal parameter, and the camera internal parameter includes focal length, principal point coordinates, radial distortion parameters, and tangential distortion parameters.

[0037] In a specific embodiment, the internal parameter module is a multi-layer perceptron. The input is a vector of camera internal parameters, such as [fx, fy, cx, cy, k1, k2, p1, p2, k3], where fx and fy are the focal lengths in the x and y directions, cx and cy are the principal point coordinates, k1, k2, and k3 are the radial distortion parameters, and p1 and p2 are the tangential distortion parameters. This vector is input into one or more fully connected layers for processing, and a feature vector with a fixed length is output.

[0038] In an optional embodiment, the spatial module is composed of at least one fully-connected neural network layer and includes a tensor expansion operation, and its input is the pixel coordinates of the corner points identified in the on-site target image. In a more specific embodiment, the spatial module is also an MLP. Assuming that N corner points are identified in the on-site target image, and each corner point has two coordinates (x, y), the coordinates of these N corner points are flattened into a 2N-dimensional vector as the input. The tensor expansion operation can be to perform some operations on this 2N-dimensional vector before inputting it into the fully-connected layer. For example, geometric features such as the distances and angles between the corner points can be calculated and added to the input vector, thereby expanding the dimension of the input tensor and enabling the network to learn richer spatial information.

[0039] In an optional embodiment, the feature reconstruction module performs tensor operations on the internal parameter features output by the internal parameter module and the spatial features output by the spatial module, and is processed by at least one fully-connected neural network layer to output the three-dimensional displacement or coordinate information of the power station pipeline. The feature reconstruction module uses a tensor splicing operation to connect the output feature vectors of the internal parameter module and the spatial module. For example, the internal parameter module outputs a feature vector of length M, and the spatial module outputs a feature vector of length K, then the length of the spliced vector is M + K. Then, the spliced vector is input into one or more fully-connected neural network layers, and these fully-connected layers are responsible for learning the final non-linear mapping to convert the fused features into the three-dimensional displacement or coordinate information of the power station pipeline.

[0040] In an optional embodiment, determining the weights of the internal parameter module of the modular displacement reconstruction network by using the corner points and the actual internal parameters specifically includes:

[0041] Input the actual internal parameters into the internal parameter module for a forward calculation once, extract and fix the output internal parameter feature tensor as the effective weight state of the internal parameter module after migration and deployment; in subsequent measurement inferences, directly use the fixed internal parameter feature tensor.

[0042] In the migration and deployment stage, when the actual internal parameters of the on-site camera are obtained, input these internal parameters into the pre-trained internal parameter module, perform a forward propagation once, and obtain the output feature vector of the module. Regard this output feature vector as the internal parameter feature of this specific camera, and in the subsequent displacement measurement inference process, fuse this fixed feature vector with the outputs of other modules such as the spatial module according to the features extracted from the on-site target image corner points to obtain the final displacement or coordinate prediction. The actual internal parameters activate the internal parameter module and use its output as a condition for subsequent calculations without the need to retrain the entire network or the internal parameter module.

[0043] In an alternative embodiment, before identifying the corner points in the on-site target image, the method further includes: applying an image restoration algorithm to the on-site target image for processing to make the characteristics of the processed on-site target image consistent with the characteristics of the simulated images in the incremental dataset. Image restoration is performed using image processing functions such as those in the OpenCV library. For example, Gaussian filtering is used to smooth the image and remove noise, histogram equalization is used to enhance the contrast of the image, or more complex denoising algorithms such as non-local means denoising are used.

[0044] In an alternative embodiment, the image restoration algorithm includes: preliminarily identifying the corner points in the on-site target image; calculating the point spread function (PSF) of the camera actually used on-site based on the preliminarily identified corner points and the slanted edge method; performing image restoration processing on the on-site target image using the point spread function (PSF). A corner detection algorithm is used to preliminarily identify the corner points in the on-site target image. Then, a clear edge region in the image is selected, and the slanted edge method is used to estimate the point spread function (PSF) of the camera. After obtaining the estimated PSF, an image deconvolution algorithm is used to apply the PSF to the original on-site target image to restore clearer details in the image, thereby improving the accuracy of corner detection.

[0045] Figure 2 A method for batch deployment of an on-line measurement device for pipeline displacement in a power station is shown, which can generally be divided into two steps: incremental learning and migration deployment.

[0046] As Figure 2 shown, incremental learning includes S101, S102, S103, S104, S105, S106. To ensure that the incremental learning can include the datasets required in the migration process, all working conditions in the migration need to be considered when setting the parameter range in S101, and the parameters to be set are the internal camera parameters and external camera parameters in S102.

[0047] The target three-dimensional structure parameters in S103 are (x w , y w , z w , G), which are the spatial coordinate points and the gray intensity at that coordinate respectively. The parameters in S102 and the target three-dimensional structure parameters in S103 are input into the stereo vision algorithm to generate the incremental dataset S104.

[0048] After identifying the corner points in the incremental dataset using the corner point recognition algorithm in S105, the corner point recognition results, internal camera parameters, and external camera parameters are input into the modular displacement network for incremental learning.

[0049] S111 - S116 is the migration and deployment process. This process involves S111 collecting on - site images and camera internal parameters, S112 inputting the on - site images into the image restoration algorithm, S113 performing corner recognition, S114 inputting the corner coordinates of the on - site images into the modular displacement reconstruction network, and S116 the modular displacement reconstruction network performing inference and outputting the reconstruction data.

[0050] The core of the method for batch deployment of the on - line measurement device for pipeline displacement in the power station in the embodiment of the present invention lies in: 1) Incremental learning: generating an incremental data set using a stereo vision algorithm, and then using the modular displacement reconstruction network to learn the parameters of all working conditions; 2) Migration and deployment: using the image restoration algorithm to restore the on - site target image so that the restored image is consistent with the image generated by the stereo vision algorithm, then deploying the modular displacement reconstruction network, inputting the corner recognition algorithm into the modular displacement reconstruction network for inference, and outputting the displacement result.

[0051] Specifically, for the camera internal parameters and camera external parameters described in S102, the internal parameters include the basic internal parameter K, the radial distortion parameter k i，i＝1…4 , and the tangential distortion parameter p i，i＝1，2 , and the camera external parameter [R|T] includes:

[0052]

[0053] In the formula, f x , f y are the scales of the focal length in the X - axis and Y - axis directions, c x , c y are the pixel coordinate positions of the principal point, f s is the tilt parameter, R is the rotation matrix, T is the translation matrix, r is the rotation matrix parameter, and t is the translation matrix parameter.

[0054] Specifically, in the process of generating the data set by the stereo vision algorithm in S103, the input data are three - dimensional structure parameters, camera internal parameters, and camera external parameters. The three - dimensional structure parameters of the target include the three - dimensional coordinates of the target color and the gray value of the color at this coordinate. The generation process is as follows:

[0055]

[0056] In the formula, u and v are the pixel coordinates of the camera, which need to be calculated separately for the left and right cameras to obtain the pixel coordinates u l , v l , u r , v r .

[0057] Specifically, in the process of generating the data set in S103, the calculation process of the gray value R(u, v) of the image in the pixel coordinates (u, v) is as follows:

[0058]

[0059] where t exp is the exposure time, N is the set of all spatial points falling on the pixel coordinates, F is the shutter size, λ

[0060] is the wavelength, I is the radiation intensity varying with the three-dimensional coordinate points and the wavelength, ξ is the transmittance of the filter, QE is the quantum conversion efficiency of the camera at different wavelengths, and Sm represents the conversion process from the electrical signal to the gray value, which includes analog-to-digital conversion and non-linear adjustment, etc.

[0061] Specifically, for the incremental learning process of the modular displacement reconstruction network in S106, the pixel coordinates of the corner points of the left and right cameras and the internal parameters of the left and right cameras are used as inputs, and the corresponding three-dimensional spatial positions of the corner points can be obtained from the external parameters of the cameras.

[0062] Specifically, for the camera internal parameter 1 described in S111, its value is the camera parameter in the actual layout process; during the migration process, there are differences in each camera internal parameter, and N is the number of units for migration and deployment.

[0063] The image restoration algorithm described in S112 is characterized by including S201 corner point recognition, S202 calculating LSF(u) and LSF(v) by the tilted edge method, S203 calculating the point spread function PSF, and S204 restoring the image using the PSF.

[0064] For the S202 tilted edge method to calculate LSF(u) and LSF(v), it is characterized by selecting the region of interest centered on the corner point coordinates, and LSF(u) can be expressed as:

[0065]

[0066] where u and v are pixel coordinates, and ESF is the edge spread function.

[0067] The method for calculating the point spread function PSF includes two steps: 1) calculating the PSF of different regions; 2) taking the average of all PSFs of a single target, and this process can be expressed as:

[0068]

[0069] where I is the number of corner points calculated on an image of a single target.

[0070] The image restoration algorithm in S204 is characterized by restoring the on-site image using the calculated PSF, and this process can be expressed as:

[0071]

[0072] where H * (u, v) is the conjugate of the PSF spectrum, and |H(u, v)| 2 is the power spectrum of the PSF, wn is the Wiener constant, and G(u, v) is the Fourier transform of the image.

[0073] Specifically, the modular displacement reconstruction network is characterized by consisting of an internal parameter module, a feature reconstruction module, and a spatial module.

[0074] Specifically, the internal parameter module consists of three fully connected neural networks, with the number of neurons in each layer being 22, 16, and 64, constituting a compressive sensing algorithm. A ReLu activation function is used for activation after each fully connected layer.

[0075] Specifically, the input of the internal parameter module is K l , K r , During the incremental learning process, the camera internal parameter module participates in the complete tensor calculation; during the migration and deployment phase, only the output tensor of the last fully connected layer of the internal parameter module is extracted after the first input of the camera internal parameters, and it is fixed as the internal parameter feature of the deployment device. During subsequent inference processes, the camera internal parameter module no longer participates in real-time tensor calculations, thereby achieving fast cross-device migration and deployment.

[0076] The spatial module consists of three fully connected layers and tensor expansion. The number of neurons in the three fully connected layers is 4, 16, and 64 respectively, and a ReLu activation function is used for activation in each fully connected layer. The output tensor of the third fully connected layer is automatically expanded to 64*64 using the broadcast mechanism.

[0077] The feature reconstruction module is characterized in that it first performs a tensor multiplication on the outputs of the internal parameter module and the spatial module, and then outputs tensors through three fully connected layers. The number of neurons in the three fully connected layers is 64, 16, and 3 respectively. Except for the direct output of the last layer, ReLu activation functions are used for activation in the other two layers.

[0078] The above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center integrating one or more available media. The available medium may be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0079] The steps of the methods or algorithms described in the embodiments of the present application may be directly embedded in hardware, a software unit executed by a processor, or a combination of the two. The software unit may be stored in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium in the art. Exemplarily, the storage medium may be connected to the processor so that the processor can read information from the storage medium and write information to the storage medium. Optionally, the storage medium may also be integrated into the processor. The processor and the storage medium may be provided in an ASIC.

[0080] These computer program instructions may also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 the steps of the functions specified in one block or multiple blocks.

[0081] Although the present application has been described in connection with specific features and their embodiments, it will be apparent that various modifications and combinations can be made without departing from the spirit and scope of the present application. Accordingly, the present specification and drawings are merely exemplary illustrations of the present application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and modifications.

Claims

1. A method for batch deployment of an on-line measurement device for pipeline displacement in a power station, characterized in that, The method includes: Generating an incremental data set based on a camera parameter range covering multiple expected working conditions; the camera parameters at least include camera internal parameters and camera external parameters; Constructing and training a modular displacement reconstruction network; the training uses the incremental data set and the corresponding camera parameters; the modular displacement reconstruction network includes at least one internal parameter module for processing information related to camera internal parameters; Obtaining the actual internal parameters of the camera actually used on site and the on-site target image captured, and identifying the corner points in the on-site target image; determining the weights of the internal parameter modules of the modular displacement reconstruction network using the corner points and the actual internal parameters to complete the migration and deployment.

2. The method according to claim 1, characterized in that, The step of generating the incremental data set includes: using the preset three-dimensional target structure parameters, the selected camera internal parameters and camera external parameters, simulating the camera imaging process through a stereo vision algorithm to generate simulated image data including the target pixel coordinates and their corresponding gray values.

3. The method according to claim 1 or 2, characterized in that, The step of training the modular displacement reconstruction network includes: using the target pixel coordinates of the simulated image in the incremental data set and the camera internal parameters used to generate the simulated image as network inputs, and using the three-dimensional spatial position of the target obtained from the camera external parameters as the training target.

4. The method according to claim 1, characterized in that, The modular displacement reconstruction network further includes a spatial module and a feature reconstruction module; the spatial module is used to extract spatial features according to the input target pixel coordinate information; the feature reconstruction module is used to combine the internal parameter features and the spatial features to output displacement or coordinate information.

5. The method according to claim 1, wherein The internal parameter module is composed of at least one fully connected neural network layer, and its input is the camera internal parameter, and the camera internal parameter includes focal length, principal point coordinates, radial distortion parameters and tangential distortion parameters.

6. The method according to claim 4, wherein The spatial module is composed of at least one fully connected neural network layer and includes a tensor expansion operation, and its input is the pixel coordinates of the corner points identified in the on-site target image.

7. The method according to claim 4, wherein The feature reconstruction module performs tensor operations on the internal parameter features output by the internal parameter module and the spatial features output by the spatial module, and processes them through at least one fully connected neural network layer to output the three-dimensional displacement or coordinate information of the power station pipeline.

8. The method according to claim 1, wherein The step of determining the weights of the internal parameter modules of the modular displacement reconstruction network using the corner points and the actual internal parameters is specifically: Inputting the actual internal parameters into the internal parameter module for a forward calculation once, extracting and fixing the output internal parameter feature tensor as the effective weight state of the internal parameter module after migration and deployment; in subsequent measurement inferences, directly use the fixed internal parameter feature tensor.

9. The method according to claim 1, wherein Before identifying the corner points in the on-site target image, the method further includes: processing the on-site target image using an image restoration algorithm so that the characteristics of the processed on-site target image are consistent with the characteristics of the simulated image in the incremental data set.

10. The method according to claim 9, characterized in that, The image restoration algorithm includes: preliminarily identifying the corner points in the on-site target image; calculating the point spread function (PSF) of the camera actually used on site based on the preliminarily identified corner points and the inclined edge method; performing image restoration processing on the on-site target image using the point spread function (PSF).

Images