An intelligent imaging device downhole and an image processing method based on phased array ultrasound

Abstract

The application discloses a downhole intelligent imaging device and an image processing method based on a phased array ultrasonic wave. The downhole intelligent imaging device comprises a downhole ultrasonic imaging tool, a ground receiving device and a signal transmission multi-core composite cable. The downhole ultrasonic imaging tool is used for penetrating into a downhole environment, emitting an ultrasonic wave beam and collecting a target echo signal; the ground receiving device is used for receiving uploaded data on the ground and displaying; and the signal transmission multi-core composite cable is used for connecting the downhole ultrasonic imaging tool and the ground receiving device, realizing carrier communication control and data transmission of the two. The downhole intelligent imaging device has the beneficial effects that accurate identification and image reconstruction of typical targets can be realized in a complex downhole environment, and reliable basis is provided for subsequent fishing operations.

Description

Technical Field

[0001] This invention relates to the field of downhole ultrasonic imaging technology, and in particular to a downhole intelligent imaging device and an image processing method based on phased array ultrasound. Background Technology

[0002] With the continuous development of oil and gas resources, wellbore accidents such as "fish getting stuck" occur frequently during well workover operations. This not only affects the efficiency of oil and gas production but also poses significant safety hazards. Especially in complex downhole operating environments, due to turbid well fluids and limited visibility, traditional downhole detection technologies struggle to directly identify the shape and location of the stuck fish. Existing detection methods, such as mechanical probes and downhole cameras, often suffer from poor penetration and low resolution, leading to many operations relying on manual experience and failing to effectively avoid "blind retrieval," potentially even causing secondary damage to downhole equipment.

[0003] To improve the efficiency and safety of downhole fishing, ultrasonic phased array imaging technology has emerged. This technology, by controlling the excitation timing of the transducer array and utilizing electronic focusing and scanning functions, can achieve high-resolution target detection in complex downhole environments. Ultrasonic detection can penetrate high-concentration well fluid media, forming clear images of the wellbore and target objects, enabling effective identification of abnormal objects even in the absence of light. However, ultrasonic phased array technology still faces some technical challenges in downhole applications, especially in complex environments where image noise is significant, and conventional denoising algorithms are insufficient.

[0004] High temperature, high pressure, mud scattering noise, mechanical vibration, and electromagnetic interference in the downhole environment can significantly affect ultrasonic images, resulting in Gaussian noise, speckle noise, and periodic artifacts. Traditional image denoising methods, such as median filtering and wavelet denoising, are ineffective at handling these complex noise types, failing to effectively preserve the edge and structural details of the target while suppressing noise. Therefore, developing an efficient image enhancement algorithm for downhole ultrasonic images has significant application value.

[0005] Currently, although ultrasonic phased array technology has made significant progress in some fields, there are still significant technological gaps in downhole workover operations, especially in the visualization and detection of "fallen fish" (lost fish) retrieval. Therefore, this invention proposes a downhole retrieval imaging solution based on ultrasonic phased arrays. Combining an improved two-dimensional variational mode decomposition denoising algorithm and deep learning enhancement technology, it can achieve high-quality target image reconstruction in complex downhole environments, providing strong technical support for intelligent retrieval decision-making. Summary of the Invention

[0006] The purpose of this invention is to provide a downhole intelligent imaging device that can accurately identify and image-reconstruct typical targets in complex downhole environments, providing a reliable basis for subsequent salvage operations.

[0007] Another objective of this invention is to provide an image processing method based on phased array ultrasound, which can accurately identify and image-reconstruct typical targets in complex downhole environments, providing a reliable basis for subsequent salvage operations.

[0008] To achieve the above objectives, the present invention adopts the following technical solution, including: Downhole ultrasonic imaging tools are used to penetrate deep into the downhole environment, emit ultrasonic beams, and acquire target echo signals. A ground receiving device, used to receive and display uploaded data on the ground; A multi-core composite cable for signal transmission is used to connect the downhole ultrasonic imaging tool and the ground receiving device to realize carrier communication control and data transmission between the two. The downhole ultrasonic imaging tool includes: A transducer array, consisting of multiple ultrasonic transducers arranged linearly or in a ring, is used to transmit forward-looking scanning sound beams and receive target echo signals. The control module, which is electrically connected to the transducer array, is used to control the array elements in the transducer array to be excited sequentially according to a set timing sequence, thereby realizing the direction control of the sound beam. The data acquisition module, which is electrically connected to the transducer array and the control module, is used to synchronously acquire and A / D convert the echo signals received by each array element to form raw echo data. The imaging and algorithm processing module is electrically connected to the data acquisition module and the control module, respectively. The imaging and algorithm processing module includes an adaptive beamforming module, a two-dimensional variational mode decomposition denoising module, an image enhancement module based on coherent superposition and multimodal information fusion, and a downhole target reconstruction module based on deep learning. The imaging and algorithm processing module is used to perform profile imaging on the raw echo data and to perform denoising and image enhancement operations.

[0009] Preferably, it also includes: A mechanical rotating mechanism, which is connected to the transducer array and the signal transmission multi-core composite cable respectively, is used to drive the transducer array to rotate so as to acquire cross-sectional images at different angles. The mechanical rotating mechanism includes: A stepper motor, the upper end of which is connected to the lower end of the signal transmission multi-core composite cable; the stepper motor is electrically connected to the control module; A drive shaft, the upper end of which is connected to the output end of the stepper motor; and the lower end of which is fixedly connected to the transducer array. A bearing assembly is disposed on the outer periphery of the drive shaft for supporting the drive shaft.

[0010] Preferably, the ground receiving device includes a display device for displaying images processed by the imaging and algorithm processing module.

[0011] An image processing method based on phased array ultrasound, using any of the above-described downhole intelligent imaging devices, is characterized by comprising the following steps: S1. The downhole ultrasonic imaging tool is lowered into the downhole observation position. The control module and data acquisition module control the transducer array to emit ultrasonic waves and acquire raw downhole echo data. The adaptive beamforming module preprocesses the raw echo data. The preprocessing is phase correction and beamforming. S2, the two-dimensional variational mode decomposition denoising module performs two-dimensional variational mode decomposition on the preprocessed raw echo data to separate the signal mode and noise mode; S3. The image enhancement module based on coherent superposition and multimodal information fusion performs three-dimensional block matching collaborative denoising on the selected signal modes to remove Gaussian noise and scattering interference. S4. The deep learning-based downhole target reconstruction module performs feature enhancement on the denoised data; S5. The image enhancement module based on coherent superposition and multimodal information fusion performs multimodal information fusion, and combines ultrasonic amplitude information and phase information to generate the final downhole image data; S6. The final downhole image data is uploaded to the ground receiving device for display via a multi-core composite cable for signal transmission.

[0012] Preferably, in step S2, the two-dimensional variational mode decomposition includes the following steps: S201, Initialization Settings Set the modal decomposition layer number K and the penalty factor α; for the spectral characteristics of downhole ultrasonic echo signals, the decomposition layer number K is 3 to 6, and the penalty factor α is 1500 to 2500. The original echo data matrix is ​​defined as the input signal f(x,y). S202, Solving variational constraints Construct a two-dimensional variational constraint model to find K eigenmode functions. This minimizes the sum of the estimated bandwidths of each mode, and the sum of the bandwidths of each mode equals the input signal. The mathematical expression of the variational constraint model is as follows: (1) (2) In equations (1)-(2), For the k-th eigenmode function obtained from the decomposition, This is the center frequency vector corresponding to the k-th mode. For gradient operators, For coordinate vectors, express The analytical signal, This represents the L2 norm, where j is the imaginary unit; S203, Alternating Updates The alternating direction multiplier method is used to alternately update each eigenmode function and its corresponding center frequency vector in the frequency domain. Among them, the center frequency vector The update is based on the centroid calculation of the current modal power spectrum, and the specific calculation formula is as follows: (3) The convergence criterion is set as follows: calculate the sum of squared relative errors of all modes in two adjacent iterations; if this sum of errors is less than a preset convergence threshold, the convergence criterion is satisfied. The iteration stops when the time is right, and the specific discriminant is: (4) In equations (3)-(4), Representing modes The Fourier transform of n, where n is the number of iterations. For frequency domain coordinate vectors, a preset threshold is typically used. The value is 10 -6 Up to 10 -7 between; S204, Modal Screening The K eigenmode functions obtained by calculation decomposition Pearson correlation coefficient with the original input signal f The calculation formula is as follows: (5) In equation (5), (x,y) are pixel coordinates, and M and N are image sizes. and These are the mean values ​​of the modal components and the original signal, respectively. A filtering threshold δ is set between 0.1 and 0.2 in order to retain weak target signals while removing noise. Eliminate <δ low correlation mode, the low correlation mode corresponds to downhole high frequency mud scattering noise; reserve Highly correlated modes ≥δ, wherein the highly correlated modes include structural features of fish falling, casing damage, wellbore structure and perforation orifice; All the retained highly correlated modes are superimposed to reconstruct a pure signal component.

[0013] Preferably, in step S3, the three-dimensional block matching collaborative denoising includes the following steps: S301, Block Matching Grouping A reference block is defined in the image, and similar blocks that are similar to the reference block are found. These similar blocks are then stacked into a three-dimensional array. S302, Collaborative Filtering A three-dimensional linear transformation is performed on the three-dimensional array, and the noise coefficient is set to zero in the transform domain by hard thresholding, thereby achieving efficient filtering of the data set. S303, Inverse Transformation and Aggregation Perform a three-dimensional inverse transform on the filtered three-dimensional array to restore the data back to the positions of similar blocks in the two-dimensional image; S304, Weighted Reconstruction Since some pixels may be covered multiple times, a weighted average method is used to fuse the pixel values ​​at overlapping locations, and the final output is a denoised image that retains edge sharpness.

[0014] Preferably, in step S4, the feature enhancement includes the following steps: S401, Dataset Construction and Pre-training Construct a training dataset containing features of typical downhole targets; the typical downhole targets include fish that have fallen into the well, casing damage, wellbore structures, and perforation holes; The training dataset consists of pairs of clear images with high signal-to-noise ratio as labels and noisy images with low signal-to-noise ratio as input; The initial convolutional neural network model is trained under supervision using the training dataset. The loss function between the network output image and the clear image with high signal-to-noise ratio is minimized through the backpropagation algorithm until the model converges, resulting in a denoising enhancement model with texture restoration capabilities. The loss function is preferably the mean squared error loss, and its expression is as follows: (6) in For the label image, Predict images for the network; S402, Network Forward Propagation The denoised image is then input into the trained convolutional neural network model. S403, Feature Extraction and Nonlinear Mapping The image passes through multiple convolutional layers, batch normalization layers, and modified linear unit activation layers, and the network automatically identifies and enhances potential target contours; S404, Residual Learning Output The network output layer employs a residual learning strategy to output a clean predicted image, significantly improving image contrast and target saliency.

[0015] Preferably, in step S5, the multimodal information fusion includes the following steps: S501, Extraction of Multiple Physical Quantities Perform Hilbert transform on the denoised image data to construct the analytic signal Z(x,y); The instantaneous amplitude information A(x,y) and instantaneous phase information P(x,y) are calculated using analytic signals. The calculation formula is as follows: (7) (8) In the formula, f(x,y) is the input image data, and H[·] represents the Hilbert transform operator; S502, Image Registration and Normalization The magnitude map and phase map are registered at the pixel level, and their gray values ​​are normalized to the [0,1] range to eliminate the difference in dimensions. S503, Weighted Fusion Strategy The Sobel operator is used to perform convolution operations on the magnitude map A and the phase map P respectively, and the local gradient value at pixel (x,y) is calculated. and This is used as a measure of pixel saliency; Construct a weighted matrix based on local gradient values. and ,in: (9) (10) S504, fused image output The final fused pixel value is calculated based on the weights: (11) In the formula, The final image is fused at coordinates (x, y) to generate downhole image data that contains both clear outlines and rich texture details.

[0016] The beneficial effects of this invention are: it enables accurate identification and image reconstruction of typical targets in complex downhole environments, providing a reliable basis for subsequent salvage operations. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a downhole intelligent imaging device according to the present invention.

[0018] Figure 2 This is a schematic diagram of a downhole intelligent imaging device according to the present invention. Detailed Implementation

[0019] The invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0020] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0021] like Figure 1-2 As shown, a downhole intelligent imaging device of the present invention includes: Downhole ultrasonic imaging tool 1, which is used to penetrate deep into the downhole environment, emit ultrasonic beams and collect target echo signals; Ground receiving device 3 is used to receive and display data uploaded on the ground; preferably, the ground receiving device is provided with a display device for displaying the image processed by the imaging and algorithm processing module. A multi-core composite cable 2 for signal transmission is used to connect the downhole ultrasonic imaging tool and the ground receiving device to realize carrier communication control and data transmission between the two. Preferably, it also includes a mechanical rotation mechanism 4, which is connected to both the transducer array and the multi-core composite cable for signal transmission, and is used to drive the transducer array to rotate, thereby acquiring profile images at different angles. The mechanical rotation mechanism 4 includes: a stepper motor, the upper end of which is connected to the lower end of the multi-core composite cable for signal transmission; the stepper motor is electrically connected to the control module; a drive shaft, the upper end of which is connected to the output end of the stepper motor; and the lower end of which is fixedly connected to the transducer array; and a bearing assembly disposed on the outer periphery of the drive shaft for supporting the drive shaft.

[0022] The downhole ultrasonic imaging tool 1 includes: Transducer array 110, which consists of multiple ultrasonic transducers arranged in a linear or ring configuration, is used to transmit forward scanning acoustic beams and receive target echo signals. The control module 120 is electrically connected to the transducer array and is used to control the array elements in the transducer array to be excited sequentially according to a set timing sequence to achieve sound beam direction control. The data acquisition module 130 is electrically connected to the transducer array and the control module, and is used to synchronously acquire and A / D convert the echo signals received by each array element to form raw echo data. The imaging and algorithm processing module 140 is electrically connected to the data acquisition module and the control module, respectively. The imaging and algorithm processing module includes an adaptive beamforming module 141, a two-dimensional variational mode decomposition denoising module 142, an image enhancement module 143 based on coherent superposition and multimodal information fusion, and a downhole target reconstruction module 144 based on deep learning. The imaging and algorithm processing module 140 is used to perform profile imaging on the raw echo data and to perform denoising and image enhancement operations.

[0023] An image processing method based on phased array ultrasound, using the aforementioned downhole intelligent imaging device, includes the following steps: S1. The downhole ultrasonic imaging tool 1 is lowered into the downhole observation position. The control module 120 and the data acquisition module 130 control the transducer array to emit ultrasonic waves and acquire the raw echo data from the downhole. The adaptive beamforming module 141 preprocesses the raw echo data. The preprocessing is phase correction and beamforming. S2, the two-dimensional variational mode decomposition denoising module 142 performs two-dimensional variational mode decomposition on the preprocessed raw echo data to separate the signal mode and the noise mode; The two-dimensional variational mode decomposition includes the following steps: S201, Initialization Settings Set the modal decomposition layer number K and the penalty factor α; for the spectral characteristics of downhole ultrasonic echo signals, the decomposition layer number K is 3 to 6, and the penalty factor α is 1500 to 2500. The original echo data matrix is ​​defined as the input signal f(x,y). S202, Solving variational constraints Construct a two-dimensional variational constraint model to find K eigenmode functions. This minimizes the sum of the estimated bandwidths of each mode, and the sum of the bandwidths of each mode equals the input signal. The mathematical expression of the variational constraint model is as follows: (1) (2) In equations (1)-(2), For the k-th eigenmode function obtained from the decomposition, This is the center frequency vector corresponding to the k-th mode. For gradient operators, For coordinate vectors, express The analytical signal, This represents the L2 norm, where j is the imaginary unit; S203, Alternating Updates The alternating direction multiplier method is used to alternately update each eigenmode function and its corresponding center frequency vector in the frequency domain. Among them, the center frequency vector The update is based on the centroid calculation of the current modal power spectrum, and the specific calculation formula is as follows: (3) The convergence criterion is set as follows: calculate the sum of squared relative errors of all modes in two adjacent iterations; if this sum of errors is less than a preset convergence threshold, the convergence criterion is satisfied. The iteration stops when the time is right, and the specific discriminant is: (4) In equations (3)-(4), Representing modes The Fourier transform of n, where n is the number of iterations. For frequency domain coordinate vectors, a preset threshold is typically used. The value is 10 -6 Up to 10 -7 between; S204, Modal Screening The K eigenmode functions obtained by calculation decomposition Pearson correlation coefficient with the original input signal f The calculation formula is as follows: (5) In equation (5), (x,y) are pixel coordinates, and M and N are image sizes. and These are the mean values ​​of the modal components and the original signal, respectively. A filtering threshold δ is set between 0.1 and 0.2 in order to retain weak target signals while removing noise. Eliminate <δ low correlation mode, the low correlation mode corresponds to downhole high frequency mud scattering noise; reserve Highly correlated modes ≥δ, wherein the highly correlated modes include structural features of fish falling, casing damage, wellbore structure and perforation orifice; All the retained highly correlated modes are superimposed to reconstruct a pure signal component.

[0024] S3. The image enhancement module 143 based on coherent superposition and multimodal information fusion performs three-dimensional block matching collaborative denoising on the selected signal modes to remove Gaussian noise and scattering interference. The 3D block matching collaborative denoising includes the following steps: S301, Block Matching Grouping A reference block is defined in the image, and similar blocks that are similar to the reference block are found. These similar blocks are then stacked into a three-dimensional array. S302, Collaborative Filtering A three-dimensional linear transformation is performed on the three-dimensional array, and the noise coefficient is set to zero in the transform domain by hard thresholding, thereby achieving efficient filtering of the data set. S303, Inverse Transformation and Aggregation Perform a three-dimensional inverse transform on the filtered three-dimensional array to restore the data back to the positions of similar blocks in the two-dimensional image; S304, Weighted Reconstruction Since some pixels may be covered multiple times, a weighted average method is used to fuse the pixel values ​​at overlapping locations, and the final output is a denoised image that retains edge sharpness.

[0025] S4. The deep learning-based downhole target reconstruction module 144 performs feature enhancement on the denoised data; The feature enhancement includes the following steps: S401, Dataset Construction and Pre-training Construct a training dataset containing features of typical downhole targets; the typical downhole targets include fish that have fallen into the well, casing damage, wellbore structures, and perforation holes; The training dataset consists of pairs of clear images with high signal-to-noise ratio as labels and noisy images with low signal-to-noise ratio as input; The initial convolutional neural network model is trained under supervision using the training dataset. The loss function between the network output image and the clear image with high signal-to-noise ratio is minimized through the backpropagation algorithm until the model converges, resulting in a denoising enhancement model with texture restoration capabilities. The loss function is preferably the mean squared error loss, and its expression is as follows: (6) in For the label image, Predict images for the network; S402, Network Forward Propagation The denoised image is then input into the trained convolutional neural network model. S403, Feature Extraction and Nonlinear Mapping The image passes through multiple convolutional layers, batch normalization layers, and modified linear unit activation layers, and the network automatically identifies and enhances potential target contours; S404, Residual Learning Output The network output layer employs a residual learning strategy to output a clean predicted image, significantly improving image contrast and target saliency.

[0026] S5. The image enhancement module 143 based on coherent superposition and multimodal information fusion performs multimodal information fusion, and combines ultrasonic amplitude information and phase information to generate the final downhole image data. The multimodal information fusion includes the following steps: S501, Extraction of Multiple Physical Quantities Perform Hilbert transform on the denoised image data to construct the analytic signal Z(x,y); The instantaneous amplitude information A(x,y) and instantaneous phase information P(x,y) are calculated using analytic signals. The calculation formula is as follows: (7) (8) In the formula, f(x,y) is the input image data, and H[·] represents the Hilbert transform operator; S502, Image Registration and Normalization The magnitude map and phase map are registered at the pixel level, and their gray values ​​are normalized to the [0,1] range to eliminate the difference in dimensions. S503, Weighted Fusion Strategy The Sobel operator is used to perform convolution operations on the magnitude map A and the phase map P respectively, and the local gradient value at pixel (x,y) is calculated. and This is used as a measure of pixel saliency; Construct a weighted matrix based on local gradient values. and ,in: (9) (10) S504, fused image output The final fused pixel value is calculated based on the weights: (11) In the formula, The final image is fused at coordinates (x, y) to generate downhole image data that contains both clear outlines and rich texture details.

[0027] S6. The final downhole image data is uploaded to the ground receiving device for display via a multi-core composite cable for signal transmission.

[0028] In summary, the present invention provides an intelligent downhole imaging device that can accurately identify and image-reconstruct typical targets in complex downhole environments, providing a reliable basis for subsequent salvage operations.

[0029] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A test device for a downhole intelligent imaging system, characterized in that, include: Downhole ultrasonic imaging tools are used to penetrate deep into the downhole environment, emit ultrasonic beams, and acquire target echo signals. A ground receiving device, used to receive and display uploaded data on the ground; A multi-core composite cable for signal transmission is used to connect the downhole ultrasonic imaging tool and the ground receiving device to realize carrier communication control and data transmission between the two. The downhole ultrasonic imaging tool includes: A transducer array, consisting of multiple ultrasonic transducers arranged linearly or in a ring, is used to transmit forward-looking scanning sound beams and receive target echo signals. The control module, which is electrically connected to the transducer array, is used to control the array elements in the transducer array to be excited sequentially according to a set timing sequence, thereby realizing the direction control of the sound beam. The data acquisition module, which is electrically connected to the transducer array and the control module, is used to synchronously acquire and A / D convert the echo signals received by each array element to form raw echo data. The imaging and algorithm processing module is electrically connected to the data acquisition module and the control module, respectively. The imaging and algorithm processing module includes an adaptive beamforming module, a two-dimensional variational mode decomposition denoising module, an image enhancement module based on coherent superposition and multimodal information fusion, and a downhole target reconstruction module based on deep learning. The imaging and algorithm processing module is used to perform profile imaging on the raw echo data and to perform denoising and image enhancement operations.

2. The downhole intelligent imaging device according to claim 1, characterized in that, Also includes: A mechanical rotating mechanism, which is connected to the transducer array and the signal transmission multi-core composite cable respectively, is used to drive the transducer array to rotate so as to acquire cross-sectional images at different angles. The mechanical rotating mechanism includes: A stepper motor, the upper end of which is connected to the lower end of the signal transmission multi-core composite cable; the stepper motor is electrically connected to the control module; A drive shaft, the upper end of which is connected to the output end of the stepper motor; and the lower end of which is fixedly connected to the transducer array. A bearing assembly is disposed on the outer periphery of the drive shaft for supporting the drive shaft.

3. The downhole intelligent imaging device according to claim 1, characterized in that: The ground receiving device is equipped with a display device for displaying images processed by the imaging and algorithm processing module.

4. An image processing method based on phased array ultrasound, using the downhole intelligent imaging device described in claims 1-3, characterized in that... Includes the following steps: S1. The downhole ultrasonic imaging tool is lowered into the downhole observation position. The control module and data acquisition module control the transducer array to emit ultrasonic waves and acquire raw downhole echo data. The adaptive beamforming module preprocesses the raw echo data. The preprocessing is phase correction and beamforming. S2, the two-dimensional variational mode decomposition denoising module performs two-dimensional variational mode decomposition on the preprocessed raw echo data to separate the signal mode and noise mode; S3. The image enhancement module based on coherent superposition and multimodal information fusion performs three-dimensional block matching collaborative denoising on the selected signal modes to remove Gaussian noise and scattering interference. S4. The deep learning-based downhole target reconstruction module performs feature enhancement on the denoised data; S5. The image enhancement module based on coherent superposition and multimodal information fusion performs multimodal information fusion, and combines ultrasonic amplitude information and phase information to generate the final downhole image data; S6. The final downhole image data is uploaded to the ground receiving device for display via a multi-core composite cable for signal transmission.

5. The image processing method based on phased array ultrasound according to claim 4, characterized in that, In step S2, the two-dimensional variational mode decomposition includes the following steps: S201, Initialization Settings Set the modal decomposition layer number K and the penalty factor α; for the spectral characteristics of downhole ultrasonic echo signals, the decomposition layer number K is 3 to 6, and the penalty factor α is 1500 to 2500. The original echo data matrix is ​​defined as the input signal f(x,y). S202, Solving variational constraints Construct a two-dimensional variational constraint model to find K eigenmode functions. This minimizes the sum of the estimated bandwidths of each mode, and the sum of the bandwidths of each mode equals the input signal. The mathematical expression of the variational constraint model is as follows: （1） （2） In equations (1)-(2), For the k-th eigenmode function obtained from the decomposition, This is the center frequency vector corresponding to the k-th mode. For gradient operators, For coordinate vectors, express The analytical signal, This represents the L2 norm, where j is the imaginary unit; S203, Alternating Updates The alternating direction multiplier method is used to alternately update each eigenmode function and its corresponding center frequency vector in the frequency domain. Among them, the center frequency vector The update is based on the centroid calculation of the current modal power spectrum, and the specific calculation formula is as follows: （3） The convergence criterion is set as follows: calculate the sum of squared relative errors of all modes in two adjacent iterations; if this sum of errors is less than a preset convergence threshold, the convergence criterion is satisfied. The iteration stops when the time is right, and the specific discriminant is: （4） In equations (3)-(4), Representing modes The Fourier transform of n, where n is the number of iterations. For frequency domain coordinate vectors, a preset threshold is typically used. The value is 10 -6 Up to 10 -7 between; S204, Modal Screening The K eigenmode functions obtained by calculation decomposition Pearson correlation coefficient with the original input signal f The calculation formula is as follows: （5） In equation (5), (x,y) are pixel coordinates, and M and N are image sizes. and These are the mean values ​​of the modal components and the original signal, respectively. A filtering threshold δ is set between 0.1 and 0.2 in order to retain weak target signals while removing noise; Eliminate <δ low correlation mode, the low correlation mode corresponds to downhole high frequency mud scattering noise; reserve Highly correlated modes ≥δ, wherein the highly correlated modes include structural features of fish falling, casing damage, wellbore structure and perforation orifice; All the retained highly correlated modes are superimposed to reconstruct a pure signal component.

6. The image processing method based on phased array ultrasound according to claim 5, characterized in that, In step S3, the three-dimensional block matching collaborative denoising includes the following steps: S301, Block Matching Grouping A reference block is defined in the image, and similar blocks that are similar to the reference block are found. These similar blocks are then stacked into a three-dimensional array. S302, Collaborative Filtering A three-dimensional linear transformation is performed on the three-dimensional array, and the noise coefficient is set to zero in the transform domain by hard thresholding, thereby achieving efficient filtering of the data set. S303, Inverse Transformation and Aggregation Perform a three-dimensional inverse transform on the filtered three-dimensional array to restore the data back to the positions of similar blocks in the two-dimensional image; S304, Weighted Reconstruction Since some pixels may be covered multiple times, a weighted average method is used to fuse the pixel values ​​at overlapping locations, and the final output is a denoised image that retains edge sharpness.

7. The image processing method based on phased array ultrasound according to claim 6, characterized in that, In step S4, the feature enhancement includes the following steps: S401, Dataset Construction and Pre-training Construct a training dataset containing features of typical downhole targets; the typical downhole targets include fish that have fallen into the well, casing damage, wellbore structures, and perforation holes; The training dataset consists of pairs of clear images with high signal-to-noise ratio as labels and noisy images with low signal-to-noise ratio as input; The initial convolutional neural network model is trained under supervision using the training dataset. The loss function between the network output image and the clear image with high signal-to-noise ratio is minimized through the backpropagation algorithm until the model converges, resulting in a denoising enhancement model with texture restoration capabilities. The loss function is preferably the mean squared error loss, and its expression is as follows: （6） in For the label image, Predict images for the network; S402, Network Forward Propagation The denoised image is then input into the trained convolutional neural network model. S403, Feature Extraction and Nonlinear Mapping The image passes through multiple convolutional layers, batch normalization layers, and modified linear unit activation layers, and the network automatically identifies and enhances potential target contours; S404, Residual Learning Output The network output layer employs a residual learning strategy to output a clean predicted image, significantly improving image contrast and target saliency.

8. The image processing method based on phased array ultrasound according to claim 7, characterized in that, In step S5, the multimodal information fusion includes the following steps: S501, Extraction of Multiple Physical Quantities Perform Hilbert transform on the denoised image data to construct the analytic signal Z(x,y); The instantaneous amplitude information A(x,y) and instantaneous phase information P(x,y) are calculated using analytic signals. The calculation formula is as follows: （7） （8） In the formula, f(x,y) is the input image data, and H[·] represents the Hilbert transform operator; S502, Image Registration and Normalization The magnitude map and phase map are registered at the pixel level, and their gray values ​​are normalized to the [0,1] range to eliminate the difference in dimensions. S503, Weighted Fusion Strategy The Sobel operator is used to perform convolution operations on the magnitude map A and the phase map P respectively, and the local gradient value at pixel (x,y) is calculated. and This is used as a measure of pixel saliency; Construct a weighted matrix based on local gradient values. and ,in: （9） （10） S504, fused image output The final fused pixel value is calculated based on the weights: （11） In the formula, The final image is fused at coordinates (x, y) to generate downhole image data that contains both clear outlines and rich texture details.

Images