Underwater oil pipeline leakage detection method based on acoustic imaging and image recognition

Abstract

The application relates to the technical field of underwater detection, and discloses an underwater oil pipeline leakage detection method based on acoustic imaging and image recognition, which comprises the following steps: firstly, a wide-beam acoustic imaging is adopted to identify a candidate region with plume morphology; then, a high-coherence narrow-beam mode is switched to, the phase of a reference point echo after penetrating through the candidate region is detected, and a phase jitter quantization index is calculated; then, a collaborative judgment is carried out on the morphology feature and the phase jitter index, and a leakage event is confirmed; finally, three-dimensional scanning is carried out on the confirmed leakage region, a phase disturbance intensity spatial distribution map is constructed, and the peak point of the map is used to determine the accurate three-dimensional coordinates of a leakage source. Through the bimodal combination of the morphology feature of the target and the physical characteristic of the acoustic phase jitter caused by the leakage turbulent flow, the application can effectively distinguish the real leakage from acoustic false targets such as fish schools and suspended silt, significantly reduce the detection false alarm rate, and realize high-precision three-dimensional positioning of the leakage source.

Description

Technical Field

[0001] This invention relates to the field of underwater detection technology, specifically to a method for detecting leaks in underwater oil pipelines based on acoustic imaging and image recognition. Background Technology

[0002] Subsea pipelines are critical infrastructure for the large-scale, long-distance transportation of oil and gas resources. These pipelines operate for extended periods in the complex marine environment, facing multiple risks including corrosion, geological activity, and third-party sabotage, and are susceptible to leaks. Oil and gas leaks not only cause enormous economic losses and energy waste, but their contaminated materials also inflict long-term and severe damage on marine ecosystems. Therefore, developing a technology capable of rapidly and accurately detecting and locating underwater pipeline leaks is of paramount importance for ensuring energy security and protecting the marine environment.

[0003] Various methods for detecting underwater pipeline leaks have been developed in the existing technology. For example, visual detection methods based on optical cameras use underwater vehicles equipped with high-definition cameras to inspect the pipeline. However, the effectiveness of such methods is heavily dependent on the cleanliness of the water. In marine environments with high turbidity, their detection range and identification capability decrease sharply. In addition, the plumes formed by some oil and gas leaks (such as natural gas) in the water are transparent or translucent, which also poses a challenge to optical identification.

[0004] Another approach is acoustic detection. Passive acoustic methods involve deploying hydrophones to listen for the characteristic acoustic signals generated by the high-pressure fluid ejection during a pipe leak. The limitation of this method is that interference from background noise in the marine environment (such as ship navigation, marine life activity, and drilling platform operations) is severe, easily drowning out weak leak signals. Furthermore, accurately locating the leak point solely through passive listening is quite challenging.

[0005] To overcome some of the aforementioned shortcomings, active acoustic methods, particularly forward-looking sonar-based imaging techniques, have been applied to leak detection. This technique generates underwater acoustic images by actively emitting sound waves and receiving the echo signals from the target. When a pipeline leaks, the ejected oil and gas mixture forms a plume in the water composed of numerous tiny bubbles or oil droplets. This plume exhibits a significant difference in acoustic impedance compared to the surrounding water, thus appearing as an anomalous target with a specific shape on the acoustic image.

[0006] Although active sonar-based imaging methods outperform optical methods in terms of detection range and resistance to turbidity, they still face a core technical challenge in practical applications: a high false alarm rate. The fundamental reason is that these methods rely heavily on the morphological characteristics of the target for identification. In complex marine environments, various acoustic pseudo-targets exist that resemble leak plumes in acoustic images, such as dense schools of fish, suspended sediment clouds formed by water currents, and hydrothermal vents. Traditional acoustic imaging systems cannot effectively distinguish these morphologically similar targets based on their physical origins, leading to frequent misidentification of false targets as real leaks. This triggers unnecessary emergency responses and reduces the reliability and operational efficiency of the detection system.

[0007] Furthermore, even if traditional active acoustic methods correctly identify the actual leak plume, accurately locating the leak source remains another technical challenge. Leak plumes drift and diffuse under the influence of environmental factors such as ocean currents, appearing as a large and irregularly shaped area on acoustic images. The areas with the strongest acoustic reflections on the image often do not precisely correspond to the actual leak point on the pipeline, which complicates subsequent maintenance work. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a method for detecting leaks in underwater oil pipelines based on acoustic imaging and image recognition. This method solves the problems of existing active acoustic detection methods that rely solely on the morphological characteristics of the leak plume for judgment, thus failing to effectively distinguish acoustic pseudo-targets (such as schools of fish or suspended sediment clouds) with different physical causes, resulting in a high false alarm rate, and the difficulty in accurately locating the leak source in three dimensions directly through acoustic images due to plume drift and diffusion.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an underwater oil pipeline leak detection method based on acoustic imaging and image recognition, the method specifically comprising the following steps: S1. Acoustic images around the underwater pipeline are acquired using a wide-beam imaging mode. The acoustic images are then processed using an image recognition algorithm to output candidate regions that are suspected of leaking and their morphological features.

[0010] In one embodiment, the detection system is mounted on an underwater vehicle and cruises along a pipeline. The system operates in wide-beam imaging mode, emitting wide-angle acoustic pulses to acquire acoustic images of the surrounding environment over a large area. An image recognition algorithm analyzes the real-time acquired acoustic images to identify targets with plume or cloud-like morphological features. These morphological features can be morphological confidence scores. When the morphological confidence score of a target is higher than a preset first threshold, the target is identified as a candidate region, and its bounding box information is recorded.

[0011] In an optional embodiment, before using an image recognition algorithm for identification, the acoustic image is further preprocessed with a fuzzy weighted average filter. This preprocessing step specifically involves: traversing all pixels in the acoustic image through a sliding window and performing fuzzy modeling on each pixel within the window; determining the membership degree of each pixel within the window through iterative optimization calculations; calculating the weight of each pixel within the window based on the final membership degree; and finally, performing a weighted average of the grayscale values ​​of all pixels within the window based on the weights to obtain the filtered center pixel value, thereby suppressing random noise in the acoustic image.

[0012] S2. After identifying the candidate region, the detection system is switched to a high coherence narrow beam detection mode, a detection beam is emitted toward the candidate region, and an echo signal from a preset reference point is received.

[0013] Once the candidate region is identified in step S1, the underwater vehicle will approach the region and switch the sonar system's operating mode from wide-beam imaging mode to high-coherence narrow-beam detection mode. In this mode, the system emits a narrow-beam detection acoustic beam with highly stable phase towards the candidate region. The detection beam passes through the candidate region and is reflected by a preset reference point with stable acoustic reflection characteristics. In one embodiment, this preset reference point is the outer wall of a pipe behind the candidate region. The system receives the echo signal returned from this reference point.

[0014] To ensure the accuracy of phase measurement, the detection beam is preferably a linear frequency modulated signal, the mathematical expression of which is: ; in, To transmit signals; The signal amplitude; This is a rectangular window function that operates within the pulse duration. Its value is 1 at the beginning and 0 at the end, which defines the transmission range of the signal. The duration of the pulse; It is a time variable; The center frequency of the signal; The modulation frequency (MCF) of linear frequency modulation is equal to the signal bandwidth and the pulse duration. The ratio is due to the existence of a quadratic phase term. Only then can the instantaneous frequency of the signal change linearly within the pulse duration; It is a cosine function.

[0015] S3. Extract the phase information of the echo signal and calculate the phase jitter quantization index of the phase information.

[0016] The physical principle behind this step is that the oil-water mixture formed by a real leak is an acoustically non-uniform, time-varying medium. When sound waves pass through this medium, their propagation path fluctuates randomly, and this fluctuation is ultimately reflected in the phase jitter of the echo signal. By quantifying this phase jitter, the physical characteristics of the medium can be determined.

[0017] The specific implementation of this step is as follows: A digital phase-locked loop is used to track the echo signal received in step S2 and extract its instantaneous phase; from this instantaneous phase, a phase jitter signal representing random fluctuations is separated; then, the variance of this phase jitter signal within a preset observation time window is calculated, and the calculated variance value is used as the phase jitter quantification index. The formula for calculating this variance is: ; in, This is a quantitative indicator of phase jitter, namely the variance of phase jitter; The length of the observation time window; This is a phase jitter signal that varies over time; In order to be in The mean value of the phase jitter signal within the time window; It represents the square of the difference between the instantaneous value of the signal and its mean at any given moment; For definite integral operations, it means that the instantaneous fluctuation power of the signal (i.e., the squared term above) is calculated over the entire observation period. Accumulate the values ​​above.

[0018] S4. The morphological features obtained in step S1 and the phase jitter quantification index obtained in step S3 are used for collaborative judgment to finally confirm the leakage event.

[0019] This step establishes a dual-modal collaborative judgment criterion. The criterion is as follows: when the morphological confidence of the candidate region obtained in step S1 is higher than a first preset threshold, and the phase jitter quantification index of the candidate region calculated in step S3 is higher than a second preset threshold, the system determines that the candidate region is a real leak event. The second preset threshold is calibrated by measuring the baseline level of phase jitter in a leak-free background environment. This collaborative judgment mechanism can effectively distinguish between real leak targets caused by oil-water mixtures and false targets caused by suspended sediment or biomass, which are only morphologically similar but lack phase perturbation characteristics.

[0020] In a further embodiment, after confirming the leakage event in step S4, the method further includes the step of locating the source of the leakage: S5. Control the high coherence narrow beam to perform gridded scanning on the candidate region that has been confirmed as a leakage event, measure the phase jitter quantization index of multiple spatial points in the scanning grid, and construct the spatial distribution of phase disturbance intensity; S6. Determine the peak point in the spatial distribution of the phase disturbance intensity as the location of the leakage source.

[0021] The principle behind this location step is that the leak source is the root cause of the most severe disturbance to the physical properties of the medium, and therefore the phase jitter quantization index of its corresponding spatial point is also the largest. This is achieved by performing a three-dimensional grid scan of the confirmed leak area and calculating the phase jitter quantization index of each grid point. The value can be used to construct a three-dimensional spatial distribution map of the phase perturbation intensity. Leakage source location. This is determined by finding the global maximum point of the distribution map: ; in, The spatial coordinates within the scanned area; This is the phase jitter quantification index corresponding to this coordinate; A quantitative indicator for phase jitter; It is a mathematical operator.

[0022] This invention provides a method for detecting leaks in underwater oil pipelines based on acoustic imaging and image recognition. It offers the following advantages: 1. This invention employs a collaborative judgment technique that integrates the morphological features of candidate regions with phase jitter quantification indicators. This technique requires the detected target to not only appear to be leaking in terms of acoustic image morphology, but also to cause significant disturbances in the acoustic wave propagation phase in terms of physical properties. This effectively distinguishes between leaks caused by real oil-water mixtures and false targets formed by suspended sediment, biological communities, etc., thereby improving the accuracy and reliability of leak event detection.

[0023] 2. This invention achieves direct tracing of the leak source by performing a gridded scan on the candidate area after confirming the leak event and determining the location of the leak source by the peak point of the spatial distribution of phase disturbance intensity. The location result directly corresponds to the physical damage point of the pipeline, avoiding the location error caused by plume drift due to ocean current influence, and obtaining high-precision location results.

[0024] 3. This invention combines a wide-beam imaging mode for large-scale initial screening with a high-coherence narrow-beam detection mode for close-range precision testing. This approach ensures the efficiency of large-scale pipeline inspection while concentrating high-precision physical verification resources on a few identified candidate areas, thus balancing detection efficiency and accuracy. This makes the overall detection method more practical and robust in real-world applications. Attached Figure Description

[0025] Figure 1 This is a block diagram of the overall architecture of the underwater pipeline leak detection system of the present invention; Figure 2This is a flowchart illustrating the overall process of the underwater pipeline leak detection method of the present invention. Figure 3 This is a schematic diagram of the physical property verification process in steps S2 and S3 of the present invention; Figure 4 This is a logic block diagram of the dual-modal collaborative judgment in step S4 of the present invention. Detailed Implementation

[0026] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] See attached document Figure 1 , Figure 1 This is a general architecture block diagram of an underwater pipeline leak detection system according to an embodiment of the present invention. The present invention provides an underwater oil pipeline leak detection system 100, which may include: an underwater vehicle 10, a dual-modal forward-looking sonar system 20, a signal processing unit 30, and a main control and decision-making unit 40.

[0028] The underwater vehicle 10 serves as the platform for the entire detection system, carrying the dual-modal forward-looking sonar system 20, the signal processing unit 30, and the main control and decision-making unit 40. The underwater vehicle 10 receives navigation commands from the main control and decision-making unit 40 and cruises along a preset pipeline path or performs actions such as approaching, hovering, and scanning within a designated area.

[0029] The dual-mode forward-looking sonar system 20, mounted at the front of the underwater vehicle 10, is used to transmit acoustic pulses and receive echo signals. The dual-mode forward-looking sonar system 20 has two operating modes that can be switched by the main control and decision unit 40: a wide-beam imaging mode and a high-coherence narrow-beam detection mode. In the wide-beam imaging mode, the system transmits wide-angle acoustic pulses to cover a large area; in the high-coherence narrow-beam detection mode, the system transmits narrow-beam acoustic pulses with highly stable phase to detect a specific area.

[0030] The signal processing unit 30, electrically connected to the dual-mode forward-looking sonar system 20, receives the raw echo signals acquired by the system and performs preliminary processing. In wide-beam imaging mode, the signal processing unit 30 constructs an acoustic image based on the time-of-flight and intensity information of the echo signals. In high-coherence narrow-beam detection mode, the signal processing unit 30 performs pulse compression processing on the echo signals and extracts their phase information. The signal processing unit 30 sends the processed acoustic image data or phase data to the main control and decision unit 40.

[0031] The main control and decision-making unit 40 establishes communication connections with the underwater vehicle 10, the dual-modal forward-looking sonar system 20, and the signal processing unit 30, respectively. The functions of the main control and decision-making unit 40 include: receiving acoustic images and phase information output by the signal processing unit 30; running an image recognition algorithm to process the acoustic images to identify candidate regions and output morphological features; controlling the dual-modal forward-looking sonar system 20 to switch between two modes according to preset conditions; running collaborative judgment logic to fuse morphological features and phase jitter quantification indicators to confirm a leakage event; after confirming the leakage event, running a precise positioning algorithm; and sending navigation commands to the underwater vehicle 10 to control its motion attitude.

[0032] In the overall workflow, system 100 first controls the underwater vehicle 10 to cruise along the pipeline, while the dual-modal forward-looking sonar system 20 operates in wide-beam imaging mode, continuously generating acoustic images. The main control and decision unit 40 analyzes the acoustic images in real time, identifying candidate regions. Subsequently, the main control and decision unit 40 controls the underwater vehicle 10 to approach the candidate region and instructs the dual-modal forward-looking sonar system 20 to switch to a high-coherence narrow-beam detection mode. The signal processing unit 30 and the main control and decision unit 40 collaboratively calculate the phase jitter quantification index and ultimately confirm the leak event. After confirming the leak, a scanning and localization procedure can be further executed to determine the precise location of the leak source.

[0033] The detection system's functions can be implemented through a series of functional modules executed in the main control and decision-making unit 40 or the signal processing unit 30. These functional modules include: an acoustic imaging module 210, an image processing module 220, a mode control module 230, an acoustic detection module 240, a signal processing module 250, a collaborative decision-making module 260, and a precise positioning module 270.

[0034] The acoustic imaging module 210 is used to control the dual-mode forward-looking sonar system 20 to emit wide-angle acoustic pulses in wide-beam imaging mode and receive raw echo data from a large area of ​​water, and transmit the raw echo data to the signal processing unit 30 to construct an acoustic image.

[0035] Image processing module 220 is used to receive and process the acoustic image constructed by signal processing unit 30. Image processing module 220 executes an image recognition algorithm to identify targets with preset shapes from the acoustic image and outputs bounding box information of one or more candidate regions, as well as morphological features corresponding to each candidate region. In one embodiment, the morphological feature is a morphological confidence score.

[0036] The mode control module 230 is used to generate and send mode switching commands. After the image processing module 220 identifies the candidate region, the mode control module 230 sends a command to the dual-modal forward-looking sonar system 20 to switch it from wide-beam imaging mode to high-coherence narrow-beam detection mode.

[0037] The acoustic detection module 240 is used to control the dual-mode forward-looking sonar system 20 to emit a detection beam toward the candidate area determined by the image processing module 220 after the system switches to the high coherence narrow beam detection mode, and to receive the echo signal reflected by the preset reference point.

[0038] Signal processing module 250 processes the echo signal received by acoustic detection module 240. Signal processing module 250 executes one or more digital signal processing algorithms to extract phase information from the echo signal and calculates a phase jitter quantization index. Signal processing module 250 outputs the calculated phase jitter quantization index to collaborative decision module 260.

[0039] The collaborative decision-making module 260 receives morphological features from the image processing module 220 and phase jitter quantization indicators from the signal processing module 250. Based on a preset collaborative judgment criterion, the collaborative decision-making module 260 fuses and judges the two input information, ultimately outputting a decision result to confirm whether a candidate region represents a real leakage event.

[0040] The precise positioning module 270 is activated after the collaborative decision-making module 260 confirms a leak event. The precise positioning module 270 controls the acoustic detection module 240 and the signal processing module 250 to perform gridded scanning and measurement on the confirmed leak area, obtain a spatial distribution of phase disturbance intensity, and calculate the three-dimensional coordinate position of the leak source based on the spatial distribution.

[0041] The detection method provided by this invention is implemented through a multi-stage, multi-modal process. This process begins with a large-scale rapid inspection and gradually focuses on a small-scale precise verification and location. The complete workflow includes a preliminary screening stage, a physical characteristic verification stage, and a leak source location stage.

[0042] In the initial screening stage, the system first executes step S1. The underwater vehicle 10, carrying the entire detection system, cruises along the pipeline. During this time, the dual-modal forward-looking sonar system 20 operates in wide-beam imaging mode. In this mode, the sonar system emits sound waves into a large area of ​​water and receives the echoes. The signal processing unit 30 continuously constructs and outputs acoustic images of the area around the underwater pipeline. The image processing module 220 in the main control and decision unit 40 processes the acoustic images in real time, runs image recognition algorithms, identifies targets with plume or cloud-like morphologies in the acoustic images, and outputs one or more candidate regions and their corresponding morphological features. The function of this stage is to complete the rapid initial screening of candidate regions that are morphologically suspected of leaks.

[0043] In the physical characteristic verification phase, after identifying the candidate region, the system proceeds to steps S2, S3, and S4. The main control and decision-making unit 40 controls the underwater vehicle 10 to approach the candidate region identified in step S1. Simultaneously, the mode control module 230 sends a command to switch the dual-modal forward-looking sonar system 20 from wide-beam imaging mode to high-coherence narrow-beam detection mode. In this mode, the acoustic detection module 240 controls the sonar to emit a detection beam towards the candidate region and receives echo signals reflected from preset reference points such as the pipe's outer wall. The signal processing module 250 extracts the phase information of the echo signal and calculates the phase jitter quantification index. Finally, the collaborative decision-making module 260 integrates the morphological features obtained in step S1 with the phase jitter quantification index obtained in step S3, and based on preset collaborative judgment criteria, performs a final confirmation of the authenticity of the candidate region, determining whether it is a genuine leak event.

[0044] In the leak source localization phase, after confirming the leak event in step S4, the system can further execute steps S5 and S6. The precise localization module 270 is activated and controls a highly coherent narrow beam to perform a three-dimensional gridded scan of the candidate region confirmed as a leak event. During the scan, the measurement process of step S3 is repeated for each grid point to obtain the phase jitter quantization index of that spatial point, thereby constructing a three-dimensional spatial distribution of phase perturbation intensity. Finally, the precise localization module 270 searches for peak points in the spatial distribution of phase perturbation intensity and outputs the three-dimensional coordinates of these peak points as the final leak source localization result.

[0045] See attached document Figure 2 , Figure 2 This is a general flowchart of the underwater pipeline leak detection method according to the present invention. In a specific embodiment, step S1 is described in detail. The main entities executing this step are the dual-modal forward-looking sonar system 20, the signal processing unit 30, and the image processing module 220 in the main control and decision unit 40.

[0046] At the start of the detection mission, the main control and decision unit 40 controls the dual-modal forward-looking sonar system 20 to operate in wide-beam imaging mode. In this mode, the transducer array of the dual-modal forward-looking sonar system 20 is activated to emit a fan-shaped sound beam with a wide opening angle in the horizontal direction and a narrow opening angle in the vertical direction. This sound beam can cover a large area in front of the underwater vehicle 10 in one go, thereby achieving rapid scanning of the pipeline and its surrounding environment.

[0047] The emitted acoustic pulses propagate in the water, interacting with underwater oil pipelines, the seabed, and suspended targets in the water, generating echo signals. The transducer array of the dual-mode forward-looking sonar system 20 receives these echo signals and converts the sound pressure signals into analog electrical signals, which are then transmitted to the signal processing unit 30.

[0048] The signal processing unit 30 amplifies, filters, and performs analog-to-digital conversion on the received multi-channel analog electrical signals to obtain digitized echo data. Next, the signal processing unit 30 performs digital beamforming processing on the digitized echo data to determine the direction of arrival of different echo signals. Simultaneously, the signal processing unit 30 records the round-trip time of each echo signal. Target distance. The calculation method is as follows: ; in, The speed of sound in water, It is a time variable.

[0049] The signal processing unit 30 calculates the target distance for each echo. , destination Together with the echo intensity I, a two-dimensional acoustic image is constructed. In this acoustic image, the position of a pixel is determined by the target distance. and direction of arrival It is confirmed that the grayscale value of a pixel is proportional to the echo intensity I.

[0050] As the underwater vehicle 10 moves continuously along the pipeline path, the aforementioned processes of sound wave emission, echo reception, and image construction are repeatedly executed at a preset frame rate, thereby generating a continuous acoustic image sequence. The signal processing unit 30 transmits this acoustic image sequence in real time to the image processing module 220 in the main control and decision-making unit 40 for subsequent recognition and processing.

[0051] In an optional embodiment, to suppress speckle noise in the acoustic image and enhance the signal-to-noise ratio of potential targets, the image processing module 220 further performs a fuzzy weighted average filtering preprocessing step before sending the acoustic image into the image recognition algorithm.

[0052] The preprocessing step iterates through every pixel in the input acoustic image using a sliding window of size M×M. For a window centered on the current pixel, all pixels within the window are treated as a local pixel set, and subsequent filtering operations are performed on this set.

[0053] Within a sliding window, blurring modeling is performed on N pixels (N=M×M) with the goal of dividing the pixel set into C categories (e.g., C=2, corresponding to the target signal class and the background noise class, respectively). This process is achieved by minimizing an objective function, the specific form of which is: ; in, This is the preset number of categories; It represents the total number of pixels within the window; It is the first one in the window The grayscale value of each pixel; It is the first Cluster centers for each category; It is the first The pixel belongs to the first The membership degree of each category satisfies ; It is a fuzzy weighting index greater than 1, used to control the degree of fuzziness; Represents pixels Feature vectors and cluster centers The square of the Euclidean distance between the eigenvectors; This means summing over all data samples (pixels); This means summing over all cluster categories.

[0054] To obtain the objective function To minimize the optimal membership degree, the image processing module 220 iteratively optimizes the membership degree of pixels within the window. In each iteration, based on the cluster centers from the previous round... Update membership Then based on the updated membership degree Calculate new cluster centers The update formulas for membership degree and cluster center are as follows: ; in, It is a pixel. With the current target cluster center The distance between them; It is the same pixel. With all cluster centers The distance; This ratio measures the distance from a pixel to the target cluster center. The distance relative to its distance to another cluster center The magnitude of the distance; It is a key index term, directly influenced by the fuzzy weighted index. Control; It is a constant greater than 1; For a fixed pixel and a target category We need to iterate through all the categories (from arrive ),calculate The value of this item, then this Add up all the calculation results.

[0055] ; in, It is a weighted sum; Clustering of all pixels The membership weights are summed, which serves as a normalization process.

[0056] The above iterative process continues until the objective function is between two iterations. The change in the number of iterations is less than a preset convergence threshold, or the preset maximum number of iterations is reached.

[0057] After the iteration is complete, each pixel within the window is obtained. For target signal category final membership degree Based on the final optimized membership degrees, the weights of each pixel within the window are calculated. In one embodiment, weight The calculation method is as follows: ; in, It is the degree of membership in a specific category; It represents the first in the image. 1 pixel; It is a specific category index; It is a fuzzy weighted index; It is a pixel The non-normalized contribution; It is the total contribution and the normalization factor.

[0058] Finally, based on the weights grayscale values ​​of all pixels within the window Perform a weighted average, and use the resulting weighted average as the filtered output value of the center pixel of the sliding window. The calculation formula is as follows: ; in, The weights are calculated using the normalized centroid. The grayscale value of a pixel; It is a weighted summation operator.

[0059] By traversing the entire acoustic image through a sliding window, a frame of acoustic image with noise suppression after fuzzy weighted average filtering preprocessing can be obtained, which can then be used for subsequent image recognition processing.

[0060] After acquiring an acoustic image or a pre-processed acoustic image, the image is transmitted to the image processing module 220 integrated in the main control and decision unit 40 for generating candidate regions.

[0061] The image processing module 220 is equipped with a pre-trained image recognition algorithm. In one embodiment, the algorithm is a target detection model based on a convolutional neural network (CNN). The model is trained on a dataset containing a large number of labeled underwater acoustic images, including samples of oil and gas plumes from real leaks as well as samples of various non-leaking targets (such as suspended sediment, biota, structures, etc.).

[0062] The image recognition algorithm receives an acoustic image frame as input and analyzes the image content to identify targets that conform to the morphological characteristics of a leaking plume.

[0063] The algorithm outputs a set of detection results, where each detection result corresponds to a recognized target.

[0064] For each identified target, the image recognition algorithm outputs its bounding box and morphological confidence score. The bounding box indicates the location and extent of the target in the acoustic image, and can be represented by a quadruple. To express.

[0065] in: These are the x and y coordinates of the top-left vertex of the bounding box in the image coordinate system, respectively. These represent the width and height of the bounding box, respectively.

[0066] Morphological confidence , is the quantitative output of the image recognition algorithm evaluating the similarity between the target's shape and the shapes of pre-trained leaked samples, with a numerical range of . A higher The value indicates that the shape of the target is highly consistent with the morphological characteristics of the leaking plume.

[0067] After obtaining all identified targets and their corresponding morphological confidence scores, the image processing module 220 determines the final candidate regions based on a preset judgment criterion. The determination criterion is: ; in, A preset first threshold is set. When the morphological confidence of a target... Higher than the first preset threshold At that time, the target is identified as a candidate region. First preset threshold. Calibration is performed based on the detection sensitivity and false alarm rate requirements of the actual application scenario.

[0068] Finally, the image processing module 220 processes the bounding box information of all candidate regions that meet the above conditions. The output guides the subsequent physical property verification phase. If no target meeting the criteria is detected in a frame of acoustic image, the system continues with wide-beam imaging scanning.

[0069] In step S1, after the image processing module 220 determines one or more candidate regions, the mode control module 230 in the main control and decision unit 40 is triggered and begins to execute step S2.

[0070] The mode control module 230 sends a mode switching command to the dual-modal forward-looking sonar system 20. Upon receiving the command, the dual-modal forward-looking sonar system 20 switches its operating mode from wide-beam imaging mode to high-coherence narrow-beam detection mode. In this mode, the sonar's transducer array is weighted by phase shifters to form a narrow beam with high directionality, concentrating acoustic energy to be emitted in a specific direction relative to the candidate region provided by the image processing module 220.

[0071] After switching to the high coherence narrow beam detection mode, the acoustic detection module 240 controls the signal generator of the dual-mode forward-looking sonar system 20 to generate and transmit a detection beam to the candidate region. In one specific embodiment, to obtain the high signal-to-noise ratio and high-precision phase information required for subsequent processing, the transmitted detection beam is a linear frequency modulated (LFM) signal.

[0072] linear frequency modulation signal The complex expression is: ; in: The amplitude of the signal; For rectangular window functions, it is defined as when The value is 1 when the time condition is met, and 0 otherwise. The duration of the signal pulse; It is a time variable; This is the starting frequency of the signal; The frequency modulation slope has a value of ,in The bandwidth of the signal; The imaginary unit; Essentially, it is a rotating unit vector; Total phase function.

[0073] The linear frequency modulated signal has a large time-bandwidth product. After pulse compression processing by the signal processing unit 30, it can obtain a narrow main lobe width and high processing gain, which provides a technical basis for the subsequent accurate extraction of phase information from the echo signal.

[0074] See attached document Figure 3 , Figure 3 This is a schematic diagram of the physical characteristic verification process in steps S2 and S3 of the present invention. After the acoustic detection module 240 controls the dual-mode forward-looking sonar system 20 to emit a detection beam, the detection beam propagates along a predetermined direction, passes through the candidate area determined in step S1, and illuminates a preset reference point located far from the candidate area.

[0075] In one specific embodiment, the preset reference point is the outer wall of the oil pipeline to be tested. The outer wall of the pipeline is selected as the reference point because it has a significant difference in acoustic impedance relative to the water body, which can form a high-strength and time-stable acoustic reflection interface.

[0076] The probe beam is reflected by a preset reference point, forming an echo signal. This echo signal passes through the candidate region again and is received by the receiving transducer array of the dual-mode forward-looking sonar system 20. The sonar's receiving channel converts the captured sound pressure signal into an analog electrical signal.

[0077] During the round trip through the candidate region, the phase of the echo signal is modulated by the random fluctuations in the acoustic refractive index of the medium (i.e., the material within the candidate region) along the propagation path. Therefore, the received echo signal... It contains information about the physical properties of the medium within the candidate region. Its signal model can be represented as: ; in: This is the total attenuation coefficient of the signal during propagation; For two-way transmission delay The transmitted signal components after that; It is a phase modulation term introduced by the medium; This refers to the random phase modulation term accumulated by the sound wave as it travels through the medium of the candidate region during its round trip. It is the sum of environmental noise and internal system noise.

[0078] The dual-modal forward-looking sonar system 20 will receive information including phase modulation. Analog echo signal The signal is transmitted to the signal processing unit 30 to execute step S3, which is used for subsequent phase information extraction and quantization index calculation.

[0079] Upon receiving the echo signal from the dual-mode forward-looking sonar system 20 Then, the signal processing unit 30 or its internally integrated signal processing module 250 begins to execute step S3, which is to extract the phase information of the echo signal and calculate the phase jitter quantization index.

[0080] First, the signal processing unit 30 performs pulse compression processing on the received digital echo signal. This processing is achieved by convolving the echo signal with a local conjugate matched filter that transmits a linear frequency modulated signal. The output of pulse compression is a signal with a narrow pulse width and a high signal-to-noise ratio, whose peak position precisely corresponds to the distance from a preset reference point (the outer wall of the pipe).

[0081] Next, the system sets a time gate at the peak of the pulse-compressed output signal and uses a digital phase-locked loop (PLL) to track the echo signal within this time gate. The PLL locks onto and tracks the phase of the echo signal reflected from a preset reference point. Within a continuous observation time window, the PLL outputs a time series representing the instantaneous phase of the echo signal. .

[0082] Instantaneous phase It contains multiple components and can be represented by the following formula: ; in: It is a fixed phase constant that is related to the initial state of the propagation path and the system parameters; It is the Doppler shift caused by the relative motion between the underwater vehicle and the pipeline; It is a linear phase drift term caused by Doppler frequency shift; It is the phase jitter signal caused by the random fluctuation of the refractive index of the medium when the sound wave passes through the candidate region (such as an oil-water mixture), and this signal is the target signal to be extracted; It is a time variable.

[0083] In order to obtain from the instantaneous phase Separate the phase jitter signal The signal processing module 250 operates within a preset observation time window. Inside, to The sequence undergoes detrending processing. In one embodiment, this processing is performed using the least squares method. Perform a first-order polynomial fitting to obtain the linear trend term. Then, subtracting this trend term from the original instantaneous phase yields the phase jitter signal: ; Obtaining phase jitter signal After the time series, the signal processing module 250 performs the same observation time window. Within this process, the variance of the sequence is calculated, and this variance is used as the final phase jitter quantization metric. For discrete sampling points, the calculation formula is: ; in: Observation time window The total number of sampling points within; This is a discrete summation operator; It is in the Phase jitter value at each sampling time; It is all within this window The arithmetic mean, after being detrended. The value is close to 0; It is the variance contribution of a single sample; : indicates the first in the time series The specific time for each sampling point.

[0084] After the calculation is completed, the signal processing module 250 quantizes the obtained phase jitter index. The output is sent to the collaborative decision-making module 260 in the main control and decision-making unit 40 for performing the collaborative judgment in step S4.

[0085] See attached document Figure 4 , Figure 4 This is a logic block diagram of the dual-modal collaborative judgment in step S4 of the present invention. The signal processing module 250 calculates and outputs the phase jitter quantization index. Then, the system begins to execute step S4, which is performed by the collaborative decision-making module 260 in the main control and decision-making unit 40.

[0086] The collaborative decision-making module 260 is used to receive two independent input data: The morphological confidence score for the current candidate region generated by the image processing module 220 in step S1 .

[0087] The phase jitter quantization index generated by the signal processing module 250 in step S3 for the same candidate region .

[0088] The collaborative decision-making module 260 is based on a preset dual-modal collaborative judgment criterion to perform fusion analysis on the two input data mentioned above, so as to output a confirmation result on whether the candidate area is a real leakage event.

[0089] In one specific embodiment, the dual-modal collaborative judgment criterion is defined as a set of logical conditions that must be satisfied simultaneously. Only when a candidate region is morphologically similar to a leaking plume and physically causes significant phase jitter in the acoustic signal is the candidate region ultimately confirmed as a real leaking event.

[0090] This judgment criterion includes the following two independent conditions: Condition 1: Morphological confidence of candidate regions It must be greater than a first preset threshold. .

[0091] ; in: It is a quantitative value output by the image recognition algorithm in step S1, representing the degree of similarity between the candidate region morphology and the leaking plume morphology. It is a pre-defined threshold used to initially filter out targets in the image that are highly suspected of being leaky in terms of morphology. Meeting this condition is a prerequisite for a target to become a candidate region.

[0092] Condition 2: Phase jitter quantification index of candidate region It must be greater than a second preset threshold. .

[0093] ; in: It is a quantized value obtained from step S3, representing the intensity of phase perturbation experienced by the sound wave as it passes through the candidate region. It is a pre-calibrated threshold whose value is determined by taking multiple measurements on a leak-free background water body to obtain the statistical distribution of background phase noise. The setting ensures that the condition is only met when the phase jitter significantly exceeds the normal background noise level.

[0094] The final decision logic of the collaborative decision-making module 260 is as follows: The collaborative decision-making module 260 outputs a "leakage event confirmed" signal only if both conditions one and two are met simultaneously. If either condition is not met, it outputs a "non-leakage event" signal and instructs the system to abandon the current candidate area and return to the large-scale cruise scan state of step S1. The confirmed leakage event information will be used to activate the precise positioning module 270 in step S5.

[0095] The application of the dual-modal collaborative judgment criterion enables the detection method of this invention to effectively distinguish between real leakage events and acoustic spurious targets. Acoustic spurious targets refer to objects that are similar in shape to a leakage plume but are not physically formed by a pipe leak, such as schools of fish or suspended sediment clouds.

[0096] For an acoustic pseudo-target (such as a suspended cloud of mud and sand), in the acoustic image generated in wide-beam imaging mode, it also presents a cloud-like or plume-like morphology without fixed boundaries. Therefore, when performing step S1, the image processing module 220 will output a high morphological confidence score from its image recognition algorithm. When this value satisfies At that time, the acoustic spurious target was incorrectly identified as a candidate region, thus triggering the subsequent physical property verification stage.

[0097] However, although the shape of this acoustic spurious target satisfies condition one, its physical composition is fundamentally different from that of a real oil and gas leak. Suspended sediment clouds are mixtures of solid particles and water. Although the acoustic refractive index of its internal medium is different from that of pure water, its internal structure is relatively stable, or its turbulence intensity is much lower than that caused by high-pressure oil and gas eruptions.

[0098] Therefore, when the system switches to the high-coherence narrow-beam detection mode and emits a detection beam towards the pseudo-target region, the phase modulation of the sound wave during its round trip through the region is weak and slow. This results in the phase jitter signal extracted by the signal processing module 250 in step S3. The amplitude is very small.

[0099] Accordingly, the calculated phase jitter quantization index The value is also low, failing to meet the preset judgment condition. That is: ; in, It is a second preset threshold used to distinguish significant phase jitter from background noise.

[0100] Ultimately, during the collaborative decision-making process in module 260, the acoustic pseudo-target fails to meet condition two. The logical AND operation of the dual-modal collaborative judgment criterion results in a false result. Based on this, the collaborative decision module 260 outputs a "non-leakage event" decision, thereby excluding the acoustic spurious target from the candidate region and preventing false alarms. This process ensures that only targets that simultaneously meet both morphological and physical characteristics (significant phase jitter) are ultimately confirmed as genuine leaks.

[0101] In step S4, after the collaborative decision-making module 260 confirms the leakage event, the precise positioning module 270 in the main control and decision-making unit 40 is activated and begins to execute step S5. The function of this step is to construct a spatial distribution of phase perturbation intensity that can characterize the location of the leakage source.

[0102] The precise positioning module 270 first defines a three-dimensional scanning space in the coordinate system of the underwater vehicle 10 based on the candidate region boundary box determined in step S1 and the leakage event information confirmed in step S4. Then, this three-dimensional scanning space is discretized into a space composed of multiple preset discrete grid points. A three-dimensional mesh is formed.

[0103] in: These are the grid points in the three-dimensional coordinate system. coordinate; These are the indices of the grid points in the three dimensions.

[0104] Next, the precise positioning module 270 begins to scan and measure the 3D mesh point by point. For each mesh point... The precise positioning module 270 performs the following operations: First, the acoustic detection module 240 is controlled to precisely guide the transmitted beam of the dual-mode forward-looking sonar system 20, ensuring that the central axis of the highly coherent narrow beam passes through the current grid point. And finally illuminate the preset reference point (i.e., the outer wall of the pipe) located behind it.

[0105] Then, the system receives and processes the echo signal reflected from the reference point. The precise positioning module 270 calls the signal processing module 250 to repeat the complete processing flow in step S3 on the echo signal, namely, pulse compression, phase extraction, detrending processing, and calculation of a phase jitter quantization index. This index corresponds one-to-one with the grid points and is denoted as... .

[0106] The precise positioning module 270 performs the above scanning and measurement process traversing all preset grid points in the three-dimensional grid. After completing the measurement of all points, a three-dimensional spatial distribution map of the phase perturbation intensity is obtained. This distribution map is a dataset containing the coordinates of each spatial grid point and the phase jitter quantization index measured at that point. Its mathematical form can be expressed as: ; in: This is a three-dimensional spatial distribution diagram of the phase perturbation intensity; For a 3D mesh point; It is a quantitative indicator of phase jitter measured when the sound beam path passes through this grid point; It is a logical symbol; It is the dimension index of the grid; This represents all valid indexes. A set of combinations.

[0107] The whole phrase means: this set It contains data pairs (coordinates, jitter intensity value) for every point within the predetermined 3D grid, with no omissions.

[0108] This three-dimensional spatial distribution map It is stored in the storage unit of the precise positioning module 270 and serves as direct input data for executing step S6, namely determining the precise coordinates of the final leak source.

[0109] In step S5, the three-dimensional spatial distribution map of the phase perturbation intensity is constructed. The system then proceeds to step S6. This step is performed by the precise positioning module 270 in the main control and decision-making unit 40, whose function is to determine the precise three-dimensional coordinates of the leak source from the three-dimensional spatial distribution map.

[0110] The physical basis for this step is that the source of the pipeline leak is the location where high-pressure oil and gas are injected into the surrounding water, thus it is the spatial point with the greatest turbulence intensity and the most drastic changes in the physical properties of the medium. This intense turbulence directly leads to the maximum random fluctuations in the acoustic refractive index along the sound wave propagation path, thereby generating the strongest phase jitter. Therefore, in the three-dimensional spatial distribution map of phase disturbance intensity... In the above, the spatial location corresponding to the peak point is the location of the leakage source.

[0111] The precise positioning module 270 receives the three-dimensional spatial distribution map generated in step S5. As input data, the precise positioning module 270 then performs a peak search operation on the dataset, which iterates through all discrete grid points in the distribution map and compares the phase jitter quantization index corresponding to each grid point. .

[0112] 3D coordinates of the leak source By finding a metric to quantify phase jitter The grid point that reaches the maximum value is used to determine this. This determination process can be expressed by the following formula: ; in: The three-dimensional spatial coordinates of the finally determined leak source; It is a function that returns the parameter (in this case, a grid point) that maximizes the subsequent expression. ); It is a discrete point in a 3D scanning mesh; It is based on grid points The quantitative index of phase jitter measured along the central path; It is generated by step S5 and contains all grid points and their corresponding values. A complete three-dimensional spatial distribution map of the values.

[0113] After the peak search operation is completed, the precise location module 270 will obtain the three-dimensional coordinates of the leak source. This coordinate information is output as the final result of the entire detection process. It can be recorded, stored, or sent to a host computer for display.

Claims

1. A method for detecting leaks in underwater oil pipelines based on acoustic imaging and image recognition, characterized in that, Includes the following steps: S1. Control the underwater vehicle to cruise along the pipeline. The dual-mode forward-looking sonar system uses a wide-beam imaging mode to acquire acoustic images around the underwater pipeline. The acoustic images are processed using an image recognition algorithm to output candidate areas that are suspected of leaking in terms of morphology and the morphological features of the candidate areas. S2. After identifying the candidate region, the dual-mode forward-looking sonar system is switched to the high coherence narrow beam detection mode, and a detection beam is emitted toward the candidate region, and the echo signal reflected from the preset reference point is received. S3. Extract the phase information of the echo signal and calculate the phase jitter quantization index of the phase information; S4. The morphological features obtained in step S1 and the phase jitter quantification index obtained in step S3 are combined for joint judgment to finally confirm the leakage event. In step S1, the image recognition algorithm outputs the bounding box and morphological confidence score of the candidate region. The condition for determining the candidate region is that the morphological confidence score is higher than a first preset threshold. The specific implementation method for extracting the phase information of the echo signal and calculating the phase jitter quantization index in step S3 is as follows: the echo signal is tracked using a phase-locked loop to obtain the instantaneous phase, and the phase jitter signal is separated from the instantaneous phase. The variance of the phase jitter signal within an observation time window is calculated, and the variance is used as the phase jitter quantification index. The specific criterion for the collaborative judgment in step S4 is: when the morphological confidence corresponding to the morphological feature is higher than the first preset threshold, and the phase jitter quantification index is higher than the second preset threshold, the candidate region is confirmed as a leakage event.

2. The underwater oil pipeline leak detection method based on acoustic imaging and image recognition according to claim 1, characterized in that, In step S2, the preset reference point is the outer wall of the pipe.

3. The underwater oil pipeline leak detection method based on acoustic imaging and image recognition according to claim 1, characterized in that, In step S2, the emitted detection beam is a linear frequency modulated signal.

4. The underwater oil pipeline leak detection method based on acoustic imaging and image recognition according to claim 1, characterized in that, After confirming the leak event in step S4, the method further includes the step of locating the source of the leak: S5. Control the high coherence narrow beam to perform gridded scanning on the candidate region that has been confirmed as a leakage event, measure the phase jitter quantization index of multiple spatial points in the scanning grid, and construct the spatial distribution of phase disturbance intensity; S6. Determine the peak point in the spatial distribution of the phase disturbance intensity as the location of the leakage source.

5. The underwater oil pipeline leakage detection method based on acoustic imaging and image recognition according to claim 1, characterized in that, The function of the wide-beam imaging mode is to perform a rapid cruise scan over a large area to complete the initial screening of the candidate areas. The function of the high-coherence narrow-beam detection mode is to perform close-range physical characteristic verification of the identified candidate areas to confirm the authenticity of the leakage event.

6. The underwater oil pipeline leak detection method based on acoustic imaging and image recognition according to claim 1, characterized in that, In step S4, the collaborative judgment can distinguish between the oil-water mixture formed by a real leak and the false target formed by suspended sediment and biomass.

7. The underwater oil pipeline leak detection method based on acoustic imaging and image recognition according to claim 1, characterized in that, In step S1, before using the image recognition algorithm for recognition, the method further includes a step of performing a fuzzy weighted average filtering preprocessing on the acoustic image. The fuzzy weighted average filtering preprocessing step specifically includes: The acoustic image is traversed through a sliding window, and the pixels are modeled with blurring within the window; The membership degree of pixels within the window is iteratively optimized and calculated. Calculate the weight of each pixel within the window based on the final optimized membership degree; The grayscale values ​​of all pixels within the window are weighted according to the weights, and the filtered pixel values ​​are output.

Images