Condensation image monitoring method and system for early identification of non-contact pipeline leaks

Abstract

The application discloses a condensation image monitoring method and system for early identification of non-contact pipeline leakage, and the method comprises the following steps: arranging a condensation observation film on a pipeline at a monitoring point, calculating a theoretical dew point temperature of a current environment; when the pipeline surface temperature is greater than the theoretical dew point temperature, if a condensation area is detected on the film surface of the condensation observation film, it is directly determined as a strong anomaly, otherwise, it is determined as a normal state; when the pipeline surface temperature is less than or equal to the theoretical dew point temperature, whether the condensation coverage area presents a grid texture or a directional flow trajectory is identified, if yes, it is determined as a shape anomaly, otherwise, it is determined as a normal state; when it is determined as the shape anomaly or the strong anomaly, a reset operation is performed on the surface of the condensation observation film, and leakage or false alarm is determined according to an anomaly score; the reset operation refers to heating and blowing operations on the surface of the condensation observation film for a preset time; the application has the advantages that the monitoring accuracy is improved, and the problems of invasiveness and cost are avoided.

Description

Technical Field

[0001] This invention relates to the field of industrial pipeline safety monitoring technology, specifically to a non-contact pipeline leakage early identification condensation image monitoring method and system. Background Technology

[0002] As a crucial infrastructure for urban water supply, gas supply, petrochemical, and industrial transportation systems, the safe operation of pipelines is directly related to public safety and economic development. With increasing pipeline service life, problems such as leakage, corrosion, and insulation damage are gradually becoming more prevalent.

[0003] Currently, pipeline monitoring technology mainly relies on pressure sensors, acoustic detection, infrared thermal imaging, and pipeline inspection robots. However, existing technologies have many limitations:

[0004] 1. Intrusiveness and cost issues: Pressure sensors need to be installed inside the pipeline, which is complex and invasive; Acoustic detection is easily interfered with by industrial background noise. 2. High environmental dependence: Infrared thermal imaging depends on significant temperature differences and is easily affected by ambient light and wind speed. For example, the method for detecting leakage in underground pipe corridors based on static infrared thermal image processing disclosed in Chinese Patent Publication No. CN110268190A is an infrared thermal imaging method, which is highly dependent on the environment and has low accuracy.

[0005] 3. False alarms in high humidity environments: When the ambient humidity is high (such as rainy or foggy days, or humid weather), condensation is easily formed on the surface of pipes. Existing visual or optical detection methods cannot distinguish between natural condensation and abnormal condensation caused by leakage, resulting in a large number of false alarms. Summary of the Invention

[0006] The technical problem to be solved by this invention is that existing pipeline leakage identification methods are not very accurate and have problems with intrusion and cost.

[0007] This invention solves the above-mentioned technical problems through the following technical means: a non-contact pipeline leakage early identification condensation image monitoring method, comprising: S1. A condensation observation membrane is installed on the pipeline at the monitoring point. The surface of the condensation observation membrane has a grid-like guide groove. The ambient temperature and relative humidity are collected in real time, and the theoretical dew point temperature of the current environment is calculated. S2. When the pipe surface temperature is higher than the theoretical dew point temperature, if a condensation area is detected on the condensation observation film, it is directly judged as a strong anomaly; otherwise, it is judged as a normal state. When the pipe surface temperature is less than or equal to the theoretical dew point temperature, it is identified whether the condensation coverage area on the condensation observation film shows a grid-like texture or directional flow trajectory. If so, it is judged as a morphological anomaly; otherwise, it is judged as a normal state. The condensation area includes water droplet areas, wet areas, and connected water trace areas. The condensation coverage area refers to the coverage area of ​​the condensation area in the image. S3. When the morphological abnormality or strong abnormality is determined, the video stream of the condensation observation membrane is captured after the reset operation is performed on the surface of the condensation observation membrane, and the abnormality score is performed. The leakage or false alarm is determined based on the abnormality score. The reset operation refers to the heating and blowing operation of the condensation observation membrane surface for a preset time.

[0008] Furthermore, the calculation process for the theoretical dew point temperature is as follows: Calculate intermediate parameters ,in, and All are fitting coefficients. For ambient temperature, Relative humidity; Calculate the theoretical dew point temperature .

[0009] Furthermore, the identification of whether the condensation-covered area on the condensation observation membrane exhibits a grid-like texture or directional flow trajectory is determined. If so, it is judged as an abnormal morphology; otherwise, it is judged as a normal state. This includes the following steps: The acquired images of the condensation observation membrane were subjected to median filtering for noise reduction and histogram equalization to obtain the preprocessed images; Using a trained semantic segmentation network, the condensation region in the preprocessed image is stripped from the background to generate a binarized mask image; After obtaining the condensation region from the binarized mask image, the flow conductivity consistency index between the condensation coverage area and the preset mesh template is calculated. ,in, Represents the number of effective condensation units. Representing the i The growth direction angle of each condensation unit; The surface of the condensation observation film represents the first... iThe direction angle of the guide groove at each condensation unit; the growth direction angle of the i-th condensation unit is obtained as follows: first, the connected condensation regions are marked in the binary mask image to obtain multiple condensation units; then, the second-order central moment matrix is ​​calculated based on the pixel coordinates within the condensation unit, and the second-order central moment matrix is ​​decomposed into eigenvalues. The angle between the eigenvector corresponding to the largest eigenvalue and the image reference axis is taken as the growth direction angle of the i-th condensation unit; for condensation units where the difference between the largest and second largest eigenvalues ​​is within a preset range, they are not counted in the number of effective condensation units; the direction angle of the guide groove at the i-th condensation unit is obtained as follows: the direction of the guide groove of the condensation observation film is given in advance. When the i-th condensation unit covers two or more sets of guide grooves at the same time, the direction of the guide groove with the largest overlap area with the i-th condensation unit is taken as the direction angle of the guide groove at the i-th condensation unit. If the consistency index of the diversion If the condensation area exceeds the first preset threshold and forms a line along the direction of the guide channel, it is determined to be an abnormal shape; otherwise, it is determined to be a normal state.

[0010] Furthermore, the identification of whether the condensation-covered area on the condensation observation membrane exhibits a grid-like texture or directional flow trajectory is determined; if so, it is judged as an abnormal morphology; otherwise, it is judged as a normal state, including: Using the RGB image of the condensation observation film in the current frame Baseline background image B and the difference map between the current frame and the background image. Construct input features The input features are fed into the MobileNetV3 network, and the output of the MobileNetV3 network is mapped to a morphological anomaly determination result through a softmax function; the baseline background image B The image shows the surface of the condensation observation film under initial conditions; the MobileNetV3 network is trained and then used to identify morphological anomalies.

[0011] Furthermore, the step of performing anomaly scoring and determining whether a leak is a false alarm based on the anomaly score includes the following steps: The condensation coverage area or condensation coverage rate of each frame in the video stream is statistically analyzed and extracted to construct a characteristic time series reflecting the condensation dynamics process; the condensation coverage rate is the ratio of the condensation coverage area to the total area of ​​the condensation observation film. The rate of change is calculated based on the characteristic time series, using the following formula:

[0012] in, This indicates the duration of the observation window after reset. Indicates the current moment. This indicates the time point after the condensation observation membrane has completed its reset. For the condensation coverage area, The ambient condensation time constant; If the rate of change If the rate of change exceeds the preset threshold, it is determined to be a leak and a formal alarm is issued; otherwise, it is determined to be a false alarm.

[0013] Furthermore, the step of performing anomaly scoring and determining leakage or false alarm based on the anomaly score also includes: The temporal characteristics of the condensation coverage and change rate of consecutive frames after reset are input into the LSTM model. The LSTM model is a single-layer or two-layer LSTM network structure. The output of the LSTM model is the anomaly confidence score. When the anomaly confidence score exceeds the second preset threshold, it is judged as leakage and a formal alarm is issued. Otherwise, it is judged as a false alarm. The physical reference loss function of the LSTM model is expressed as follows: ,in, The current temperature of the pipe surface. This is the theoretical dew point temperature at the current moment; When training the LSTM model, its parameters are adjusted until the preset number of iterations or the maximum value of the physical reference loss function is reached, at which point the iteration stops, resulting in a trained LSTM model. The trained LSTM model is then used to determine leakage or false alarms.

[0014] This invention also provides a non-contact condensation image monitoring system for early identification of pipeline leaks, comprising: The condensation observation membrane has a prefabricated grid-like array of micron-sized flow channels on its surface. The surface of the condensation observation membrane has hydrophilic and hydrophobic regions. The hydrophilic regions coincide with the array of micron-sized flow channels in spatial position, while the hydrophobic regions are distributed on the flat surface between the grooves of the array of micron-sized flow channels. The protective housing assembly has its edge connected to the bottom of the condensation observation membrane, forming a closed cavity. The protective housing assembly is installed on the outer wall of the pipe and the condensation observation membrane is attached to the outer wall of the pipe to be monitored. An image acquisition unit, located inside the protective housing assembly, is used to acquire image data of the condensation observation film surface; An environmental sensing unit is used to collect ambient temperature and relative humidity data in real time. The edge computing module is communicatively connected to the image acquisition unit and the environmental sensing unit, respectively, to execute the above-mentioned non-contact pipeline leakage early identification condensation image monitoring method.

[0015] Furthermore, the condensation image monitoring system also includes an active reset module, which consists of a heating element and a fan. The fan outlet is aligned with the condensation observation film, and the heating element is positioned opposite to the surface of the micron-level flow channel array on the condensation observation film. The active reset module is connected to the edge computing module. The active reset module responds to the control commands of the edge computing module to dry the surface of the condensation observation film.

[0016] Furthermore, the protective housing assembly is an opaque hollow cavity structure with a transparent isolation window inside, dividing the cavity into a sensing area at the bottom and a device area at the top. The sensing area includes a condensation observation membrane and an active reset module, which is fixed to the rear wall of the protective housing assembly. The device area, from top to bottom, includes an edge computing module, an image acquisition unit, and a ring illumination unit, all of which are fixed to the rear wall of the protective housing assembly. An environmental sensing unit is located at the vent of the protective housing assembly. The active reset module, edge computing module, image acquisition unit, environmental sensing unit, and ring illumination unit are all connected to a power source.

[0017] Furthermore, the bottom left and right sides of the protective shell assembly are configured with arc-shaped structures that match the outer wall of the pipe. The edges of the condensation observation membrane are fixed to the bottom of the protective shell assembly, forming the bottom surface of the protective shell assembly, and the shape of the bottom surface of the protective shell assembly matches the shape of the outer wall of the pipe. A front baffle and a rear baffle extending downward are also provided on the front and rear sides of the bottom of the protective shell assembly. The front baffle, the rear baffle, and the arc-shaped structure form a hollow area. When the protective shell assembly is placed downward at a preset position on the pipe, the pipe is stuck in the hollow area, and the condensation observation membrane is attached to the outer wall of the pipe.

[0018] The advantages of this invention are: (1) This invention calculates the theoretical dew point temperature of the current environment, uses the pipe surface temperature and the theoretical dew point temperature to determine whether there is an anomaly, and when it is determined to be a morphological anomaly or a strong anomaly, it performs a reset operation on the surface of the condensation observation film and captures the video stream of the condensation observation film, performs anomaly scoring, and determines leakage or false alarm based on the anomaly score. Using the physical criteria of dew point, it effectively distinguishes between environmental condensation and leakage condensation. Using the grid-like guide channel combined with relevant criteria, it transforms disordered natural water droplets into ordered artificial textures, improves the accuracy of recognition, and reduces false alarms in high humidity environments. At the same time, the test process does not require the deployment of sensors inside the pipe, and there are no invasiveness or cost issues.

[0019] (2) This invention provides a non-contact monitoring scheme that can accurately isolate abnormal condensation features induced by minute leaks from complex natural condensation backgrounds and eliminate interference from light and wind speed. Through the design of a closed darkroom and active reset verification, the robustness of the system under harsh outdoor conditions is greatly improved, and the monitoring accuracy is greatly enhanced.

[0020] (3) The theoretical dew point temperature calculation process of this invention first uses intermediate parameters. It expresses the saturation state of water vapor under the current environment, and then uses a constant. and The nonlinear proportional transformation is used to inversely determine the critical temperature at which water vapor in the air reaches saturation and begins to condense. This transforms the raw temperature and humidity data collected by the environmental sensing unit into a theoretical dew point temperature with thermodynamic significance. This provides a unified thermodynamic criterion for determining whether condensation should occur under current operating conditions, enabling the system to adaptively adjust the judgment benchmark according to environmental changes. This reduces reliance on fixed empirical thresholds, thereby improving the stability and applicability of leakage identification under different seasons and humid and hot conditions.

[0021] (4) The present invention calculates the flow consistency index between the condensation region and the preset grid template. This index quantifies the degree of matching between the actual growth direction of condensation and the direction of the micro-channel structure. A value closer to 1 indicates that condensation strictly follows the channel, while a value closer to 0 indicates random condensation. Since natural condensation has no direction, while leakage condensation has a direction consistent with a specific design structure, this index can accurately reflect the degree to which the overall condensation behavior is guided by the microstructure. A higher value indicates that the condensation is not naturally and randomly generated, but driven by a local, continuous water vapor source. It serves as a core morphological indicator to distinguish between natural condensation and leakage-induced condensation. By constructing a channel consistency index, the condensation morphology is transformed from a manual qualitative judgment into a quantifiable evaluation indicator. This effectively distinguishes between random natural condensation and abnormal condensation growing along a preset channel structure, improving the objectivity, repeatability, and portability of the judgment results. It also facilitates the unified threshold deployment in engineering sites, significantly improving monitoring accuracy and reducing false alarm rates.

[0022] (5) By performing time-series discrimination on the growth rate of the condensation coverage area or condensation coverage rate after reset, the present invention can identify the dynamic difference between the continuous leakage source and the natural humidity of the environment, identify the continuous leakage source in a short time, rather than relying solely on a single frame of static image for judgment, thereby improving the early detection capability of micro-leakage and reducing false alarms caused by transient interference such as environmental humidity and occasional attached water droplets.

[0023] (6) The present invention can also use a lightweight MobileNetV3 network for morphological anomaly recognition and use an LSTM model for temporal behavior verification. While ensuring recognition accuracy, it also takes into account the real-time and low power consumption requirements of edge computing scenarios, making it easy to deploy online at on-site monitoring nodes and improving the engineering practicality and early warning response efficiency of the system. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the monitoring device in the non-contact pipeline leakage early identification condensate image monitoring method disclosed in the embodiments of the present invention; Figure 2 This is a schematic diagram of the microstructure of the condensation observation membrane in the non-contact pipeline leakage early identification condensation image monitoring method disclosed in the embodiments of the present invention; Figure 3 This is a flowchart of a non-contact pipeline leakage early identification condensation image monitoring method disclosed in an embodiment of the present invention; Figure 4 This is a schematic diagram of the active reset verification process in the condensation image monitoring method for early identification of non-contact pipeline leakage disclosed in the embodiments of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0026] Example 1 Embodiment 1 of this invention provides a non-contact method for early identification of pipeline leaks using condensation image monitoring, the hardware architecture of which is as follows: Figure 1 The monitoring device shown.

[0027] The monitoring device includes a condensation observation film 1 (COF), which is the core sensing medium of the system. It is made of a transparent polymer film, such as PET, PC, or a highly flexible TPU substrate film, with a thickness of 0.05–0.5 mm, and maintains an optical transmittance of over 90% to ensure clear imaging. The surface of the transparent polymer film can be further coated with a hydrophilic coating containing silica nanoparticles to form hydrophilic regions, and with fluorinated silanes or fluorinated polymers to form hydrophobic regions. The hydrophilic regions spatially overlap with a pre-fabricated grid-like array of micron-sized flow channels, while the hydrophobic regions are distributed on the flat surface between the grooves of the micron-sized flow channel array. The hydrophilic and hydrophobic regions can be achieved through mask spraying, screen printing, or plasma treatment combined with patterned coating; the micron-sized flow channel array can be formed in one step using laser etching or embossing molds to create a grid-like groove structure. Figure 2 As shown, there are hydrophilic meshes with a spacing of 0.5 mm, and the remaining areas are hydrophobic surfaces. Figure 2 From left to right, the diagrams show the structure of the guide channel, the distribution of disordered water droplets due to natural condensation, and the leakage condensation of the condensation area connected along the direction of the guide channel.

[0028] The working principle of the condensation observation membrane 1 is that under natural condensation conditions, water droplets tend to be randomly distributed; while when leakage occurs, the continuously overflowing micro-airflow or micro-droplets will preferentially fill the hydrophilic guide channel, forming a clear grid-like texture on the image as a strong visual feature.

[0029] The monitoring device also includes a protective housing assembly 2, which is an opaque hollow cavity structure. Its bottom is sealed to the outer wall of the pipe 3, forming a relatively closed, photothermally stable cavity. A transparent isolation window 4 (such as optical-grade tempered glass) is provided inside the cavity, dividing it into a lower sensing area (containing the condensation observation membrane 1) and an upper equipment area. The sensing area includes the condensation observation membrane 1 and an active reset module 5. The edges of the condensation observation membrane 1 are fixed to the bottom of the protective housing assembly 2, forming the bottom surface of the protective housing assembly 2. The active reset module 5 is fixed to the rear wall of the protective housing assembly 2. The active reset module 5 is an integrated small ceramic heating element and a micro fan. The fan outlet is directed towards the condensation observation membrane 1 for rapid drying. The heating element indirectly heats the condensation observation membrane 1 through heat conduction to achieve rapid evaporation of water droplets on the membrane surface. The heating element is positioned opposite the surface of the micron-level flow channel array of the condensation observation membrane 1, without directly contacting the flow channel structure area, to avoid interference with the micron-level flow channel array and condensation behavior.

[0030] The device area is arranged from top to bottom as follows: an edge computing module 6, an image acquisition unit 7, and a ring illumination unit 8. All three are fixed to the rear wall of the protective housing assembly 2. The ring illumination unit 8 is a ring-shaped light source providing uniform, controlled light with adjustable brightness. The light source outlet is aligned with the transparent isolation window 4. The image acquisition unit 7 is equipped with a macro lens, located above the ring illumination unit 8, with its lens aligned with the transparent isolation window 4 through the central blank area of ​​the ring illumination unit 8, to acquire images of the condensation observation film 1 surface within the sensing area under light conditions. The edge computing module 6 is located above the image acquisition unit 7 or fixed in another position within the device area that does not obstruct the image acquisition unit 7 and the ring illumination unit 8. An environmental sensing unit 9 is installed at the vent of the protective housing assembly 2 to collect ambient temperature and relative humidity in real time.

[0031] The image acquisition unit 7 and the environment sensing unit 9 are both communicatively connected to the edge computing module 6. The active reset module 5, the edge computing module 6, the image acquisition unit 7, the environment sensing unit 9, and the ring illumination unit 8 are all connected to a power supply.

[0032] The bottom left and right sides of the protective housing assembly 2 are designed with arc-shaped structures 10 that match the outer wall of the pipe 3. The edges of the condensation observation membrane 1 are fixed to the bottom of the protective housing assembly 2, forming the bottom surface of the protective housing assembly 2, and the shape of the bottom surface of the protective housing assembly 2 matches the shape of the outer wall of the pipe 3. To prevent the entire device from being unstable during installation, a front baffle 11 and a rear baffle extending downwards are provided on the front and rear sides of the bottom of the protective housing assembly 2. The front baffle 11, the rear baffle, and the arc-shaped structure 10 form a hollow area. In actual application, the protective housing assembly 2 is placed downwards at a preset position on the pipe 3, and the pipe 3 is just stuck in the area enclosed by the front baffle 11, the rear baffle, and the arc-shaped structure 10. The condensation observation membrane 1 at the bottom of the protective housing assembly 2 is attached to the outer wall of the pipe 3. The working principle of each component inside the protective housing assembly 2 is to form an optical and thermal darkroom. It shields external wind speed (wind speed will accelerate evaporation and cover up minor leaks) and external sunlight (changes in light will interfere with image recognition), ensuring that the internal environment is only controlled by the surface state of the pipe 3. Figure 1 This is a simplified diagram of the equipment space relationship. To make it easier to understand the positional relationship of each component, the front panel of the protective housing assembly 2 is removed and part of the side panel is cut out. In actual application, the protective housing assembly 2 is a closed structure, and these components are enclosed inside.

[0033] The edge computing module 6 runs the relevant algorithms of the condensation image monitoring method for early identification of non-contact pipeline leaks and controls the actions of each hardware module. For example... Figure 3 As shown, the specific process is as follows: Step S1: Dew point calculation The system operates in low-power polling mode (e.g., sampling once per minute). Edge computing module 6 reads the ambient temperature collected by environmental sensing unit 9 in real time. and relative humidity RH Calculate the theoretical dew point temperature of the current environment. The calculation process is as follows: (1) Calculate intermediate parameters :

[0034] In the formula, and These are all fitting coefficients. Intermediate parameters. This expresses the saturation state of water vapor under the current environment. Because in actual sensor measurements, relative humidity... RH It is usually expressed as a percentage. Therefore, first... RH Dividing by 100 restores the decimal form. Since the change of water vapor with temperature follows an exponential law, by taking the natural logarithm, the complex exponential relationship can be transformed into an operand that can be linearly added and subtracted. For the temperature compensation term, the fitting coefficients used are... b =17.625 and c =243.04℃ is a standard fitting coefficient recognized in the field of meteorology. This set of constants exhibits extremely high thermodynamic fitting accuracy within the environmental temperature range of -45℃ to 60℃, ensuring that the formula maintains extremely high fitting accuracy within the Earth's normal operating temperature range. This term reflects the current environmental temperature. The baseline capacity of the air to hold water vapor.

[0035] (2) Calculate the theoretical dew point temperature :

[0036] This formula achieves the result from intermediate parameters. The inverse mapping to the temperature space. Given the current atmospheric saturation state (from... Under the premise of expression, through constants and The nonlinear proportional transformation is used to inversely determine the critical temperature at which water vapor in the air reaches saturation and begins to condense.

[0037] Through the above calculations, the edge computing module 6 transforms the raw temperature and humidity data collected by the environmental sensing unit 9 into a theoretical dew point temperature with thermodynamic significance. Simultaneously, the surface temperature of pipe 3 is obtained through an infrared probe or thermocouple. .

[0038] Step S2: Dual anomaly detection (initial screening) Edge computing module 6 acquires an image of the condensation observation film 1 and performs logical judgments based on temperature data: Operating condition A (strong criterion) > At this point, the temperature of pipe 3 is higher than the dew point, and theoretically, natural condensation should not occur. If the image recognition algorithm detects a condensation area on the membrane surface, the system directly determines it as a strong anomaly and sets the confidence level to the highest; otherwise, it is determined to be a normal state.

[0039] Operating condition B (weak criterion) ≤ At this point, the environment allows condensation. The system initiates morphological analysis, focusing on identifying whether the condensation-covered area exhibits [certain characteristics]. Figure 2 The aforementioned grid-like texture (guided by microstructures) or directional flow trajectory. If such non-random features are detected, it is determined to be a morphological abnormality; otherwise, it is determined to be a normal state.

[0040] In this embodiment, the condensation area refers to the condensation foreground area on the surface of the condensation observation film obtained by image recognition or image segmentation, including the water droplet area, the wet area, and the connected water trace area; the condensation coverage area refers to the coverage range of the condensation area in the image.

[0041] The specific process of morphological analysis to identify grid-like textures is as follows: 1. Image preprocessing: Median filtering and histogram equalization are performed on the acquired images of the condensation observation membrane 1 to enhance the contrast between water droplets and the background membrane surface. 2. Semantic Segmentation Extraction: Using a trained semantic segmentation network, the condensation region in the image is extracted from the background to generate a binary mask. Specifically, the RGB or grayscale image of the condensation observation film 1 is normalized and size-standardized before being input into the semantic segmentation network to extract multi-scale features. The output is a binary probability map with the same size as the input. For any pixel in the binary probability map, if the predicted probability that the pixel belongs to the condensation region is greater than or equal to 0.5, the pixel is recorded as 1; if the predicted probability that the pixel belongs to the condensation region is less than 0.5, the pixel is recorded as 0. Here, 1 represents the foreground region of the condensation droplets, and 0 represents the background region, thus obtaining the binary mask.

[0042] 3. Calculate texture regularity and texture regularity score: After obtaining the condensation region in the binarized mask image, calculate the flow consistency index between the condensation coverage area and the preset mesh template. If the index exceeds a first preset threshold (e.g., 0.8), and the condensation area forms a continuous line along the direction of the guide channel, it is determined to be a non-random specific geometric texture, indicating an abnormal shape. The guide consistency index between the condensation coverage area and the preset mesh template is calculated. The formula is shown below:

[0043] in, Represents the number of effective condensation units, which indicates the number of effective condensation regions identified in the current image; Representing the i The growth direction angle of each condensation unit; The surface of the condensation observation film 1 represents the first... i The direction and angle of the guide groove at each condensation unit; This value is used to quantify the degree of matching between the actual growth direction of condensation and the direction of the micro-guide structure. The closer the value is to 1, the more it indicates that the condensation develops strictly along the guide groove, and the closer it is to 0, the more it indicates that the condensation occurs in a random direction.

[0044] The growth direction angle of the i-th condensation unit is obtained as follows: First, connected regions of the condensation area are marked in the binarized mask image to obtain multiple condensation units; then, the second-order central moment matrix of the condensation unit is calculated based on the pixel coordinates within the condensation unit, and eigenvalue decomposition is performed on the second-order central moment matrix. The angle between the eigenvector corresponding to the largest eigenvalue and the image reference axis is taken as the growth direction angle of the i-th condensation unit. Principal component analysis is an equivalent implementation of the above eigenvalue decomposition. For condensation units where the largest and second-largest eigenvalues ​​are close (the difference between the largest and second-largest eigenvalues ​​is within a preset range), they are determined to be approximately circular and have no obvious directionality, and are not included in the number of effective condensation units.

[0045] The method for obtaining the direction angle of the guide channel at the i-th condensation unit is as follows: the direction of the guide channel of the condensation observation membrane 1 is given in advance by the structural design parameters; when the i-th condensation unit covers two or more sets of guide channels at the same time, the direction of the guide channel with the largest overlap area with the i-th condensation unit is taken as the direction angle of the guide channel at the i-th condensation unit.

[0046] Since natural condensation usually does not have stable directional growth characteristics, while leakage-induced condensation is more likely to extend along the direction of the pre-set flow channel, the flow consistency index can reflect the degree to which the overall condensation behavior is guided by the microstructure. The higher the value, the more likely the condensation is not naturally and randomly generated, but driven by a local continuous water vapor source, thus it can be used as a morphological discrimination index to distinguish between natural condensation and leakage-induced condensation.

[0047] Step S3: Active reset verification (closed-loop confirmation), this is a key step in the invention to eliminate false alarms. For example... Figure 4 The diagram illustrates the entire process from detecting a suspected leak to forced reset and then confirming a persistent leak. Specifically, when step S2 determines a strong anomaly and a morphological anomaly, the system does not immediately trigger an alarm but instead initiates a verification process: 1. Cleaning Phase: Edge computing module 6 triggers a control signal, driving active reset module 5 to perform short-term forced processing on the surface of condensation observation film 1 (e.g., continuous heating and blowing operation for 30 seconds). Through local thermophysical intervention or forced convection, the existing condensate droplets on the surface of condensation observation film 1 are completely evaporated, restoring the image features to a dry baseline state.

[0048] 2. Observation Phase: After the active reset module 5 stops working, the image acquisition unit 7 immediately switches to a high sampling rate mode (e.g., 10fps–30fps) to capture the dynamic video stream. The edge computing module 6 processes the video stream frame by frame, specifically including: (1) Real-time segmentation: The morphological recognition path (i.e., the method of morphological analysis to identify grid-like textures mentioned above) is used to segment the condensation region of each frame of the image; (2) Sequence generation: The condensation coverage area or condensation coverage rate of each frame in the video stream is statistically extracted, thereby constructing a characteristic time series reflecting the condensation dynamics process in time sequence; the condensation coverage rate is the ratio of the condensation coverage area to the total area of ​​the condensation observation film.

[0049] 3. Time-series behavior verification and judgment: Edge computing module 6 performs anomaly scoring on the above-mentioned feature time series. The specific process is as follows: Scenario 1 (Confirmed Leakage): If video analysis shows that the condensation coverage area or condensation coverage rate in a local area within the observation window after reset exhibits a non-linear, abrupt increase (i.e., the rate of change) If the rate of change significantly exceeds the normal environmental level (the normal environmental level is a preset rate of change threshold), it indicates the presence of a persistent internal leakage source, and the system issues a formal alarm. (Rate of change) The calculation formula is as follows:

[0050] in, This indicates the duration of the observation window after reset. Indicates the current moment. The time point after the condensation observation membrane 1 has been reset (forced drying, restored to baseline state) is used as a unified starting reference for recondensation behavior; This indicates a short time interval during which the system focuses on monitoring recondensation behavior; condensation caused by leakage usually occurs rapidly within this time window. For the condensation coverage area, It represents the rate of increase of the condensation coverage area per unit time, and is used to measure whether the water vapor supply is continuous and whether it has an internal source. The environmental condensation time constant represents the characteristic time scale required for condensation to be caused solely by changes in ambient humidity under leak-free conditions. This parameter can be obtained through baseline calibration or historical statistical data. This is an exponential decay term used to suppress the slow environmental condensation behavior far from the reset time.

[0051] Scenario 2 (excluding false alarms): If the video shows that the membrane surface remains dry, or the changes in the characteristic time series conform to a slow and consistent increase caused by the environmental dew point balance (i.e., the rate of change) If the temperature is below the normal environmental level, it is determined to be environmental interference, and the system will automatically return to its normal position.

[0052] This invention, considering the economic efficiency of monitoring long-distance pipelines 3, adopts a strategy of key node coverage + regional flow guidance. Specifically, a condensation observation membrane 1 is deployed at high-risk locations, and its surface array of micron-level flow channels guides leaking water vapor in a directional manner, thus transforming point-based monitoring into localized coverage monitoring of high-risk areas. Furthermore, through RS485 or LoRa networking, multiple monitoring nodes can share a single edge computing module 6, achieving collaborative early warning of risk points along the entire pipeline.

[0053] Through the above technical solutions, this invention effectively distinguishes between environmental condensation and leakage condensation by utilizing dew point physical criteria. By using a micron-level flow channel array combined with relevant algorithms, disordered natural water droplets are transformed into ordered artificial textures, thereby improving the accuracy of recognition. Through the closed darkroom design and active reset verification, the robustness of the system under harsh outdoor conditions is greatly improved.

[0054] Example 2 The difference between Embodiment 2 and Embodiment 1 is that in this embodiment, the morphological anomaly discrimination in step S2 can be implemented using a trained MobileNetV3 network; and in this embodiment, the anomaly scoring and leakage confirmation in step S3 can be implemented using an LSTM model. The implementation methods of deep learning models (MobileNetV3 network and LSTM model) are parallel optional implementation methods to the rule-based and physical criterion-based implementation method in Embodiment 1.

[0055] In this embodiment, the morphological analysis for identifying the mesh-like texture extraction is implemented using a lightweight MobileNetV3 network, specifically trained to recognize the mesh texture guided by the micrometer-scale channel array. Considering that condensation droplets exhibit characteristics such as localized highlight enhancement, edge blurring due to refraction, regional brightness gradient changes, and morphological changes over time in imaging, while the micrometer-scale channel array is a fixed periodic texture with a stable spatial frequency distribution, the MobileNetV3 network can achieve accurate segmentation through feature differences. Specifically, the input to MobileNetV3 is a three-channel image feature tensor, including the RGB image of the condensation observation film 1 in the current frame. Baseline background image B(Image of the surface of condensation observation film 1 under initial conditions), difference image between the current frame and the background image. The final input format is: The output of the MobileNetV3 network is mapped to the determination result of whether the shape is abnormal through the softmax function. For example, the output 1 indicates a shape abnormality and 0 indicates a normal state.

[0056] In this embodiment, the temporal behavior verification and judgment process uses a single-layer or double-layer LSTM network structure to model the dynamic changes of condensation behavior over time. Since leakage-induced condensation after reset exhibits rapid growth in a short period, concentrated expansion in localized areas, and an exponential or step-like growth curve, while natural environmental condensation exhibits slow linear growth, uniform damping across the entire area, and no obvious abrupt changes, the LSTM can distinguish between these two types of temporal evolution patterns by comparing the differences between the previous and current states. Specifically, the LSTM input consists of the temporal characteristics of the condensation coverage rate (the ratio of the condensation coverage area to the total area of ​​the condensation observation film 1) and the rate of change in consecutive frames after reset, used to identify recondensation patterns with abrupt growth in a short period. The LSTM output is an anomaly confidence score S∈[0,1]. When the score exceeds a second preset threshold (e.g., 0.8), it is determined to be a persistent physical leakage, and the system issues a formal alarm; otherwise, it is in a normal state. The physical reference loss function of the LSTM can be expressed as... The calculation method is as follows:

[0057] in, The current temperature of the pipe surface. The theoretical dew point temperature at the current moment. This represents the thermodynamic range in which condensation should not occur.

[0058] When training an LSTM, its parameters are adjusted until the preset number of iterations or the maximum value of the physical reference loss function is reached, at which point the iteration stops, resulting in a trained LSTM model.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A non-contact method for early identification of pipeline leaks using condensation image monitoring, characterized in that, include: S1. A condensation observation membrane is installed on the pipeline at the monitoring point. The surface of the condensation observation membrane has a grid-like flow guide groove. Real-time collection of ambient temperature and relative humidity; calculation of the theoretical dew point temperature of the current environment. S2. When the pipe surface temperature is higher than the theoretical dew point temperature, if a condensation area is detected on the condensation observation film, it is directly judged as a strong anomaly; otherwise, it is judged as a normal state. When the pipe surface temperature is less than or equal to the theoretical dew point temperature, it is identified whether the condensation coverage area on the condensation observation film shows a grid-like texture or directional flow trajectory. If so, it is judged as a morphological anomaly; otherwise, it is judged as a normal state. The condensation area includes water droplet areas, wet areas, and connected water trace areas. The condensation coverage area refers to the coverage area of ​​the condensation area in the image. S3. When an abnormal shape or strong abnormality is detected, after resetting the surface of the condensation observation membrane, the video stream of the condensation observation membrane is captured, and an anomaly score is performed. Based on the anomaly score, leakage or false alarm is determined. The resetting operation refers to heating and blowing the surface of the condensation observation membrane for a preset time. The anomaly score and the determination of leakage or false alarm based on the anomaly score include the following steps: The condensation coverage area or condensation coverage rate of each frame in the video stream is statistically analyzed and extracted to construct a characteristic time series reflecting the condensation dynamics process; the condensation coverage rate is the ratio of the condensation coverage area to the total area of ​​the condensation observation film. The rate of change is calculated based on the characteristic time series, using the following formula: in, This indicates the duration of the observation window after reset. Indicates the current moment. This indicates the time point after the condensation observation membrane has completed its reset. For the condensation coverage area, The ambient condensation time constant; If the rate of change If the rate of change exceeds the preset threshold, it is determined to be a leak and a formal alarm is issued; otherwise, it is determined to be a false alarm.

2. The non-contact pipeline leakage early identification condensation image monitoring method according to claim 1, characterized in that, The calculation process for the theoretical dew point temperature is as follows: Calculate intermediate parameters ,in, and All are fitting coefficients. For ambient temperature, Relative humidity; Calculate the theoretical dew point temperature .

3. The non-contact pipeline leakage early identification condensation image monitoring method according to claim 1, characterized in that, The identification process involves determining whether the condensation-covered area on the condensation observation membrane exhibits a grid-like texture or directional flow trajectory. If so, it is considered an abnormal morphology; otherwise, it is considered normal. This process includes the following steps: The acquired images of the condensation observation membrane were subjected to median filtering for noise reduction and histogram equalization to obtain the preprocessed images; Using a trained semantic segmentation network, the condensation region in the preprocessed image is stripped from the background to generate a binarized mask image; After obtaining the condensation region from the binarized mask image, the flow conductivity consistency index between the condensation coverage area and the preset mesh template is calculated. ,in, Represents the number of effective condensation units. Representing the i The growth direction angle of each condensation unit; The surface of the condensation observation film represents the first... i The direction angle of the guide groove at each condensation unit; the growth direction angle of the i-th condensation unit is obtained as follows: first, the connected condensation regions are marked in the binary mask image to obtain multiple condensation units; then, the second-order central moment matrix is ​​calculated based on the pixel coordinates within the condensation unit, and the second-order central moment matrix is ​​decomposed into eigenvalues. The angle between the eigenvector corresponding to the largest eigenvalue and the image reference axis is taken as the growth direction angle of the i-th condensation unit; for condensation units where the difference between the largest and second largest eigenvalues ​​is within a preset range, they are not counted in the number of effective condensation units; the direction angle of the guide groove at the i-th condensation unit is obtained as follows: the direction of the guide groove of the condensation observation film is given in advance. When the i-th condensation unit covers two or more sets of guide grooves at the same time, the direction of the guide groove with the largest overlap area with the i-th condensation unit is taken as the direction angle of the guide groove at the i-th condensation unit. If the consistency index of the diversion If the condensation area exceeds the first preset threshold and forms a line along the direction of the guide channel, it is determined to be an abnormal shape; otherwise, it is determined to be a normal state.

4. The non-contact pipeline leakage early identification condensation image monitoring method according to claim 1, characterized in that, The identification method determines whether the condensation-covered area on the condensation observation membrane exhibits a grid-like texture or directional flow trajectory. If so, it is judged as an abnormal morphology; otherwise, it is judged as a normal state, including: Using the RGB image of the condensation observation film in the current frame Baseline background image B and the difference map between the current frame and the background image. Construct input features The input features are fed into the MobileNetV3 network, and the output of the MobileNetV3 network is mapped to a morphological anomaly determination result through a softmax function; the baseline background image B The image shows the surface of the condensation observation film under initial conditions; the MobileNetV3 network is trained and then used to identify morphological anomalies.

5. The non-contact pipeline leakage early identification condensation image monitoring method according to claim 1, characterized in that, The process of performing anomaly scoring and determining whether a leak is a false alarm based on the anomaly score also includes: The temporal characteristics of the condensation coverage and change rate of consecutive frames after reset are input into the LSTM model. The LSTM model is a single-layer or two-layer LSTM network structure. The output of the LSTM model is the anomaly confidence score. When the anomaly confidence score exceeds the second preset threshold, it is judged as leakage and a formal alarm is issued. Otherwise, it is judged as a false alarm. The physical reference loss function of the LSTM model is expressed as follows: ,in, The current temperature of the pipe surface. This is the theoretical dew point temperature at the current moment; When training the LSTM model, its parameters are adjusted until the preset number of iterations or the maximum value of the physical reference loss function is reached, at which point the iteration stops, resulting in a trained LSTM model. The trained LSTM model is then used to determine leakage or false alarms.

6. A non-contact pipeline leakage early detection condensation image monitoring system, characterized in that, include: The condensation observation membrane has a prefabricated grid-like array of micron-sized flow channels on its surface. The surface of the condensation observation membrane has hydrophilic and hydrophobic regions. The hydrophilic regions coincide with the array of micron-sized flow channels in spatial position, while the hydrophobic regions are distributed on the flat surface between the grooves of the array of micron-sized flow channels. The protective housing assembly has its edge connected to the bottom of the condensation observation membrane, forming a closed cavity. The protective housing assembly is installed on the outer wall of the pipe and the condensation observation membrane is attached to the outer wall of the pipe to be monitored. An image acquisition unit, located inside the protective housing assembly, is used to acquire image data of the condensation observation film surface; An environmental sensing unit is used to collect ambient temperature and relative humidity data in real time. The edge computing module is communicatively connected to the image acquisition unit and the environmental sensing unit, respectively, and executes the condensation image monitoring method for early identification of non-contact pipeline leakage as described in any one of claims 1-5.

7. The non-contact pipeline leakage early identification condensation image monitoring system according to claim 6, characterized in that, It also includes an active reset module, which consists of a heating element and a fan. The fan outlet is aligned with the condensation observation film, and the heating element is positioned opposite to the surface of the micron-level guide groove array of the condensation observation film. The active reset module is connected to the edge computing module. The active reset module responds to the control commands of the edge computing module to dry the surface of the condensation observation film.

8. The condensation image monitoring system for early identification of pipeline leaks according to claim 7, characterized in that, The protective housing assembly is an opaque hollow cavity structure with a transparent isolation window inside, dividing the cavity into a sensing area at the bottom and a device area at the top. The sensing area includes a condensation observation membrane and an active reset module, which is fixed to the rear wall of the protective housing assembly. The device area, from top to bottom, includes an edge computing module, an image acquisition unit, and a ring illumination unit, all of which are fixed to the rear wall of the protective housing assembly. An environmental sensing unit is located at the vent of the protective housing assembly. The active reset module, edge computing module, image acquisition unit, environmental sensing unit, and ring illumination unit are all connected to a power source.

9. The non-contact pipeline leakage early identification condensation image monitoring system according to claim 6, characterized in that, The bottom left and right sides of the protective shell assembly are set into arc-shaped structures that match the outer wall of the pipe. The edges of the condensation observation membrane are fixed to the bottom of the protective shell assembly to form the bottom surface of the protective shell assembly, and the shape of the bottom surface of the protective shell assembly matches the shape of the outer wall of the pipe. A front baffle and a rear baffle extending downward are also set on the front and rear sides of the bottom of the protective shell assembly. The front baffle, the rear baffle, and the arc-shaped structure form a hollow area.

Images