Cooling tower sump liquid leakage visual monitoring method, device, equipment and storage medium

Abstract

The present application relates to the technical field of image processing, in particular to a kind of cooling tower sump leakage visual monitoring method, device, equipment and computer storage medium.The present application obtains video image sequence, carries out defogging and anti-glare enhancement based on dark channel prior, extracts region of interest according to monitoring area mask;The region of interest is input into semantic segmentation model, and water body static semantic mask is extracted;Dense optical flow field is calculated, and dynamic optical flow region with directional continuous motion characteristics is screened based on the fluid motion characteristics under the action of gravity;Static semantic mask and dynamic optical flow region are fused to obtain leakage feature map at pixel level, and time sequence trajectory tracking is carried out on connected region, and leakage event is judged and alarm signal is generated according to the duration of track and cumulative displacement.The present application effectively solves the problem of false alarm under the environment of high humidity steam and water splash, significantly improves the accuracy, anti-interference ability and all-weather adaptability of leakage monitoring under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a visual monitoring method, apparatus, equipment, and computer storage medium for monitoring leakage in cooling tower water collection tanks. Background Technology

[0002] Currently, for leak monitoring in cooling tower water collection tanks, existing technologies mainly rely on background modeling difference methods, color space threshold segmentation methods, or manual video inspections. However, cooling towers continuously experience extreme conditions such as high humidity steam, strong glare, and water splashing in the packing area. This leads to fundamental defects in traditional solutions: First, the detection logic based on pixel changes cannot distinguish between "random and disordered splashing water" and "continuous leakage guided by gravity," easily causing false alarms around the clock, forcing maintenance personnel to disable the automatic alarm function, and severely weakening the ability to detect minor leaks; Second, threshold segmentation based on fixed colors or brightness is easily affected by fog and specular reflection, frequently failing in low-light or high-reflectivity environments; Third, existing systems can only output binary results of "present / absent," lacking spatiotemporal consistency verification of the leakage trajectory and analysis of fluid physics mechanisms. They cannot accurately locate specific leak points such as cracks and welds, nor can they assess the severity of the leak, making it difficult to meet the practical needs of industrial sites for all-weather, high-reliability monitoring under complex conditions. Summary of the Invention

[0003] Therefore, the technical problem to be solved by the present invention is to overcome the problem that the prior art cannot distinguish between environmental interference and actual leakage under complex operating conditions of cooling towers, resulting in low monitoring accuracy, poor robustness and lack of leakage situation awareness.

[0004] To solve the above-mentioned technical problems, the present invention provides a visual monitoring method for leakage in a cooling tower water collection tank, comprising:

[0005] Acquire video image sequences of the monitored area; The video image sequence is subjected to image enhancement processing, and the region of interest image is extracted based on a preset monitoring area mask; The image of the region of interest is input into a pre-trained semantic segmentation model to extract a static semantic mask representing the distribution of water substances. Calculate the dense optical flow field between consecutive frames of images, and based on the fluid motion characteristics under gravity, filter out dynamic optical flow regions with directional continuous motion characteristics from the dense optical flow field; The static semantic mask and the dynamic optical flow region are fused at the pixel level to obtain a leakage feature map. The connected regions in the leakage feature map are tracked in a temporal sequence. The leakage event is determined based on the duration and cumulative displacement of the trajectory, and an alarm signal is generated.

[0006] Preferably, the image enhancement processing of the video image sequence includes: Based on the dark channel prior principle, the atmospheric light value and transmittance of the scene are estimated to restore the fog-free image and achieve defogging and anti-glare enhancement processing. The extraction of the region of interest image based on the preset monitoring area mask includes: Based on the preset division of high-risk monitoring areas and environmental interference areas, a binary mask is generated, and the pixel values ​​outside the binary mask in the video image sequence after image enhancement are forcibly set to zero to obtain the region of interest image.

[0007] Preferably, the extraction of the static semantic mask characterizing the distribution of substances in the water body includes: A lightweight semantic segmentation network containing a multi-scale feature extraction module is used to output a probability map of each pixel belonging to a water body. Threshold filtering is performed on the probability map, and the filtered pixel set is used to generate the static semantic mask.

[0008] Preferably, calculating the dense optical flow field between consecutive frame images includes: A series of consecutive frames of images are selected as input data, and a dense optical flow field is calculated through an optical flow network. The dense optical flow field describes the instantaneous motion velocity of the pixels. The dynamic optical flow region with directional continuous motion characteristics selected from the dense optical flow field based on the fluid motion characteristics under gravity includes: The motion direction of the optical flow vector in the dense optical flow field is calculated, a directional constraint interval and a velocity modulus threshold matching the direction of gravity are set, and a set of pixels that simultaneously satisfy the directional constraint interval and the velocity modulus threshold are extracted as the dynamic optical flow region.

[0009] Preferably, the step of fusing the static semantic mask and the dynamic optical flow region at the pixel level to obtain the leakage feature map includes: A pixel-level logical AND operation is performed on the static semantic mask and the dynamic optical flow region, so that only pixels that simultaneously possess water material properties and directional motion characteristics are retained in the leakage feature map.

[0010] Preferably, the step of performing time-series trajectory tracking on the connected regions in the leakage feature map, and determining the leakage event based on the duration and cumulative displacement of the trajectory, includes: The connected regions in the leakage feature map are tracked across multiple frames using a filter, and the number of consecutive frames and cumulative displacement distance of each tracked target trajectory are counted. When the number of continuous frames exceeds a preset frame threshold and the cumulative displacement distance exceeds a preset distance threshold, the trajectory is confirmed to be a real physical fluid flow.

[0011] Preferably, the generation of the alarm signal includes: The leakage severity index is calculated based on the area of ​​the confirmed leakage area and the average optical flow velocity. Different levels of alarm signals are generated based on the different ranges of the leakage severity index.

[0012] The present invention also provides a visual monitoring device for leakage in a cooling tower water collection tank, comprising: The image acquisition and preprocessing module is used to acquire video image sequences of the monitored area, perform image enhancement processing on the video image sequences, and extract the region of interest image based on a preset monitoring area mask. The static feature extraction module is used to input the image of the region of interest into a pre-trained semantic segmentation model to extract a static semantic mask that represents the distribution of water substances. The dynamic feature extraction module is used to calculate the dense optical flow field between consecutive frames of images, and based on the fluid motion characteristics under the action of gravity, to filter out dynamic optical flow regions with directional continuous motion characteristics from the dense optical flow field. The fusion and determination module is used to perform pixel-level fusion of the static semantic mask and the dynamic optical flow region to obtain a leakage feature map, and to perform temporal trajectory tracking on the connected regions in the leakage feature map. Based on the duration and cumulative displacement of the trajectory, the module determines the leakage event and generates an alarm signal.

[0013] The present invention also provides a visual monitoring device for leakage in a cooling tower water collection tank, comprising: Memory, used to store computer programs; A processor is used to implement the steps of the above-described visual monitoring method for leakage in a cooling tower water collection tank when executing the computer program.

[0014] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described visual monitoring method for leakage in a cooling tower water collection tank.

[0015] The technical solution of the present invention has the following advantages compared with the prior art: The visual monitoring method for cooling tower water collection tank leakage described in this invention effectively overcomes the fundamental defect of existing technologies that cannot distinguish between environmental interference and actual leakage under complex operating conditions of cooling towers by constructing a multimodal fusion monitoring mechanism of "static water body semantic features + gravity-constrained dynamic flow field features + temporal consistency verification". It significantly improves the monitoring accuracy and robustness in high humidity, strong glare and water splashing environments. At the same time, through pixel-level fusion and trajectory tracking, it realizes the situational awareness upgrade from "binary alarm" to "precise location of leakage and quantitative assessment of leakage severity". Based on the hierarchical alarm mechanism, it links with the industrial control system, providing a high-reliability guarantee for the safe operation of cooling towers with all-weather, low false alarm and closed-loop capability. Attached Figure Description

[0016] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart illustrating the implementation of a visual monitoring method for leakage in a cooling tower water collection tank provided by the present invention. Figure 2 This is a structural block diagram of a visual monitoring device for leakage in a cooling tower water collection tank provided in an embodiment of the present invention. Detailed Implementation

[0017] The core of this invention is to provide a visual monitoring method, device, equipment, and computer storage medium for leakage in cooling tower water collection tanks, which effectively solves the problem of false alarms in leakage under high humidity steam and water splashing environments, and realizes accurate identification, location, and graded early warning of leakage events under complex working conditions.

[0018] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely 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.

[0019] Please refer to Figure 1. Figure 1 The flowchart illustrates the implementation of a visual monitoring method for leakage in a cooling tower water collection tank provided by this invention; the specific operation steps are as follows: S101: Acquire video image sequences of the monitored area; S102: Perform image enhancement processing on the video image sequence, and extract the region of interest image based on the preset monitoring area mask; S103: Input the image of the region of interest into a pre-trained semantic segmentation model to extract a static semantic mask representing the distribution of water substances; S104: Calculate the dense optical flow field between consecutive frame images, and based on the fluid motion characteristics under gravity, filter out the dynamic optical flow region with directional continuous motion characteristics from the dense optical flow field; S105: The static semantic mask and the dynamic optical flow region are fused at the pixel level to obtain a leakage feature map, and the connected regions in the leakage feature map are tracked in a time sequence. The leakage event is determined based on the duration and cumulative displacement of the trajectory, and an alarm signal is generated.

[0020] Based on the above embodiments, in some embodiments, the image enhancement processing of the video image sequence includes: Based on the dark channel prior principle, the atmospheric light value and transmittance of the scene are estimated to restore the fog-free image and achieve defogging and anti-glare enhancement processing. The extraction of the region of interest image based on the preset monitoring area mask includes: Based on the preset division of high-risk monitoring areas and environmental interference areas, a binary mask is generated, and the pixel values ​​outside the binary mask in the video image sequence after image enhancement are forcibly set to zero to obtain the region of interest image.

[0021] In some embodiments, the "image enhancement processing" described in step S102 is intended to address the image degradation problem under conditions of high humidity steam and strong glare in cooling towers. The enhancement processing is based on the dark channel prior principle.

[0022] In other embodiments, in order to reduce the subsequent computational load while enhancing the image, the "extracting the region of interest image based on the preset monitoring area mask" in step S102 specifically includes: dividing the monitoring screen into a high-risk monitoring area and an environmental interference area using preset polygon coordinates, generating a binary mask, and removing environmental interference from the data source by forcibly setting the pixel values ​​outside the mask area to zero.

[0023] In one specific embodiment, step S102 specifically includes: First, receiving the original video frame as the input image I(x). The processing begins with the calculation of the dark channel map. Each pixel of the input image I(x) is traversed, and the minimum intensity value among the three RGB color channels is selected within its neighborhood window to construct a dark channel map reflecting the scene's fog concentration. Based on the dark channel prior principle, the global atmospheric light value A and the initial transmittance t(x) of the scene are estimated using this dark channel map, where the atmospheric light value A represents the background light intensity at infinity, and the initial transmittance t(x) characterizes the degree of light attenuation during transmission. Subsequently, a guided filtering algorithm is used to refine the initial transmittance map. Using the input image I(x) as the guide map, the initial transmittance t(x) is filtered and smoothed, removing noise while preserving edge details, resulting in a refined transmittance map. Finally, the fog-free image is recovered using an atmospheric scattering physical model. The previously estimated atmospheric light value A and the refined transmittance map are substituted into the recovery formula for pixel-level calculations to obtain a video image sequence enhanced with defogging and anti-glare. Specifically, the recovery process follows the formula... Where I(x) is the input hazy image, and J(x) is the recovered hazy-free image. To prevent the denominator from being too small and causing numerical overflow, and to retain a small amount of natural fog to maintain the realism of the image, a micro-fog threshold is set.

[0024] Specifically: After dehazing and enhancement, the system reads a pre-configured set of polygon coordinates and strictly divides the pixel space of the monitoring image into a high-risk monitoring area and an environmental interference area. The high-risk monitoring area includes the outer wall of the water collection tank, pipe welds, and expansion joints, while the environmental interference area includes the internal filling area and the water spray area. Based on the above coordinate division, the system generates a binary mask Mask_ROI with the same resolution as the input image. In this mask, the grayscale value of pixels belonging to the high-risk monitoring area is set to 255, and the grayscale value of pixels belonging to the environmental interference area and the background area is set to 0. Subsequently, the system performs a pixel-level masking operation, forcibly setting the pixel values ​​corresponding to all positions in the dehazing and enhanced video image sequence that are located at the position of the binary mask Mask_ROI value of 0 to zero, thus obtaining a region of interest image containing only the content of the high-risk monitoring area.

[0025] It should be noted that this implementation method effectively solves the image degradation problem caused by dense fog and strong glare interference in the high humidity environment of the cooling tower by introducing guided filtering to optimize transmittance estimation and combining it with physical scattering model to restore the image. At the same time, by introducing a binary mask shielding mechanism based on physical location at the front end of the algorithm, the path of normal water spray and splash water into the core detection logic is effectively blocked, and the false alarm rate is significantly reduced from the data source.

[0026] Based on the above embodiments, in some embodiments, the extraction of the static semantic mask characterizing the distribution of substances in the water body includes: A lightweight semantic segmentation network containing a multi-scale feature extraction module is used to output a probability map of each pixel belonging to a water body. Threshold filtering is performed on the probability map, and the filtered pixel set is used to generate the static semantic mask.

[0027] In other embodiments, the semantic segmentation network adopts a DeepLabV3+ architecture with MobileNetV3 as the backbone, and its ASPP module at the head is used to capture water features at different scales.

[0028] In one specific embodiment: Step S103 specifically includes: using the region of interest image output in step S102 as input data, and feeding it into the DeepLabV3+ semantic segmentation model with MobileNetV3 as the backbone network. This model utilizes the depthwise separable convolutional structure in MobileNetV3 to extract features from the image. The extracted feature map is passed to the ASPP module at the model head. This module constructs a multi-scale receptive field by setting dilated convolutions with different dilation rates in parallel, to simultaneously capture the local details of watermarks in fine cracks and the global contextual features of large water stains. The multi-scale features processed by the ASPP module undergo fusion and upsampling operations, outputting the probability value of each pixel belonging to the water body category, forming a probability map Pwater. Next, a threshold filtering operation is performed on the probability map Pwater, marking pixels with probability values ​​greater than a preset threshold as potential water body pixels. Further, a morphological opening operation is performed on the marked set of potential water body pixels, specifically using a sequence of operations of erosion followed by dilation, utilizing structuring elements to remove isolated noise points and smooth the boundaries of the water body region. Finally, the pixel set processed by morphological opening operation is used to generate a preliminary water semantic mask M_seg, which includes static water accumulation, dynamic water flow, and some reflective areas.

[0029] Specifically, during the model training phase, an industrial scene dataset containing categories such as "dry concrete", "wet water stains", "flowing water columns", "splashing water", and "metal reflection" was constructed. The weighted cross-entropy loss function was used as the optimization objective, and the loss weights for difficult-to-classify samples (such as faint water marks whose color is close to the background) were increased to force the network to focus on subtle features.

[0030] It should be noted that this implementation method significantly improves the recognition accuracy of water features of different shapes by adopting a lightweight network architecture and a multi-scale feature extraction mechanism while ensuring the real-time performance of edge computing devices. Combined with the post-processing strategy of threshold filtering and morphological opening operation, it effectively filters out high-frequency noise interference in industrial scenarios, ensuring that the generated static semantic mask has high confidence.

[0031] Based on the above embodiments, in some embodiments, calculating the dense optical flow field between consecutive frame images includes: A series of consecutive frames of images are selected as input data, and a dense optical flow field is calculated through an optical flow network. The dense optical flow field describes the instantaneous motion velocity of the pixels. The dynamic optical flow region with directional continuous motion characteristics selected from the dense optical flow field based on the fluid motion characteristics under gravity includes: The motion direction of the optical flow vector in the dense optical flow field is calculated, a directional constraint interval and a velocity modulus threshold matching the direction of gravity are set, and a set of pixels that simultaneously satisfy the directional constraint interval and the velocity modulus threshold are extracted as the dynamic optical flow region.

[0032] In other embodiments, a lightweight optical flow network is used to perform inference computation on a sequence of three consecutive image frames.

[0033] In one specific embodiment: step S104 specifically includes: based on the region of interest image sequence output in the previous stage, selecting a sequence of three temporally consecutive images. As input data, Indicates the current frame. and These are frames from the previous time step and two time steps prior, respectively. Inputting these three frames into a lightweight optical flow network deployed on an edge computing device, the output is a dense optical flow field. The vector components u and v at coordinates (x, y) precisely describe the instantaneous velocity of the pixel in the horizontal and vertical directions, respectively. Subsequently, a selection process is performed based on the fluid motion characteristics under gravity: traversing each pixel in the optical flow field, the motion direction angle of its optical flow vector is calculated. arctan2(v,u). Defines the direction constraint interval. To match the ideal physical direction of motion vertically downwards, while setting a velocity modulus threshold. This is used to eliminate static background noise and slowly spreading water accumulation interference. The system further traverses the dense optical flow field, extracts a set of pixels that simultaneously satisfy the directional constraint interval and the velocity modulus threshold, and marks this set of pixels as a directional motion region. And it is output as the final dynamic optical flow region.

[0034] Specifically, the directional constraint range can be set to 70° to 110° to allow for deviations caused by wall tilt.

[0035] It should be noted that this implementation method obtains high-precision dense motion vectors while ensuring real-time performance by introducing three-frame time-series input and lightweight network inference. Combined with the dual screening mechanism of gravity direction angle constraint and velocity threshold, it accurately distinguishes disordered splashing water from ordered leakage water flow at the physical level, significantly reducing the false alarm rate under complex working conditions.

[0036] Based on the above embodiments, in some embodiments, the step of fusing the static semantic mask and the dynamic optical flow region at the pixel level to obtain the leakage feature map includes: A pixel-level logical AND operation is performed on the static semantic mask and the dynamic optical flow region, so that only pixels that simultaneously possess water material properties and directional motion characteristics are retained in the leakage feature map.

[0037] In one specific embodiment: the fusion part in step S105 specifically includes: obtaining the static semantic mask M_seg output in step S103 and the dynamic optical flow set generated in step S104. As the input data source, M_seg represents the pixel distribution in the image that possesses water texture and color features, while This marks the set of moving pixels that meet the gravity direction constraint and whose velocity magnitude exceeds the threshold. The processing unit performs a pixel-by-pixel logical AND operation on these two binary matrices, that is, only if the pixel at coordinates (x,y) is within the range of M_seg and M_seg... Only when both are marked as valid is the location identified as a high-confidence leak point. This processing follows the formula. The resulting leakage feature map It effectively eliminates artifacts caused by stagnant water and interference from the downward movement of non-aquatic substances.

[0038] It should be noted that this implementation method significantly improves the accuracy of leakage detection through dual logical verification of static material properties and dynamic motion characteristics, effectively solving the problem of false alarms caused by water splashing or changes in light and shadow in complex industrial environments.

[0039] Based on the above embodiments, in some embodiments, the step of performing time-series trajectory tracking on the connected regions in the leakage feature map, and determining the leakage event based on the duration and cumulative displacement of the trajectory, includes: The connected regions in the leakage feature map are tracked across multiple frames using a filter, and the number of consecutive frames and cumulative displacement distance of each tracked target trajectory are counted. When the number of continuous frames exceeds a preset frame threshold and the cumulative displacement distance exceeds a preset distance threshold, the trajectory is confirmed to be a real physical fluid flow.

[0040] In other embodiments, the filter is a Kalman filter.

[0041] In one specific embodiment: the trajectory tracking and determination part in step S105 specifically includes: for the connected regions extracted from M_leak, starting the temporal trajectory tracking module, and using a Kalman filter to perform association matching on the same fluid flow target in multiple consecutive frames. In this process, the state vector of the filter is specifically configured as a five-dimensional array containing horizontal coordinate x, vertical coordinate y, horizontal velocity vx, vertical velocity vy, and area area. This allows the system to predict and correct the target's position in the next frame. Based on the tracking results, the system counts the number of consecutive frames for each target in real time. The cumulative displacement distance is obtained by summing the displacement vectors. Finally, the statistics obtained... With preset frame rate threshold (For example, 100 frames corresponding to a 4-second duration) are compared, and at the same time... and preset distance threshold (For example, a pixel distance of 30cm in physical space) is double-checked. Only when both conditions are met simultaneously is the connected area determined to be a real physical liquid flow and an alarm signal is triggered, thus logically eliminating transient interference caused by light flickering or instantaneous water droplet splashing.

[0042] It should be noted that this implementation method effectively eliminates transient interference caused by light flickering or instantaneous water droplet splashing through the timing smoothing mechanism of Kalman filtering, thus achieving accurate differentiation between minute leaks and continuous leaks.

[0043] Based on the above embodiments, in some embodiments, generating the alarm signal includes: The leakage severity index is calculated based on the area of ​​the confirmed leakage area and the average optical flow velocity. Different levels of alarm signals are generated based on the different ranges of the leakage severity index.

[0044] In other embodiments, the alarm signal includes a primary alarm signal and a secondary alarm signal, wherein the primary alarm signal corresponds to a minor leakage state and the secondary alarm signal corresponds to a severe leakage state.

[0045] In one specific embodiment, the step of generating an alarm signal specifically includes: calculating a leakage severity index based on the pixel area S of the confirmed leakage area and the average optical flow velocity V_mean, using the formula Index = α·S + β·V_mean, where α and β are weighting coefficients. When the leakage severity index is in the low range and the trajectory is in the form of thin lines, a level one alarm signal containing the text "Suspected leakage found in monitoring area X" is generated, and a yellow box is marked on the monitoring screen to alert maintenance personnel. When the leakage severity index exceeds a high threshold and the trajectory is in the form of wide bands or jets, a level two alarm signal containing the text "Severe leakage has occurred in monitoring area X" is generated, a red flashing box is marked on the monitoring screen, and the PLC control system is linked to issue a shutdown command or close the water supply valve to prevent the accident from escalating.

[0046] Specifically, the method further includes a visualization enhancement step: superimposing the calculated optical flow vector field onto the original video image in the form of a color flow field map. When this step is specifically executed, its input source is the dense optical flow field calculated in step S104. And the original video image sequence after dehazing and anti-glare enhancement processing output in step S102. The processing action first involves the mapping and transformation of the optical flow vector field. The system converts the motion velocity magnitude of each pixel in the optical flow field. with the direction angle of motion Mapped to a specific color space. Specifically, for regions that conform to the consistency constraint of the direction of gravity. And the velocity modulus exceeds the threshold. For each pixel, the system sets its hue value to the red band, and the saturation or brightness component of red varies with the velocity modulus. The optical flow vectors are linearly enhanced as the flow rate increases, creating a visual gradient effect where high-speed downward-flowing water appears deep red or bright red, while low-speed flow appears dark red. For background areas or random splashes that do not meet the above-mentioned directional continuous motion characteristics, their corresponding optical flow vectors are rendered as transparent or cool tones (such as blue or green) in the color flow field map to achieve significant visual distinction. Subsequently, the system uses the generated color flow field map as a foreground layer and blends it into the corresponding coordinate position of the original video image in a semi-transparent overlay using an alpha blending algorithm. This ensures that the texture details of the original scene are not completely obscured while highlighting areas of abnormal flow. The final output of this processing is an enhanced video frame with a dynamic color optical flow overlay effect, in which the leak point stands out from the complex background due to the obvious red downward-flowing stripes. By observing this enhanced video, monitoring personnel can intuitively identify the specific flow direction, relative speed, and precise physical location of the leak without relying on abstract data reports, thereby quickly confirming the authenticity of the alarm signal and assessing the urgency of the leak.

[0047] It should be noted that this implementation method achieves intelligent assessment of the severity of leakage through a multi-level alarm mechanism and enhanced visualization, providing maintenance personnel with an intuitive and reliable monitoring interface. This solution is based entirely on external visual observation, requires no system shutdown or welding during installation, has low maintenance costs, and eliminates the risk of sensor corrosion or scale buildup leading to sensor failure.

[0048] Please refer to Figure 2 , Figure 2 A structural block diagram of a visual monitoring device for leakage in a cooling tower water collection tank, provided in an embodiment of the present invention; the specific device may include: The image acquisition and preprocessing module 100 is used to acquire a video image sequence of the monitored area, perform image enhancement processing on the video image sequence, and extract the region of interest image based on a preset monitoring area mask. The static feature extraction module 200 is used to input the image of the region of interest into a pre-trained semantic segmentation model to extract a static semantic mask that represents the distribution of water substances. The dynamic feature extraction module 300 is used to calculate the dense optical flow field between consecutive frame images, and based on the fluid motion characteristics under gravity, to filter out dynamic optical flow regions with directional continuous motion characteristics from the dense optical flow field. The fusion and determination module 400 is used to perform pixel-level fusion of the static semantic mask and the dynamic optical flow region to obtain a leakage feature map, and to perform temporal trajectory tracking on the connected regions in the leakage feature map, and to determine the leakage event and generate an alarm signal based on the duration and cumulative displacement of the trajectory.

[0049] In one specific embodiment: This device is deployed in a three-layer "end-edge-cloud" architecture. The front-end perception layer includes an explosion-proof network camera (with IP67 protection rating, equipped with a self-cleaning wiper and infrared supplementary lighting or low-light starlight level sensor), the edge computing layer includes an AI edge computing box, and the central application layer includes a central control room server or a cloud-based IoT platform. The image acquisition and preprocessing module, static feature extraction module, dynamic feature extraction module, and fusion and judgment module are deployed in the edge computing layer.

[0050] In another specific embodiment, the fusion and judgment module is further used to: when a secondary alarm signal is generated, link the PLC control system to issue a stop command or close the water supply valve. The device also includes a visualization enhancement module, used to overlay the calculated optical flow vector field as a color flow field map onto the original video image, where red is used to represent high-speed downward water flow, allowing monitoring personnel to intuitively see the direction and speed of the water flow to enhance the credibility of the alarm.

[0051] The cooling tower water collection tank leakage visual monitoring device of this embodiment is used to implement the aforementioned cooling tower water collection tank leakage visual monitoring method. Therefore, the specific implementation of the cooling tower water collection tank leakage visual monitoring device can be found in the embodiment section of the cooling tower water collection tank leakage visual monitoring method above. For example, the image acquisition and preprocessing module 100, the static feature extraction module 200, the dynamic feature extraction module 300, and the fusion and judgment module 400 are respectively used to implement steps S101-S105 in the above-mentioned cooling tower water collection tank leakage visual monitoring method. Therefore, its specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0052] It should be further explained that the terms involved in this invention are defined as follows: "Dark channel prior principle" refers to the statistical law that in most non-sky local regions, the intensity value of at least one color channel is very low; "Guided filtering" is an edge-preserving filtering algorithm that uses a guide image to filter the input image; "Morphological opening operation" refers to an image processing operation sequence of erosion followed by dilation; "Dense optical flow field" refers to an optical flow representation method that calculates a motion vector for each pixel in the image; "ASPP module" is the dilated spatial pyramid pooling module, which captures multi-scale contextual information through dilated convolution with different dilation rates; "Kalman filter" is a recursive algorithm that uses the state equation of a linear system to make an optimal estimate of the system state through observation data.

[0053] A specific embodiment of the present invention also provides a visual monitoring device for leakage in a cooling tower water collection tank, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the above-described visual monitoring method for leakage in a cooling tower water collection tank.

[0054] A specific embodiment of the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described visual monitoring method for leakage in a cooling tower water collection tank.

[0055] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0056] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0057] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0058] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A visual monitoring method for leakage in a cooling tower water collection tank, characterized in that, Includes the following steps: Acquire video image sequences of the monitored area; The video image sequence is subjected to image enhancement processing, and the region of interest image is extracted based on a preset monitoring area mask; The image of the region of interest is input into a pre-trained semantic segmentation model to extract a static semantic mask representing the distribution of substances in the water body; Calculate the dense optical flow field between consecutive frames of images, and based on the fluid motion characteristics under gravity, filter out dynamic optical flow regions with directional continuous motion characteristics from the dense optical flow field; The static semantic mask and the dynamic optical flow region are fused at the pixel level to obtain a leakage feature map. The connected regions in the leakage feature map are tracked in a temporal sequence. The leakage event is determined based on the duration and cumulative displacement of the trajectory, and an alarm signal is generated.

2. The method according to claim 1, characterized in that, The image enhancement processing of the video image sequence includes: Based on the dark channel prior principle, the atmospheric light value and transmittance of the scene are estimated to restore the fog-free image and achieve defogging and anti-glare enhancement processing. The extraction of the region of interest image based on the preset monitoring area mask includes: Based on the preset division of high-risk monitoring areas and environmental interference areas, a binary mask is generated, and the pixel values ​​outside the binary mask in the video image sequence after image enhancement are forcibly set to zero to obtain the region of interest image.

3. The method according to claim 1, characterized in that, The static semantic mask for extracting and characterizing the distribution of substances in water bodies includes: A lightweight semantic segmentation network containing a multi-scale feature extraction module is used to output a probability map of each pixel belonging to a water body. Threshold filtering is performed on the probability map, and the filtered pixel set is used to generate the static semantic mask.

4. The method according to claim 1, characterized in that, The calculation of the dense optical flow field between consecutive frames of images includes: A series of consecutive frames of images are selected as input data, and a dense optical flow field is calculated through an optical flow network. The dense optical flow field describes the instantaneous motion velocity of the pixels. The dynamic optical flow region with directional continuous motion characteristics selected from the dense optical flow field based on the fluid motion characteristics under gravity includes: The motion direction of the optical flow vector in the dense optical flow field is calculated, a directional constraint interval and a velocity modulus threshold matching the direction of gravity are set, and a set of pixels that simultaneously satisfy the directional constraint interval and the velocity modulus threshold is extracted as the dynamic optical flow region.

5. The method according to claim 1, characterized in that, The step of fusing the static semantic mask with the dynamic optical flow region at the pixel level to obtain the leakage feature map includes: A pixel-level logical AND operation is performed on the static semantic mask and the dynamic optical flow region, so that only pixels that simultaneously possess water material properties and directional motion characteristics are retained in the leakage feature map.

6. The method according to claim 1, characterized in that, The step of performing time-series trajectory tracking on the connected regions in the leakage feature map, and determining the leakage event based on the duration and cumulative displacement of the trajectory, includes: The connected regions in the leakage feature map are tracked across multiple frames using a filter, and the number of consecutive frames and cumulative displacement distance of each tracked target trajectory are counted. When the number of continuous frames exceeds a preset frame threshold and the cumulative displacement distance exceeds a preset distance threshold, the trajectory is confirmed to be a real physical fluid flow.

7. The method according to claim 6, characterized in that, The generated alarm signal includes: The leakage severity index is calculated based on the area of ​​the confirmed leakage area and the average optical flow velocity. Different levels of alarm signals are generated based on the different ranges of the leakage severity index.

8. A visual monitoring device for leakage in a cooling tower water collection tank, characterized in that, include: The image acquisition and preprocessing module is used to acquire video image sequences of the monitored area, perform image enhancement processing on the video image sequences, and extract the region of interest image based on a preset monitoring area mask. The static feature extraction module is used to input the image of the region of interest into a pre-trained semantic segmentation model to extract a static semantic mask that represents the distribution of water substances. The dynamic feature extraction module is used to calculate the dense optical flow field between consecutive frames of images, and based on the fluid motion characteristics under the action of gravity, to filter out dynamic optical flow regions with directional continuous motion characteristics from the dense optical flow field. The fusion and determination module is used to perform pixel-level fusion of the static semantic mask and the dynamic optical flow region to obtain a leakage feature map, and to perform temporal trajectory tracking on the connected regions in the leakage feature map. Based on the duration and cumulative displacement of the trajectory, the module determines the leakage event and generates an alarm signal.

9. A visual monitoring device for leakage in a cooling tower water collection tank, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the visual monitoring method for leakage in a cooling tower water collection tank as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the visual monitoring method for leakage in a cooling tower water collection tank as described in any one of claims 1 to 7.

Images