A reservoir spillway water passing monitoring method and system based on image recognition

Abstract

The application provides a reservoir spillway water monitoring method and system based on image recognition, and relates to the technical field of earth and rockfill dam water monitoring, wherein the method comprises the following steps: synchronously collecting a sequence of mid-wave infrared polarized images and a sequence of long-wave infrared polarized images of the connecting part of the dam body and the concrete spillway of the earth and rockfill dam; calculating the polarized physical characteristics of the sequence of mid-wave infrared polarized images and the sequence of long-wave infrared polarized images, and performing multi-band polarized feature fusion to generate an enhanced fused image sequence, so as to suppress water mist interference and highlight water flow and structure boundaries; extracting kinetic parameters for characterizing the flow pattern of the spillway water flow based on the generated fused image sequence; constructing a scouring risk index based on the extracted kinetic parameters, and performing graded early warning according to the scouring risk index. The method of the application can realize the effect of real-time, active and graded early warning of the scouring risk of the connecting part under the poor visual environment of the earth and rockfill dam spillway water.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of earth-rock dam water monitoring, and in particular to a reservoir spillway water monitoring method and system based on image recognition. BACKGROUND

[0002] Currently, for the overflow of the spillway of the earth-rock dam, there are some specific problems different from the concrete dam, and the core root is that the dam body of the earth-rock dam is earth-rock filling, and the spillway is generally poured with concrete (to meet the long-term flood discharge demand), and the two materials are directly connected together. The fundamental difference in structure (relative to the concrete dam) brings unique risks and monitoring difficulties.

[0003] Among them, the specific problems mainly include: seepage and bypass at the connection (the contact surface is easy to become a seepage channel, causing seepage deformation, and even causing the dam body and the spillway to separate), structural interaction (uneven settlement may cause the spillway bottom plate to crack, water stop to fail, and seepage path to change), and particularity during overflow (high-speed water flow erodes the connection part of the dam body and the spillway, which may directly wash away the earth-rock dam slope, especially in the case of line-of-sight obstruction, low visibility during rainfall, or strong atomization phenomenon during flood discharge, resulting in low visibility, so that the existing visible light camera cannot clearly capture the contact flow state of the connection part. Therefore, the monitoring method mainly relies on manual inspection or post-flood topographic survey, which cannot realize early warning, and the response is lagging, which may cause dam failure).

[0004] For the above specific problems, the existing solutions include: using a pressure pipe laser tracking water level monitoring technology to address the seepage and bypass at the connection (which is a non-contact measurement that can avoid sensor failure due to water quality problems, and the accuracy can reach ±1mm. In addition, there is an air-ground patrol sensing technology, which can quickly detect abnormal seepage areas around the dam body and the spillway by using a UAV carrying infrared thermal imaging equipment), and using displacement and settlement monitoring points to deal with structural deformation.

[0005] However, for the particularity of the overflow, the current field is still blank. SUMMARY

[0006] In order to solve the problems in the related art, the present application provides a reservoir spillway water monitoring method and system based on image recognition.

[0007] In order to achieve the above purpose, the technical scheme adopted by the present application includes:

[0008] According to a first aspect of the present application, a reservoir spillway water monitoring method based on image recognition is provided, comprising the following steps:

[0009] Step S1: synchronously collect a sequence of mid-wave infrared polarized images and a sequence of long-wave infrared polarized images at a connection part between a dam body and a concrete spillway of an earth-rock dam;

[0010] Step S2: based on the sequence of mid-wave infrared polarized images and the sequence of long-wave infrared polarized images, calculate polarized physical characteristics thereof, and perform multi-band polarized feature fusion to generate an enhanced fused image sequence, so as to suppress water mist interference and highlight water flow and structure boundaries;

[0011] Step S3: based on the generated fused image sequence, extract a kinetic parameter for representing a flow pattern of water flow of the spillway;

[0012] Step S4: based on the extracted kinetic parameter, construct a scouring risk index, and perform graded early warning according to the scouring risk index.

[0013] Optionally, in the step S1, a synchronous controller is used to control the mid-wave infrared polarized camera and the long-wave infrared polarized camera to perform synchronous image collection, so as to obtain the sequence of mid-wave infrared polarized images and the sequence of long-wave infrared polarized images that are aligned in time.

[0014] Optionally, in the step S2, the calculation of the polarized physical characteristics specifically includes:

[0015] For each band of images, based on images at at least three different polarized angles, a Stokes vector of each pixel is calculated;

[0016] and based on the Stokes vector, a degree of polarization and a polarization angle of each pixel are further calculated.

[0017] Optionally, in the step S2, the process of generating the enhanced fused image sequence includes:

[0018] normalizing and weighting the degree of polarization and the light intensity of the mid-wave infrared polarized images, and normalizing and weighting the degree of polarization, the polarization angle and the light intensity of the long-wave infrared polarized images, and fusing the two to generate the fused image sequence;

[0019] wherein, preset weight coefficients are used in the fusion process, and a difference between the polarization angle and a reference polarization angle is compensated.

[0020] Optionally, in the step S3, the kinetic parameter is extracted by using an improved polarized constraint optical flow method, and an image optical flow field is calculated by constructing an optical flow energy function containing a degree of polarization constraint.

[0021] Optionally, after the image optical flow field is obtained, a scale factor obtained through camera calibration is used to convert an image velocity into a physical velocity field; and based on the physical velocity field, an average flow velocity, a flow velocity standard deviation and a vorticity average value are calculated as the kinetic parameter.

[0022] Optionally, the kinetic parameters extracted in the step S3 further include: a temperature gradient modulus from the long-wave infrared polarization image and a polarization degree difference between the current frame and the background frame.

[0023] Optionally, in the step S4, the process of constructing the flushing risk index includes:

[0024] Step S4-1: constructing a feature vector containing the kinetic parameters;

[0025] Step S4-2: based on historical data, normalizing each component of the feature vector;

[0026] Step S4-3: weighting and combining the normalized feature vector with the weight vector obtained by machine learning training to obtain the flushing risk index.

[0027] Optionally, the graded early warning in the step S4 is specifically:

[0028] presetting a first threshold value and a second threshold value;

[0029] when the flushing risk index is between the first threshold value and the second threshold value, issuing a first-level early warning information;

[0030] when the flushing risk index exceeds the second threshold value, issuing a second-level early warning information.

[0031] According to the second aspect of the present application, a reservoir spillway water monitoring system based on image recognition is also provided, which is used to implement the reservoir spillway water monitoring method based on image recognition in any of the technical solutions of the first aspect of the present application, and includes an image acquisition module, a data processing module and an early warning output module.

[0032] The image acquisition module includes at least one set of multi-band infrared polarization image acquisition unit deployed around the spillway, and the acquisition unit includes a mid-wave infrared polarization camera, a long-wave infrared polarization camera and a synchronization controller for controlling the synchronous acquisition of the two.

[0033] The data processing module is configured to: receive and store image data from the image acquisition module; based on the mid-wave infrared polarization image sequence and the long-wave infrared polarization image sequence, calculate the polarization physical characteristics thereof, perform multi-band polarization feature fusion, generate an enhanced fusion image sequence to suppress water mist interference and highlight the water flow and structure boundary; based on the generated fusion image sequence, extract kinetic parameters for representing the flow pattern of the spillway water flow; based on the extracted kinetic parameters, construct a flushing risk index, and perform graded early warning according to the flushing risk index.

[0034] The pre-warning output module is configured to send corresponding pre-warning information to a monitoring center or a user terminal according to the judgment result of the data processing module. Advantages

[0035] 1. By the technical scheme, first, the problem of imaging in a low-visibility environment can be effectively solved. Specifically, the infrared band has strong penetration ability for suspended water droplet particles such as rain and fog. Thus, under the low-visibility conditions of the rain visual obstruction and floodwater fog that cause visible light to fail as indicated in the background technology, the method of the present application can still obtain image information of the connecting part of the dam body and the spillway.

[0036] Second, the image quality can be effectively improved, laying a foundation for subsequent accurate analysis. Specifically, step S2 directly produces two levels of effects, i.e., suppressing water mist interference and highlighting water flow and structure boundaries, by calculating the polarization physical characteristics and performing multi-band polarization feature fusion. As for suppressing water mist interference, polarization information is sensitive to non-polarized and chaotic scattered ambient light (such as the scattering of fog-like water droplets), and the image signal-to-noise ratio can be effectively improved by polarization feature processing. As for highlighting water flow and structure boundaries, there is a difference in polarization characteristics between the reflected light of the water flow surface and the reflected light of the dam body and concrete structure. By utilizing this difference for feature fusion, the boundaries of the water flow and the washed structure can be made more clear and sharp in the image, so that they are more easily identified and tracked.

[0037] Third, the quantitative perception of the water flow state can be effectively realized, and the image information can be converted into physical parameters. Specifically, step S3 extracts dynamic parameters for representing the flow regime of the water flow based on the fused high-quality image sequence. This step improves the visual image information to quantifiable data representing the motion state of the water flow, thereby providing data support for objective risk assessment.

[0038] Fourth, step S4 constructs a scouring risk index based on the extracted dynamic parameters and performs graded pre-warning. In this way, the quantitative data obtained in the foregoing steps are directly associated with the most core security threat (scouring risk) of the connecting part of the spillway, so that the monitoring mode is changed from post-revision (damage found by manual inspection) or periodic measurement to a real-time and risk threshold-based active pre-warning mechanism.

[0039] Overall, the technical scheme of the present application realizes a technical effect of real-time, active and graded pre-warning of the scouring risk of the connecting part under the poor visual environment of the spillway of the earth-rock dam through the progressive technical chain of step S1 (ensuring visibility) → step S2 (ensuring clarity) → step S3 (ensuring quantification) → step S4 (ensuring pre-warning).

[0040] 2, Other benefits or advantages of the present application will be described in detail in the specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiment description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0042] Among them:

[0043] Figure 1 is a step flow schematic diagram of the image recognition-based reservoir spillway water monitoring method provided by an exemplary embodiment of the present application;

[0044] Figure 2 is a schematic diagram of the actual arrangement of the image recognition-based reservoir spillway water monitoring system in the earth-rock dam provided by an exemplary embodiment of the present application;

[0045] Figure 3 is a schematic diagram of the arrangement of the image recognition-based reservoir spillway water monitoring system provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to make the technical personnel of the related art have a clearer and more accurate understanding of the technical solutions of the present application, the prior art and the technical problems thereof will be described in more detail below.

[0047] As one of the most widely used dam types in the world, the safety operation of the earth-rock dam is of great importance. The spillway, as the most important flood discharge structure of the dam, the safety of the connection part between the spillway and the dam body is the key to flood control and safety. Unlike the concrete dam whose whole body is poured with concrete, the dam body of the earth-rock dam is filled with earth and stone materials, and the spillway is usually made of concrete structure to meet the requirements of long-term flood discharge. The physical connection between the "earth and stone materials" and the "rigid concrete structure" constitutes a unique structural weak link of the earth-rock dam, and thus derives a unique spillway safety monitoring problem that is different from the concrete dam.

[0048] These specific problems mainly focus on three aspects:

[0049] First, seepage and around seepage problems at the connection. The contact surface between the earth-rock dam body and the concrete spillway side wall and bottom plate is prone to form a preferential seepage channel due to the great difference in material properties. Under the action of reservoir water pressure, water seeps along the contact surface, which easily causes seepage deformation such as contact scouring and piping, and in severe cases, may lead to the loss of dam filling, causing the dam body and the spillway structure to separate, forming a safety hazard.

[0050] Second, structural interaction problems. The earth-rock dam body will produce continuous settlement and deformation under the action of self-weight and water level change, while the concrete spillway as a rigid structure has a much smaller settlement than the dam body. This uneven settlement will cause a large differential deformation at the joint, which may cause the spillway bottom plate to crack and the water stop system to fail, thereby changing the original seepage and drainage path and exacerbating the risk of seepage damage.

[0051] Third, the particularity problem during flow. This is the biggest challenge currently faced by the monitoring field and the core problem that the present application aims to solve. Its particularity is reflected in:

[0052] Directness of water flow dynamics: During flood discharge, high-speed water flow directly scours and erodes the connection between the dam body and the spillway. Especially when the water flow is large, its huge kinetic energy will quickly strip away and carry away the earth-rock material of the dam body.

[0053] Monitoring failure in low-visibility environment: This problem is particularly prominent during heavy rain or flood discharge. Specifically, during rainfall, the rain curtain and low light conditions will severely obstruct the line of sight of the visible light camera; and during flood discharge, the impact of high-speed water flow will produce a diffuse and intense water mist (atomization phenomenon), forming a "white barrier". Both of these situations result in the complete failure of traditional visible light-based video monitoring systems, which cannot capture the water flow pattern, vortex formation, and subtle signs of initial scouring at the connection site.

[0054] For example, after a period of heavy rain, the reservoir begins to discharge water through the spillway. Due to the intense atomization phenomenon, the visible light camera at the monitoring center shows a blurred picture, and the connection between the spillway and the left bank soil slope cannot be seen. At this time, there is a small weak point at the connection site due to structural differences. High-speed water flow forms a vortex at this point, starting to slowly and continuously scour the dam body. Without any effective real-time monitoring means, this dangerous situation cannot be detected in the early stages. After several hours, when the rain subsides or the discharge volume decreases, the water mist dissipates, and the inspection personnel may be able to observe the part of the dam body that has been scoured out with the naked eye, but by then the best opportunity for rescue may have been missed. If the scouring continues to develop, it is likely to cause the slope to collapse, thereby endangering the entire spillway structure and even causing a dam breach accident, with disastrous consequences.

[0055] For the above problems, the current industry solutions have obvious limitations. For seepage and around seepage problems, non-contact seepage monitoring techniques such as pressure tube laser tracking water level monitoring or unmanned aerial vehicles carrying infrared thermal imaging equipment for air-ground patrol to find abnormal seepage areas can be used. For structural deformation, displacement and settlement monitoring points are mainly relied on.

[0056] However, for overcurrent, especially in low visibility conditions, the connecting part is directly at risk of washing out, and there is currently a lack of effective real-time monitoring and early warning means. The existing methods mainly rely on manual inspection or topographic survey after flood discharge, which is a "post-check" mode with serious response lag and cannot achieve early warning and active prevention of dangerous situations.

[0057] Therefore, developing a technology that can monitor the flow pattern and washing risk of the connecting part of the spillway and the earth and rock dam in real time and accurately in low visibility environments such as rain and fog has become a technical gap that needs to be solved in the field.

[0058] Therefore, the present application provides a brand new solution, i.e. the reservoir spillway overwater monitoring method based on image recognition of the present application, the technical concept of the present application is: by synchronously collecting the mid-wave and long-wave infrared polarization images of the connecting part of the dam body and the spillway, calculating and fusing the polarization physical characteristics (such as polarization degree, polarization angle), generating a fused image sequence that can effectively suppress water mist interference and enhance the water flow boundary features; further, based on this fused image, the improved polarization constrained optical flow method is used to accurately extract the kinetic parameters of the water flow, and a comprehensive washing risk index is constructed, and finally the real-time and accurate grading warning of the washing risk of the connecting part is realized, thereby filling the gap of the prior art in active early warning in poor visibility.

[0059] The technical solutions of the present application will be described in detail below with reference to the accompanying drawings.

[0060] As shown in Figure 1 According to the first aspect of the present application, a reservoir spillway overwater monitoring method based on image recognition is provided, comprising the following steps:

[0061] Step S1: synchronously collecting mid-wave infrared polarization image sequences and long-wave infrared polarization image sequences of the connecting part of the dam body and the concrete spillway of the earth and rock dam;

[0062] Step S2: based on the mid-wave infrared polarization image sequences and the long-wave infrared polarization image sequences, calculating the polarization physical characteristics and performing multi-band polarization feature fusion to generate enhanced fused image sequences to suppress water mist interference and highlight the water flow and structure boundary;

[0063] Step S3: based on the generated fused image sequence, extracting kinetic parameters for characterizing the flow pattern of the spillway water flow;

[0064] Step S4: based on the extracted kinetic parameters, construct the washing risk index, and conduct graded early warning according to the washing risk index.

[0065] Through the above technical scheme, first, the problem of imaging in low visibility environment can be effectively solved. Specifically, the infrared band has strong penetration ability to suspended water droplet particles such as rain and fog. Thus, under the low visibility conditions such as rain visual obstruction and flood water mist that cause visible light failure as indicated in the background technology, the method of the present application can still obtain image information of the dam and spillway connection part.

[0066] Second, the image quality can be effectively improved, laying a foundation for subsequent accurate analysis. Specifically, step S2 directly produces two levels of effects by calculating the polarization physical characteristics and performing multi-band polarization feature fusion, namely suppressing water mist interference and highlighting water flow and structure boundary. As for suppressing water mist interference, polarization information is sensitive to non-polarized and chaotic scattered ambient light (such as the scattering of fog-like water droplets), and through polarization feature processing, these background noises can be effectively suppressed to improve the image signal-to-noise ratio. As for highlighting the water flow and structure boundary, the reflected light of the water flow surface and the reflected light of the dam and concrete structure differ in polarization characteristics. By utilizing this difference for feature fusion, the boundary between the water flow and the washed structure can be made more clear and sharp in the image, so that it is easier to be identified and tracked.

[0067] Third, the quantified perception of the water flow state can be effectively realized, and the image information can be converted into physical parameters. Specifically, step S3 extracts kinetic parameters for representing the flow regime based on the fused high-quality image sequence. This step improves the visual image information to quantifiable data representing the water flow motion state, thereby providing data support for objective risk assessment.

[0068] Fourth, step S4 constructs the washing risk index based on the extracted kinetic parameters and conducts graded early warning. In this way, the quantified data obtained in the foregoing steps can be directly associated with the most core security threat (washing risk) of the spillway connection part, so that the monitoring mode is changed from post-recheck (damage found by manual inspection) or periodic measurement to a real-time and risk threshold-based active early warning mechanism.

[0069] Overall, the technical scheme of the present application realizes a real-time, active and graded early warning of the washing risk of the connection part under the poor visual environment of the spillway of the earth-rock dam through the progressive technical chain of step S1 (ensure visibility) → step S2 (ensure clarity) → step S3 (ensure quantification) → step S4 (ensure early warning).

[0070] The technical solutions of the present application are further described below in combination with an exemplary embodiment.

[0071] Project example: implementation process of a reservoir spillway water monitoring method based on image recognition

[0072] This embodiment is applied to a right side slope spillway of a certain earth-rock dam. On the solid observation pier on one side of the spillway, a set of image acquisition modules is deployed, which includes a medium wave infrared polarization camera (waveband 3-5 μm), a long wave infrared polarization camera (waveband 8-12 μm), and a synchronous controller.

[0073] Step 1, synchronously acquiring image sequences

[0074] During the flood discharge, the synchronous controller synchronously triggers the two cameras to acquire images at a frame rate of 30 Hz, obtaining a medium wave infrared polarization image sequence and a long wave infrared polarization image sequence that are accurately aligned in time. Each camera acquires images at three polarization angles of 0°, 45°, and 90° at t1, t2, and t3 through rotating the built-in polarizer, respectively, to form a complete polarization image group.

[0075] In this embodiment, by fusing the features of the medium wave and long wave infrared polarization images, it can effectively ensure that each frame in the fused image sequence is a true and consistent description of the same physical moment, ensuring the physical authenticity of multi-modal information fusion.

[0076] Step 2, calculating polarization features and image fusion

[0077] Step 2.1, calculating polarization physical features

[0078] For each pixel point , according to the formulas , and , the Stokes vector and polarization parameters of each waveband are calculated, respectively.

[0079] Taking a pixel in the long wave band as an example, the light intensity values at three polarization angles are: =150 DN, =100 DN, =50 DN (DN, Digital Number, digital value).

[0080] Calculate the Stokes vector:

[0081] =150 DN+50 DN=200 DN;

[0082] =150DN-50DN=100DN;

[0083] =2*100DN-200DN=0DN;

[0084] Calculate the degree of polarization DoP and the polarization angle AoP:

[0085] =0.5;

[0086] =0 rad;

[0087] Similarly, all polarization parameters of this pixel in the mid-wave band can be calculated. , wait.

[0088] Step 2.2: Multi-band polarization feature fusion

[0089] According to the formula Generate an enhanced fused image sequence .

[0090] Setting parameters: , ;

[0091] And set a reference value: =1.0 (dimensionless), =4095DN, =1.0 (dimensionless), =4095DN, =0rad (assuming the polarization angle is 0 when water surface reflection is dominant).

[0092] Using the pixel values ​​from the previous step, and assuming that... =3000DN, =0.8 (dimensionless).

[0093] Then, the medium wave component=0.6*(0.8 / 1.0)*(3000 / 4095)≈0.6*0.8*0.732≈0.351;

[0094] Long-wavelength component = 0.4*(0.5 / 1.0)*((π / 2-|0-0|) / (π / 2))*(150 / 4095) = 0.4*0.5*1*0.037≈0.007 (dimensionless);

[0095] The grayscale value of this pixel in the merged image is: =0.351+0.007=0.358 (normalized dimensionless value, which can be mapped to 0-255 or 0-4095 for display).

[0096] This fusion process suppresses the water mist interference (as the water mist scatters light with low DoP, which is weakened in the fusion), and highlights the water flow and structure boundary (as the difference in polarization properties between water flow and concrete is enhanced in the fusion image).

[0097] This step 2 is the core of the whole system. The original, basic infrared intensity image is converted into a feature space that can reveal the intrinsic properties of the objects through a specific physical information extraction process. This is because the original infrared image only records the total intensity of the infrared light emitted or reflected by the object (a scalar field). The processing defined in this step extends each pixel from a single intensity value to a set of physical features that can describe the polarization state of the light wave. This up-conversion from one-dimensional information to multi-dimensional features can greatly enrich the information capacity of the image, making it possible to distinguish objects with similar gray levels but different materials in ordinary infrared images (such as water mist, water flow, concrete surface, and soil). At the same time, water mist, water flow, and dam structure may have blurred boundaries in ordinary infrared images due to similar temperatures. By extracting polarization physical features, the system can distinguish different materials by utilizing the differences in their polarization modulation characteristics of infrared light. For example, the specular reflection of a smooth water surface has strong polarization characteristics, while the scattering light inside the water body or the diffuse scattering light of the water mist has very low polarization. This sensitivity to surface properties and scattering characteristics enables the processing in this step to effectively highlight the water flow and structure boundary and suppress the water mist interference, which is a prerequisite for reliable monitoring.

[0098] In addition, the polarization physical features calculated in this step are a set of parameters with clear physical definitions. They are less affected by absolute intensity errors such as environmental light changes and sensor non-uniformity than the original light intensity, and the features are more stable. This provides a standardized, comparable, and physically meaningful input for the multi-band feature fusion performed in the following steps. If fusion is performed directly on the original light intensity image, its effectiveness will be severely dependent on the instantaneous temperature and emissivity of the scene, and the robustness will be poor.

[0099] Step 3, extract dynamic parameters

[0100] Step 3.1, calculate physical velocity field using improved polarization-constrained optical flow method

[0101] Using the improved polarization-constrained optical flow method, the image optical flow field (u, v) is calculated by minimizing the energy function E(u, v):

[0102]

[0103] Set parameters: = 0.5 (dimensionless), = 0.01 (pixel local average polarization degree ), = 0.6.

[0104] By solving Euler-Lagrange equation iteratively, the optical flow velocity of this pixel is finally obtained as u = 2.5 pixel / frame, v = 1.0 pixel / frame.

[0105] By camera calibration, the scale factor is known as = 0.05 m / pixel.

[0106] The physical velocity of this point is then:

[0107] = 2.5 * 0.05 = 0.125 m / s;

[0108] = 1.0 * 0.05 = 0.05 m / s;

[0109] Step 3.2, calculate the kinetic parameters

[0110] The physical velocity field (U, V) of the whole connection region is calculated, and the results are:

[0111] The average flow velocity = 2.1 m / s, the flow velocity standard deviation = 0.8 m / s (reflecting the uniformity of flow velocity distribution), the average vorticity (reflecting the vortex intensity);

[0112] At the same time, the temperature gradient modulus from the long-wave infrared polarization image and the polarization degree difference between the current frame and the background frame are extracted:

[0113] Temperature gradient modulus : calculated from the long-wave infrared polarization image (essentially a temperature field), the average value of this region is 3.5 °C / pixel.

[0114] Polarization degree difference : calculate the regional average value of the absolute difference between the polarization degree of the current image and the polarization degree of the pre-stored water-free background image , and get = 0.25 (dimensionless).

[0115] The feature vector is obtained as: = [2.1 m / s, 0.8 m / s, , 3.5 °C / pixel, 0.25].

[0116] Step 4, Construction of the Brushing Risk Index and Graded Early Warning

[0117] Step 4.1: Feature Standardization

[0118] According to the formula , the mean u = [1.5, 0.3, 0.05, 1.0, 0.1] and the standard deviation σ = [0.5, 0.2, 0.05, 0.5, 0.05] of each feature in the historical normal flow data are called.

[0119] Standardize :

[0120] = (2.1-1.5) / 0.5 = 1.2 (dimensionless);

[0121] = (0.8-0.3) / 0.2 = 2.5 (dimensionless);

[0122] = (0.15-0.05) / 0.05 = 2.0 (dimensionless);

[0123] = (3.5-1.0) / 0.5 = 5.0 (dimensionless);

[0124] = (0.25-0.1) / 0.05 = 3.0 (dimensionless);

[0125] Get the standard feature vector: = [1.2, 2.5, 2.0, 5.0, 3.0] (dimensionless).

[0126] Step 4.2, Calculate the Brushing Risk Index R

[0127] According to the weight vector W = [0.2, 0.25, 0.25, 0.2, 0.1] trained by historical catastrophic data, and according to the formula , the brushing risk index is calculated.

[0128] = 0.2*1.2 + 0.25*2.5 + 0.25*2.0 + 0.2*5.0 + 0.1*3.0 = 0.24 + 0.625 + 0.5 + 1.0 + 0.3 = 2.665.

[0129] Step 4.3: Graded Early Warning

[0130] The first threshold = 1.5 and the second threshold = 3.0.

[0131] Since R = 2.665, the following is satisfied Therefore, the early warning output module sends the first level early warning information to the monitoring center, prompting that "the overflow spillway and the dam body connection part appears the risk sign of washing, please pay more attention".

[0132] In the above embodiment, it should be noted that,

[0133] First, for the calculation formula of the Stokes vector , it is a mathematical tool for completely describing the polarization state of a light beam. It avoids using phase information and directly uses four measurable light intensity values to represent polarized light, which is very suitable for imaging applications. Among them, represents the total light intensity of the light beam. It is calculated from the sum of the light intensity of the two orthogonal polarization directions and , which reflects the total energy of the infrared radiation received by the pixel point. describes the energy difference between the horizontal (0°) and vertical (90°) polarization directions in the light wave. represents that the horizontal polarization is dominant; represents that the vertical polarization is dominant; represents that the energy in the two directions is balanced. describes the energy difference between the 45° and 135° polarization directions in the light wave. It captures the polarization information at an angle of 45 degrees to the axis described by S1. Through three images with different polarization angles , , , this formula encapsulates the polarization information of each pixel point into a vector, laying a foundation for subsequent calculation of more intuitive polarization parameters.

[0134] Second, for the calculation formula of the degree of polarization , the purpose is to quantify the degree of polarization of the light beam, that is, what proportion of the light is completely polarized. Among them, the numerator represents the intensity of the completely polarized light part, and the denominator is the total light intensity. The value range of DoP is 0 to 1, and DoP is approximately equal to 1 indicates that the light is almost completely polarized (such as mirror reflection of smooth surface), and DoP is approximately equal to 0 indicates that the light is almost non-polarized (such as diffuse reflection from rough surface or body scattering of water mist), in the present application, DoP is a key feature, high DoP area usually corresponds to water flow surface mirror reflection or structure boundary, and low DoP area corresponds to interfering water mist. Therefore, DoP is used in image fusion and optical flow calculation to suppress interference and enhance targets.

[0135] Third, for the calculation formula of the polarization angle The purpose of the calculation formula is to represent the principal direction of the polarization photoelectric field vector vibration. This formula calculates the angle between the principal polarization axis and the reference direction (usually the 0° direction). For example, if... and If the polarization direction is horizontal, then AoP = 0, indicating that the polarization direction is horizontal. AoP provides additional information about surface orientation and microstructure. In the fusion formula, the current AoP is compared with a reference polarization angle. (For example, assuming the expected angle of water reflection) can further enhance the image region that matches the expected target characteristics.

[0136] Fourth, for fused image sequences The aim is to intelligently fuse information from the mid-wave and long-wave infrared bands into an enhanced image, designed to suppress water mist interference and highlight water flow and structural boundaries. Specifically, , For medium wave components and This component is determined by the degree of polarization of the mid-wave image ( ) and original light intensity ( Together, these factors determine that it will give higher weight to regions with high polarization (i.e., less interference and reliable information) and high light intensity. For long-wavelength components and In addition to the degree of polarization and light intensity, this component also introduces the polarization angle (…). The degree of matching, The term is a matching factor. When observed Reference angle as expected When they are completely identical, this term is 1 (maximum value); when they differ by 90°, this term is 0. This allows the fusion process to prioritize regions whose polarization direction matches the target (such as water reflection). This formula is a fusion strategy guided by physical characteristics. It is not a simple averaging, but rather believes that regions with good polarization characteristics (high DoP) and whose polarization direction matches the expected direction (AoP matching) provide more reliable and relevant information, thus highlighting these regions in the final image to achieve dehazing and boundary enhancement.

[0137] Fifth, for minimizing the energy function Regarding the calculation formula, its purpose is to define an objective function to be minimized for the improved polarization-constrained optical flow method, in order to calculate an optical flow field that satisfies both image brightness variation constraints and is smooth and physically reasonable. Among these, the data terms... This is a fundamental constraint of classical optical flow, namely the brightness constancy assumption. It requires that the brightness of a point should remain constant as it moves between two frames. This term penalizes optical flow vectors that violate this assumption. Smoothing term This assumption is that the optical flow field varies smoothly in space. and is the gradient of the optical flow field in x and y direction, which penalizes the sharp change of the optical flow field to prevent noise from generating chaotic motion vectors. The polarization constraint term is the key improvement of the present invention, is the square of the optical flow velocity, which directly penalizes the excessive motion amplitude. is the local average polarization degree, which plays the role of a spatial modulator. For example, in low DoP regions (e.g. water mist), there should not be large amplitude of meaningful real water flow motion, so this term will impose a strong constraint to suppress large optical flow values in these regions. While in high DoP regions (e.g. clear water surface), the data term and the smoothness term are allowed to play a greater role to calculate the possibly larger and more detailed motion vectors. Overall, the energy function By introducing the polarization constraint, the optical flow method has the intelligence of automatically suppressing noise in low SNR regions (water mist) and accurately calculating motion in high SNR regions (clear boundary), greatly improving the robustness of motion estimation in poor visual environments.

[0138] Sixth, for the calculation formula of the physical velocity field , the purpose is to convert the motion velocity calculated in the image coordinate system and in pixel units into the physical velocity in the real world and in meters per second. Among them, is obtained through camera calibration, which establishes the correspondence between the distance of one pixel in the image and the number of meters in the real world. This is a key step from image perception to physical cognition. Only after this conversion, the subsequent calculations of the average flow velocity , the flow velocity standard deviation and the average value of vorticity have clear physical meaning and engineering value, and can be directly compared and analyzed with hydraulic models, design specifications and historical data.

[0139] Seventh, for the calculation formula of the standard feature vector , it is the Z-score standardization method in statistics. The standardized feature value indicates how many standard deviations the current observation value deviates from its historical normal level. The larger this value (positive or negative), the higher the degree of abnormality.

[0140] Eighth, for the calculation formula of the scouring risk index R, the purpose is to fuse the standardized multi-dimensional feature vector into a single, comprehensive risk indicator. The formula is a weighted linear combination, is a weight vector trained by machine learning. The principle is: the machine learning algorithm (such as logistic regression, support vector machine) will automatically learn this set of weights on the historical data containing normal and dangerous cases. The goal is to let the weight vector be able to distinguish the safe state and the dangerous state to the greatest extent. The size of the weight reflects the importance of the corresponding feature to the prediction of the flushing risk. The flushing risk index R formula simplifies a complex multi-parameter risk assessment problem into an intuitive scalar R. The monitoring personnel only need to focus on the value of R and compare it with the preset threshold to trigger the corresponding level of early warning, realizing automatic and intelligent safety decision-making.

[0141] According to the second aspect of the present application, as shown in Figure 2 and Figure 3 , a reservoir spillway water monitoring system based on image recognition is also provided, which is used to realize the image recognition-based reservoir spillway water monitoring method of any one of the technical solutions in the first aspect of the present application, and includes an image acquisition module, a data processing module, and an early warning output module.

[0142] The image acquisition module includes at least one set of multi-band infrared polarization image acquisition unit deployed around the spillway. The acquisition unit includes a medium-wave infrared polarization camera, a long-wave infrared polarization camera, and a synchronization controller for controlling the synchronous acquisition of the two. It can be understood that the image acquisition module can be arranged on a fixed building (for example, a reservoir management room) around the spillway, or a separate fixed building (for example, a fixed pile or a fixed monitoring end) can be arranged.

[0143] The data processing module is configured to receive and store image data from the image acquisition module; and perform polarization physical feature calculation based on the medium-wave infrared polarization image sequence and the long-wave infrared polarization image sequence, and perform multi-band polarization feature fusion to generate an enhanced fusion image sequence to suppress water mist interference and highlight the water flow and structure boundary; and perform extraction of dynamic parameters for characterizing the flow pattern of the spillway water flow based on the generated fusion image sequence; and perform construction of a flushing risk index based on the extracted dynamic parameters, and perform hierarchical early warning according to the flushing risk index;

[0144] The early warning output module is used to send corresponding early warning information to the monitoring center or the user terminal according to the judgment result of the data processing module.

[0145] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present application should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for monitoring water flow through a reservoir spillway based on image recognition, characterized in that, Includes the following steps: Step S1: Synchronously acquire mid-wave infrared polarization image sequences and long-wave infrared polarization image sequences of the connection between the earth-rock dam body and the concrete spillway. Specifically, the mid-wave infrared polarization camera and the long-wave infrared polarization camera are controlled by a synchronization controller to acquire images synchronously, thereby obtaining mid-wave infrared polarization image sequences and long-wave infrared polarization image sequences that are aligned in time. The mid-wave infrared polarization camera and the long-wave infrared polarization camera continuously acquire at least three images with different polarization angles to form a complete polarization image group. Step S2: Based on the mid-wave infrared polarization image sequence and the long-wave infrared polarization image sequence, calculate their polarization physical characteristics. Including the polarization degree of each pixel in the mid-wave band polarization angle The degree of polarization of each pixel in the long-wave band and polarization angle And according to the formula: ； Generate enhanced fused image sequence To suppress water mist interference and highlight the water flow and structural boundaries, in the formula, , , , The values ​​represent the light intensity of pixels in the mid-wave infrared polarization image and the long-wave infrared polarization image, respectively. This represents the reference maximum value used for normalization of wave polarization. This represents the reference maximum value used for normalizing the light intensity of mid-wave images. This represents the reference maximum value used to normalize the long-wavelength polarization degree. When water surface reflection dominates, the polarization angle is... This represents the maximum reference value used to normalize the light intensity of long-wavelength images; Step S3: Based on the generated fused image sequence, an improved polarization-constrained optical flow method is used to extract dynamic parameters, including average flow velocity, to characterize the flow regime of the spillway. Flow velocity standard deviation , average vorticity Simultaneously, the temperature gradient mode from the long-wave infrared polarization image is extracted. and the difference in polarization between the current frame and the background frame. Obtain the feature vector According to the formula right Standardization is carried out, among which, , These are the mean and standard deviation of each feature in the historical normal overcurrent data, respectively; Step S4: Based on the extracted dynamic parameters, according to the formula Calculate the risk index of Taobao fraud, among which, The weight vector is trained based on historical disaster data. Represents the standard eigenvector The transpose is then used, and a graded warning is issued based on the Taobao brushing risk index.

2. The method for monitoring water flow through a reservoir spillway based on image recognition according to claim 1, characterized in that, In step S2, calculating the polarization physical characteristics specifically includes: For each band of image, calculate the Stokes vector for each pixel based on images at at least three different polarization angles; The polarization degree and polarization angle of each pixel are further calculated based on the Stokes vector.

3. The method for monitoring water flow through a reservoir spillway based on image recognition according to claim 1, characterized in that, In step S3, the dynamic parameters are extracted using an improved polarization-constrained optical flow method, and the image optical flow field is calculated by constructing an optical flow energy function that includes polarization degree constraints.

4. The method for monitoring water flow through a reservoir spillway based on image recognition according to claim 3, characterized in that, After obtaining the optical flow field of the image, the image velocity is converted into a physical velocity field by combining the scale factor obtained from camera calibration; and the average flow velocity, flow velocity standard deviation and vorticity average value are calculated as dynamic parameters based on the physical velocity field.

5. The method for monitoring water flow through a reservoir spillway based on image recognition according to claim 1, characterized in that, The graded early warning in step S4 specifically refers to: Preset a first threshold and a second threshold; When the risk index of Taobao fraud is between the first threshold and the second threshold, a first-level warning message is issued. When the risk index of Taobao fraud exceeds the second threshold, a second-level warning message is issued.

6. A reservoir spillway water flow monitoring system based on image recognition, used to implement the reservoir spillway water flow monitoring method based on image recognition as described in any one of claims 1-5, characterized in that, It includes an image acquisition module, a data processing module, and an early warning output module; The image acquisition module includes at least one multi-band infrared polarization image acquisition unit deployed around the spillway. The acquisition unit includes a mid-wave infrared polarization camera, a long-wave infrared polarization camera, and a synchronization controller for controlling the synchronous acquisition of both. The data processing module is configured to: receive and store image data from the image acquisition module; calculate the polarization physical characteristics based on the mid-wave infrared polarization image sequence and the long-wave infrared polarization image sequence, and perform multi-band polarization feature fusion to generate an enhanced fused image sequence to suppress water mist interference and highlight the water flow and structural boundaries; extract dynamic parameters to characterize the spillway flow pattern based on the generated fused image sequence; and construct a scour risk index based on the extracted dynamic parameters, and perform graded early warning according to the scour risk index. The early warning output module is used to send corresponding early warning information to the monitoring center or user terminal based on the judgment result of the data processing module.

Images