Open channel flow monitoring method and system based on Karman vortex street visual identification and adaptive structure

Abstract

The invention discloses an open channel flow monitoring method and system based on Karman vortex street visual identification and a self-adaptive structure, and relates to the technical field of open channel flow monitoring. The underwater depth of a rod body is automatically adjusted through a telescopic structure, and the formation of a Karman vortex street is stably induced; acquiring a vortex image by adopting a camera and extracting the shedding frequency of the vortex image in real time; and accurately inverting the water velocity and the flow rate by combining a Slodhaar relationship and a learning type model. The open channel flow monitoring method has good water level adaptability, sediment disturbance adaptability and model generalization ability, can realize high-efficiency, low-cost and intelligent open channel flow monitoring under complex field conditions, and provides powerful support for irrigation area water-saving management and control and fine management of water resources.

Description

Technical Field

[0001] The present invention relates to the technical field of open channel flow monitoring, and more particularly to an open channel flow monitoring method and system based on Karman vortex street visual recognition and adaptive structure. Background Art

[0002] Currently, open channels are the most common form of water transfer for irrigation agriculture, water resource scheduling, and aquatic ecological management. They offer advantages such as simple structure, low cost, and flexible deployment. However, due to their largely natural flow pattern, lack of pressure, significant influence from climate and topographic conditions, frequent water level fluctuations, and dramatic changes in sediment concentration, accurate and stable monitoring of flow velocity and flow rate is difficult, presenting a challenge in the current development of water conservancy informatization and intelligent irrigation systems. The development of real-time, non-contact, high-precision, low-cost, and low-maintenance open channel velocity and flow rate monitoring technology has become a research hotspot in the fields of water conservancy, water services, and smart agriculture.

[0003] Existing research has made active explorations in irrigation area water quantity testing systems, and the main methods include contact speed measuring instruments, ultrasonic / radar non-contact methods, and water level methods based on image recognition. For example, the open channel flow monitoring system and method based on the cylindrical flow principle disclosed by Lanzhou University of Technology (CN202310061629.6) mainly uses a camera to shoot the water level gauge and perform pixel processing, so as to calculate the depth of the fluid in the actual open channel using the number of pixels. At the same time, the flow velocity of the fluid can be calculated by the resistance and torque of the fluid to the water level gauge and the projected area of ​​the water level gauge in the direction perpendicular to the fluid, thereby obtaining the open channel flow. A monitoring system for automatic water measurement in open channels in irrigation areas disclosed by Zhejiang Water Conservancy and Estuary Research Institute (CN2015101967670) mainly uses an ultrasonic water level meter to detect the water level depth of the channel, and converts the water level signal into flow data through a flow integrator to obtain the open channel flow. Although the above-mentioned technologies have achieved certain application results in specific areas, the following problems still exist: First, the interference of sand-containing water bodies on ultrasonic signals and image quality has not been fully considered, making it difficult to operate stably under conditions of medium and high sand concentrations; second, the existing sensing methods (such as radar and ultrasound) are relatively expensive, which is not conducive to promotion and deployment in large-scale irrigation areas; third, traditional physical models have poor real-time performance and weak adaptability, while data-driven models generally lack physical constraints and interpretability, are prone to deviations, and are difficult to support precise scheduling and dynamic decision-making, especially in irrigation area environments with insufficient equipment and missing data.

[0004] Therefore, how to provide an open channel flow monitoring method with good water level adaptability, sediment disturbance adaptability and model generalization ability, which can realize efficient, low-cost and intelligent open channel flow monitoring under complex field conditions is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0005] In light of this, the present invention provides a method and system for open channel flow monitoring based on Karman vortex street visual recognition and adaptive structure. This method integrates key technologies such as vortex street induction, video recognition, frequency extraction, and machine learning modeling. A retractable structure automatically adjusts the depth of the rod's immersion in the water, stably inducing the formation of a Karman vortex street. A camera is used to capture vortex images and extract their shedding frequency in real time. Furthermore, the system combines the Strouhal relation and a learning model to accurately invert water velocity and flow rate. This method exhibits excellent water level adaptability, sediment disturbance adaptability, and model generalization capabilities, enabling efficient, low-cost, and intelligent open channel flow monitoring in complex field conditions. This system provides strong support for water conservation and refined water resource management in irrigation areas.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] An open channel flow monitoring method based on Karman vortex street visual recognition and adaptive structure, comprising:

[0008] A retractable guide body with automatic height adjustment function is deployed in the open channel to monitor water level changes in real time and obtain real-time water depth data;

[0009] Continuously capture a sequence of water flow images in the downstream area of ​​the retractable inducer and extract the Karman vortex shedding frequency using the Fourier transform method;

[0010] Taking the Strouhal relationship between the Kármán vortex shedding frequency and flow velocity as a constraint, the image features are introduced to construct an open channel flow velocity inversion model. Based on the open channel flow velocity inversion model, the current open channel predicted flow velocity is obtained.

[0011] Combining the real-time water depth data and the current open channel predicted flow rate, the open channel cross-sectional flow is calculated to obtain the open channel cross-sectional flow.

[0012] Optionally, the telescopic induction body includes a motor, a motor traction drive mechanism, a first-layer induction rod body, a water level sensor, a second-layer induction rod body, a third-layer induction rod body, a fourth-layer induction rod body, and a base; the base serves as the foundation of the entire device, and the output shaft of the motor is connected to the motor traction drive mechanism; the motor traction drive mechanism is connected to the bottom end of the first-layer induction rod body, and the second-layer induction rod body, the third-layer induction rod body, and the fourth-layer induction rod body are nested in sequence inside the lower-layer induction rod to achieve layer-by-layer telescopic movement; the water level sensor is fixed inside the first-layer induction rod body and changes its position with the telescopic movement of the induction rod.

[0013] Optionally, the construction of the open channel flow velocity inversion model is specifically as follows:

[0014] As a periodic flow structure, the shedding frequency of the Karman vortex street satisfies the following relationship with the incoming flow velocity:

[0015]

[0016] Where V is the incoming flow velocity; f is the vortex shedding frequency; D is the size of the inducer; St is the Strouhal number;

[0017] The open channel flow velocity inversion model uses the relationship as a constraint and introduces image features to build a machine learning regression model:

[0018] V pred =f θ (X)

[0019] And must meet

[0020] Among them, V pred is the current open channel predicted flow velocity; ∈ is the error tolerance;

[0021] Based on the collected physical-image composite feature set, the prediction function is constructed as follows:

[0022]

[0023] Among them, f θ (X) is the machine learning regression function; is the physical consistency loss function, defined as:

[0024]

[0025] in, is the time smoothing constraint; λ1, λ2 are the regularization coefficients.

[0026] Optionally, the feature system of the physical-image composite feature set includes:

[0027] a. Frequency characteristics:

[0028] Vortex main frequency f: extracted from the grayscale time series by Fourier transform or autocorrelation analysis;

[0029] Spectral energy density: the energy ratio at the main frequency, which measures the clarity of the disturbance;

[0030] Spectral width: reflects disturbance stability;

[0031] b. Image texture features:

[0032] Local entropy: reflects the complexity of regional information and is associated with vortex entrainment disturbances;

[0033] Grayscale mean / variance: reflects changes in illumination and disturbance;

[0034] Edge density / sharpness: the clarity of vortex edges in high-frequency areas;

[0035] Structural similarity SSIM: measures the periodic consistency of vortex images;

[0036] c. Image dynamic features:

[0037] Average optical flow velocity: Calculate flow field motion based on the optical flow algorithm;

[0038] Cycle Strength Index: a measure of the consistency of frequencies across multiple cycle windows;

[0039] High-frequency jitter factor: used to remove frames affected by strong reflections or disturbances;

[0040] d. External features:

[0041] Current water level information;

[0042] The current depth of the inducer in the water.

[0043] Optionally, the specific formula for obtaining the open channel cross-sectional flow rate is:

[0044] Q=V·A

[0045] A=B·h

[0046] Where V is the current predicted open channel flow velocity, B is the width of the bottom of the irrigation branch channel, and h is the water depth in the channel.

[0047] An open channel flow monitoring system based on Karman vortex street visual recognition and adaptive structure, including a camera, a camera pole, a retractable guide body with automatic height adjustment function, an open channel, a solar panel, and an edge computing device;

[0048] A camera continuously captures a sequence of water flow images in the downstream area of ​​the inducer;

[0049] A camera pole is used to support the camera and place it above the open channel;

[0050] The retractable inducer with automatic height adjustment function is used to stably induce water flow to generate Karman vortex street. The water level sensor is integrated into the inducer to monitor water level changes in real time.

[0051] The image processing module extracts the dominant frequency of vortex shedding through Fourier transform. Subsequently, the extracted frequency is converted to the current open channel flow velocity using the Strouhal relationship between the Karman vortex street frequency and the flow velocity.

[0052] Edge computing devices extract the dominant frequency of vortex shedding through methods such as Fourier transform. They then convert this frequency into the current open channel flow velocity using the Strouhal relationship between the Karman vortex street frequency and flow velocity. Furthermore, they deploy an open channel flow velocity inversion model to calculate the cross-sectional flow rate of the open channel, achieving simultaneous estimation and output of both flow velocity and flow rate.

[0053] Solar panels provide electricity to various components.

[0054] Optionally, the telescopic induction body includes a motor, a motor traction drive mechanism, a first-layer induction rod body, a water level sensor, a second-layer induction rod body, a third-layer induction rod body, a fourth-layer induction rod body, and a base; the base serves as the foundation of the entire device, and the output shaft of the motor is connected to the motor traction drive mechanism; the motor traction drive mechanism is connected to the bottom end of the first-layer induction rod body, and the second-layer induction rod body, the third-layer induction rod body, and the fourth-layer induction rod body are nested in sequence inside the lower-layer induction rod to achieve layer-by-layer telescopic movement; the water level sensor is fixed inside the first-layer induction rod body and changes its position with the telescopic movement of the induction rod.

[0055] It can be seen from the above technical solution that compared with the existing technology, the present invention discloses a method and system for monitoring open channel flow based on Karman vortex street visual recognition and adaptive structure, which automatically adjusts the water entry depth by deploying a retractable inducer to stably induce the Karman vortex street, uses a camera to capture water flow images and extracts the vortex street shedding frequency, and combines the Strouhal relationship and machine learning model to construct an open channel flow velocity inversion model to calculate the open channel cross-sectional flow; the present invention has many advantages: the retractable inducer can automatically adjust its height according to the water level, monitor the water depth in real time, and improve the adaptability to water level fluctuations; the Karman vortex street frequency is extracted through visual recognition, and the flow velocity is inverted by combining physical constraints and machine learning models, which reduces the interference of sand-containing water bodies on monitoring and enhances the model generalization ability and monitoring accuracy; it can realize efficient, low-cost and intelligent open channel flow monitoring under complex field conditions, and provide strong support for water-saving control and refined management of water resources in irrigation areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0057] Figure 1 A schematic flow chart of the method provided by the present invention;

[0058] Figure 2A schematic diagram of the system structure provided by the present invention;

[0059] Figure 3 This is a structural diagram of the retractable inducer provided by the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] The embodiment of the present invention discloses an open channel flow monitoring method based on Karman vortex street visual recognition and adaptive structure, such as Figure 1 Shown, including:

[0062] A retractable guide body with automatic height adjustment function is deployed in the open channel to monitor water level changes in real time and obtain real-time water depth data;

[0063] Continuously capture a sequence of water flow images in the downstream area of ​​the retractable inducer and extract the Karman vortex shedding frequency using the Fourier transform method;

[0064] Taking the Strouhal relationship between the Kármán vortex shedding frequency and flow velocity as a constraint, the image features are introduced to construct an open channel flow velocity inversion model. Based on the open channel flow velocity inversion model, the current open channel predicted flow velocity is obtained.

[0065] Combining the real-time water depth data and the current open channel predicted flow rate, the open channel cross-sectional flow is calculated to obtain the open channel cross-sectional flow.

[0066] In a specific embodiment, the retractable inducer with automatic height adjustment function is an intelligent flow disturbance device designed for Karman vortex induction and water level adaptation in open channels. Figure 3As shown, its overall structure consists of a motor 2-1, a motor traction drive mechanism 2-2, a first-layer induction rod body 2-3, a water level sensor 2-4, a second-layer induction rod body 2-5, a third-layer induction rod body 2-6, a fourth-layer induction rod body 2-7, and a base 2-8. The length of each induction rod layer ranges from 5 to 30 cm, making it suitable for channel water level testing with a depth of 5 to 120 cm. The base 2-8 serves as the foundation of the entire device. The output shaft of the motor 2-1 is connected to the motor traction drive mechanism 2-2; the motor traction drive mechanism 2-2 is connected to the bottom end of the first-layer induction rod body 2-3, while the second-layer induction rod body 2-5, the third-layer induction rod body 2-6, and the fourth-layer induction rod body 2-7 are sequentially nested inside the lower-layer induction rod, achieving layer-by-layer expansion and contraction. The water level sensor 2-4 is fixed inside the first-layer induction rod body 2-3 and changes position as the induction rod expands and contracts.

[0067] The induction rod body is made of a cylindrical, corrosion-resistant material and serves as a flow barrier to induce vortex formation. The rod's outer diameter is typically 1-10 cm and can be adjusted to suit the target flow rate range. A water level sensor, typically an ultrasonic or pressure sensor, is located in the middle of the rod body to monitor changes in the open channel water depth in real time. The telescopic adjustment mechanism is the core of this structure. It utilizes a micro electric push rod, electric cylinder, or a buoy linkage to drive the rod up and down, maintaining its top slightly below the water surface to ensure stable Karman vortex formation. This adjustment mechanism can be controlled by a PLC or edge computing device and receives feedback from the water level sensor for closed-loop regulation. A support and positioning device is fixed to the bottom or sidewall of the open channel, providing vertical positioning and lateral stability for the rod body. A protective sealing system protects the electronic components from sediment and water vapor corrosion during long-term underwater operation. The induction rod structure is characterized by its anti-deposition and corrosion resistance, automatic adaptation to water level changes, height adjustability, and symmetrical stability to disturbances. It is a key component for achieving stable open channel vortex induction and flow velocity identification.

[0068] In a specific embodiment, constructing an open channel flow velocity inversion model is specifically as follows:

[0069] This model aims to achieve non-contact, intelligent inversion of instantaneous flow velocity in open channels by extracting vortex frequency and visual dynamic features from image sequences, combined with the Karman vortex theory from hydraulics. Its core advantage lies in the fusion of physical mechanism constraints (white box) and data-driven learning (black box), improving the model's adaptability and accuracy while ensuring physical rationality.

[0070] As a periodic flow structure, the shedding frequency of the Karman vortex street satisfies the following relationship with the incoming flow velocity:

[0071]

[0072] Where V is the incoming flow velocity; f is the vortex shedding frequency (extracted from the image sequence); D is the size of the inducer (fixed); St is the Strouhal number;

[0073] The open channel flow velocity inversion model uses the relationship as a constraint and introduces image features to build a machine learning regression model:

[0074] V pred =f θ (X)

[0075] And must meet

[0076] Among them, V pred is the current open channel predicted flow velocity; ∈ is the error tolerance;

[0077] Construct a feature system for the physical-image composite feature set, specifically including:

[0078] a. Frequency characteristics:

[0079] Vortex main frequency f: extracted from the grayscale time series by Fourier transform (FFT) or autocorrelation analysis;

[0080] Spectral energy density: the energy ratio at the main frequency, which measures the clarity of the disturbance;

[0081] Spectrum width: reflects the stability of disturbance (the narrower the better);

[0082] b. Image texture features (based on grayscale frames):

[0083] Local entropy: reflects the complexity of regional information and is associated with vortex entrainment disturbances;

[0084] Grayscale mean / variance: reflects changes in illumination and disturbance;

[0085] Edge density / sharpness: the clarity of vortex edges in high-frequency areas;

[0086] Structural similarity SSIM: measures the periodic consistency of vortex images;

[0087] c. Image dynamic features (based on time series images):

[0088] Average optical flow velocity: Calculates flow field motion based on optical flow algorithms such as Farneback / RAFT;

[0089] Cycle Strength Index: a measure of the consistency of frequencies across multiple cycle windows;

[0090] High-frequency jitter factor: used to remove frames affected by strong reflections or disturbances;

[0091] d. Optional external features:

[0092] Current water level information (from water level sensor);

[0093] The current depth of the inducer in the water;

[0094] Ambient brightness (LUX): can be estimated from the image exposure.

[0095] Based on the constructed physical-image composite feature set, the constructed prediction function is as follows:

[0096]

[0097] Among them, f θ (X) is the machine learning regression function (SVR / RF / CNN-LSTM); is the physical consistency loss function, defined as:

[0098]

[0099] in, is the temporal smoothing constraint (to reduce inter-frame jumps); λ1 and λ2 are the regularization coefficients.

[0100] Specifically, in terms of model selection, the present invention provides adaptation strategies for multiple regression structures based on the differences in actual deployment environment and data complexity: when the number of samples is limited, the feature dimension is medium and it needs to be quickly deployed on edge devices, it is recommended to use the support vector regression (SVR) model, which has stable training, high prediction accuracy, and good generalization ability for small samples; when the image feature dimensions are large and there is a certain amount of noise interference in the data, it is suitable to use the random forest (RF) model, which has strong nonlinear modeling ability, excellent noise resistance, and strong interpretability; for platforms with significant image temporal characteristics, complex scenes or large computing resources, a lightweight convolutional neural network (CNN) or a CNN-LSTM combination structure can be used to achieve joint modeling of image spatial texture and vortex street periodic dynamics. Overall, the model structure selection is based on the comprehensive consideration of accuracy, stability, deployment feasibility and reasoning efficiency to ensure the adaptability and practicality of the system in different scenarios.

[0101] Open channel suspended sediment inversion model

[0102] This model aims to non-contactly infer the suspended sediment concentration (SSC) in open channel water bodies based on the changes in water surface texture, grayscale distribution and dynamic disturbance characteristics observed in video image sequences. It is suitable for complex water bodies such as irrigation channels, ecological open channels and water ditches in mountainous areas under unmanned, real-time monitoring conditions.

[0103] a. Basic Principle: This model is based on the influence of suspended sediment on the optical properties of water. It leverages the mapping relationship between grayscale, texture, and dynamic disturbance features in an image and sediment concentration to achieve non-contact inversion of suspended sediment concentration (SSC) in open channels. As sediment concentration increases, the water's reflective properties, grayscale distribution, and disturbance patterns change significantly, resulting in, for example, an increase in overall image grayscale, blurred textures, weakened edges, and unstable optical flow. By extracting these changing features and combining them with regression modeling, effective real-time estimation of sediment concentration can be achieved.

[0104] b. Model Structure: This model consists of four core modules: an image acquisition module, an image preprocessing and feature extraction module, a regression modeling module, and a result output module. The image acquisition module acquires water body image sequences in real time using top or side cameras. The image preprocessing module performs image stabilization, cropping, denoising, and illumination normalization. The feature extraction module extracts multiple quantitative metrics from the image, including grayscale statistics, texture entropy, edge density, spectral energy, and optical flow perturbation. The regression modeling module inputs the extracted image features into algorithms such as SVR, random forest, or lightweight CNN to construct an SSC prediction model and outputs the estimated results in real time for monitoring and early warning.

[0105] c. Input Features: Model input features fall into two categories: static image features and dynamic disturbance features. Static features such as grayscale mean, variance, image entropy, edge density, and spectral energy ratio reflect changes in image brightness and texture distribution of sediment-laden water bodies. Dynamic features such as average optical flow velocity and optical flow fluctuation index describe the uneven motion of water disturbances as sediment concentration changes. These features can be extracted algorithmically from image frame sequences and are characterized by high stability, strong interpretability, and sensitivity to changes in sediment concentration.

[0106] d. Training Set Calibration Process: Model training relies on a "image data - measured SSC" comparison sample. First, video images are collected under varying sediment concentrations, and corresponding water samples are simultaneously collected for laboratory testing to form a training dataset. Next, the image sequences are preprocessed and feature extracted to construct a standardized feature matrix. An appropriate regression model is then selected for training, with cross-validation used to optimize parameters and evaluate performance. Finally, the trained model is deployed to edge computing devices to enable automated estimation and dynamic monitoring of open channel sediment concentrations.

[0107] In a specific embodiment, the open channel flow rate is calculated as follows:

[0108] By integrating Karman vortex frequency inversion with water level measurement technology, real-time calculation of open channel flow is achieved. First, the top camera captures a sequence of vortex images downstream of the induction rod. The vortex shedding frequency is extracted through Fourier transform, and the instantaneous flow velocity is inferred by combining the scale of the induction body and the Strouhal relation. At the same time, the built-in water level sensor on the rod obtains the water depth in real time, and the cross-sectional area of ​​the water flow is calculated based on the channel geometry. The instantaneous flow rate Q of the open channel can be expressed as:

[0109] Q=V·A

[0110] That is, the volume of water flowing through the section per unit time (unit m 3 The system can refresh the data every second or every minute to output real-time flow values, or accumulate or average the data in time series to generate a flow curve.

[0111] An open channel flow monitoring system based on Karman vortex street visual recognition and adaptive structure, such as Figure 2 As shown, it includes a high frame rate camera 1, a retractable guide body 2 (spoiler rod body) with automatic height adjustment function, a water flow 3, an open channel 4, a camera pole 5, a solar panel 6, an edge computing device and an image processing module (not shown in the figure);

[0112] High frame rate camera 1, continuously capturing a sequence of water flow images in the downstream area of ​​the inducer;

[0113] The camera pole 5 is used to support the camera and place the camera above the open channel 4; the water flow 3 flows in the open channel 4;

[0114] The retractable inducer 2, which has an automatic height adjustment function, is used to stably induce water flow to generate a Karman vortex street. The water level sensor is integrated into the inducer to monitor water level changes in real time.

[0115] The image processing module extracts the dominant frequency of vortex shedding through Fourier transform. Subsequently, the extracted frequency is converted to the current open channel flow velocity using the Strouhal relationship between the Karman vortex street frequency and the flow velocity.

[0116] Edge computing devices deploy open channel flow velocity inversion models to calculate cross-sectional flow in open channels, enabling simultaneous estimation and output of flow velocity and flow rate.

[0117] The solar panel 6 provides electrical energy for each component.

[0118] The retractable guide body, with automatic height adjustment, is an intelligent flow disturbance device designed specifically for Karman vortex induction and water level adaptation in open channels. Its overall structure consists of a motor 2-1, a motor traction drive mechanism 2-2, a first-layer guide rod body 2-3, a water level sensor 2-4, a second-layer guide rod body 2-5, a third-layer guide rod body 2-6, a fourth-layer guide rod body 2-7, and a base 2-8. Each guide rod body ranges in length from 5 to 30 cm, suitable for measuring water levels in channels with depths ranging from 5 to 120 cm. The base 2-8 serves as the foundation of the entire device. The output shaft of the motor 2-1 is connected to the motor traction drive mechanism 2-2. The motor traction drive mechanism 2-2 is connected to the bottom end of the first-layer guide rod body 2-3. The second-layer guide rod body 2-5, the third-layer guide rod body 2-6, and the fourth-layer guide rod body 2-7 are nested within the lower guide rod body, enabling layer-by-layer expansion and contraction. The water level sensor 2-4 is fixed within the first-layer guide rod body 2-3 and changes position as the guide rod expands and contracts.

[0119] The method of the present invention is specifically described below with reference to a typical open channel application scenario in an irrigation area.

[0120] The system of the present invention was deployed in a branch canal (1.5 meters wide, 1.2 meters deep, and rectangular in cross-section) in an irrigation district. A cylindrical, retractable induction rod with a diameter of 4 cm was vertically installed in the center of the canal bottom. The rod housed a high-precision pressure-type water level sensor and a micro-electric lifting mechanism. An industrial camera with a resolution of 1920×1080 and a frame rate of 60 fps was installed above the system to capture images of the water surface in the wake of the induction rod. The camera was connected to an image processing module via an edge computing device.

[0121] When the system is working, the camera collects a real-time sequence of water surface images in the area downstream of the rod, and uses background subtraction and Fourier transform algorithms to extract the vortex shedding frequency f = 1.6 Hz. Based on the inducer diameter D = 0.04 m and the Strouhal number St = 0.2, the current flow velocity is calculated:

[0122]

[0123] At the same time, the water level sensor measured the water depth in the channel to be h = 0.9m. According to the rectangular channel cross-section formula A = B·h = 1.5·0.9 = 1.35m 2 , then the instantaneous flow rate of the open channel is:

[0124] Q=V·A=0.32·1.35=0.432m 3 / s

[0125] Specifically, the results can also be displayed in real time on the control platform interface, stored in the local database, and pushed to the irrigation district dispatching center through the 4G module.

[0126] After a week of continuous operation, data was compared with a flow meter located downstream, showing a consistent error within ±3%, validating the system's monitoring accuracy and stability. The system boasts 24 / 7 automated operation and is widely applicable to typical scenarios such as irrigation areas, ecological waterways, and hydraulic test open channels, enabling intelligent, low-maintenance, and high-frequency monitoring of open channel flows.

[0127] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0128] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for monitoring open channel flow based on Karman vortex street visual recognition and adaptive structure, characterized in that: include: A retractable guide body with automatic height adjustment function is deployed in the open channel to monitor water level changes in real time and obtain real-time water depth data; Continuously capture a sequence of water flow images in the downstream area of ​​the retractable inducer and extract the Karman vortex shedding frequency using the Fourier transform method; Taking the Strouhal relationship between the Kármán vortex shedding frequency and flow velocity as a constraint, the image features are introduced to construct an open channel flow velocity inversion model. Based on the open channel flow velocity inversion model, the current open channel predicted flow velocity is obtained. Combining the real-time water depth data and the current open channel predicted flow rate, the open channel cross-sectional flow is calculated to obtain the open channel cross-sectional flow.

2. The open channel flow monitoring method based on Karman vortex street visual recognition and adaptive structure according to claim 1 is characterized in that: The retractable induction body includes a motor, a motor traction drive mechanism, a first-layer induction rod body, a water level sensor, a second-layer induction rod body, a third-layer induction rod body, a fourth-layer induction rod body, and a base; the base serves as the foundation of the entire device, and the output shaft of the motor is connected to the motor traction drive mechanism; the motor traction drive mechanism is connected to the bottom end of the first-layer induction rod body, and the second-layer induction rod body, the third-layer induction rod body, and the fourth-layer induction rod body are nested in sequence inside the lower-layer induction rod to achieve layer-by-layer expansion and contraction; the water level sensor is fixed inside the first-layer induction rod body and changes its position with the expansion and contraction movement of the induction rod.

3. The open channel flow monitoring method based on Karman vortex street visual recognition and adaptive structure according to claim 1 is characterized in that: The construction of the open channel velocity inversion model is specifically as follows: As a periodic flow structure, the shedding frequency of the Karman vortex street satisfies the following relationship with the incoming flow velocity: Where V is the incoming flow velocity; f is the vortex shedding frequency; D is the size of the inducer; St is the Strouhal number; The open channel flow velocity inversion model uses the relationship as a constraint and introduces image features to build a machine learning regression model: V pred =f θ (X) And must meet Among them, V pred is the current open channel predicted flow velocity; ∈ is the error tolerance; Based on the collected physical-image composite feature set, the prediction function is constructed as follows: Among them, f θ (X) is the machine learning regression function; is the physical consistency loss function, defined as: in, is the time smoothing constraint; λ1, λ2 are the regularization coefficients.

4. The open channel flow monitoring method based on Karman vortex street visual recognition and adaptive structure according to claim 3 is characterized in that: The feature system of the physical-image composite feature set includes: a. Frequency characteristics: Vortex main frequency f: extracted from the grayscale time series by Fourier transform or autocorrelation analysis; Spectral energy density: the energy ratio at the main frequency, which measures the clarity of the disturbance; Spectral width: reflects disturbance stability; b. Image texture features: Local entropy: reflects the complexity of regional information and is associated with vortex entrainment disturbances; Grayscale mean / variance: reflects changes in illumination and disturbance; Edge density / sharpness: the clarity of vortex edges in high-frequency areas; Structural similarity SSIM: measures the periodic consistency of vortex images; c. Image dynamic features: Average optical flow velocity: Calculate flow field motion based on the optical flow algorithm; Cycle Strength Index: a measure of the consistency of frequencies across multiple cycle windows; High-frequency jitter factor: used to remove frames affected by strong reflections or disturbances; d. External features: Current water level information; The current depth of the inducer in the water.

5. The open channel flow monitoring method based on Karman vortex street visual recognition and adaptive structure according to claim 1 is characterized in that: The specific formula for obtaining the open channel cross-sectional flow rate is: Q=V·A A=B·h Where V is the current predicted open channel flow velocity, B is the width of the bottom of the irrigation branch channel, and h is the water depth in the channel.

6. An open channel flow monitoring system based on Karman vortex street visual recognition and adaptive structure, characterized in that: Including cameras, camera poles, retractable guide bodies with automatic height adjustment function, open channel channels, solar panels, and edge computing equipment; A camera continuously captures a sequence of water flow images in the downstream area of ​​the inducer; A camera pole is used to support the camera and place it above the open channel; The retractable inducer with automatic height adjustment function is used to stably induce water flow to generate Karman vortex street. The water level sensor is integrated into the inducer to monitor water level changes in real time. The image processing module extracts the dominant frequency of vortex shedding through Fourier transform. Subsequently, the extracted frequency is converted to the current open channel flow velocity using the Strouhal relationship between the Karman vortex street frequency and the flow velocity. Edge computing devices extract the dominant frequency of vortex shedding through methods such as Fourier transform. They then convert this frequency into the current open channel flow velocity using the Strouhal relationship between the Karman vortex street frequency and flow velocity. Furthermore, they deploy an open channel flow velocity inversion model to calculate the cross-sectional flow rate of the open channel, achieving simultaneous estimation and output of both flow velocity and flow rate. Solar panels provide electricity to various components.

7. The open channel flow monitoring system based on Karman vortex street visual recognition and adaptive structure according to claim 6 is characterized in that: The retractable induction body includes a motor, a motor traction drive mechanism, a first-layer induction rod body, a water level sensor, a second-layer induction rod body, a third-layer induction rod body, a fourth-layer induction rod body, and a base; the base serves as the foundation of the entire device, and the output shaft of the motor is connected to the motor traction drive mechanism; the motor traction drive mechanism is connected to the bottom end of the first-layer induction rod body, and the second-layer induction rod body, the third-layer induction rod body, and the fourth-layer induction rod body are nested in sequence inside the lower-layer induction rod to achieve layer-by-layer expansion and contraction; the water level sensor is fixed inside the first-layer induction rod body and changes its position with the expansion and contraction movement of the induction rod.

Images