Multi-sensor fused intelligent detection method for water flow direction and flow velocity

Abstract

The invention discloses a water flow direction and flow velocity intelligent detection method based on multi-sensor fusion, and belongs to the field of underwater sensors, and the method comprises the following steps: constructing a piezoelectric sensor array based on a plurality of piezoelectric sensors; acquiring sensor array actual measurement data and sensor array fluid simulation data based on the piezoelectric sensor array to construct a training data set; preprocessing the training data set to obtain preprocessed data; and inputting the preprocessed data into the CNN-LSTM hybrid neural network for data processing to obtain the flow direction and the flow velocity of the horizontal plane. The invention provides an efficient and reliable flow velocity and flow direction detection means for the fields of underwater environment monitoring, marine equipment control and the like.

Description

Technical Field

[0001] This invention belongs to the field of underwater sensor technology, and particularly relates to an intelligent detection method for water flow direction and velocity using multi-sensor fusion. Background Technology

[0002] Detecting the direction and velocity of water flow is crucial for numerous fields, including water resource management, ecological environmental protection, flood control and disaster reduction, and marine equipment development. Currently, common detection methods primarily measure flow velocity by the rotational speed of a rotor or propeller propelled by the water flow, using a magnetic compass to determine the flow direction. However, these methods suffer from drawbacks such as only being able to measure one-dimensional velocity, being prone to stalling at low flow rates, interfering with the flow field, and being susceptible to corrosion of moving parts. Furthermore, they are bulky and difficult to integrate into other marine equipment. Electromagnetic and acoustic detection devices also suffer from large size and high cost, and cannot simultaneously detect flow direction and velocity. Marine robots play a vital role in marine exploration. When moving underwater, marine robots are easily affected by ocean currents; therefore, accurately determining the direction and velocity of currents is essential for stable robot control. Existing equipment either cannot simultaneously detect flow velocity and direction or is bulky, expensive, and difficult to install. Therefore, there is a need for novel sensors that are small, inexpensive, highly reliable, easy to integrate, and capable of simultaneously detecting flow velocity and direction, providing real-time flow velocity and direction information for the reliable operation of marine equipment. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a multi-sensor fusion method for intelligent detection of water flow direction and velocity, thereby resolving the issues present in the prior art.

[0004] To achieve the above objectives, the present invention provides a multi-sensor fusion method for intelligent detection of water flow direction and velocity, comprising:

[0005] Construct a piezoelectric sensor array based on several piezoelectric sensors;

[0006] A training dataset is constructed based on the measured data and fluid simulation data of the piezoelectric sensor array.

[0007] The training dataset is preprocessed to obtain preprocessed data;

[0008] The preprocessed data is input into a CNN-LSTM hybrid neural network for data processing to obtain the flow direction and velocity of the horizontal surface.

[0009] Optionally, the process of constructing a piezoelectric sensor array includes:

[0010] A sensor array is constructed based on five piezoelectric sensors. Four of the piezoelectric sensors are arranged horizontally on the same plane. Furthermore, the four piezoelectric sensors are orthogonally distributed along the positive X-axis, negative X-axis, positive Y-axis, and negative Y-axis, respectively, to detect changes in water flow pressure in four orthogonal directions within the horizontal plane.

[0011] The fifth piezoelectric sensor is arranged vertically in the XY plane to detect the static pressure value in the direction perpendicular to the plane.

[0012] Optionally, the piezoelectric sensor array is connected to an embedded processor via an IIC communication interface, and the embedded processor transmits sensor data to a host computer via an RS485 bus.

[0013] Optionally, the process of constructing the training dataset includes:

[0014] Fluid simulation data of sensor array under different flow velocities and flow directions were obtained through fluid simulation experiments;

[0015] The sensor array measured data under different flow velocities and flow directions was obtained through a physical experiment in a wind-wave-flow experimental tank.

[0016] The measured data and fluid simulation data of the sensor array are merged and processed using data augmentation methods such as random noise addition, random scaling, and random offset to form a training dataset.

[0017] Optionally, the process of preprocessing the training dataset to obtain preprocessed data includes:

[0018] The training dataset is normalized to form grayscale images;

[0019] The grayscale images are organized into time series data, with each time series containing 10 sets of sensor pressure value data.

[0020] Optionally, the expression for normalizing the training dataset is:

[0021]

[0022] In the formula, P is the original pressure value measured by the sensor; P min The global minimum value among all pressure sensor data; P max P is the global maximum value among all pressure sensor data. norm This is the normalized pressure value.

[0023] Optionally, the CNN-LSTM hybrid neural network includes: a first convolutional layer, a second convolutional layer, a first pooling layer, a second pooling layer, an LSTM layer, a first fully connected layer, and a second fully connected layer;

[0024] The process of obtaining the flow direction and velocity of the horizontal surface through data processing based on the CNN-LSTM hybrid neural network includes:

[0025] The grayscale image is input into the first convolutional layer, and the spatial feature map is extracted using a (2,2) convolutional kernel and a ReLU activation function to obtain the output of the first convolutional layer.

[0026] The first pooling layer performs max pooling on the output of the first convolutional layer using a (2,2) pooling kernel to reduce the dimension of the feature map and obtain the output of the first pooling layer.

[0027] The output of the first pooling layer is input into the second convolutional layer to extract high-level spatial features and obtain the output of the second convolutional layer.

[0028] The second pooling layer performs global average pooling on the output of the second convolutional layer to obtain the output of the second pooling layer;

[0029] The output of the second pooling layer is input into the LSTM layer to output the final time step result;

[0030] The first fully connected layer output is obtained by performing feature mapping on the final time step result based on the first fully connected layer.

[0031] The output of the first fully connected layer is input into two independent branches to obtain the predicted flow velocity and the predicted flow direction angle.

[0032] Optionally, the piezoelectric sensor is an MS5837 sensor.

[0033] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned intelligent detection method for water flow direction and velocity fusion based on multi-sensor fusion.

[0034] The present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned intelligent detection method for water flow direction and velocity fusion based on multi-sensor fusion.

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] This invention achieves comprehensive and accurate detection of water flow parameters by employing a multi-piezoelectric sensor array design, effectively overcoming the shortcomings of traditional mechanical sensors, such as limited measurement dimensions and susceptibility to interference. The innovative CNN-LSTM hybrid neural network architecture fully leverages the advantages of spatial feature extraction and time series analysis, enabling the system to accurately capture complex dynamic characteristics of water flow. This scheme significantly improves the model's environmental adaptability and robustness by fusing simulation and measured data to construct a training set. The entire system has a compact structure, facilitating integration into various underwater devices, and requires no moving parts, offering advantages such as high reliability and low maintenance costs. In particular, the vertically positioned static pressure sensor effectively eliminates measurement errors caused by water depth variations, further improving detection accuracy. This technical solution provides an efficient and reliable means of detecting flow velocity and direction for underwater environmental monitoring, marine equipment control, and other fields. Attached Figure Description

[0037] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 This is a piezoelectric sensor array according to an embodiment of the present invention;

[0039] Figure 2 This is a piezoelectric sensor array data processing system according to an embodiment of the present invention;

[0040] Figure 3 These are pressure variation curves at different flow rates according to an embodiment of the present invention;

[0041] Figure 4 The pressure change curves of each sensor under different flow directions are shown in the embodiments of the present invention.

[0042] Figure 5 This describes the organization of network input data in an embodiment of the present invention.

[0043] Figure 6 This is a grayscale image used for network input in an embodiment of the present invention;

[0044] Figure 7 This is a network structure diagram of an embodiment of the present invention;

[0045] Figure 8 This is the total loss curve for the training set and validation set in an embodiment of the present invention. Detailed Implementation

[0046] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0048] This embodiment provides a multi-sensor fusion method for intelligent detection of water flow direction and velocity, including the following steps: constructing a piezoelectric sensor array based on several piezoelectric sensors; acquiring measured data and fluid simulation data of the sensor array based on the piezoelectric sensor array to construct a training dataset; preprocessing the training dataset to obtain preprocessed data; and inputting the preprocessed data into a CNN-LSTM hybrid neural network for data processing to obtain the flow direction and velocity at the horizontal surface.

[0049] like Figure 1 As shown, this invention proposes a method for constructing a flow direction and velocity sensor based on five piezoelectric sensors. Four of these piezoelectric sensors are sensitive to pressure changes caused by water flow in four orthogonal directions on an underwater plane (sensors 1 and 3 sense pressure from the positive and negative X-axis, respectively; sensors 2 and 4 sense pressure from the positive and negative Y-axis, respectively; and sensor 5 senses pressure from the positive Z-axis, perpendicular to the XY plane, and is used to eliminate the static pressure from the other four piezoelectric sensors). The pressure values ​​acquired by all five piezoelectric sensors are processed by a deep learning neural network to simultaneously obtain the flow direction and velocity on the horizontal plane. The five piezoelectric sensors are denoted as P1 to P5.

[0050] like Figure 2 As shown, this invention constructs a data processing system for a piezoelectric sensor array. To calculate flow velocity and direction based on the values ​​of the five pressure sensors, the host computer uses a CNN-LSTM (Convolutional Neural Network and Long Short-Term Memory Network) hybrid neural network for data analysis. This fully utilizes the spatial feature extraction capability of the Convolutional Neural Network (CNN) and the temporal modeling capability of the Long Short-Term Memory Network (LSTM).

[0051] 1. Dataset Construction. To train the CNN-LSTM hybrid neural network for flow velocity and direction prediction, a dataset needs to be constructed for training and testing. To construct the dataset, simulation and physical experiments were designed to obtain data samples. A fluid simulation experiment scenario with a sensor array was built on the SolidWorks platform. Fluid simulation data was obtained through the FlowSimulation module. During the simulation experiments, the piezoelectric sensor array structure was kept stationary, the water flow direction was set to a constant value, and the water flow velocity was changed. The water flow velocity was increased by 0.1 m / s in each experiment to simulate different flow field environmental changes. For each different flow field environment, the pressure value data on the surface of each biomimetic lateral line sensor was recorded. This includes pressure contour maps for flow velocities of 0.1 m / s, 0.2 m / s, 0.3 m / s, 0.4 m / s, 0.5 m / s, 0.6 m / s, 0.7 m / s, 0.8 m / s, and 0.9 m / s. Figure 3 These are pressure change curves for each sensor under different flow rates, which conform to the laws of fluid mechanics.

[0052] To obtain the pressure changes of each piezoelectric sensor under different flow directions, pressure contour maps under different flow directions were obtained by changing the sensor array angle while keeping the flow velocity constant in a simulation environment. The pressure change curves of each sensor under different flow directions are shown below. Figure 4 As shown in the figure. Through simulation experiments, pressure change data of each piezoelectric sensor under different flow velocities and flow directions can be obtained, which can be used as part of the dataset.

[0053] The physical experiment was conducted in the laboratory using a wave-current experimental tank to generate a stable, continuous water flow, which was then measured using a pressure sensor. The wave-current experimental tank is an important tool for simulating the flow characteristics of the actual marine environment, allowing for the generation of water flows at different velocities for experimental research.

[0054] A piezoelectric sensor array was placed in a stable flow area within the water tank. The flow velocity was altered by changing the vibration frequency of the wave source in the experimental water tank. Changing the horizontal placement angle of the piezoelectric sensors simulated changes in flow direction. The flow velocity was measured using an LS1206B velocimeter. Under each different combination of angle and flow velocity, the pressure changes caused by the water flow at each piezoelectric sensor were recorded. Simultaneously, the static pressure at position P5 was recorded. By calculating the difference between P1-P4 and P5, the actual pressure change caused by relative motion was obtained, effectively eliminating the interference of static pressure on the experimental data. This method helps to accurately capture the influence of fluid motion on the sensors, ensuring the reliability and accuracy of the experimental results.

[0055] Through the above-mentioned physical experiment, pressure change data of each piezoelectric sensor under different flow rates and flow directions can also be obtained, and this data can be used as part of the dataset.

[0056] The simulation and physical experiment results were combined to form the training dataset for the neural network. To increase the sample size and account for variations in the actual ocean current environment and sensor errors, three data augmentation methods were introduced:

[0057] (1) Random noise addition: Add random perturbations following a normal distribution N(0,0.1) to each pressure value (p1, p2, p3, p4) to enhance the robustness of the data.

[0058] (2) Random scaling: Apply a random scaling factor with a range of 0.9 to 1.1 uniformly distributed to the pressure value to simulate small changes in the sensor sensitivity.

[0059] (3) Random offset: Apply a random quantity with a range of -5 to 5 evenly distributed to the pressure value to simulate the offset of water flow pressure as the environment changes.

[0060] After the above processing, a dataset of sensor pressure values ​​from p1 to p4 was generated, containing flow velocities ranging from 0 to 1 m / s, with intervals of 0.1 m / s, and flow directions ranging from 0 to 355 degrees, with intervals of 15 degrees for each flow velocity. There are 200 sets of sensor pressure values ​​for each flow velocity and each flow direction.

[0061] 2. Organization of Network Input Data. To construct a time series representation that reflects uncertain disturbances in underwater flow, 10 sets of sensor pressure values ​​were randomly selected from 200 sets for each flow velocity and direction. Each set consisted of pressure values ​​from four sensors, P1-P4. Figure 5 As shown.

[0062] To construct images suitable for input to a CNN network, Figure 5 The pressure value matrix data shown is normalized to normalize all pressure values ​​to the range of 0 to 255, forming a grayscale image. The normalization formula is as follows:

[0063]

[0064] Where P is the raw pressure value measured by the sensor, i.e. Figure 7 In Pn-m; P min The global minimum value among all pressure sensor data; P max P is the global maximum value among all pressure sensor data. norm This is the normalized pressure value, ranging from 0 to 255. Figure 5It can generate grayscale images. Figure 6 This is an example of a grayscale image. The network structure used is as follows: Figure 7 As shown:

[0065] The input to the model is Figure 6 The grayscale image shown has input parameters (N, 10, 4, 10, 1) representing (N samples, each containing 10 frames of data; each frame of data comes from 4 sensors, with each sensor's measurement values ​​within 10 time steps, stored in single-channel format).

[0066] The CNN-LSTM hybrid neural network structure includes a first convolutional layer, a second convolutional layer, a first pooling layer, a second pooling layer, an LSTM layer, a first fully connected layer, and a second fully connected layer, ultimately outputting flow velocity and flow direction. The specific implementation process includes: inputting a grayscale image into the first convolutional layer, using a (2,2) convolutional kernel and a ReLU activation function to extract spatial feature maps to obtain the output of the first convolutional layer; performing max pooling on the output of the first convolutional layer using a (2,2) pooling kernel to reduce the dimension of the feature maps to obtain the output of the first pooling layer; inputting the output of the first pooling layer into the second convolutional layer to extract high-level spatial features to obtain the output of the second convolutional layer; performing global average pooling on the output of the second convolutional layer to obtain the output of the second pooling layer; inputting the output of the second pooling layer into the LSTM layer to output the final time-step result; performing feature mapping on the final time-step result based on the first fully connected layer to obtain the output of the first fully connected layer; and inputting the output of the first fully connected layer into two independent branches to obtain the predicted flow velocity and predicted flow direction angle.

[0067] Furthermore, as a specific implementation of this embodiment, the process of obtaining the flow direction and velocity of the horizontal plane through data processing based on the CNN-LSTM hybrid neural network includes: the input data is processed through a first TimeDistributed two-dimensional convolutional layer to output 64 feature maps, with a convolutional kernel size of (2,2) and the ReLU activation function used. The TimeDistributed method allows the convolution operation to be applied to each frame in the sequence, extracting local features of the spatial distribution of the sensors. The output shape of this layer is (None, 10, 4, 10, 64), representing 10 frames, 4 sensors, 10 time steps, and 64 feature maps.

[0068] After the first convolutional layer, max pooling is performed with a pooling kernel size of (2, 2), which halves the feature map size to (None, 10, 2, 5, 64). Max pooling reduces computational complexity, enhances the model's robustness to spatial translation invariance, and preserves the main features.

[0069] The second convolutional layer uses the same (2, 2) kernel parameters and outputs 64 feature maps to further extract high-level spatial features. The output shape is maintained at (None, 10, 2, 5, 64), and the feature representation capability is enhanced through multiple convolutions.

[0070] The second convolutional layer, followed by a global average pooling layer, performs average pooling on each feature map, compressing the spatial dimension and outputting a shape of (None, 10, 64), transforming the time-series features into a format suitable for LSTM input. This operation reduces the risk of overfitting while preserving temporal dependencies.

[0071] The LSTM layer contains 128 neurons and outputs only the result of the last time step, with a shape of (None, 128). LSTM captures the long-term dynamic changes of 10 frames of pressure data in a sequence through input gates, forget gates, and output gates, making it particularly suitable for processing nonlinear time patterns caused by water flow.

[0072] The first fully connected layer contains 128 neurons, uses the ReLU activation function and adds L2 regularization (0.001) to prevent overfitting, and then applies a 30% Dropout layer to enhance generalization ability.

[0073] The model has two output branches: flow velocity, which is output as a linear value after passing through a fully connected layer of 64 neurons, predicting the flow velocity; and flow direction, which is output as the predicted angle after passing through a fully connected layer of 64 neurons. The system uses the Adam optimizer for parameter optimization.

[0074] Model training employed EarlyStopping (monitoring val_loss, patiently for 50 epochs, recovering optimal weights, min_delta = 0.0005) and ReduceLROnPlateau (monitoring val_loss, learning rate decay factor 0.5, patiently for 5 epochs, min_lr = 0.000001) callback functions to ensure convergence and stability. The batch size was 128, and the number of iterations was 500. 64 randomly generated samples were used for validation. The total loss curves for the training and validation sets are shown below. Figure 8 As shown.

[0075] Training and validation results show that the mean absolute error (MAE) of flow velocity is 0.03 m / s on both the training and validation sets. The MAE of flow direction is 1.21 on the training set and 0.84 on the validation set. After model deployment, the single prediction time on the host computer is 0.052 seconds, demonstrating good real-time performance.

[0076] The host computer hardware configuration is as follows: ① Processor: Intel(R) Core i9-13900H 2.60GHz, with 14 cores and a base frequency of 2.60GHz; ② Memory: 16GB; ③ Storage: 1TB SSD; ④ Graphics card: Intel(R) Iris(R) Xe Graphics.

[0077] The software configuration is as follows: ① Operating system: Windows 11 64-bit; ② Runtime environment: running in PyCharm, with Python 3.12 as the interpreter; ③ Dependencies: Numpy: 2.1.3, Pandas: 2.2.3, Tensorflow: 2.19.0, Matplotlib: 3.10.3, Keras: 3.9.2.

[0078] As a specific implementation process of this embodiment, the piezoelectric sensor array is installed on the underwater equipment. The piezoelectric sensor adopts the MS5837 sensor. The MS5837 is an ultra-small high-precision pressure sensor suitable for water depth measurement. It can achieve accurate pressure and depth detection at greater water depths. At the same time, it provides a temperature compensation function, which can maintain highly stable readings under different temperature conditions and ensure the reliability of the data.

[0079] The MS5837 supports the IIC communication interface, enabling simple and efficient data transmission. An embedded processor (such as the STM32 series) installed in the pressure chamber reads pressure data from five piezoelectric sensors via the IIC bus and transmits it to a host computer via the RS485 bus. The host computer then runs the trained deep neural network model and outputs the current water flow velocity and direction.

[0080] This invention designs an array structure consisting of five piezoelectric sensors, which can simultaneously achieve accurate measurement of flow velocity and flow direction. This sensor array has the advantages of small size, low cost, high reliability, and is easy to integrate and install.

[0081] This invention employs a CNN-LSTM hybrid neural network architecture for data processing, significantly improving the accuracy and efficiency of data processing. By constructing a training sample library using dual-source data from both simulation and physical experiments, this invention effectively enhances the comprehensiveness and effectiveness of model training, laying a solid foundation for model performance optimization.

[0082] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent detection of water flow direction and velocity using multi-sensor fusion, characterized in that, Includes the following steps: Construct a piezoelectric sensor array based on several piezoelectric sensors; A training dataset is constructed based on the measured data and fluid simulation data of the piezoelectric sensor array. The training dataset is preprocessed to obtain preprocessed data; The preprocessed data is input into a CNN-LSTM hybrid neural network for data processing to obtain the flow direction and velocity of the horizontal surface.

2. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 1, characterized in that, The process of constructing a piezoelectric sensor array includes: A sensor array is constructed based on five piezoelectric sensors. Four of the piezoelectric sensors are arranged horizontally on the same plane. Furthermore, the four piezoelectric sensors are orthogonally distributed along the positive X-axis, negative X-axis, positive Y-axis, and negative Y-axis, respectively, to detect changes in water flow pressure in four orthogonal directions within the horizontal plane. The fifth piezoelectric sensor is arranged vertically in the XY plane to detect the static pressure value in the direction perpendicular to the plane.

3. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 2, characterized in that, The piezoelectric sensor array is connected to the embedded processor via an IIC communication interface, and the embedded processor transmits sensor data to the host computer via an RS485 bus.

4. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 1, characterized in that, The process of constructing the training dataset includes: Fluid simulation data of sensor array under different flow velocities and flow directions were obtained through fluid simulation experiments; The sensor array measured data under different flow velocities and flow directions was obtained through a physical experiment in a wind-wave-flow experimental tank. The measured data and fluid simulation data of the sensor array are merged and processed using data augmentation methods such as random noise addition, random scaling, and random offset to form a training dataset.

5. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 1, characterized in that, The process of preprocessing the training dataset to obtain preprocessed data includes: The training dataset is normalized to form grayscale images; The grayscale images are organized into time series data, with each time series containing 10 sets of sensor pressure value data.

6. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 5, characterized in that, The expression for normalizing the training dataset is: In the formula, P is the original pressure value measured by the sensor; P min The global minimum value among all pressure sensor data; P max P is the global maximum value among all pressure sensor data. norm This is the normalized pressure value.

7. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 6, characterized in that, The CNN-LSTM hybrid neural network includes: a first convolutional layer, a second convolutional layer, a first pooling layer, a second pooling layer, an LSTM layer, a first fully connected layer, and a second fully connected layer; The process of obtaining the flow direction and velocity of the horizontal surface through data processing based on the CNN-LSTM hybrid neural network includes: The grayscale image is input into the first convolutional layer, and the spatial feature map is extracted using a (2,2) convolutional kernel and a ReLU activation function to obtain the output of the first convolutional layer. The first pooling layer performs max pooling on the output of the first convolutional layer using a (2,2) pooling kernel to reduce the dimension of the feature map and obtain the output of the first pooling layer. The output of the first pooling layer is input into the second convolutional layer to extract high-level spatial features and obtain the output of the second convolutional layer. The second pooling layer performs global average pooling on the output of the second convolutional layer to obtain the output of the second pooling layer; The output of the second pooling layer is input into the LSTM layer to output the final time step result; The first fully connected layer output is obtained by performing feature mapping on the final time step result based on the first fully connected layer. The output of the first fully connected layer is input into two independent branches to obtain the predicted flow velocity and the predicted flow direction angle.

8. The intelligent detection method for water flow direction and velocity using multi-sensor fusion according to claim 7, characterized in that, The piezoelectric sensor used is the MS5837 sensor.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent detection method for water flow direction and velocity using multi-sensor fusion as described in claim 1.

10. A storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the intelligent detection method for water flow direction and velocity using multi-sensor fusion as described in claim 1.

Images