Water body motion state recognition method and device based on spectral moment characteristics and readable storage medium thereof

Abstract

The invention provides a water body motion state recognition method and device based on spectral moment characteristics and a readable storage medium thereof, and belongs to the technical field of hydrographic survey and intelligent analysis. According to the method, an acoustic Doppler flow meter is used for collecting a flow velocity sequence, the flow velocity sequence is converted into frequency domain power spectral density through short-time Fourier transform, multi-dimensional features including zero-order to fourth-order spectral moments, frequency band energy, statistics and spectrum features are extracted, a feature time heat map is constructed and input into a multi-scale residual convolutional neural network, and the multi-scale residual convolutional neural network is obtained. And identification of laminar flow, turbulent flow and backflow is realized. The method accurately depicts essential differences of three types of flow states through frequency domain features, retains weak signals in combination with multi-scale convolution kernel and residual connection, solves the problems that a traditional method is poor in interpretability and insensitive in distinguishing, and has the advantages of being high in recognition accuracy, high in robustness and suitable for multiple scenes such as rivers and dams.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological measurement and intelligent analysis, and in particular to a method and device for identifying the motion state of a water body based on spectral moment characteristics, and a readable storage medium thereof. Background Art

[0002] Accurately identifying water flow states (such as laminar flow, turbulent flow, and backflow) is crucial for hydrological monitoring, water conservancy project safety assessment, and urban drainage system regulation. Traditional identification methods rely primarily on temporal statistical characteristics (such as mean and variance) or spatial distribution characteristics of raw flow velocity, using shallow models such as support vector machines (SVM) for classification.

[0003] However, water flow is a typical non-stationary signal, and its core characteristics (such as energy distribution and frequency components) are more reflected in the frequency domain: laminar flow energy is concentrated in low frequencies and has a single peak in the spectrum, turbulent flow spectrum is broadened and the center of gravity frequency shifts upward, and backflow exhibits multi-peak distribution characteristics. Existing technologies have two major limitations because they do not deeply explore frequency domain characteristics: First, the data is poorly interpretable, making it difficult to reveal the spectral nature behind fluid motion; Second, it is insensitive to the frequency domain differences between laminar flow, turbulent flow, and backflow, resulting in insufficient recognition accuracy and robustness, especially in complex water environments (such as high noise interference in urban drainage systems and natural disturbances in rivers).

[0004] Therefore, there is an urgent need for a water motion state identification method based on frequency domain characteristics, strong physical interpretation, and robust to noise to meet the high-precision requirements of actual hydrological monitoring. Summary of the Invention

[0005] The embodiments of the present invention provide a method, device and readable storage medium for identifying the motion state of a water body based on spectral moment characteristics. The existing technologies currently available rely on time domain or spatial characteristics, making it difficult to reveal the spectral characteristics of water body motion, and are insensitive to the frequency domain differences between laminar flow, turbulent flow and backflow, resulting in insufficient recognition accuracy and robustness.

[0006] The core technology of this invention is to collect flow velocity data through an acoustic Doppler current meter (ADCP), convert it into frequency domain power spectral density through short-time Fourier transform (STFT), extract 12-dimensional features such as spectral moment (zero to fourth order), frequency band energy, power gradient, and construct a time heat map, and input it into a convolutional neural network with a multi-scale residual structure (1×1, 3×3, 5×5 convolution kernels parallel feature extraction + residual connection) to achieve accurate identification of laminar flow, turbulent flow, and backflow.

[0007] In a first aspect, the present invention provides a method for identifying water body motion state based on spectral moment features, the method comprising the following steps: Step 1: Use an acoustic Doppler flow meter to collect flow velocity series as input data; Step 2: Convert the time domain signal of the velocity series into frequency domain power spectrum density through frequency domain transformation; Step 3: Extract features from the frequency domain power spectral density. The features include spectral moment features, frequency band energy, statistical features, and spectral features. Spectral moment features include zero-order spectral moment, first-order spectral moment, second-order spectral moment, third-order spectral moment, and fourth-order spectral moment. Frequency band energy includes low-frequency band energy, mid-frequency band energy, and high-frequency band energy. Statistical features include maximum spectral power and spectral power standard deviation. Spectral features include frequency peak position and spectral power gradient. Step 4: Construct the extracted features into a feature vector, and input the feature vector into a preset classifier model. The classifier model is used to identify the current motion state of the water body, which includes laminar flow, turbulent flow, and backflow.

[0008] Furthermore, in step 2, the frequency domain transform is a short-time Fourier transform, and the process of converting the time domain signal into the frequency domain power spectral density through the short-time Fourier transform includes: dividing the original time domain signal into short time windows, performing Fourier transform on each short time window to obtain a time-varying spectrum, averaging the square of the amplitude of each frequency point to estimate the power spectral density, and then averaging the power spectral density along the time axis to obtain the overall power spectral density.

[0009] Furthermore, the window function used by the short-time Fourier transform is a Hanning window.

[0010] Furthermore, in step 3, the zero-order spectral moment is the total energy of the power spectral density, the first-order spectral moment is the spectral centroid, the second-order spectral moment is the dispersion, the third-order spectral moment is the skewness, and the fourth-order spectral moment is the kurtosis.

[0011] Furthermore, in step 3, the low-frequency band energy is the integrated value of the power spectrum density in the low-frequency range, the mid-frequency band energy is the integrated value of the power spectrum density in the mid-frequency range, and the high-frequency band energy is the integrated value of the power spectrum density in the high-frequency range.

[0012] Furthermore, in step 3, the maximum spectral power is the maximum power density value corresponding to a single frequency point in the spectrum, and the spectral power standard deviation is the degree of dispersion of the power spectral density value in the entire frequency range.

[0013] Furthermore, in step 3, the frequency peak position is the frequency point corresponding to the maximum value in the power spectrum density, and the spectrum power gradient is the average rate of change of the power spectrum density along the frequency.

[0014] Furthermore, in step 4, the preset classifier model is a convolutional neural network.

[0015] Furthermore, the convolutional neural network includes a multi-scale residual extraction module, which is composed of multiple multi-scale residual structures connected in series. Multiple parallel convolution branches are set inside each multi-scale residual structure, and the multiple parallel convolution branches respectively use convolution kernels with different receptive fields.

[0016] Furthermore, convolution kernels with different receptive fields include 1×1, 3×3, and 5×5 convolution kernels.

[0017] Furthermore, in each multi-scale residual structure, the outputs of all convolution branches are channel-wise spliced, fused through a 1×1 convolution and batch normalized, and then fused with the input signal using a residual connection.

[0018] Furthermore, in step 4, the feature vectors are arranged in chronological order to form a two-dimensional feature time heat map, and then the feature time heat map is input into the convolutional neural network.

[0019] Furthermore, the identification results of water body motion states include laminar state, turbulent state and backflow state. Among them, the spectrum of the laminar state has a clear single peak, concentrated energy, and skewness close to 0; the spectrum of the turbulent state is broadened, the center of gravity frequency shifts upward, and the skewness is significant; the spectrum of the backflow state has a multi-peak distribution and a high kurtosis.

[0020] In a second aspect, the present invention provides a device for identifying water body motion state based on spectral moment features, comprising: The acquisition module uses an acoustic Doppler flow meter to collect flow velocity sequences as input data; Frequency domain conversion module, which converts the time domain signal of the velocity sequence into frequency domain power spectrum density through frequency domain transformation; The feature extraction module extracts features from the frequency domain power spectral density. The features include spectral moment features, frequency band energy, statistical features, and spectral features. Spectral moment features include zero-order spectral moment, first-order spectral moment, second-order spectral moment, third-order spectral moment, and fourth-order spectral moment. Frequency band energy includes low-frequency band energy, mid-frequency band energy, and high-frequency band energy. Statistical features include maximum spectral power and spectral power standard deviation. Spectral features include frequency peak position and spectral power gradient. The processing module constructs the extracted features into a feature vector and inputs the feature vector into a preset classifier model to identify the current motion state of the water body through the classifier model. The motion state of the water body includes laminar flow, turbulent flow and backflow; Output module, outputs the recognition results of the classifier model.

[0021] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the above-mentioned method for identifying water body motion state based on spectral moment features.

[0022] In a fourth aspect, the present invention provides a readable storage medium storing a computer program, wherein the computer program includes a program code for controlling a process to execute a process, wherein the process includes the water body motion state identification method based on the above-mentioned spectral moment feature.

[0023] The main contributions and innovations of the present invention are as follows: 1. Strong feature interpretation: Based on frequency domain spectral moment characteristics (e.g., the zero-order spectral moment reflects total energy, and the third-order spectral moment reflects skewness), it directly depicts the essential differences between laminar flow (single peak, concentrated energy), turbulent flow (spectral broadening), and recirculation (multiple peaks), solving the problem of poor interpretability of traditional time domain features. 2. High recognition accuracy: The multi-scale residual convolutional neural network uses 1×1, 3×3, and 5×5 convolution kernels to capture frequency domain features at different scales in parallel. Combined with residual connections, it preserves weak signals (such as multi-peak reflux). The F1-score for all three categories on the validation set exceeded 95%, significantly outperforming the traditional time-domain statistics + SVM (which achieved an average F1-score of only 62%). 3. Excellent robustness: It maintains an accuracy rate of 86.5% even at a low signal-to-noise ratio (SNR = 0dB) and is applicable to multiple scenarios such as rivers, dams, and urban drainage (with an accuracy rate of ≥ 93%), with strong resistance to noise and environmental interference. 4. No need for high-density sampling: Frequency domain features can be extracted based on single-point or small-scale ADCP sampling data, reducing dependence on spatial sampling density and being suitable for complex hydrological environments; 5. Efficient and flexible model: End-to-end training does not rely on pre-trained models, the number of parameters is only 3.2M (far lower than AlexNet's 57M), the convergence speed is fast, and it is easy to deploy in engineering.

[0024] The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below so that other features, objects, and advantages of the invention are more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 is a flow chart of a method for identifying water body motion state based on spectral moment features according to an embodiment of the present invention; Figure 2 2. This is a comparison chart of the loss rate of the validation set according to an embodiment of the present invention; Figure 3 is a schematic diagram of a multi-scale residual structure according to an embodiment of the present invention; Figure 4is a block diagram of a neural network structure used in the present invention according to an embodiment of the present invention; Figure 5 is a characteristic temporal heat map of laminar flow according to an embodiment of the present invention; Figure 6 is a characteristic temporal heat map of turbulence according to an embodiment of the present invention; Figure 7 is a characteristic time heat map of reflow according to an embodiment of the present invention; Figure 8 is a schematic diagram of a loss curve of a training process according to an embodiment of the present invention; Figure 9 is a schematic diagram of an accuracy curve of a training process according to an embodiment of the present invention; Figure 10 is a schematic diagram of a confusion matrix according to an embodiment of the present invention; Figure 11 is a schematic diagram of recognition accuracy under different signal-to-noise ratios according to an embodiment of the present invention; Figure 12 FIG. 4 is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The implementations described in the following exemplary embodiments are not intended to represent all implementations consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with certain aspects of one or more embodiments of this specification, as detailed in the appended claims.

[0027] It should be noted that in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than those described in this specification. In addition, a single step described in this specification may be broken down into multiple steps for description in other embodiments, and multiple steps described in this specification may be combined into a single step for description in other embodiments.

[0028] Existing technologies rely on time domain or spatial characteristics, making it difficult to reveal the spectral characteristics of water movement, and are insensitive to the frequency domain differences between laminar flow, turbulent flow, and backflow, resulting in insufficient recognition accuracy and robustness.

[0029] Based on this, the present invention solves the problems existing in the prior art based on deep learning.

[0030] Example 1 The present invention aims to propose a method for identifying the motion state of water bodies based on spectral moment characteristics. Specifically, Figure 1 , the method comprises the following steps: Step 1: Use an acoustic Doppler current meter (ADCP) to collect flow velocity series as input data; In this embodiment, ADCP is based on the acoustic Doppler principle and can measure water flow velocity non-contact in a variety of water environments such as rivers, waterways, dams, and urban drainage systems. The collected flow velocity sequence can reflect the dynamic changes in water movement, providing basic data support for the subsequent conversion of time domain signals into frequency domain features and extraction of key features such as spectral moments. It is the primary data source for achieving accurate identification of water movement status.

[0031] Step 2: Convert the time domain signal of the velocity series into frequency domain power spectrum density through frequency domain transformation; In this embodiment, the time domain signal is converted into frequency domain power spectral density by short-time Fourier transform (STFT); In the short-time Fourier transform, the original time domain signal is divided into short time windows, and the Fourier transform is performed on each segment to obtain the time-varying spectrum. Then, the square of the amplitude of each frequency point is averaged to estimate the power spectral density. The specific formula is:

[0032] in is the original time domain signal; is the window function (Hanning window is selected in this method), and represents a sliding window centered at t; is the sliding window position; Indicates frequency; is the power spectrum estimate of the frequency at that moment, and the power spectrum estimate Averaging along the time axis gives the overall power spectral density (PSD).

[0033] Step 3: Extract features from the frequency domain power spectral density. The features include spectral moment features, frequency band energy, statistical features, and spectral features. Spectral moment features include the zero-order spectral moment (total energy), the first-order spectral moment (spectral center of gravity), the second-order spectral moment (dispersion), the third-order spectral moment (skewness), and the fourth-order spectral moment (kurtosis). Frequency band energy includes low-frequency band energy, mid-frequency band energy, and high-frequency band energy. Statistical features include maximum spectral power and spectral power standard deviation. Spectral features include frequency peak position and spectral power gradient. In this embodiment, the calculation formula of the spectral moment is based on the power spectral density P(f) (i.e., the power distribution per unit frequency). The power spectrum density within is integrated to quantify the energy distribution characteristics of the spectrum, as follows: 1. The formula for the zero-order spectral moment (M0) is:

[0034] It means integrating the power spectral density in the entire frequency range. The result is the total energy (or total power) of the signal, which reflects the total energy of the flow velocity signal at all frequencies.

[0035] 2. The first-order spectral moment (M1, spectral center of gravity) is as follows:

[0036] It is the weighted average of the frequency f with the power spectrum density P(f) as the weight, that is, the spectrum center frequency, which reflects the center frequency position where the spectrum energy is concentrated (the energy is mainly distributed near this frequency).

[0037] 3. The second-order spectral moment (M2) is as follows:

[0038] It measures the degree of dispersion of frequency components relative to the spectral center M1, that is, the spectrum dispersion. The larger the value, the more dispersed the spectrum energy distribution.

[0039] 4. The third-order spectral moment (M3, skewness) is as follows:

[0040] Reflects the asymmetry (skewness) of the spectrum distribution: a positive value indicates that the spectrum is biased towards the high frequency side, a negative value indicates that it is biased towards the low frequency side, and a value close to 0 indicates a symmetrical distribution.

[0041] 5. The fourth-order spectral moment (M4, kurtosis) is as follows:

[0042] Characterizes the sharpness (kurtosis) of the spectrum. A higher value indicates that there are obvious peaks in the spectrum (such as the multi-peak structure of reflux) and the distribution is steeper.

[0043] In summary, these spectral moments comprehensively characterize the frequency domain characteristics of the power spectrum from the dimensions of total energy, concentrated location, dispersion, symmetry and sharpness, and provide a key quantitative basis for distinguishing the motion states of laminar flow, turbulent flow and backflow.

[0044] Step 4: Construct the extracted features into a feature vector and input the feature vector into the preset classifier model (temporal and spatial feature heat map multi-scale residual convolutional neural network model). The classifier model is used to identify the current motion state of the water body, which includes laminar flow, turbulent flow, and backflow.

[0045] In this embodiment, a feature vector is constructed and input into a classifier model (such as SVM, CNN, etc.) to identify the current motion state of the water body.

[0046] Among them, the final feature vector includes: (1) Five spectral moment characteristics ( ), Given by step 3.

[0047] (2) Three frequency band energy ( ), is a quantitative feature of the energy distribution of the power spectral density (PSD) in different frequency ranges, which is used to characterize the energy proportion of the water velocity signal in the low-frequency, medium-frequency and high-frequency bands, as follows: Low frequency energy ( ): refers to the power spectrum density in the low frequency range The integral value within , reflecting the total energy in the low-frequency range. In the laminar state, the spectrum energy is concentrated in the low frequency range, so Usually larger.

[0048] Mid-band energy ( ): refers to the power spectrum density in the intermediate frequency range The integral value within , reflecting the total energy in the mid-frequency range. Due to the multi-peak distribution of the spectrum, the reflux state often has multiple energy peaks in the mid-frequency band. There will be significant fluctuations.

[0049] High frequency energy ( ): refers to the power spectrum density in the high frequency range The integral value within , reflecting the total energy in the high-frequency range. The spectrum of the turbulent state is broadened and the high-frequency components are rich, so Typically above laminar flow.

[0050] (3) Two statistics: maximum spectral power (max_power) and standard deviation of spectral power (std_power).

[0051] Specifically, the maximum spectral power (max_power) is: set up The frequency is The power spectral density value at (unit: power / Hz), then:

[0052] This means taking the maximum value from the power spectral density of all frequency points. Its physical meaning is the power density of the single frequency point with the highest energy in the spectrum, corresponding to the energy intensity of the dominant frequency in the signal. The maximum spectral power represents the maximum power density value corresponding to a single frequency point in the spectrum, reflecting the energy intensity of the main frequency component. It usually corresponds to the location of the dominant frequency of the signal. For laminar flow, the spectrum is concentrated at low frequencies, and the max_power value is large. For turbulent flow or backflow, the spectrum is broadened and the max_power value is reduced.

[0053] Assume a total of frequency points , the corresponding power spectral density is , average power for:

[0054] Specifically, the standard deviation of the power spectrum (std_power) is:

[0055] std_power represents the degree of dispersion (the square root of the variance) of the power spectral density across the entire frequency range, reflecting the flatness or drastic variations of the spectral distribution. For laminar flow, where spectral energy is concentrated, std_power is typically small. For turbulent or backflow flow, where energy distribution is more dispersed, std_power is larger.

[0056] In summary, the eigenvector is:

[0057] Two spectral features (frequency peak position freq_at_max and spectral power gradient power_grad): Where freq_at_max represents the frequency point corresponding to the maximum value in the power spectrum density, that is, the position of the dominant frequency in the signal. It reflects the frequency component that contributes most energy to water movement.

[0058] set up is the frequency The power spectral density at the position of the maxima is at index Department:

[0059] but:

[0060] Here, power_grad represents the average rate of change of the power spectral density along the frequency, which is used to measure the slope trend of the spectrum curve and is an important indicator of whether the spectrum changes smoothly or decays rapidly.

[0061] For laminar flow, freq_at_max is low and the dominant frequency is in the low frequency band. For turbulent flow, freq_at_max is in the mid-frequency band (the spectrum is broadened and the dominant frequency is shifted upward). For backflow, freq_at_max shows a multi-peak distribution characteristic, and the dominant frequency is not obvious or unstable.

[0062] Assume that the frequency sequence is , the corresponding power spectrum density is , then the power gradient is:

[0063] If the sampling is uniform, that is, is constant, it simplifies to:

[0064] For laminar flow, the energy is concentrated in the low frequency, and the energy at the tail of the spectrum is low. is negative and decreases slowly. For turbulence, high-frequency energy decreases quickly, indicating that the spectrum is broadened. is negative and drops steeply. For reflux, the spectrum has multiple peaks, large local changes, and the overall trend is unclear. Fluctuations or approaching zero.

[0065] In this embodiment, by classifying and identifying the spectral moment feature vectors, the water body state can be effectively divided into the following three categories based on the spatiotemporal feature heat map CNN model proposed by this method: (1) Laminar flow state: the spectrum has a distinct single peak, concentrated energy, and skewness close to 0; (2) Turbulent state: spectrum broadening, center of gravity frequency upward shift, and significant skewness; (3) Reflux state: The spectrum has multi-peak distribution and high kurtosis.

[0066] Preferably, in the spatiotemporal feature heat map multi-scale residual convolutional neural network model proposed in this method, the main steps are as follows: Step 1: First, transform the above multidimensional feature vector Arrange them in chronological order to form a two-dimensional matrix, which constitutes a feature time heat map. The horizontal axis is the time index, the vertical axis is the feature dimension index, and the matrix element value is the numerical value under the corresponding time and feature dimension.

[0067] Step 2: In order to improve the classification accuracy of the classical convolutional neural network for spatiotemporal feature heat maps, this method also proposes a multi-scale residual block (MSRB) structure to extract and enhance multi-scale features of the input two-dimensional feature map, which is particularly suitable for the frequency domain heat map structure that represents the motion state of water bodies, such as Figure 3As shown in the figure, BN stands for batch normalization, and ReLU () represents the ReLU activation function. This architecture introduces multiple convolutional channels with different receptive fields (kernel sizes of 1×1, 3×3, and 5×5) to process the same input feature map in parallel, thereby extracting feature information at different spatial scales. Each convolutional branch can capture local (small kernel) and global (large kernel) variation patterns, such as spectral energy concentration, energy diffusion, frequency shift, and other complex features.

[0068] At the output, this structure concatenates and fuses features at each scale in the channel dimension, then compresses the number of channels through a 1×1 convolution operation, keeping the overall output dimension consistent with a single branch. To further optimize information flow and prevent feature degradation, the structure employs a residual connection mechanism, directly adding the input feature map element-by-element to the multi-scale convolution output, resulting in end-to-end information preservation and enhancement. This multi-scale residual structure combines the ability to extract diverse features with the stable training capabilities of deep networks. It exhibits excellent generalization and robustness when processing input data with multi-scale pattern variations, such as water velocity spectra and feature temporal heat maps.

[0069] Step 3: The multi-scale residual convolutional neural network of spatiotemporal feature heatmap mainly includes the following functional modules and structural levels: (3-1) Input layer module This module receives as input a two-dimensional feature heatmap tensor generated by the data preprocessing module. The horizontal axis of the heatmap represents the time step dimension, and the vertical axis represents the feature dimension. Each pixel in the heatmap reflects the quantitative value of a frequency domain feature at a specific time step. The input heatmap has dimensions H × WH × WH × W. Typically, HHH is 12 (corresponding to 12 spectral statistical features) and WWW is 60 (indicating 60 time steps). The input tensor shape is [B, C, H, W], where B is the batch size and C = 1 indicates a single channel.

[0070] To further illustrate the exclusive optimization of the model proposed in this method, the classic AlexNet was used for training and testing on the same dataset. The test data is shown in Table 1 below: Table 1

[0071] It can be seen that under the same input data, the method of the present invention improves the F1-score by about 4.3% compared with AlexNet; the number of model parameters is reduced by about 94.3%, which is significantly lighter; the number of convergence rounds is reduced by more than 60%, and the training efficiency is greatly improved.

[0072] (3-2) Multi-scale residual extraction module (backbone feature extractor) This module, the core component of the present invention, is responsible for extracting multi-level, multi-scale spectral variation features from the input heat map. It consists of multiple multi-scale residual blocks (MSRBs) connected in series. Each MSRB module contains multiple parallel convolution branches, each employing convolution kernels with different receptive fields (e.g., 1×1, 3×3, and 5×5). These branches are used to extract features such as local edge changes, energy concentrations, and spectral mutation trends within different spatial perception ranges. The outputs of all branches are concatenated through channels, fused through a 1×1 convolution, and batch normalized. They are then fused with the input signal using a residual connection to preserve low-level features while introducing high-level feature expression capabilities.

[0073] Multiple MSRBs are stacked sequentially, and pooling operations are inserted in between to gradually compress the image space dimensions while increasing the number of channels (i.e., feature dimensions), thereby achieving deeper semantic understanding.

[0074] An example hierarchy is: MSRB1 (input channel 1 → output channel 16) + maximum pooling; MSRB2 (16→32) + max pooling; MSRB3 (32→64) + max pooling.

[0075] Preferably, to illustrate the irreplaceable nature of multi-scale design, the frequency domain feature time heat map constructed by ADCP is used as input, and convolutional neural networks (CNNs) with single convolution kernel sizes (1×1, 3×3, and 5×5) are constructed to classify and identify three types of flow regimes: laminar flow, turbulent flow, and recirculation. The model structure is shown in Table 2: Table 2

[0076] The final experimental results are compared in Table 3: Table 3

[0077] In practical applications, the spectral characteristics of different flow states have structural differences, and a single convolution kernel is difficult to adapt to these three different characteristic patterns at the same time. Experimental comparisons found that: ① Advantages and limitations of 1×1 convolution kernel: The 1×1 convolution kernel has a minimal receptive field and can only capture intensity changes at a single point. This makes it particularly well-suited for identifying single-peak structures with concentrated energy in low-frequency regions of laminar flow. In experiments, its accuracy for laminar flow recognition reached 93.2%, but its ability to identify turbulent flow and backflow was weaker, at only 68.4% and 72.3%, respectively, demonstrating a lack of awareness of changes in spectral structure.

[0078] ②Sensitivity of 3×3 convolution kernel to reflow structure: The 3×3 convolutional kernel has a medium receptive field and can simultaneously perceive relative changes between two or three local frequency bands, demonstrating excellent recognition of multi-peak oscillation structures in the recirculation spectrum. Its recognition accuracy reaches 91.2%, outperforming other single-scale networks. It also exhibits balanced performance for both laminar and turbulent flows, demonstrating its excellent local information fusion capabilities.

[0079] ③5×5 convolution kernel’s ability to perceive turbulent energy diffusion: The 5×5 convolution kernel has a large receptive field and is suitable for extracting global diffusion features such as spectral broadening and sideband enhancement in turbulence. Its accuracy in identifying turbulence reaches 92.4%, but it performs poorly in laminar flow scenarios, misinterpreting concentrated spectral energy as disturbances, resulting in an accuracy of only 82.1%.

[0080] The above results show that: A single-scale convolution kernel has obvious target bias; laminar flow requires a 1×1 kernel to extract the concentrated main peak energy; turbulent flow requires a 5×5 kernel to integrate and broaden the structure; backflow requires a 3×3 kernel to analyze the intermediate frequency disturbance and multi-peak information; any kernel size is difficult to take into account the three types of spectral characteristics, resulting in significant performance loss.

[0081] Therefore, the multi-scale convolutional fusion structure (1×1+3×3+5×5) proposed in this paper can extract multiple scale feature responses in parallel under a single feature heat map input, and through subsequent feature fusion and residual enhancement mechanism, significantly improve the model's recognition accuracy and robustness for complex water body spectral structures.

[0082] (3-3) Classification mapping module (fully connected layer) The multi-scale residual feature map extracted by the convolution module is flattened and sent to the fully connected layer. This part usually includes: A high-dimensional fully connected layer, responsible for integrating multi-channel convolutional features; Dropout random deactivation mechanism to prevent overfitting; The output layer is a fully connected layer with Softmax activation, and the output dimension is the number of predefined classification categories (3 types of flow state).

[0083] (3-4) Output module The network output is the predicted probability distribution of each input heat map sample in each category, which can be used to make the maximum probability judgment, thereby realizing the automatic classification of water body status.

[0084] The proposed neural network structure diagram is as follows Figure 4 shown.

[0085] Step 4. Neural network training method and usage process: (a) Training data preparation Perform short-time Fourier transform (STFT) on the original flow velocity signal; extract frequency domain statistical features (such as spectral moment, energy ratio, maximum spectral power, frequency offset, etc.) to form a multidimensional feature vector; after extracting features at each moment, combine them into a feature time heat map in chronological order to form a two-dimensional tensor sample; mark the flow state label corresponding to each sample and use it as a training supervision signal.

[0086] (b) Model initialization Initialize neural network parameters, including convolution kernel weights, BN parameters, attention module weights, etc.; define the loss function (such as cross entropy loss); select the optimizer (such as Adam) and learning rate strategy.

[0087] (c) Training iterations During each round of training, a batch of samples is fed into the network; it passes through the multi-scale residual module, pooling module, and fully connected module in sequence to output the predicted probability; the loss value between the predicted result and the actual label is calculated; the network parameters are updated through the backpropagation algorithm; and the iteration is repeated until the loss converges or the accuracy meets the requirements.

[0088] (d) Model evaluation and inference After training is complete, the model is placed in evaluation mode and fed with new heatmap sample data for recognition. The output category prediction results can be used for further flow regime identification, hydraulic monitoring, or control decision-making systems.

[0089] In this embodiment, in order to illustrate the effectiveness of the method proposed by the present invention, an application example analysis is performed as follows: Assume that the original flow velocity data is obtained by an ADCP installed on a river. After using this method, the water movement state is identified based on the collected flow velocity data, and the results are compared and analyzed with the expert judgment results to illustrate the effectiveness of this method. The specific data are as follows: The total data level consists of 1 million sampling moments of data, which are manually labeled and divided into three water flow states: laminar flow, turbulent flow, and backflow. Each set of data has 5,000 items. Some of the data are plotted as characteristic time heat maps as shown below. Figure 5-7 As shown, the loss curve of the training process is as follows Figure 8 As shown, the accuracy curve is as Figure 9 shown.

[0090] After training with the method of the present invention, the accuracy of various indicators on the validation set is shown in Table 4 below: Table 4

[0091] The confusion matrix is ​​as follows Figure 10As shown, by combining the above indicators and confusion matrix, it can be obtained that the method of the present invention can accurately distinguish the three states and is effective and feasible in actual applications.

[0092] Preferably, in order to illustrate the advancement of this method over traditional methods, the following two classification methods are compared in terms of their effectiveness in identifying water body motion states (laminar flow / turbulent flow / backflow) as shown in Table 5: Table 5

[0093] The experimental results are compared in Table 6: Table 6

[0094] The comparative analysis of experimental results reveals the limitations of traditional time domain methods: 1. Weak feature expression ability: Time-domain statistics (such as mean and standard deviation) only capture the overall signal trend and cannot describe the spectral structure. For reflux, which has a multi-peaked and dispersed spectrum, the F1-score is only 50%.

[0095] 2. Serious information loss: The differences in high-frequency / low-frequency / medium-frequency disturbances in water bodies, especially the key spectral line shapes such as energy mutation points, are ignored.

[0096] 3. Limited model learning ability: SVM is a shallow model that can only perform classification under linear or kernel function mapping and has poor perception of nonlinear spectral structures.

[0097] The significant advantages of the method of the present invention are: 1. Automatic multi-scale feature extraction: Multi-scale convolution kernels (1×1, 3×3, 5×5) can respectively learn spectral features such as local peaks, sideband broadening, and multi-peak oscillations; they are particularly suitable for distinguishing laminar flow (main peak concentration), turbulent flow (spectral width dispersion), and backflow (multi-peak disturbance) states.

[0098] 2. Residual connection enhances weak feature retention: It effectively solves the problem of "weak peak energy being submerged by the mainstream" such as backflow; shallow low-frequency signals can be directly transmitted to deep layers, improving discrimination.

[0099] 3. Better training stability and generalization ability: The multi-scale structure avoids overfitting at a single scale; mechanisms such as BatchNorm and Dropout ensure that the network has good adaptability to new samples.

[0100] In this embodiment, in order to verify the anti-noise performance of the present invention, as shown in FIG. Figure 11 As shown in the figure, under different signal-to-noise ratio conditions, the proposed method always maintains a high classification accuracy, especially when SNR=0dB, it still reaches an accuracy of 86.5%, which is much higher than the underwater acoustic signal model in Document 2 (only 60.4%). This fully demonstrates the following key advantages: The multi-scale convolutional design forms a complementary structure that effectively models key features across a wide range of perturbation frequency distributions, demonstrating excellent information redundancy and structural robustness. Detailed information representing water characteristics in the original spectrum (such as local peaks and valleys) is easily overwhelmed by noise. Residual connections allow shallow-level features (such as areas of concentrated low-frequency energy) to propagate directly to higher levels, effectively preserving the original spectral structure.

[0101] In this example, to verify whether the method of the present invention is applicable to actual water scenarios with different physical conditions, flow rates, and boundary interference strengths, rather than being limited to a specific test tank or experimental environment, the following experiments were conducted: (1) Test scenario and description: River scenario: Open river basins with natural laminar flow, bank backflow, and turbulence after rain. The water body is greatly affected by wind speed disturbances, and the signal-to-noise ratio is unstable.

[0102] Dam spillway: high velocity, obvious turbulence, common high-frequency scattering disturbances; backflow intervals and downflow waves repeatedly interweave.

[0103] Urban drainage system: The pipeline structure is complex, with obvious switching between "intermittent laminar flow-backflow-interference flow"; the channel space is narrow and there is a lot of interference.

[0104] (2) Summary of recognition accuracy results The recognition accuracy in each scenario is shown in Table 7: Table 7

[0105] The analysis of the above recognition accuracy results shows that the accuracy rate remains stable at above 93% in all scenarios. In particular, in urban drainage systems, this method still has high robustness despite the signal being interfered with by factors such as high noise and echo multipath.

[0106] Example 2 Based on the same concept, the present invention also proposes a water body motion state recognition device based on spectral moment characteristics, comprising: The acquisition module uses an acoustic Doppler flow meter to collect flow velocity sequences as input data; Frequency domain conversion module, which converts the time domain signal of the velocity sequence into frequency domain power spectrum density through frequency domain transformation; The feature extraction module extracts features from the frequency domain power spectral density. The features include spectral moment features, frequency band energy, statistical features, and spectral features. Spectral moment features include zero-order spectral moment, first-order spectral moment, second-order spectral moment, third-order spectral moment, and fourth-order spectral moment. Frequency band energy includes low-frequency band energy, mid-frequency band energy, and high-frequency band energy. Statistical features include maximum spectral power and spectral power standard deviation. Spectral features include frequency peak position and spectral power gradient. The processing module constructs the extracted features into a feature vector and inputs the feature vector into a preset classifier model to identify the current motion state of the water body through the classifier model. The motion state of the water body includes laminar flow, turbulent flow and backflow; Output module, outputs the recognition results of the classifier model.

[0107] Preferably, in order to solve the problem of "weak signal drowning" of water characteristics (such as the energy dispersion of multi-peak signals of backflow, residual connection can retain shallow low-frequency characteristics). The present invention solves the problem of "weak signal drowning" through the following three methods: (1) Retain low-frequency and low-amplitude characteristics: Weak signals, such as the multi-peak spectrum of the reflux state, often have weak oscillations in the low-frequency region of the spectrum. These features can be well detected in the initial convolution layer (smaller convolution kernels or 1×1 convolutions are sensitive to local peaks), but after multiple downsampling (pooling) and high-order convolutions, this information often disappears.

[0108] Residual connections can directly add and inject these shallow low-frequency features into deep outputs, preserving the original information of these features in space, providing multi-perspective references in the channel dimension, and solving the attenuation of information along the propagation path.

[0109] (2) Avoid gradient vanishing and strengthen the gradient signal to penetrate weak feature paths: From a training perspective, the residual connection forms an "identity mapping path", which allows the gradients of weak feature areas to be directly transmitted back to the previous layer without nonlinear transformation, thereby increasing the probability of updating the weights of these areas. The network with residual structure is more robust to learning low-amplitude inputs (such as background changes and small peaks).

[0110] This is extremely critical for those weak multi-peak forms that are "masked" by high-power mainstream signals in the early stages of training, and can ensure that the model does not fall into the local optimum of "only learning strong signals".

[0111] (3) Multi-scale perspective enhancement of detail expression: In the multi-scale residual structure of the present invention, convolution branches of different scales are responsible for capturing: 1×1 convolution: sensitive to pixel-level energy (difference in strength); 3×3 convolution: local fluctuation features (such as two neighboring frequency peaks); 5×5 convolution: medium frequency structural changes (such as the main / secondary peak spacing of the reflow spectrum); By retaining the overall input information through residual connections and then fusing these scale features, the spectral position information of multi-peak weak signals can be effectively retained, avoiding the problems of "being averaged" and "being compressed".

[0112] Preferably, in order to verify the comparison of model loss curves with and without residual connections, such as Figure 2 As shown in Figure 2, it can be seen that with residual connections, the verification set accuracy is improved by 6.35%.

[0113] Example 3 This embodiment also provides an electronic device, referring to Figure 12 , includes a memory 404 and a processor 402, wherein the memory 404 stores a computer program, and the processor 402 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0114] Specifically, the processor 402 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits for implementing the embodiments of the present invention.

[0115] Memory 404 may include a large-capacity memory 404 for data or instructions. By way of example, and not limitation, memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 404 may include removable or non-removable (or fixed) media. Where appropriate, memory 404 may be internal or external to the data processing device. In certain embodiments, memory 404 is non-volatile memory. In certain embodiments, memory 404 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. In appropriate circumstances, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM may be a fast page mode dynamic random access memory 404 (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0116] The memory 404 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 402 .

[0117] The processor 402 reads and executes computer program instructions stored in the memory 404 to implement any one of the water body motion state recognition methods based on spectral moment features in the above embodiments.

[0118] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408 , wherein the transmission device 406 is connected to the processor 402 , and the input / output device 408 is connected to the processor 402 .

[0119] Transmission device 406 can be used to receive or transmit data via a network. Specific examples of such networks may include wired or wireless networks provided by the electronic device's communications provider. In one embodiment, the transmission device includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission device 406 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0120] The input / output device 408 is used to input or output information.

[0121] Example 4 This embodiment also provides a readable storage medium, in which a computer program is stored. The computer program includes a program code for controlling a process to execute a process, and the process includes the water body motion state identification method based on spectral moment features according to the first embodiment.

[0122] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be repeated here.

[0123] In general, various embodiments may be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention may be implemented in hardware, while other aspects may be implemented in firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flow charts, or using some other graphical representation, it should be understood that, as non-limiting examples, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0124] Embodiments of the present invention can be implemented by computer software, which is executable by the data processor of the mobile device, such as in the processor entity, or is implemented by hardware, or is implemented by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets and / or macros can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. The computer program product can include one or more computer executable components configured to perform the embodiment when the program is running. One or more computer executable components can be at least one software code or a part thereof. In addition, at this point, it should be noted that any box of the logic flow in the figure can represent a program step, or interconnected logical circuits, boxes and functions, or a combination of program steps and logical circuits, boxes and functions. The software can be stored in physical media such as memory chips or storage blocks implemented in the processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs. Physical media is non-transient media.

[0125] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0126] The above embodiments merely illustrate several embodiments of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of the present invention. Therefore, the scope of the present invention shall be determined by the appended claims.

Claims

1. The water body motion state recognition method based on spectral moment characteristics is characterized by: The following steps are involved: Step 1: Use an acoustic Doppler flow meter to collect flow velocity series as input data; Step 2: Convert the time domain signal of the velocity sequence into frequency domain power spectrum density through frequency domain transformation; Step 3: extracting features from the frequency domain power spectral density, the features including spectral moment features, frequency band energy, statistical features and spectral features, wherein the spectral moment features include zero-order spectral moment, first-order spectral moment, second-order spectral moment, third-order spectral moment and fourth-order spectral moment, the frequency band energy includes low-frequency band energy, mid-frequency band energy and high-frequency band energy, the statistical features include maximum spectral power and spectral power standard deviation, and the spectral features include frequency peak position and spectral power gradient; Step 4: Construct the extracted features into a feature vector, and input the feature vector into a preset classifier model, and use the classifier model to identify the current motion state of the water body, where the motion state of the water body includes laminar flow, turbulent flow and backflow.

2. The method for identifying water body motion state based on spectral moment features according to claim 1, characterized in that: In step 2, the frequency domain transformation is short-time Fourier transform, and the process of converting the time domain signal into frequency domain power spectral density by the short-time Fourier transform includes: The original time domain signal is divided into short time windows, and a Fourier transform is performed on each short time window to obtain a time-varying spectrum. The square of the amplitude of each frequency point is averaged to estimate the power spectral density, and then the power spectral density is averaged along the time axis to obtain the overall power spectral density.

3. The method for identifying water body motion state based on spectral moment features according to claim 1, characterized in that: In step 3, the zero-order spectral moment is the total energy of the power spectral density, the first-order spectral moment is the spectral centroid, the second-order spectral moment is the dispersion, the third-order spectral moment is the skewness, and the fourth-order spectral moment is the kurtosis.

4. The method for identifying water body motion state based on spectral moment characteristics according to claim 1, characterized in that: In step 3, the low-frequency band energy is the integrated value of the power spectrum density in the low-frequency range, the mid-frequency band energy is the integrated value of the power spectrum density in the mid-frequency range, and the high-frequency band energy is the integrated value of the power spectrum density in the high-frequency range; The maximum spectral power is the maximum power density value corresponding to a single frequency point in the spectrum, and the spectral power standard deviation is the degree of dispersion of the power spectral density value in the entire frequency range; The frequency peak position is the frequency point corresponding to the maximum value in the power spectrum density, and the spectrum power gradient is the average rate of change of the power spectrum density along the frequency.

5. The method for identifying water body motion state based on spectral moment characteristics according to claim 1, characterized in that: In step 4, the preset classifier model is a convolutional neural network, which includes a multi-scale residual extraction module. The multi-scale residual extraction module is composed of multiple multi-scale residual structures connected in series. Multiple parallel convolution branches are set inside each multi-scale residual structure, and the multiple parallel convolution branches respectively use convolution kernels with different receptive fields.

6. The method for identifying water body motion state based on spectral moment characteristics according to claim 5, characterized in that: The convolution kernels of different receptive fields include 1×1, 3×3 and 5×5 convolution kernels; In each of the multi-scale residual structures, the outputs of all convolution branches are channel-joined, fused through a 1×1 convolution, and batch normalized, and then fused with the input signal using a residual connection.

7. The method for identifying water body motion state based on spectral moment features according to any one of claims 1 to 6, characterized in that: The identification results of the water body movement state include laminar state, turbulent state and backflow state. Among them, the spectrum of the laminar state has a clear single peak, concentrated energy, and a skewness close to 0; the spectrum of the turbulent state is broadened, the center of gravity frequency shifts upward, and the skewness is significant; the spectrum of the backflow state has a multi-peak distribution and a high kurtosis.

8. A water body motion state recognition device based on spectral moment characteristics, characterized in that: include: The acquisition module uses an acoustic Doppler flow meter to collect flow velocity sequences as input data; A frequency domain conversion module converts the time domain signal of the flow velocity sequence into a frequency domain power spectrum density through frequency domain transformation; a feature extraction module, extracting features from the frequency domain power spectral density, wherein the features include spectral moment features, frequency band energy, statistical features, and spectral features, wherein the spectral moment features include zero-order spectral moment, first-order spectral moment, second-order spectral moment, third-order spectral moment, and fourth-order spectral moment, the frequency band energy includes low-frequency band energy, mid-frequency band energy, and high-frequency band energy, the statistical features include maximum spectral power and spectral power standard deviation, and the spectral features include frequency peak position and spectral power gradient; a processing module, constructing the extracted features into a feature vector, and inputting the feature vector into a preset classifier model, and identifying the current motion state of the water body through the classifier model, wherein the motion state of the water body includes laminar flow, turbulent flow and backflow; Output module, outputs the recognition results of the classifier model.

9. An electronic device comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is configured to run the computer program to execute the method for identifying water body motion state based on spectral moment features according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes a program code for controlling a process to execute a process, wherein the process includes the method for identifying water body motion state based on spectral moment features according to any one of claims 1 to 7.

Images