A method and apparatus for detecting fluid concentration in pipelines based on millimeter-wave radar

Abstract

The application discloses a pipeline fluid concentration detection method and device based on a millimeter wave radar, emits a linear frequency modulation signal to pipeline fluid at a specific distance through the millimeter wave radar, and obtains data of reflected intermediate frequency signals. The obtained intermediate frequency signal information is subjected to fast Fourier transform processing to obtain liquid and reflection surface position information. A neural network pre-trained for specific liquid concentration recognition extracts features from the expanded radar data, and simultaneously performs attention mechanism processing on the liquid and reflection surface information at a feature level to realize fine detection of liquid concentration in the whole range. The liquid concentration in the pipeline fluid can be accurately reflected. Meanwhile, information is transmitted to a management terminal through a specific data protocol.

Description

Technical Field

[0001] This invention relates to the field of pipeline fluid detection, and in particular to a method and apparatus for detecting pipeline fluid concentration based on millimeter-wave radar. Background Technology

[0002] With the continuous development of industrial technology, the demand for liquid detection technology in fields such as chemical, petroleum, and pharmaceutical industries is increasing. Liquid detection in these fields typically involves the accurate measurement and monitoring of fluid concentration. Traditional methods for liquid concentration detection include chemical analysis, conductivity measurement, and optical measurement. These methods rely on the photoelectric properties and various chemical properties of liquids and usually require sophisticated instruments such as Raman analyzers. However, these specialized instruments are expensive and complex to operate, limiting their widespread application in large-scale production. Furthermore, traditional methods are often offline, exhibiting latency and making real-time adjustments during production difficult. This detection approach lacks the real-time judgment capability at multiple points in systems engineering, typically only performing intermittent detection at certain points, thus affecting the efficiency and accuracy of production monitoring.

[0003] In recent years, with the continuous development of wireless sensing technology, detecting liquid properties by analyzing changes in wireless signals after they penetrate the liquid has become a fast, convenient, and low-cost liquid detection solution. Its non-contact measurement characteristic, in particular, has attracted increasing attention from researchers. The application of wireless sensing technology in liquid detection overcomes many limitations of traditional methods, enabling rapid detection without contact with the liquid, thus reducing detection costs and complexity.

[0004] However, current wireless sensing technologies primarily focus on liquid type identification, with limited research on liquid concentration detection. Furthermore, existing methods still face challenges in terms of accuracy and robustness; for example, shifts in liquid position can lead to unstable recognition results. These issues limit the effectiveness of wireless sensing technologies in liquid concentration detection. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and device for detecting pipeline fluid concentration based on millimeter-wave radar.

[0006] The objective of this invention is achieved through the following technical solution: a method for detecting fluid concentration in pipelines based on millimeter-wave radar, the method comprising the following steps:

[0007] S1. In a closed environment, a millimeter-wave radar is used to transmit linear frequency modulated signals to both ends of the sight glass of the pipe to be inspected, and the reflected signals are received to obtain raw data. Based on the raw data, radar information with strong correlation between liquid and reflective surface reflection is extracted.

[0008] S2. Construct a concentration dataset of the full range of pipeline fluids; classify them according to the minimum concentration to be detected and generate corresponding real labels;

[0009] S3. The neural network for recognizing concentrations across the entire range is trained using a concentration dataset. The neural network performs one-dimensional feature extraction on strongly correlated radar information, extracts high-dimensional features from the original data using a complex domain multilayer perceptron, performs feature fusion using an attention mechanism, and decodes the fused features to obtain the concentration prediction result.

[0010] S4. Input the raw data and strongly correlated radar information into the trained neural network that identifies concentrations across the entire range to obtain concentration prediction results.

[0011] Furthermore, the method of using millimeter-wave radar to transmit linear frequency modulated signals to both ends of the sight glass of the pipe to be detected in a closed environment specifically includes: installing a box-type structure at both ends of the liquid sight glass of the pipe to transmit linear frequency modulated signals. The box-type structure is equipped with millimeter-wave radar, corresponding power adapter and embedded equipment. The transceiver antenna of the millimeter-wave radar is exposed in the field of view of the sight glass, and the other parts are isolated by metal plates.

[0012] Furthermore, acquiring the raw data includes mixing the received reflected signal using a mixer to generate the raw data for the intermediate frequency signal seating radar.

[0013] Furthermore, the extraction of strongly correlated radar information between liquid and reflective surface reflection includes: performing a fast Fourier transform on the original data to obtain a frequency domain signal; sorting the frequency domain signal intensities within a set distance range to obtain three peak signals as the radar signal that first reaches the sight mirror surface, the radar signal that first penetrates the sight mirror, and the radar signal that first reaches the reflective surface; calculating the positions of the three target points based on the subscript of the signal location and the radar's range resolution; using the distance of the millimeter-wave radar to the reflective surface as a reference, dividing the entire radar information into multiple segments; and stacking the three signals in the same dimension within these segments to obtain strongly correlated radar information between liquid and reflective surface reflection.

[0014] Furthermore, the neural network for recognizing concentrations across the entire range uses modules from a deep complex network model as the basic modules in the network, including: a data expansion layer, a feature extraction layer, an attention mechanism, and a decoding part;

[0015] The data expansion layer expands the radar data in the original data from two dimensions to three dimensions by copying, so that the original data changes from the number of antennas × number of samples to the number of antennas × number of samples × number of samples.

[0016] The feature extraction layer includes a multilayer perceptron for high-dimensional feature extraction of the radar's raw data, and a one-dimensional perceptron for information extraction of radar information with strong correlation between liquid and reflective surface reflections.

[0017] In the attention mechanism, the high-dimensional feature part of the original radar data is divided into H×1×C and 1×W×C parts, which are respectively combined with the reflection of the reflector to form strongly correlated radar information features. The multi-head attention mechanism is then used to perform dimension mapping, mapping to the required number of attention heads, and then broadcasting the results. The two results are then expanded to the same dimension and added together.

[0018] The decoding part includes multiple complex fully connected layers to reduce the dimensionality of the result to one dimension, then transforming the result in the complex domain to the real domain by taking the modulus, and finally obtaining the probability of the corresponding concentration gradient through the softmax function.

[0019] Furthermore, the multilayer perceptron model consists of several 2D 3*3 complex convolution modules, complex activation function modules, complex pooling modules, and complex batch normalization modules.

[0020] Furthermore, the complex activation function module employs the CReLU function, and the complex batch standardization module performs a 0-mean, 1-variance operation on the input:

[0021]

[0022] Where x is the input vector and V is the correlation matrix, the correlation matrix V is calculated as follows:

[0023]

[0024] The final BN has two more parameters, γ and β, therefore the final batch normalization result is as follows.

[0025]

[0026] Furthermore, the cross-entropy function is selected as the loss function during the training process, and its calculation method is as follows:

[0027]

[0028] Where N is the number of samples, C is the number of categories obtained based on the minimum concentration resolution, and x y Let x be the probability that the i-th sample belongs to the true class y. c Let c be the probability of the i-th sample data category c.

[0029] On the other hand, the present invention also provides a pipeline fluid concentration detection device based on millimeter-wave radar, including a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, it implements the pipeline fluid concentration detection method based on millimeter-wave radar.

[0030] On the other hand, this specification also provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the aforementioned method for detecting pipeline fluid concentration based on millimeter-wave radar.

[0031] The beneficial effects of this invention are as follows: The pipeline fluid concentration detection method and system based on millimeter-wave radar described in this invention utilizes a fully enclosed metal external detection environment system. The system is externally attached to both ends of the liquid pipeline to be detected, forming a closed metal environment. The millimeter-wave radar emits a linear frequency modulated signal within this enclosed metal environment. The signal penetrates the liquid and reaches the metal reflective surface, undergoing multiple reflections. This process yields data on the millimeter wave's multiple penetrations through the liquid and its arrival at the reflective surface, significantly increasing the proportion of effective data received. At the data level, the accuracy and robustness of the neural network are improved by superimposing the data from multiple penetrations. Furthermore, the data dimension is expanded through replication, transforming two-dimensional data into three-dimensional data, enabling the network to undergo more complex training.

[0032] The neural network model established in this invention for detecting fluid concentration in pipelines primarily utilizes complex modules within a deep complex network to efficiently and rationally leverage radar intermediate frequency complex data. A perceptron model composed of complex modules serves as a feature extraction module to extract data from the original data. An attention mechanism is used to train the high-dimensional features of the original data, combining data from multiple penetrations of the liquid and the reflection surface by millimeter waves with those from the original data, through a parameter matrix. This achieves long-distance sequential feature fusion and correlation fusion. This approach enables refined, full-range, stepped identification of liquid concentration, capable of recognizing data within a 0.5% concentration range. Furthermore, it can identify concentrations not included in the network training as corresponding stepped concentrations, resulting in more stable and smooth concentration recognition.

[0033] The present invention relates to a pipeline fluid concentration detection method and system based on millimeter-wave radar. This system combines millimeter-wave radar with an embedded system and an external system, allowing for simple and efficient installation on the outside of the sight glass of the pipeline containing the liquid to be detected. It is easy to install and remove, and online detection avoids the cumbersome process and complex training required for manual inspection. It is easy to learn, has a wide range of applications, and is particularly effective in detecting high-risk liquid concentrations, preventing potential safety hazards. Attached Figure Description

[0034] Figure 1A flowchart illustrating the main steps of a pipeline fluid concentration detection method based on millimeter-wave radar provided in an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram of the fully enclosed metal texture detection environment system described in this invention;

[0036] Figure 3 This is a schematic diagram of radar signal echoes in liquids of different concentrations. Taking 1% and 10% saline solutions as examples, there are eight echo curves representing the dimension data of the transmitting antenna and the receiving antenna.

[0037] Figure 4 This is a schematic diagram of the neural network structure described in this invention;

[0038] Figure 5 This is a schematic diagram of the attention mechanism module using multi-feature fusion in the neural network of this invention;

[0039] Figure 6 This is a schematic diagram illustrating the accuracy of the present invention at 0.5% saline concentration intervals.

[0040] Figure 7 This is a schematic diagram illustrating the accuracy of the present invention with 1% concentration intervals as training samples and 0%-1% random concentration intervals as test samples, wherein the accuracy is calculated by rounding the concentrations in the test set.

[0041] Figure 8 This is a schematic diagram of a pipeline fluid concentration detection device based on millimeter-wave radar provided in an embodiment of the present invention. Detailed Implementation

[0042] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0043] like Figure 1 As shown, the present invention provides a method and apparatus for detecting pipeline fluid concentration based on millimeter-wave radar.

[0044] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention are merely one application example and are not intended to limit the scope of the invention.

[0045] First, linear frequency modulated signals are transmitted towards the sight glass via multiple antennas of the millimeter-wave radar. In this embodiment, a special structure is used for signal transmission and reception. Two symmetrical box-type structures are tightly connected to both ends of the sight glass of the pipe liquid to be detected by screws, thus ensuring a closed detection environment. One side of the system's housing contains the millimeter-wave radar, corresponding power adapter, and embedded equipment, with space reserved for the millimeter-wave radar housing. The other side of the housing has a metal plate as a millimeter-wave reflector. The transceiver antennas of the millimeter-wave radar are exposed within the sight glass's field of view, while other parts are isolated by the metal plate to reflect the millimeter waves. One side of the embedded plate is coated with insulating material to prevent short circuits. To ensure the system's corrosion resistance and durability, 316 austenitic stainless steel, widely used in the chemical industry, is selected. This stainless steel not only possesses excellent corrosion resistance but can also be used for extended periods in harsh industrial environments, ensuring the stability and reliability of the detection system. Specific embodiments are available in [link to specific embodiments]. Figure 2 .

[0046] The reflected signal is received through multiple receiving antennas. The received signal is mixed by a mixer to generate an intermediate frequency (IF) signal, which serves as the raw data for the reflected signal, i.e., the time-domain signal of the reflection. To demonstrate that the reflected time-domain signal exhibits different results for liquids of different concentrations, [further details are needed]. Figure 3 The diagram shows the echo pattern of radar signals under different liquid concentrations. Figure 3 The left side shows the echo data for a 1% saline solution. Figure 3 The right side shows the echo data of a 10% saline solution, with eight echo curves representing the dimension data of the transmitting antenna and the receiving antenna. It can be seen that the reflected signals differ under different saline concentrations, which is the basis for this invention.

[0047] A Fast Fourier Transform (FFT) is performed on the time-domain signal to obtain the frequency-domain signal. Different peak regions in the frequency-domain signal represent the locations of objects detected by the radar. In practice, three peak signals are considered: the radar signal that first reaches the sight mirror surface, the radar signal that first penetrates the sight mirror, and the radar signal that first reaches the reflecting surface. The specific steps are as follows: First, the frequency-domain signal strengths within a set distance range are sorted, and the locations of the three strongest signal values ​​are selected. These locations are the three signal locations of interest, with the strongest signal generally originating from the reflecting surface. Based on the index of the signal location and the radar's range resolution, the locations of the three target points are calculated. Using the distance from the millimeter-wave radar to the reflecting surface as a reference, the entire radar information is divided into multiple segments. Within these segments, radar signals that repeatedly reach the sight mirror surface, repeatedly penetrate the sight mirror, and repeatedly reach the reflecting surface are stacked on the same dimension to obtain the main liquid information data. Due to the limitations of millimeter-wave radar data resolution, the location may not be precise. In operation, it is often necessary to stack 1 to 2 additional sampling points to improve accuracy and robustness.

[0048] A concentration dataset of the full range of pipeline fluids was constructed using the method described above. The data was classified according to the minimum concentration required for detection, and corresponding ground truth labels were generated. When constructing the dataset, radar data from a single chirp was used as the raw input data. The original data file was saved during sample set construction, while the specific liquid data information—including radar signals arriving at the sight glass surface multiple times, radar signals penetrating the sight glass multiple times, and radar signals arriving at the reflector multiple times—was loaded via parsing during data loading.

[0049] Considering that the raw data from a single chirp can only form a two-dimensional dataset of (number of transmit antennas × number of receive antennas) × number of samples, only one-dimensional convolution operations can be performed in neural networks, making it difficult to perform deeper feature mining. Therefore, after loading radar data, the raw data needs to be augmented. By copying, the raw radar data is augmented into a three-dimensional dataset of (number of transmit antennas × number of receive antennas) × number of samples × number of samples. This augmentation method allows for general two-dimensional convolution operations in neural networks, thereby overcoming the limitations of network depth and enhancing the network's fitting and learning capabilities.

[0050] For the neural network model, its basic architecture primarily employs the fundamental structure of deep complex networks to process the complex data from millimeter-wave radar, thereby efficiently and rationally utilizing the complex data from radar reflections. Its architecture is as follows: Figure 4 As shown. Specifically, for the convolution operation, assuming a learnable convolution kernel W = A + iB and an input vector h = x + iy, its convolution operation is expressed as the following formula:

[0051] W*h=(A*xB*y)+i(B*xA*y)

[0052] Where A and B are the real and imaginary parts of the complex convolution kernel, respectively, and x and y are the real and imaginary parts of the input vector, respectively.

[0053] The activation function is CReLU, i.e.

[0054]

[0055] in Represents taking the real part. Represents the imaginary part.

[0056] For batch standardization of the complex domain, the input is subjected to a mean of 0 and a variance of 1.

[0057]

[0058] Where x is the input vector, E[x] is the expected value of the input batch data (in practice, the mean is used instead of the expected value), and V is the correlation matrix, which is calculated as follows:

[0059]

[0060] The final batch normalization (BN) has two more parameters, γ and β, where γ is the scaling factor and β is the offset. These parameters enable the network to recover and adjust the normalized data distribution. Therefore, the final batch normalization result is as follows:

[0061]

[0062] Furthermore, complex pooling operations are similar to activation functions, and fully connected linear layers are similar to convolutional layers, which will not be elaborated here.

[0063] The above basic modules form a complex domain multilayer perceptron model. Each multilayer perceptron model consists of a 3x3 convolutional layer, an activation function, an average pooling layer, and a regularization layer. Multiple multilayer perceptron models form a feature extraction layer to extract high-dimensional features from the raw radar data. For the strongly correlated radar information resulting from the stacked liquid and reflective surfaces, feature extraction is performed using a single-layer one-dimensional perceptron.

[0064] At the attention mechanism level, firstly, the high-dimensional features of the original radar data and the strongly correlated radar information features of the liquid and reflector surfaces are dimension-mapped using learnable parameter matrices and vectors to the required number of attention heads. The high-dimensional features of the original radar data serve as the query, the strongly correlated radar information features of the reflector surfaces serve as the key, and the value is formed by processing the high-dimensional features of the original radar data through the parameter matrix. At the attention mechanism computation level, since the high-dimensional features of the original radar data are three-dimensional data and the strongly correlated radar information features of the reflector surfaces are two-dimensional data, to achieve dimension matching and reduce computational load, the high-dimensional features of the original radar data are divided into H×1×C and 1×W×C parts, respectively, and multi-head attention mechanism operations are performed with the strongly correlated radar information features of the reflector surfaces. The results are then broadcast, and the two results are expanded to the same dimension before being added together. The specific implementation can be seen in [link to implementation details]. Figure 5 A schematic diagram of the attention mechanism.

[0065] The decoding operation of the neural network mainly involves dimensionality expansion using a multilayer perceptron, followed by dimensionality reduction to one dimension using multiple complex fully connected layers. Then, the result from the complex domain is transformed to the real domain using modulo operation. Finally, the probability of the corresponding concentration gradient is obtained using the softmax function. During training, this probability is used for loss propagation. During prediction, the concentration with the highest probability is selected as the true liquid concentration under the corresponding concentration gradient.

[0066] During training, the cross-entropy function is selected as the loss function, and its calculation method is as follows:

[0067]

[0068] Where N is the number of samples, C is the number of categories obtained based on the minimum concentration resolution, and x y Let x be the probability that the i-th sample belongs to the true class y. c Let c be the probability of the i-th sample data category c.

[0069] Similarly, Adam is used as the optimizer during training, with the learning rate initialized to 0.001, eps set to 1e-8 to prevent division by zero, and weight_decay set to 0.0001 to achieve L2 regularization. The model parameters are decayed on each update to prevent overfitting.

[0070] During the training phase, the number of epochs should be no less than 20 to ensure the network converges to near its optimal value while maintaining accurate performance. Furthermore, the average test loss on the test set should be below 0.01, which is generally sufficient for practical applications.

[0071] After training, select the model and export it in ONNX mode. Deploy the model in an embedded system and achieve online real-time concentration detection of pipeline fluid through automated configuration. Simultaneously, transmit the predicted liquid concentration back to the control center via serial communication using the Modbus protocol.

[0072] To demonstrate the feasibility of the full-concentration step detection method described in this invention, a saline concentration test experiment was first conducted with concentration intervals of 0.5%. The accuracy of the confusion matrix results is as follows: Figure 6 As shown, the horizontal axis represents the predicted concentration, and the vertical axis represents the actual concentration. Furthermore, we trained using a 1% saline concentration as a step, and tested with random concentration intervals from 0% to 1%. The results are as follows. Figure 7 As shown, even without a sample concentration gradient, the network can generally classify the liquid into the correct concentration range, achieving detection of the entire liquid concentration range. For ease of display, the horizontal and vertical axes are not labeled with concentration, but only with concentration steps, starting at 0% and increasing stepwise based on 0.5% or 1%.

[0073] Corresponding to the aforementioned embodiment of a pipeline fluid concentration detection method based on millimeter-wave radar, the present invention also provides an embodiment of a pipeline fluid concentration detection device based on millimeter-wave radar.

[0074] See Figure 8 The present invention provides a pipeline fluid concentration detection device based on millimeter-wave radar, comprising a memory and one or more processors. The memory stores executable code, and when the processor executes the executable code, it is used to implement a pipeline fluid concentration detection method based on millimeter-wave radar in the above embodiment.

[0075] The embodiment of the pipeline fluid concentration detection device based on millimeter-wave radar provided by this invention can be applied to any device with data processing capabilities, such as a computer. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of any data processing device loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 8 The diagram shown is a hardware structure diagram of any device with data processing capabilities, including a pipeline fluid concentration detection device based on millimeter-wave radar provided by the present invention. (Except for...) Figure 8In addition to the processor, memory, network interface, and non-volatile memory shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0076] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0077] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0078] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements a pipeline fluid concentration detection method based on millimeter-wave radar as described in the above embodiments.

[0079] The computer-readable storage medium can be an internal storage unit of any data processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0080] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for detecting pipeline fluid concentration based on millimeter-wave radar.

[0081] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0082] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application. This application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A millimeter wave radar based method of detecting concentration of a fluid in a pipe, the method comprising: The method comprises the following steps: S1, using a millimeter wave radar to emit a linear frequency modulation signal at both ends of a pipeline sight glass to be detected in a closed environment, and receiving a reflected signal to obtain original data, and extracting strong correlation radar information of liquid and reflection surface reflection according to the original data; The step of emitting a linear frequency modulation signal at both ends of a pipeline sight glass to be detected in a closed environment comprises: installing a box structure at both ends of the pipeline sight glass to emit a linear frequency modulation signal, wherein the box structure is provided with a millimeter wave radar, a corresponding power adapter and an embedded device, the transmitting and receiving antenna of the millimeter wave radar is exposed in the field of view of the sight glass, and the other parts are isolated by a metal plate; The step of extracting strong correlation radar information of liquid and reflection surface reflection comprises: performing fast Fourier transform on the original data to obtain a frequency domain signal, sorting the intensity of the frequency domain signal in a set distance range to obtain three peak signals as the first radar signal reaching the surface of the sight glass, the first radar signal penetrating the sight glass, and the first radar signal reaching the reflection surface, calculating the positions of the three target points according to the index of the position of the signal and the distance resolution of the radar, taking the distance from the millimeter wave radar to the reflection surface as a reference, dividing the entire radar information into multiple sections, and stacking the three signals in the same dimension in these sections to obtain the strong correlation radar information of liquid and reflection surface reflection; S2, constructing a concentration data set of a full-range pipeline fluid, classifying according to a minimum concentration to be detected, and generating a corresponding true label; S3, training a neural network for identifying a full-range concentration using the concentration data set, the neural network performing one-dimensional feature extraction on the strong correlation radar information, extracting high-dimensional features from the original data through a complex domain multi-layer perception, performing feature fusion using an attention mechanism, and performing decoding operation on the fused features to obtain a concentration prediction result; S4, inputting the original data and the strong correlation radar information into the trained neural network for identifying a full-range concentration to obtain a concentration prediction result.

2. The method of claim 1, wherein, The step of obtaining original data comprises mixing the received reflected signal through a mixer to generate an intermediate frequency signal as the original data of the radar.

3. The method of claim 1, wherein, The neural network for identifying a full-range concentration adopts a module in a deep complex network model as a basic module in the network, comprising: a data expansion layer, a feature extraction layer, an attention mechanism and a decoding part; The data expansion layer expands the radar data in the original data from two dimensions to three dimensions through copying, so that the original data changes from an antenna number X a sampling number dimension to an antenna number X a sampling number X a sampling number; The feature extraction layer comprises a multi-layer perception for extracting high-dimensional features from the original data of the radar, and a one-dimensional perception for extracting information from the strong correlation radar information of liquid and reflection surface reflection; In the attention mechanism, the high-dimensional feature part of the original radar data is divided into H 1 C and 1 W The C part and the strong correlation radar information feature composed of the reflection of the reflection surface are operated by multi-head attention mechanism, dimension mapping is performed, the required number of attention heads is mapped, the obtained results are broadcast, the two results are expanded to the same dimension, and then added. The decoding part comprises a plurality of complex fully connected layers for reducing and expanding the result to one dimension, then converting the result in the complex domain to the real number domain through a modulo method, and finally obtaining the probability of the corresponding concentration gradient through a softmax function.

4. The method of claim 3, wherein the method further comprises: The multi-layer perception is composed of a plurality of 2-dimensional 3*3 complex convolution modules, complex activation function modules, complex pooling modules and complex batch normalization modules.

5. The millimeter wave radar-based method of claim 4, wherein, The complex activation function module adopts a CReLU function, and the complex batch normalization module performs 0 mean value and 1 variance operation on the input: ; where x is an input vector, is the expectation of the input batch data, is the correlation matrix, where the correlation matrix V is calculated as follows ； The final BN has two parameters With where γ is a scaling factor and β is a shift. Thus the final batch normalization result is as follows 。 6. The millimeter wave radar-based method of claim 1, wherein, In the training process, a cross-entropy function is selected as a loss function, and the calculation method is as follows: ； where N is the number of samples, C is the number of classes obtained according to the minimum concentration resolution, is the probability that the ith sample belongs to the true class y, is the probability of the data class c for the ith sample. 7.A millimeter wave radar based pipeline fluid concentration detection apparatus comprising a memory and one or more processors, the memory having stored therein executable code to, The processor executes the executable code to implement the millimeter wave radar-based pipeline fluid concentration detection method in any one of claims 1-6.

8. A computer-readable storage medium having stored thereon a program, characterized in that, The processor executes the executable code to implement the millimeter wave radar-based pipeline fluid concentration detection method in any one of claims 1-6.

Images