CN117056794A - Non-contact liquid component recognition model training method, recognition method, system and device - Google Patents

Non-contact liquid component recognition model training method, recognition method, system and device Download PDF

Info

Publication number
CN117056794A
CN117056794A CN202310918630.6A CN202310918630A CN117056794A CN 117056794 A CN117056794 A CN 117056794A CN 202310918630 A CN202310918630 A CN 202310918630A CN 117056794 A CN117056794 A CN 117056794A
Authority
CN
China
Prior art keywords
liquid
liquid component
signal
frequency
sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310918630.6A
Other languages
Chinese (zh)
Inventor
马华东
梁雨萌
周安福
石璞
蒲凌宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN202310918630.6A priority Critical patent/CN117056794A/en
Publication of CN117056794A publication Critical patent/CN117056794A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The application provides a non-contact liquid component recognition model training method, a non-contact liquid component recognition model recognition method, a non-contact liquid component recognition system and a non-contact liquid component recognition model recognition device. The multi-band reflection characteristics are processed through the customized neural network, the reflection characteristics related to the liquid to be detected are extracted from the multi-band reflection characteristics, and classification and identification of liquid components are performed, so that the identification precision of the liquid components can be improved, and the identification of liquid with finer granularity is realized. The neural network focuses on the reflection characteristics of the distance units where the liquid to be identified is located in the reflection characteristics, and can achieve accurate identification results when the liquid to be identified is randomly placed at random positions and angles within a set distance range.

Description

Non-contact liquid component recognition model training method, recognition method, system and device
Technical Field
The application relates to the technical field of liquid sensing, in particular to a non-contact liquid component identification model training method, a non-contact liquid component identification system and a non-contact liquid component identification device.
Background
Liquid sensing technology refers to a class of technology for detecting, measuring and monitoring liquid properties, states and changes. These liquid sensing techniques typically use sensors or detectors to sense physical or chemical properties of the liquid, etc. With the frequent occurrence of food and beverage safety cases in recent years and the case needs of specific places, liquid sensing technology is widely applied to production and living. This technique identifies the kind and composition of a liquid by analyzing the characteristics of the liquid by an instrument. For example, a user allergic to alcohol may detect a small amount of alcohol that may be contained in the beverage, or identify the content of liquid components in different white spirits to identify counterfeit and inferior products.
Conventional liquid detection methods generally need to be performed in a laboratory by relying on expensive and large-scale professional equipment, and generally realize sensing of liquid components such as alcohol by using a spectrometer based on absorption and scattering characteristics of different liquids for light with different frequencies. However, these methods are difficult to be widely used in people's daily lives due to the high cost of specialized equipment and the complexity of operation.
Methods for sensing liquid components using wireless sensing technology have been developed in recent years, and analysis of liquid characteristics has been achieved by analyzing changes that occur when wireless signals reflect or penetrate in a liquid. Although wireless sensing can realize non-contact nondestructive sensing, namely, the liquid components in the container are identified under the conditions of not opening the container and damaging the liquid sample, the problems of limited identification range, insufficient fine identification granularity, unstable identification effect and the like still exist.
Disclosure of Invention
In view of this, the embodiments of the present application provide a non-contact liquid component recognition model training method, recognition method, system, and apparatus, so as to eliminate or improve one or more drawbacks existing in the prior art, and solve the problems of limited recognition range, insufficient fine recognition granularity, and greater dependence on specific detection positions and angles in the existing liquid wireless sensing technology.
One aspect of the present application provides a non-contact liquid component identification model training method comprising the steps of:
detecting sample liquid in a set distance range by adopting sweep frequency signals with increasing frequencies through a millimeter wave radar, and mixing a receiving signal and a reflected signal to obtain an intermediate frequency signal; dividing the intermediate frequency signal into a plurality of frequency bands, and respectively performing fast Fourier transform to obtain frequency domain signals, wherein each frequency band corresponds to each frequency in the frequency domain signals and is associated with a distance unit in the set distance range; extracting the intensity and the phase of a received signal in a frequency domain signal corresponding to each frequency band, and constructing reflection parameters for classifying and identifying sample liquid;
constructing a training sample set, wherein the training sample set comprises a plurality of samples, each sample comprises the reflection parameters obtained by acquiring and processing single sample liquid within the set distance range through the millimeter wave radar, and the categories of the components of the corresponding sample liquid are added as labels;
acquiring an initial neural network model for identifying liquid components, wherein the initial neural network model comprises a multi-distance unit feature extraction module and a liquid component classification module; the multi-distance unit feature extraction module adopts a shared convolution kernel with the same learnable parameters to extract the reflection features of the corresponding reflection parameters of the sample liquid on different distance units in a translation unchanged mode; the liquid component classification module comprises a plurality of first full-connection layers, and after flattening the reflection characteristics into one dimension, the liquid component classification module inputs the plurality of first full-connection layers and outputs a component identification result of the corresponding sample liquid;
and training the initial neural network model by adopting the training sample set, and updating parameters of the initial neural network model based on a cross entropy loss function to obtain a liquid component identification model.
In some embodiments, the sample liquids are randomly placed within the set distance range, the set distance range is 20 cm to 2 m, and the sample liquids are arranged in a plurality according to different compositions and concentrations.
In some embodiments, the multi-distance unit feature extraction module comprises a continuous two-branch module, a first activation function layer, an attention module and a temporary fallback mechanism layer; the two-branch module comprises a main branch and a residual branch which are parallel, wherein the main branch consists of a first continuous one-dimensional convolution layer, a first regularization layer, a second activation function layer, a second one-dimensional convolution layer and a second regularization layer, the convolution kernel of the first one-dimensional convolution layer is 1 multiplied by 1, and the convolution kernel of the second one-dimensional convolution layer is 1 multiplied by 3; the residual branches comprise a third one-dimensional convolution layer and a third regularization layer in succession, wherein the convolution kernel of the third one-dimensional convolution layer is 1×1.
In some embodiments, the attention module averages the input characteristic parameters in each distance unit dimension to obtain an average response of each characteristic channel, obtains weights of each characteristic channel through two second full-connection layers, and finally multiplies the weights with the original input characteristic parameters and outputs the multiplied weights.
In some embodiments, the first fully-connected layer is a three-layer stack structure.
In some embodiments, the cross entropy loss function is calculated as:
wherein N represents the number of samples, x y Representing the probability that the ith sample liquid belongs to the true class y, x c Representing the probability that the i-th sample liquid belongs to category c。
In another aspect, the present application also provides a method for identifying a non-contact liquid component, including:
transmitting a sweep frequency signal to the liquid to be detected within a set distance range based on the millimeter wave radar, and mixing a received signal and a reflected signal to obtain an intermediate frequency signal; dividing the intermediate frequency signal into a plurality of frequency bands, respectively performing fast Fourier transform to obtain frequency domain signals, extracting the received signal strength and the phase of the frequency domain signals corresponding to each frequency band, and constructing reflection parameters;
inputting the reflection parameters into a liquid component identification model in the non-contact liquid component identification model training method, and outputting the component identification result of the liquid to be tested.
In another aspect, the present application also provides a non-contact liquid component identification system, including:
the millimeter wave radar is used for transmitting sweep frequency signals to the liquid to be detected within a set distance range and mixing the receiving signals and the reflected signals to obtain intermediate frequency signals;
and the processor is used for executing the non-contact liquid component identification method so as to output the component identification result of the liquid to be tested.
In some embodiments, the processor is a personal mobile terminal device, and the personal mobile terminal device is connected to the millimeter wave radar through USB or WIFI; the personal mobile terminal device is a smart phone or a tablet computer.
In another aspect, the present application also provides a computer readable storage medium having stored thereon a computer program, characterized in that the program when executed by a processor implements the steps of the above method.
The application has the advantages that:
according to the liquid component identification model training method, the liquid component identification system and the liquid component identification model training device, in the detection process, the millimeter wave radar is utilized to send the sweep frequency signals with the frequency ranging from low to high to the liquid within the set distance range, the intermediate frequency signals are divided into a plurality of frequency bands with different starting frequencies and ending frequencies, fast Fourier transformation is respectively carried out on the intermediate frequency signals, reflection parameters of millimeter wave signals with various frequency bands within the set distance range are obtained, and reflection characteristics of units with different distances are captured based on the signals of the frequency domain. The reflection characteristics of the multi-band reflection parameters and the liquid to be detected are collected through the customized neural network, the liquid classification and identification are carried out, the identification precision of liquid components is improved, and the liquid identification with finer granularity is realized. The neural network focuses on the reflection characteristics of the distance units where the liquid to be identified is located in the reflection characteristics, and can achieve accurate identification results when the liquid to be identified is randomly placed at random positions and angles within a set distance range. The liquid component identification system has high identification precision and can identify the liquid component difference of 0.2 percent; the equipment is small in size, supports random placement of the target position to be detected, and is convenient to deploy; is expected to be widely applied in various daily life scenes: such as wine identification, blood glucose concentration change monitoring, food safety detection, etc.
Additional advantages, objects, and features of the application will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the application. The objectives and other advantages of the application may be realized and attained by the structure particularly pointed out in the written description and drawings.
It will be appreciated by those skilled in the art that the objects and advantages that can be achieved with the present application are not limited to the above-described specific ones, and that the above and other objects that can be achieved with the present application will be more clearly understood from the following detailed description.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate and together with the description serve to explain the application. In the drawings:
fig. 1 is a flow chart of a non-contact liquid component identification method according to an embodiment of the application.
Fig. 2 is a schematic diagram illustrating the extraction of multi-frequency reflection characteristics in a non-contact liquid component recognition method according to an embodiment of the application.
Fig. 3 is a diagram illustrating a neural network used in a non-contact liquid component identification method according to an embodiment of the present application.
FIG. 4 is a schematic diagram of a feature extraction scheme with translational invariance employed in the neural network of FIG. 3.
Fig. 5 is a schematic diagram of the structure of the attention module in the neural network shown in fig. 3.
Detailed Description
The present application will be described in further detail with reference to the following embodiments and the accompanying drawings, in order to make the objects, technical solutions and advantages of the present application more apparent. The exemplary embodiments of the present application and the descriptions thereof are used herein to explain the present application, but are not intended to limit the application.
It should be noted here that, in order to avoid obscuring the present application due to unnecessary details, only structures and/or processing steps closely related to the solution according to the present application are shown in the drawings, while other details not greatly related to the present application are omitted.
It should be emphasized that the term "comprises/comprising" when used herein is taken to specify the presence of stated features, elements, steps or components, but does not preclude the presence or addition of one or more other features, elements, steps or components.
It is also noted herein that the term "coupled" may refer to not only a direct connection, but also an indirect connection in which an intermediate is present, unless otherwise specified.
Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.
At present, schemes for identifying liquid components by using a smart phone, LED, RFID, wi-Fi equipment, millimeter wave radar and the like as low-cost liquid component sensors exist, but the existing methods have obvious limitations in practical application. Smart phone based methods require the user to use a special container to identify the liquid depending on the specific ripple created on the surface of the liquid as it oscillates. The requirement for special containers limits the usability of the method. LED-based approaches require the sensor to be immersed in the liquid for sensing and are therefore unsuitable for sealing containers. Wireless signal non-contact and non-invasive liquid component sensing using RFID, UWB, wi-Fi signals typically requires placement of the container in a fixed area to sense the liquid target, resulting in poor mobility and complex deployment in different scenarios. The wireless sensing method is coarse in liquid component sensing precision, is easily affected by target position changes, such as distance changes and sensing angle changes, and is difficult to stably and finely identify liquid components in real life. In daily use, displacement and angular rotation of the liquid container are unavoidable, and these changes affect the reflected signal of millimeter waves and are confused with signal changes caused by liquid composition changes, so that the recognition accuracy is lowered.
According to the application, the millimeter wave reflection of the liquid is utilized to realize the refined sensing and identification of the superfine liquid component difference, the reduction of the identification performance caused by the distance change and the sensing angle change between the millimeter wave radar and the target liquid in actual use is overcome, and the characteristic of the liquid reflection signal under the random angle of the random position is captured by combining the customized neural network, so that the stable, accurate and fine-grained liquid component identification is realized.
Specifically, the application provides a non-contact liquid component recognition model training method, which comprises the following steps of S101 to S104:
step S101: detecting sample liquid in a set distance range by adopting sweep frequency signals with increasing frequencies through a millimeter wave radar, and mixing a receiving signal and a reflected signal to obtain an intermediate frequency signal; dividing the intermediate frequency signal into a plurality of frequency bands, and respectively performing fast Fourier transform to obtain frequency domain signals, wherein each frequency band corresponds to a distance unit in a set distance range of each frequency association in the frequency domain signals; and extracting the received signal strength and the phase of the frequency domain signal corresponding to each frequency band, and constructing reflection parameters for classifying and identifying the sample liquid.
Step S102: the method comprises the steps of constructing a training sample set, wherein the training sample set comprises a plurality of samples, each sample comprises reflection parameters obtained by acquiring and processing single sample liquid within a set distance range through millimeter wave radar, and categories of corresponding sample liquid components are added as labels.
Step S103: acquiring an initial neural network model for identifying liquid components, wherein the initial neural network model comprises a multi-distance unit feature extraction module and a liquid component classification module; the multi-distance unit feature extraction module adopts a shared convolution kernel with the same learnable parameters to extract the reflection features of the corresponding reflection parameters of the sample liquid on different distance units in a translation unchanged mode; the liquid component classification module comprises a plurality of first full-connection layers, and after flattening the reflection characteristics into one dimension, the liquid component classification module inputs the plurality of first full-connection layers and outputs component identification results of corresponding sample liquid.
Step S104: and training the initial neural network model by adopting a training sample set, and updating parameters of the initial neural network model based on the cross entropy loss function to obtain a liquid component identification model.
In step S101, sample data for training a model is first prepared. Specifically, the embodiment adopts the FMCW millimeter wave radar to emit millimeter waves to a sample special body in a limited space range, and the same liquid has different dielectric characteristics and reflection characteristics to millimeter waves with different frequencies, so the embodiment utilizes the millimeter wave radar to emit sweep signals so as to provide diversified frequency millimeter wave signals for detection. A Sweep Signal (Sweep Signal) is a Signal of continuously varying frequency whose frequency varies over a period of time in steps over a range of frequencies. In the period of one sweep frequency signal, the frequency of the signal (Tx) transmitted by the millimeter wave radar transmitting antenna is gradually increased, and the signal reflected signal (Rx) received by the receiving antenna is mixed with the Tx signal to generate an intermediate frequency signal. The intermediate frequency signal is divided into a plurality of frequency bands to mine the reflection characteristics of the sample liquid at different frequencies. Further, if the intermediate frequency signal in the time domain is converted into the frequency domain by using the fast fourier transform, each frequency component corresponds to the reflected signal of each distance unit.
Note that Range bin (Range bin) is a term used in radar or other sensor systems to represent the Range resolution of a target. A range of distances is divided into discrete intervals, each interval representing a range cell. The distance units may be of equal size or may be non-equidistantly spaced as defined by the application requirements. Each range cell contains target information within a specific range of distances, which enables the radar system to resolve and locate targets over distances. The size of the range bin determines the range resolution of the radar system, i.e., the ability of the system to distinguish between two closely spaced targets.
Therefore, through the form of millimeter wave radar emission sweep frequency signal and frequency division to carry out fast Fourier transform, the reflection characteristic of sample liquid to the multifrequency can be gathered in the range of the multi-distance unit, and the space stability and the recognition fine granularity of liquid component recognition are ensured at first from the data plane.
In some embodiments, the sample liquids are randomly placed within the set distance range, the set distance range is 20 cm to 2 m, and a plurality of sample liquids are arranged according to different compositions and concentrations.
In step S102, the training sample set takes the frequency domain reflection parameters of the multi-distance multi-band as input, and the components of the sample liquid as labels. The sample liquid can be classified according to different components, the proportion and the concentration of the components.
In step S103, reflection characteristics of the reflection parameters in both distance and frequency band are mined by constructing a neural network, and a classification task is performed. The multi-distance unit feature extraction module can extract the reflected signal features of the same liquid target on different distance units in a translation unchanged mode by using the shared convolution kernel and sliding on the feature patterns of different distance units through the attention module on the basis of feature extraction through convolution. The design allows the same learning parameters to be used at different positions to capture the commonality of the targets, focuses on the channel with the most distinguishing capability, thereby enhancing the detection and recognition capability of the model on the targets, reducing the dependence on specific distance and angle in the detection process, ensuring that the sample liquid can accurately focus on the reflection characteristics presented by the sample liquid whenever being placed within a set distance range.
In some embodiments, the multi-distance unit feature extraction module comprises a continuous two-branch module, a first activation function layer, an attention module and a temporary back mechanism layer; the two-branch module comprises a main branch and a residual branch which are parallel, wherein the main branch consists of a first continuous one-dimensional convolution layer, a first regularization layer, a second activation function layer, a second one-dimensional convolution layer and a second regularization layer, the convolution kernel of the first one-dimensional convolution layer is 1 multiplied by 1, and the convolution kernel of the second one-dimensional convolution layer is 1 multiplied by 3; the residual branches comprise a third one-dimensional convolution layer and a third regularization layer, wherein the convolution kernel of the third one-dimensional convolution layer is 1 multiplied by 1.
It should be noted that the first, second and third embodiments of the present application are not limited to ordinal numbers, but should be construed as distinguishing the same type of structure at different positions.
In some embodiments, the attention module averages the input characteristic parameters in each distance unit dimension to obtain an average response of each characteristic channel, obtains weights of each characteristic channel through two second full-connection layers, and finally multiplies the weights with the original input characteristic parameters and outputs the multiplied weights. This approach may enhance the neural network's attention to the characteristic channels that are most capable of characterizing the liquid composition.
In the liquid component classification module, the first full-connection layer is a three-layer stacked structure. The final output is the probability that the sample fluid belongs to each class.
In step S104, in the classification task, parameters of the model are updated with the cross entropy loss function. Illustratively, ten liquids with alcohol degrees of 0.1 degree, 0.2 degree, 0.3 degree, … and 1.0 degree are distinguished, and the sample liquids are randomly placed in a range of 25 cm to 60 cm from the millimeter wave radar.
The signal is transmitted using 1 transmitting antenna of the millimeter wave radar, and the signal is received by 4 antennas, and 64 points are sampled in one sweep period (chirp). The 64 sample points are first divided into 8 millimeter wave signals with different start and stop frequencies using a sliding window of length 42 and step size 3. The signal of each segment is then converted to a reflected signal of a different range bin using fourier transform, the resolution of the corresponding range bin being 3.81 cm. Further, in each frequency segment, signal intensity and phase characteristics are extracted from the reflected signals of the distance units corresponding to the liquid target placement range (25-60 cm). At this range resolution, 10 range bins are obtained, 64 features per range bin: 4 antennas, and signal strength and phase corresponding to 8 frequency bands. The data feature with the dimension of 10 x 64 is input into the neural network, resulting in corresponding probabilities for 10 liquid species, and the species of which the highest probability is selected as the predicted liquid component. During training, the adopted loss function is a cross entropy function. The cross entropy loss function is calculated as:
wherein N represents the number of samples, x y Representing the probability that the ith sample liquid belongs to the true class y, x c Indicating the probability that the i-th sample liquid belongs to category c.
Finally, the liquid component identification model obtained through training in the steps S101-S104 can be based on the millimeter wave radar to emit the medium frequency signal acquired by the sweep frequency signal in the set distance range, and high-precision identification of liquid fine granularity, random position and angle can be realized.
On the other hand, the application also provides a non-contact liquid component identification method, which comprises the following steps of S201 to S202:
step S201: transmitting a sweep frequency signal to the liquid to be detected within a set distance range based on the millimeter wave radar, and mixing a received signal and a reflected signal to obtain an intermediate frequency signal; dividing the intermediate frequency signal into a plurality of frequency bands, respectively performing fast Fourier transform to obtain frequency domain signals, and extracting the received signal intensity and the phase in the frequency domain signals corresponding to each frequency band to construct reflection parameters.
Step S202: and (3) inputting the reflection parameters into a liquid component identification model in the non-contact liquid component identification model training method in the steps S101 to S104, and outputting a component identification result of the liquid to be detected.
In another aspect, the present application also provides a non-contact liquid component identification system, including:
the millimeter wave radar is used for transmitting sweep frequency signals to the liquid to be detected within a set distance range, receiving signals and mixing the received signals with reflected signals to obtain intermediate frequency signals;
and the processor is used for executing the non-contact liquid component identification method in the steps S201 to S202 so as to output the component identification result of the liquid to be tested.
In some embodiments, the processor is a personal mobile terminal device, and the personal mobile terminal device is connected to the millimeter wave radar through USB or WIFI; the personal mobile terminal equipment is a smart phone, a tablet personal computer and the like.
In another aspect, the present application also provides a computer readable storage medium having stored thereon a computer program, characterized in that the program when executed by a processor implements the steps of the above method.
The application is illustrated below in connection with specific examples:
the embodiment provides a non-contact liquid component identification method and designs a device composed of a smart phone and a millimeter wave radar. The device transmits millimeter wave signals to the liquid target through the millimeter wave radar connected with the smart phone, receives and analyzes the millimeter wave signals reflected by the liquid, and therefore the identification of liquid components is achieved. It is capable of finely sensing subtle differences in the liquid composition, for example, differences in alcohol content of only 0.2 degrees. Meanwhile, the method can overcome the serious interference of the distance change and the perception angle change of the liquid target on the identification performance, and can stably identify the liquid components under any displacement and rotation of the liquid target.
FMCW millimeter wave radars connected to smart phones are capable of emitting and sensing millimeter wave signals reflected from objects, and the signal strength (Received Signal Strength, RSS) of the reflected signals may be different due to differences in the characteristics of the target liquids (e.g., different liquid dielectric constants). Therefore, the liquid reflection signals received by the millimeter wave radar are processed and analyzed by the smart phone, and the component information of the liquid is obtained according to the millimeter wave reflection differences caused by different liquid components.
Specifically, referring to fig. 1, the present embodiment includes the following 2 parts:
1) Signal feature extraction based on multiple frequency channels
The present embodiment transmits millimeter wave signals to a target liquid and collects liquid-reflected millimeter wave signals by an FMCW millimeter wave radar connected to a smart phone. Next, by utilizing the frequency characteristics of the diversification of millimeter wave signals transmitted by the FMCW millimeter wave radar, the intermediate frequency (intermediate frequency) signal sampled in one FMCW sweep signal (chirp) is divided into a plurality of perceived frequency bands having different start and end frequencies. Then, the signal of each frequency band is subjected to fast fourier transformation, converted into a frequency domain signal, and the reflection characteristics of the specific region where the liquid is located are extracted. Finally, these features are combined to produce a reflection signature that is rich in the liquid target for millimeter wave signals in multiple frequency bands. This feature implies unique reflection characteristics caused by different liquid components.
The prior art relies on the reflective properties of a liquid target for a single frequency wireless signal to extract relevant features. However, this approach is difficult to learn about the characteristics of the finer-grained liquid component. Inspired by the phenomenon observed in the experiment, the same liquid has different dielectric characteristics and reflection characteristics for millimeter waves with different frequencies, so that the embodiment obtains the reflection characteristics of liquid components for millimeter waves with different frequencies by utilizing millimeter wave signals with diversified frequencies emitted by the FMCW millimeter wave radar, and generates the characteristic of fine-grained liquid components with more distinguishing force.
As shown in fig. 2, the frequencies of the transmitting antenna and the receiving antenna gradually increase in one sweep signal period, and the present embodiment first divides the signal samples sampled in one sweep signal period into a plurality of frequency bands, each frequency band having a different start frequency and end frequency. For the sampled signal points within each frequency band, the time domain signal is converted to a frequency domain signal by applying a fast fourier transform. Then, the present embodiment extracts the Received Signal Strength (RSS) of the region where the liquid is located on each frequency band, respectively, to obtain the reflection characteristics of the liquid target for millimeter waves of different frequencies. By the method, the multi-frequency signal characteristics of the FMCW millimeter wave radar are fully utilized, and the reflection characteristics of a plurality of different frequency bands can be obtained simultaneously by only using the sampling data obtained in one sweep frequency signal period without increasing additional signal receiving and transmitting expenditure. The present embodiment then uses the resulting set of multi-frequency characteristics as input information to the neural network to distinguish between different liquid components.
2) Construction of neural networks for liquid component identification
The embodiment designs a customized neural network, extracts the reflection characteristics related to the liquid components from the multi-frequency reflection characteristics obtained from different positions, and overcomes the signal interference caused by the position change and the perception angle change of the liquid target. In the neural network, the embodiment designs a characteristic extraction module with a translation invariant characteristic, which can automatically extract component characteristics of liquid targets placed at different positions, and the characteristics can be sent to a liquid component classification module to distinguish different liquid components. According to the method, the neural network is trained by collecting the data of the liquid target at different placement positions and rotation angles, so that the neural network learns how to remove signal interference caused by position change and perception angle change from diversified data samples, and reliable and stable liquid component identification is realized.
As shown in fig. 3, the neural network designed in this embodiment has two modules: a multi-range bin (range bin) feature extraction module and a liquid component classification module. Firstly, acquiring respective multi-frequency reflection information from a plurality of distance units as input of a neural network, so that the neural network has perception capability on the plurality of distance units to cope with different possible placement positions of liquid targets; the reflection information comprises signal intensity and phase characteristics respectively received by the millimeter wave radar multi-receiving antenna so as to respectively provide reflection information and position information of the target liquid.
The multi-distance unit feature extraction module in fig. 3 has a 6-layer structure, each of which is composed of a two-branch module (composed of a main branch and a residual branch), an activation function (ReLU), an attention module, and a temporary-back mechanism layer (Dropout). The main branch in the two-branch module consists of one-dimensional convolution (the convolution kernel size is 1x 1), a regularization layer (BatchNorm), an activation function, one-dimensional convolution (the convolution kernel size is 1x 3), and the residual branch consists of additional one-dimensional convolution (the convolution kernel size is 1x 1) and the regularization layer. After the dual-branch structure, we use the attention module to focus on the channel most having the distinguishing component in the multi-channel feature map.
As shown in fig. 4, the present embodiment uses a shared convolution kernel with the same learnable parameters to slide over the feature maps of different distance units to extract the reflected signal features of the same liquid target over the different distance units in a manner with translational invariance. This way, the dependency of the feature extraction process on the position of the liquid target (i.e. the distance cell in which it is located) is reduced, and a stable and reliable liquid identification is achieved.
The structure of the attention module is shown in fig. 5, firstly, the average response of each characteristic channel is obtained by averaging under different distance unit dimensions, then the weight of each channel is obtained through two fully-connected layers, and finally, the weight is multiplied by the input characteristics, so that the attention of the neural network to the characteristic channel with the liquid component characterization capability is enhanced.
At the end of the neural network, this embodiment uses a liquid component classification module that first flattens the two-dimensional features (features at multiple distances and frequency channels) into one dimension, then processes the features through three stacked fully connected layers, and outputs a corresponding probability score for each liquid component.
And finishing training and parameter updating of the neural network based on the classification task, and executing liquid component identification by using a model obtained by training.
The method provided by the embodiment breaks through the limitation of the existing wireless sensing method on the identification granularity, and can stably and reliably realize the fine-granularity liquid component identification effect with high accuracy in different environments. The technical scheme of the application can distinguish very similar liquids, such as 0.2-degree alcohol concentration difference, and overcomes the dependence on the placement position of the target liquid. As a non-contact liquid identification scheme, the technology can be applied to realize nondestructive liquid composition analysis in various living scenes, such as detection of trace alcohol possibly contained in a beverage, counterfeit and inferior wine with high fraud property, and the like.
This embodiment enhances the perceptibility of subtle differences in liquid composition: the prior art can only support the identification of 1 ° alcohol differences, and the present embodiment can support the identification of finer liquid differences, such as 0.2 ° alcohol differences. In addition, the prior art only supports the position difference of the liquid target of about 3cm, and the embodiment can stably identify the liquid under the random displacement and rotation of the liquid target of tens of centimeters.
Accordingly, the present application also provides an apparatus/system comprising a computer device including a processor and a memory, the memory having stored therein computer instructions for executing the computer instructions stored in the memory, the apparatus/system implementing the steps of the method as described above when the computer instructions are executed by the processor.
The embodiments of the present application also provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the edge computing server deployment method described above. The computer readable storage medium may be a tangible storage medium such as Random Access Memory (RAM), memory, read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disk, a removable memory disk, a CD-ROM, or any other form of storage medium known in the art.
In summary, according to the liquid component identification model training method, the liquid component identification model training system and the liquid component identification model training device, in the detection process, the millimeter wave radar is utilized to send the sweep frequency signals with the frequencies ranging from low to high to the liquid within the set distance range, the received intermediate frequency signals are divided into a plurality of frequency bands with different starting frequencies and ending frequencies, and fast Fourier transformation is respectively carried out on the frequency bands, so that reflection parameters of millimeter wave signals with various frequency bands within the set distance range are obtained, and reflection characteristics of different distance units are captured based on the signals of the frequency domain. The reflection characteristics of the multi-band reflection parameters and the liquid to be detected are collected through the customized neural network, and the liquid classification and identification are carried out, so that finer granularity detection can be realized, and the detection capability is improved. The neural network focuses on the reflection characteristic presented by the distance unit where the liquid to be detected is located in the reflection characteristic by using the attention module, and can accurately detect the random angle of the random position of the liquid to be detected in the set distance range.
Those of ordinary skill in the art will appreciate that the various illustrative components, systems, and methods described in connection with the embodiments disclosed herein can be implemented as hardware, software, or a combination of both. The particular implementation is hardware or software dependent on the specific application of the solution and the design constraints. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, a plug-in, a function card, or the like. When implemented in software, the elements of the application are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine readable medium or transmitted over transmission media or communication links by a data signal carried in a carrier wave.
It should be understood that the application is not limited to the particular arrangements and instrumentality described above and shown in the drawings. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present application are not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between steps, after appreciating the spirit of the present application.
In this disclosure, features that are described and/or illustrated with respect to one embodiment may be used in the same way or in a similar way in one or more other embodiments and/or in combination with or instead of the features of the other embodiments.
The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, and various modifications and variations can be made to the embodiments of the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims (9)

1. A non-contact liquid component identification model training method, characterized in that the method comprises the following steps:
detecting sample liquid in a set distance range by adopting sweep frequency signals with increasing frequencies through a millimeter wave radar, and mixing a receiving signal and a reflected signal to obtain an intermediate frequency signal; dividing the intermediate frequency signal into a plurality of frequency bands, and respectively performing fast Fourier transform to obtain frequency domain signals, wherein each frequency band corresponds to each frequency in the frequency domain signals and is associated with a distance unit in the set distance range; extracting the intensity and the phase of a received signal in a frequency domain signal corresponding to each frequency band, and constructing reflection parameters for classifying and identifying sample liquid;
constructing a training sample set, wherein the training sample set comprises a plurality of samples, each sample comprises the reflection parameters obtained by acquiring and processing single sample liquid within the set distance range through the millimeter wave radar, and the categories of the components of the corresponding sample liquid are added as labels;
acquiring an initial neural network model for identifying liquid components, wherein the initial neural network model comprises a multi-distance unit feature extraction module and a liquid component classification module; the multi-distance unit feature extraction module adopts a shared convolution kernel with the same learnable parameters to extract the reflection features of the corresponding reflection parameters of the sample liquid on different distance units in a translation unchanged mode; the liquid component classification module comprises a plurality of first full-connection layers, and after flattening the reflection characteristics into one dimension, the liquid component classification module inputs the plurality of first full-connection layers and outputs component identification results of corresponding sample liquid;
and training the initial neural network model by adopting the training sample set, and updating parameters of the initial neural network model based on a cross entropy loss function to obtain a liquid component identification model.
2. The method according to claim 1, wherein the sample liquid is placed at random within the set distance range, and a plurality of sample liquids are arranged according to different compositions and concentrations.
3. The non-contact liquid component recognition model training method according to claim 1, wherein the multi-distance unit feature extraction module comprises a continuous two-branch module, a first activation function layer, an attention module and a temporary withdrawal mechanism layer; the two-branch module comprises a main branch and a residual branch which are parallel, wherein the main branch consists of a first continuous one-dimensional convolution layer, a first regularization layer, a second activation function layer, a second one-dimensional convolution layer and a second regularization layer, the convolution kernel of the first one-dimensional convolution layer is 1 multiplied by 1, and the convolution kernel of the second one-dimensional convolution layer is 1 multiplied by 3; the residual branches comprise a third one-dimensional convolution layer and a third regularization layer in succession, wherein the convolution kernel of the third one-dimensional convolution layer is 1×1.
4. The training method of the non-contact liquid component recognition model according to claim 1, wherein the attention module averages the input characteristic parameters in each distance unit dimension to obtain an average response of each characteristic channel, obtains weights of each characteristic channel through two second full-connection layers, and finally multiplies the weights with the original input characteristic parameters and outputs the multiplied weights.
5. The method of claim 1, wherein the cross entropy loss function is calculated by:
wherein N represents the number of samples, x y Representing the probability that the ith sample liquid belongs to the true class y, x c Indicating the probability that the i-th sample liquid belongs to category c.
6. A method for identifying a non-contact liquid component, comprising:
transmitting a sweep frequency signal to the liquid to be detected within a set distance range based on the millimeter wave radar, and mixing a received signal and a reflected signal to obtain an intermediate frequency signal; dividing the intermediate frequency signal into a plurality of frequency bands, respectively performing fast Fourier transform to obtain frequency domain signals, extracting the received signal strength and the phase of the frequency domain signals corresponding to each frequency band, and constructing reflection parameters;
inputting the reflection parameters into a liquid component identification model in the non-contact liquid component identification model training method according to any one of claims 1 to 5, and outputting the component identification result of the liquid to be tested.
7. A non-contact liquid component identification system, comprising:
the millimeter wave radar is used for transmitting sweep frequency signals to the liquid to be detected within a set distance range and mixing the received signals and the reflected signals to obtain intermediate frequency signals;
a processor for executing the non-contact liquid component recognition method according to claim 6 to output a component recognition result of the liquid to be measured.
8. The non-contact liquid component identification system of claim 7, wherein the processor is a personal mobile terminal device connected to the millimeter wave radar via USB or WIFI; the personal mobile terminal device is a smart phone or a tablet computer.
9. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any one of claims 1 to 6.
CN202310918630.6A 2023-07-25 2023-07-25 Non-contact liquid component recognition model training method, recognition method, system and device Pending CN117056794A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310918630.6A CN117056794A (en) 2023-07-25 2023-07-25 Non-contact liquid component recognition model training method, recognition method, system and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310918630.6A CN117056794A (en) 2023-07-25 2023-07-25 Non-contact liquid component recognition model training method, recognition method, system and device

Publications (1)

Publication Number Publication Date
CN117056794A true CN117056794A (en) 2023-11-14

Family

ID=88654440

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310918630.6A Pending CN117056794A (en) 2023-07-25 2023-07-25 Non-contact liquid component recognition model training method, recognition method, system and device

Country Status (1)

Country Link
CN (1) CN117056794A (en)

Similar Documents

Publication Publication Date Title
US8717146B2 (en) Methods and systems for integrated interrogation of RFID sensors
Zhang et al. Latern: Dynamic continuous hand gesture recognition using FMCW radar sensor
US20190087654A1 (en) Method and system for csi-based fine-grained gesture recognition
CN111399642B (en) Gesture recognition method and device, mobile terminal and storage medium
US11391817B2 (en) Radio frequency (RF) object detection using radar and machine learning
WO2023029390A1 (en) Millimeter wave radar-based gesture detection and recognition method
US11899132B2 (en) Super-resolution enhancement techniques for radar
Liang et al. FG-LiquID: A contact-less fine-grained liquid identifier by pushing the limits of millimeter-wave sensing
CN113376609B (en) Liquid identification method based on millimeter wave radar
CN109407083A (en) The Weighted adaptive detector that a kind of pair of mismatch subspace signal flexibly controls
CN116027294A (en) Single-bit radar micro-motion target identification method and device based on decision-level fusion
Lee et al. Digit recognition in air-writing using single millimeter-wave band radar system
Rather et al. Evaluation of machine learning models for a chipless rfid sensor tag
CN213904292U (en) Gesture recognition system
CN117056794A (en) Non-contact liquid component recognition model training method, recognition method, system and device
Skaria et al. Machine-Learning Methods for Material Identification Using mmWave Radar Sensor
CN110554439B (en) Article detection method and apparatus
CN110174711A (en) Terminal device and its foreign bodies detection evaluation of result method, computer storage medium
CN115754956A (en) Millimeter wave radar gesture recognition method based on envelope data time sequence
CN113074718B (en) Positioning method, device, equipment and storage medium
EP4222485A1 (en) An analytical device for quantitative measurement of the concentration of analytes in a liquid sample
CN110618465B (en) Article detection method and apparatus
CN112595734A (en) Liquid identification method and device based on Wi-Fi signal
CN116597521B (en) Human track and trunk activity recognition method and system
CN116699590B (en) FMCW multi-target ranging method and system based on 5.8G microwave radar

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination