A method and system for detecting a single component content of urine

Abstract

The application discloses a method and system for detecting the content of a single component in urine. Since electromagnetic wave signals and light signals have similar propagation characteristics in a medium, the application directly represents the relationship between the strength of a radio frequency signal and the concentration of urine based on the Beer-Lambert law, and builds a double-tag sensing system based on passive Internet of Things communication technology, namely radio frequency identification. The application uses a designed convolutional neural network model to train and test the collected data, and realizes the sensing of the slight change in the concentration of a component in urine. The method greatly improves the identification granularity of RFID applied to material sensing, and makes it possible to monitor urine health in a home environment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of liquid identification, and relates to a method and system for detecting the content of a single component in urine based on radio frequency identification technology. BACKGROUND

[0002] Urine analysis can effectively reflect the health status of the human body and provide early diagnosis basis for chronic diseases. For example, the determination of inorganic salts in urine is of great significance for understanding the water and salt metabolism in the body, the acid-base balance and the endocrine function of the kidney; the content of urine sugar can monitor diabetes; and the level of uric acid can distinguish gouty nephropathy from renal tubular dysfunction. However, the existing urine analysis relies on optical spectrum technology to realize quantitative analysis of components, and is limited to specific scenes such as hospitals or laboratories. Urine routine detection must be carried out in hospitals, which is complicated and time-consuming. Therefore, it is of great significance to realize rapid and low-cost urine component detection.

[0003] Using wireless signals to identify urine does not require carrying any large equipment and does not cause damage to the body, and has the advantages of being fast, convenient and non-intrusive. If people can detect key indicators in urine at home earlier and faster, they can diagnose as early as possible to improve treatment efficiency.

[0004] Existing liquid identification schemes based on wireless signals, such as radio frequency signals, UWB signals, WiFi signals, millimeter wave signals and sound signals, are based on the principle that due to the inherent complex permittivity properties of liquids, the wireless signals will produce different changes after being reflected or penetrated by different liquids, which are reflected in the strength and phase of the signals. By extracting and analyzing these differences, the unique characteristics of a certain liquid can be obtained to realize the identification of the liquid. However, the existing method relies on the large difference in complex permittivity between different materials, and it is difficult to detect trace components in urine. Urine is a mixed solution composed of water, protein, urea, uric acid, inorganic salt and other components. The existing method focuses on distinguishing the types of solutions rather than exploring the components and concentrations of the mixed solution. Therefore, it is still a challenge to use wireless signal sensing methods to detect the components in urine. SUMMARY

[0005] In view of the defects or deficiencies of the prior art, the application provides a method and system for detecting the content of a single component in urine.

[0006] Therefore, the method for detecting the content of a single component in urine provided by the application comprises the following steps:

[0007] A standard concentration characteristic curve graph of known concentration gradient solutions is generated, the standard concentration characteristic curve graph being composed of concentration characteristic curves of a plurality of known concentration solutions located on the same coordinate; the known gradient concentration solutions are prepared by adding different amounts of the single component to human urine or artificial urine;

[0008] generating a concentration characteristic curve of the to-be-detected urine under the coordinates, and determining the concentration of the measured component in the to-be-detected urine according to a position of the concentration characteristic curve of the to-be-detected urine;

[0009] The concentration characteristic curve of each known concentration solution or the concentration characteristic curve of the to-be-detected urine is acquired by using an antenna, a reader, a first RFID tag and a second RFID tag, and the method comprises the following steps:

[0010] (1) placing the to-be-detected solution in a container; the first RFID tag and the second RFID tag are attached to the container, and the first RFID tag and the second RFID tag are located on two sides of the to-be-detected solution, and the first RFID tag, the second RFID tag and the antenna are located on the same straight line;

[0011] (2) the antenna transmits a frequency signal, the frequency signal is received by the first RFID tag, and then passes through the to-be-detected solution and is received by the second RFID tag; the first RFID tag and the second RFID tag receive the signal and backscatter the signal to the reader, the reader reports a first initial RSSI according to the signal backscattered by the first RFID tag, and reports a second initial RSSI according to the signal backscattered by the second RFID tag; the antenna transmitting frequency range corresponding to each solution is the same; the sampling period depends on the sampling frequency of the reader; the frequency range is 860-960 MHz;

[0012] (3) the first amplitude and the second amplitude corresponding to each frequency are calculated according to formula I; the difference between the first amplitude and the second amplitude corresponding to each frequency and the ratio of the first amplitude are the concentration characteristic values under the corresponding frequencies;

[0013]

[0014] wherein A is the first amplitude or the second amplitude under the corresponding frequency; RSSI is the first initial RSSI or the second initial RSSI under the corresponding frequency; that is, when RSSI takes the first initial RSSI under the corresponding frequency, A is the first amplitude under the corresponding frequency, and when RSSI takes the second initial RSSI under the corresponding frequency, A is the second amplitude under the corresponding frequency;

[0015] (4) generating a frequency-concentration characteristic curve S1 by taking the frequency as the horizontal coordinate and the concentration characteristic value as the vertical coordinate;

[0016] (5) performing smoothing processing on the frequency-concentration characteristic curve S1 to obtain a concentration characteristic curve S2.

[0017] Optionally, the single component is sodium chloride, glucose or uric acid.

[0018] Optionally, the reader is an RFID reader.

[0019] Optionally, the frequency range of the frequency signal transmitted by the antenna is 860-960 MHz.

[0020] Optionally, the smoothing processing adopts an S-G smoothing filtering algorithm.

[0021] Optionally, the concentration of the measured component in the to-be-detected urine is determined according to a position of a concentration characteristic curve of the to-be-detected urine and a similarity between the concentration characteristic curve of the to-be-detected urine and standard concentration characteristic curves of the known concentration gradient solutions. Further optionally, a CNN model is trained by using a plurality of standard concentration characteristic curve graphs of the measured component, and a concentration recognition model is constructed; and the concentration characteristic curve of the to-be-detected urine is determined according to the concentration recognition model.

[0022] The application also provides a urine single-component content detection system, which comprises the antenna, the two RFID tags, the reader and the processor.

[0023] The antenna, the reader, the two RFID tags and the processor detect the concentration of the single component in the to-be-detected urine by using the above method.

[0024] The artificial urine component concentration detection method based on the radio frequency identification technology is a fine-grained solution concentration sensing method, which senses a target solution by means of an RFID device, extracts a concentration characteristic curve by analogy with Beer's law, and classifies by means of a designed convolutional neural network model, and can perform fine-grained identification on concentration similar artificial urine with high accuracy.

[0025] The application uses an RFID device to perform passive urine identification, and is a fine-grained liquid identification method based on wireless sensing. Unlike the scene limitation and high cost of traditional technologies such as spectrometers, the method uses an RFID reader to perform data transmission and reception, etc., which facilitates scientific research and daily life needs, and provides a new idea for home urine health detection. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 The hardware components of the radio frequency identification system used in the application.

[0027] Figure 2 The upper and lower label display diagrams of the application.

[0028] Figure 3 The CNN model processing process diagram of the application.

[0029] Figure 4The frequency-initial concentration characteristic map of artificial urine with different glucose, sodium chloride and uric acid concentrations in the embodiments of the present application; the curves ①-⑨ in figure (a) correspond to glucose solutions with concentrations of 58 g / L, 56 g / L, 54 g / L, 52 g / L, 50 g / L, 48 g / L, 46 g / L, 44 g / L and 42 g / L, respectively; the curves ①-⑨ in figure (b) correspond to sodium chloride solutions with concentrations of 58 g / L, 56 g / L, 54 g / L, 52 g / L, 50 g / L, 48 g / L, 46 g / L, 44 g / L and 42 g / L, respectively; the curves ①-⑨ in figure (c) correspond to uric acid solutions with concentrations of 18 g / L, 16 g / L, 14 g / L, 12 g / L, 10 g / L, 8 g / L, 6 g / L, 4 g / L and 2 g / L, respectively.

[0030] Figure 5 The frequency-concentration characteristic map of artificial urine with different glucose, sodium chloride and uric acid concentrations in the embodiments of the present application; the curves ①-⑨ in figure (a) correspond to glucose solutions with concentrations of 58 g / L, 56 g / L, 54 g / L, 52 g / L, 50 g / L, 48 g / L, 46 g / L, 44 g / L and 42 g / L, respectively; the curves ①-⑨ in figure (b) correspond to sodium chloride solutions with concentrations of 58 g / L, 56 g / L, 54 g / L, 52 g / L, 50 g / L, 48 g / L, 46 g / L, 44 g / L and 42 g / L, respectively; the curves ①-⑨ in figure (c) correspond to uric acid solutions with concentrations of 18 g / L, 16 g / L, 14 g / L, 12 g / L, 10 g / L, 8 g / L, 6 g / L, 4 g / L and 2 g / L, respectively.

[0031] Figure 6 The confusion matrix diagram of the classification accuracy of the CNN model for artificial urine with different glucose, sodium chloride and uric acid concentrations. DETAILED DESCRIPTION

[0032] Unless otherwise defined, scientific and technical terms used herein are understood according to the common usage of the same in the relevant art.

[0033] When the radio frequency signal passes through the liquid, the signal will be decelerated and attenuated, which is reflected in the phase and intensity of the output signal, and the degree of deceleration and attenuation will be different due to the difference of the target characteristics. In addition, according to the Beer-Lambert law, there is a certain relationship between the strength of the liquid absorption of light of a certain wavelength and the concentration of the light-absorbing substance. At the same time, both electromagnetic waves and light waves have wave-particle duality, and the propagation in the medium is very similar, so when the electromagnetic wave signal propagates in the dilute solution, it also satisfies the theorem. It is known that 92% to 99% of the composition of urine is water, and the remaining elements are trace elements, which meets the relevant conditions of the dilute solution. Based on this, the present application uses the correlation between the concentration of the composition of the urine and the strength of the radio frequency signal in the radio frequency identification system to detect the concentration of a single component in the urine or artificial urine.

[0034] In order to characterize the relationship between the absorption signal intensity of the solution and the total incident intensity, the present application finds that the ratio of the absorption signal intensity to the total incident intensity of the solution to be tested with different component concentrations is linearly related to the component concentration, so that the relationship can be used for the detection of a single component. During the specific detection, the two sides (i.e. the two sides of the solution) of the container containing the solution to be detected, such as the top and the bottom (or the left and the right) are respectively pasted with RFID tags (no interference material is present beside the container tag during the detection), and the detection is carried out based on the change of the signal reflected by the two RFID tags and the relationship between the solution concentration.

[0035] The main hardware components of the radio frequency identification system used in the present application include an antenna, two RFID tags, a reader, as shown in Figure 1 , and the two tag indication diagrams are as shown in Figure 2 ( Figure 1 The two tags shown are of the same specification and are respectively pasted on the top and the bottom of the beaker). During the signal transmission, the signal attenuation in the air between the upper tag and the liquid surface can be ignored, and the RSSI of the upper and lower tags can approximately represent the total intensity before the signal penetration and the transmission intensity after the penetration of the urine, so that the signal intensity attenuated in the urine can be represented as the difference between the two, that is, the absorption signal intensity is the RSSI of the upper tag minus the RSSI of the lower tag. Therefore, the concentration and the difference between the RSSI of the upper and lower tags and the ratio of the RSSI of the upper tag are linearly related. The linear relationship still holds after converting the RSSI value into the amplitude value.

[0036] In the present application, the frequency-concentration characteristic curve is drawn with the sampling frequency of the radio frequency signal as the abscissa and the corresponding concentration characteristic value as the ordinate. Since the resolution of the RSSI value collected by the RFID is relatively low, the characteristic curve of the graph is not smooth, and the present application further smoothes the curve. The concentration characteristic curve is obtained after the smoothing.

[0037] In a preferred solution, to automatically determine the concentration of the solution to be tested, the concentration recognition model can be determined according to the concentration of the component to be tested. Specifically, a plurality of standard concentration characteristic curve graphs of the solution containing the component to be tested (human urine or artificial urine) with a known concentration gradient are generated according to the method of the present application (a plurality of curve graphs can be obtained by repeatedly measuring the same group of gradient concentrations using the method of the present application), and the concentration recognition model of the component to be tested is obtained by training the CNN model using the plurality of generated standard concentration characteristic curve graphs; then the concentration of the component to be tested in the urine to be tested is determined by using the concentration recognition model according to the concentration characteristic curve of the solution to be tested, specifically by obtaining the probability of the position and linear similarity of the concentration characteristic curve of the solution to be tested and each curve in the standard concentration characteristic curve, and determining the concentration of the solution to be tested according to the concentration corresponding to the curve with the maximum probability.

[0038] The CNN model extracts and learns the value features and shape features of the frequency-concentration characteristic curve during training. The value features include the absorption characteristics of electromagnetic waves of different concentrations of liquid, and the shape features include the absorption characteristics of the same component concentration for different frequency electromagnetic waves. Through the learning of these two features, the CNN model has more accurate distinguishing ability for the content of the urine component. Different solutes are used to train the model, and after training, the respective network parameters (such as the weights in the convolution network) are obtained, and then the respective trained network is used to recognize the concentration of the corresponding solute.

[0039] The method of the present application obtains the linear relationship between the concentration of the urine component and the radio frequency signal intensity according to the Beer-Lambert law, thereby designing a double-tag sensing system based on radio frequency identification technology. Unlike previous methods for identifying liquid types, the present application directly represents the concentration characteristics, effectively detects the content of the urine component, and breaks through the limitations of the wireless sensing method for liquid recognition granularity. Unlike traditional urine detection methods, the present application is a fast, convenient and inexpensive solution, and provides a new idea and method for the detection of the content of the urine component.

[0040] The antenna model used below is EN-9028P, and the working frequency range is 902-928 MHz; the RFID electronic tag model is IMPINJ M4QT, and the working frequency range is 860-960 MHz, and the working frequency used in the following examples is 902.75-927.25 MHz; the RFID reader model is Impinj Speedway R420, and the frequency is sampled once every 0.5 MHz interval. To facilitate the representation of the standard concentration characteristic curve equation, the horizontal coordinate is represented as the count value of 50 frequency channels.

[0041] The known gradient concentration solution of the present application is a gradient concentration solution prepared by adding a component to be measured to an artificial urine as a base solution. The artificial urine is a simulation test solution developed through scientific research and strictly according to the composition of real human urine, and is a solution artificially configured and having chemical components similar to human urine, and the similarity of the chemical components and the content to real human urine is greater than 99.5%. Since natural human urine will deteriorate rapidly without refrigeration and will emit a foul odor, in addition, the urine of different individuals is different, and scientific experiments have strict requirements for the consistency of the urine before and after, therefore, artificial urine similar to human urine is selected for experimental research.

[0042] The specification of the artificial urine used in the following examples of the present application is pH = 6.0-6.7 (containing creatinine), and is specifically configured as: 3 g of disodium hydrogen phosphate, 0.75 g of potassium dihydrogen phosphate, 10 g of sodium chloride, 10 g of urea, 10 g of glucose, 1 g of thiomersal, and 250 mL of white blood cell solution. The examples add the component of interest (sodium chloride, glucose, or uric acid) to the above formulation to explain and illustrate the scheme of the present application.

[0043] Example 1:

[0044] In this example, three groups of samples to be measured are set up, and 0.1 g, 0.2 g, 0.3 g, 0.4 g, 0.5 g, 0.6 g, 0.7 g, 0.8 g, and 0.9 g of glucose, sodium chloride, and uric acid are added to 50 mL of artificial urine in each group, respectively. The concentration of glucose and sodium chloride in the sample to be measured after addition is in the range of 2-18 g / L, and the concentration of uric acid is in the range of 40-58 g / L.

[0045] Referring to Figure 2 In this example, the prepared solution is poured into a beaker, the upper label is pasted on the plastic wrap and covered on the mouth of the beaker, and the lower label is pasted on the bottom of the beaker. The reader emits a signal of a certain frequency through an antenna, the RFID tag receives the signal and backscatters the signal back to the reader, the reader receives and identifies the information sent back by the RFID tag, and detects the concentration of the solution according to the RSSI information received by the RFID reader; the calculation work of each step of this example is carried out in Matlab software.

[0046] Since the concentration characteristics depend on the signal strength received by the upper and lower labels, the distance between the reader antenna and the label, the attenuation of the signal in the solution, etc. will interfere with the accuracy of identification, for example, the closer the reader antenna is to the label, the greater the RSSI value of the label. In addition, the rotation of the RFID tag will also have some influence, therefore, the distance between the antenna and the beaker, the height of the solution, and the position of the label are kept unchanged during each data collection process to prevent errors. Six groups of data are collected for the solution at each concentration with three different solvents added respectively.

[0047] The frequency-raw concentration characteristic curve S1 obtained in this example is shown in Figure 4 ; the frequency-concentration characteristic curve S2 obtained by S-G filter smoothing processing is shown in Figure 5 ; and the concentration of each curve is the average of the 6 groups of data collected above. As can be seen from the results shown in Figure 5 , the concentration and the height of the curve position are positively related, and for a solution of unknown concentration, the concentration can be determined according to its position in the standard concentration characteristic curve graph of the same solute to be measured.

[0048] Example 2:

[0049] This example determines the concentration of the test solution by constructing a concentration identification model.

[0050] The CNN model used in this example is shown in Figure 3 , and its structure and principle are as follows: a 3-layer convolutional layer structure is used to satisfy the deep feature extraction, and the concentration and linear features contained in the input data are extracted; then two fully connected layers are used to realize the classification task: in each fully connected layer, first, a two-dimensional convolutional layer is used for feature extraction, then a BN layer is used for data normalization to speed up the convergence of the model training and make the model training process more stable; then a ReLU layer is used to enhance the non-linear expression ability of the model. Since the input of the two-dimensional convolutional layer needs a four-dimensional tensor, the amplitude bit feature is converted into a four-dimensional tensor and input into the CNN model, and after the last convolutional layer structure, the classification task is completed by two fully connected layers FC.

[0051] The CNN model training data set used in the following examples is: for each component (sodium chloride, glucose, uric acid), 50 standard concentration characteristic curve graphs of known concentration gradient solutions (repeated 50 times under the same concentration gradient) are collected, which are used as training sets to train the CNN model shown in Figure 4 .

[0052] In this example, the corresponding solute is added to the artificial urine used in the example when training the model. The concentration gradient of the glucose training data is set to: 58 g / L, 56 g / L, 54 g / L, 52 g / L, 50 g / L, 48 g / L, 46 g / L, 44 g / L, 42 g / L; the concentration gradient of the sodium chloride training data is set to: 58 g / L, 56 g / L, 54 g / L, 52 g / L, 50 g / L, 48 g / L, 46 g / L, 44 g / L, 42 g / L; and the concentration gradient of the uric acid training data is set to: 18 g / L, 16 g / L, 14 g / L, 12 g / L, 10 g / L, 8 g / L, 6 g / L, 4 g / L, 2 g / L.

[0053] The specific training process is as follows: maximum cross-entropy is used as the training loss in the training, and in the training process, the cross-entropy loss function is minimized, that is, the similarity between the true label and the model prediction result is maximized, so as to improve the classification accuracy of the model. In order to minimize the cross-entropy loss function, the gradient descent algorithm is used to update the parameter model. In each training iteration, first calculate the model prediction value by forward propagation, then calculate the cross-entropy loss function, if the value is greater than the expected loss value, then calculate the gradient by the back propagation algorithm and update the model parameters, until the loss value is equal to or less than the expected value, the training is ended.

[0054] The CNN model is trained by Adam optimizer during training. Adam iteratively updates the neural network weights based on the training data, and can quickly achieve good results. The initial learning rate is set to 0.001, and during the training process, the algorithm adaptively adjusts the learning rate according to the actual training situation. The weight decay value is set to 5 -4 During training, the amount of data captured by one training is 10, and the number of training iterations is 200, so as to ensure sufficient learning of the data features.

[0055] Further, the 6 groups of data obtained in Example 1 are classified by using the CNN model trained in this example, and the classification accuracy results are as shown in Figure 6 The greater the value on the diagonal, the higher the classification accuracy of the CNN model for the concentration.

Claims

1. A method for detecting the content of a single component in urine, the method comprising: generating a standard concentration characteristic curve of a known concentration gradient solution, the standard concentration characteristic curve being composed of concentration characteristics of a plurality of known concentration solutions located on the same coordinate; the known concentration gradient solution being prepared by adding different amounts of the single component to human urine or artificial urine; generating a concentration characteristic curve of the urine to be detected on the coordinate, and determining the concentration of the measured component in the urine to be detected according to the position of the concentration characteristic curve of the urine to be detected; the concentration characteristic curve of each known concentration solution or the concentration characteristic curve of the urine to be detected being obtained by using an antenna, a reader, a first RFID tag and a second RFID tag, the method comprising: (1) placing the solution to be detected in a container; the first RFID tag and the second RFID tag being attached to the container, and the first RFID tag and the second RFID tag being located on both sides of the solution to be detected, while the first RFID tag, the second RFID tag and the antenna are located on the same straight line; (2) the antenna emits a frequency signal, which is received by the first RFID tag, and then passes through the solution to be detected and is received by the second RFID tag; the first RFID tag and the second RFID tag receive the signal and backscatter it to the reader, and the reader reports a first initial RSSI according to the signal backscattered by the first RFID tag, and reports a second initial RSSI according to the signal backscattered by the second RFID tag; the antenna emission frequency range of each solution is the same; the sampling period depends on the sampling frequency of the reader; the frequency range is 860-960 MHz; (3) calculating the first amplitude and the second amplitude corresponding to each frequency according to formula I; the difference between the first amplitude and the second amplitude corresponding to each frequency and the ratio of the first amplitude are the concentration characteristic values at the corresponding frequencies; wherein: A is the first amplitude or the second amplitude at the corresponding frequency; RSSI is the first initial RSSI or the second initial RSSI at the corresponding frequency; (4) generating a frequency-concentration characteristic curve S1 with frequency as the horizontal coordinate and concentration characteristic value as the vertical coordinate; (5) performing smoothing processing on the frequency-concentration characteristic curve S1 to obtain a concentration characteristic curve S2. The single component is sodium chloride, glucose or uric acid. The reader is an RFID reader. The frequency range of the frequency signal emitted by the antenna is 860-960 MHz. The smoothing processing adopts an S-G smoothing filtering algorithm. The concentration of the measured component in the urine to be detected is determined according to the position of the concentration characteristic curve of the urine to be detected and the similarity between the concentration characteristic curve of the urine to be detected and the standard concentration characteristic curves of the known concentration gradient solutions. A CNN model is trained using a plurality of standard concentration characteristic curves of the measured component to construct a concentration recognition model, and the concentration characteristic curve of the urine to be detected is determined according to the concentration recognition model. The detection system provided comprises the antenna, the two RFID tags, the reader and the processor; The antenna, the reader, the RFID tags and the processor detect the concentration of the single component in the urine to be detected by using the method of claim 1. ​ 2. The method of claim 1, wherein the urine single constituent concentration is selected from the group consisting of: blood, bilirubin, urobilinogen, glucose, ketones, protein, nitrites, leukocytes, specific gravity, pH, and combinations thereof. ​ 3. The method of claim 1, wherein the method is for detecting a single component content of urine. ​ 4. The method of claim 1, wherein the method is used for detecting a single component content of urine. ​ 5. The method of claim 1, wherein the method is used for detecting a single component content of urine. ​ 6. The method of claim 1, wherein the method is used for detecting a single component content of urine. ​ 7. The method of claim 6, wherein the urine single constituent concentration is selected from the group consisting of: blood, bilirubin, urobilinogen, glucose, ketones, protein, nitrites, leukocytes, specific gravity, pH, and combinations thereof. ​ 8. A urine single constituent concentration detection system, comprising: ​ ​

Images