An infusion medicine category identification method, system, device and storage medium

Abstract

The application discloses a kind of infusion medicine class identification method, system, equipment and storage medium, infusion medicine class identification method includes by the mixing of millimeter wave transmission signal and reflected signal, obtain intermediate frequency signal, based on intermediate frequency signal, by fast fourier transform determine first data pair sequence, by all window first data pair sequence of euclidean distance function is deleted to abnormal data, second data pair sequence is input into the generator of well-trained, obtain the first feature vector output by the generator of well-trained;First feature vector is input into the well-trained classifier, obtains the class identification result of the infusion medicine to be identified output by the well-trained classifier, realizes lossless identification, improves the efficiency and accuracy of infusion medicine class identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of infusion medicine category identification, and in particular to an infusion medicine category identification method, system, device and storage medium. BACKGROUND

[0002] At present, since the solute component cannot be directly judged, only artificial marking can be relied on, which not only easily causes confusion or medicine deterioration, but also is difficult to realize non-destructive identification. The existing technology mostly adopts random sampling, and uses a vector network analyzer or a liquid chromatograph for destructive detection, which not only causes medicine pollution and waste, but also has low identification accuracy. SUMMARY

[0003] The present application aims to at least solve the technical problems existing in the prior art. To this end, the present application provides an infusion medicine category identification method, system, device and storage medium, which can realize non-destructive identification and improve the efficiency and accuracy of infusion medicine category identification.

[0004] In a first aspect, the present application provides an infusion medicine category identification method, comprising the following steps:

[0005] In the case of obtaining the millimeter wave transmission signal and the reflection signal of the infusion medicine to be identified, the millimeter wave transmission signal and the reflection signal are mixed to obtain an intermediate frequency signal;

[0006] Based on the intermediate frequency signal, a first data pair sequence is determined by fast Fourier transform, wherein the first data pair sequence includes a first center frequency value sequence and a first amplitude value sequence of all windows, and the window is obtained by using a window function on the intermediate frequency signal;

[0007] All the first data pair sequences of the windows are subjected to abnormal data deletion by an Euclidean distance function to obtain a second data pair sequence;

[0008] The second data pair sequence is input into a trained generator to obtain a first feature vector output by the trained generator;

[0009] The first feature vector is input into a trained classifier to obtain a category identification result of the infusion medicine to be identified output by the trained classifier. The infusion medicine category identification method according to the embodiments of the present application has at least the following beneficial effects:

[0010] The method firstly mixes the millimeter wave transmitting signal and the reflected signal to obtain an intermediate frequency signal, which can provide more accurate and stable data basis for subsequent steps; then based on the intermediate frequency signal, a first data pair sequence is determined by fast Fourier transform, which can extract frequency information and improve subsequent calculation efficiency; then the first data pair sequence of all windows is deleted by the Euclidean distance function to improve the accuracy of the data; finally, the second data pair sequence is input into the trained generator to obtain the first feature vector output by the trained generator; the first feature vector is input into the trained classifier to obtain the class recognition result of the to-be-identified infusion medicine output by the trained classifier, realizing lossless identification and improving the efficiency and accuracy of infusion medicine class recognition.

[0011] According to some embodiments of the present application, the second data pair sequence includes a second center frequency value sequence and a second amplitude value sequence, and the inputting of the second data pair sequence into the trained generator to obtain the first feature vector output by the trained generator includes:

[0012] The second data pair sequence is input into the trained generator, so that the trained generator determines a first signal intensity sequence according to the second center frequency value sequence; so that the trained generator normalizes the second amplitude value sequence to obtain a first normalized amplitude sequence; and so that the activation layer of the trained generator extracts a second feature vector based on the second center frequency value sequence and the second amplitude value sequence.

[0013] The trained generator determines the first feature vector based on the first signal intensity sequence, the first normalized amplitude sequence and the second feature vector, and outputs the first feature vector.

[0014] According to some embodiments of the present application, the determination of the first feature vector based on the first signal intensity sequence, the first normalized amplitude sequence and the second feature vector includes:

[0015] The dimension of the second feature vector is obtained.

[0016] The first signal intensity sequence is expanded based on the dimension of the second feature vector to obtain a first expanded intensity sequence.

[0017] The first feature vector is obtained by feature fusion based on the second feature vector, the first normalized amplitude sequence and the first expanded intensity sequence.

[0018] According to some embodiments of the present application, before the inputting the second data pair sequence into the trained generator to enable the trained generator to determine a first signal intensity sequence according to the second center frequency value sequence, the method comprises:

[0019] obtaining a relative dielectric constant of each medium layer, wherein the medium layer comprises an air layer, a container layer and a liquid layer;

[0020] determining a total signal intensity function based on the relative dielectric constant of each medium layer, a preset propagation phase delay parameter and a Fresnel reflection formula;

[0021] constructing a preset base function based on the total signal intensity function;

[0022] The inputting the second data pair sequence into the trained generator to enable the trained generator to determine a first signal intensity sequence according to the second center frequency value sequence comprises:

[0023] inputting the second data pair sequence into the trained generator to enable the trained generator to determine the first signal intensity sequence based on the second center frequency value sequence through the preset base function.

[0024] According to some embodiments of the present application, the removing abnormal data from all the first data pair sequences of the windows through the Euclidean distance function to obtain a second data pair sequence comprises:

[0025] calculating the Euclidean distance between each two first data pairs in the first data pair sequence through the Euclidean distance function;

[0026] calculating the average value of the Euclidean distance between each first data pair and all other first data pairs in the first data pair sequence as an average Euclidean distance;

[0027] determining the median value of all the average Euclidean distances in the first data pair sequence and taking the median value as an abnormal number threshold;

[0028] removing the first data pairs with the average Euclidean distance greater than the abnormal number threshold in the first data pair sequence to obtain the second data pair sequence.

[0029] According to some embodiments of the present application, the training process of the generator comprises:

[0030] constructing a first adversarial generative network model and a training data set, wherein the first adversarial generative network model comprises a first generator, a first discriminator and a first classifier, and the training data set is a labeled historical data pair sequence, and the historical data pair sequence comprises a third center frequency value sequence and a third amplitude value sequence;

[0031] inputting the training data set into the first generator, so that the trained generator determines a second signal intensity sequence according to the third sequence of center frequency values and the preset basis function; so that the first generator normalizes the third sequence of amplitude values to obtain a second normalized amplitude sequence; so that an activation layer of the first generator extracts a third feature vector based on the third sequence of center frequency values and the third sequence of amplitude values;

[0032] obtaining a dimension of the third feature vector;

[0033] based on the dimension of the third feature vector, expanding the second signal intensity sequence to obtain a second expanded intensity sequence;

[0034] performing feature fusion based on the third feature vector, the second normalized amplitude sequence and the second expanded intensity sequence to obtain a fourth feature vector;

[0035] determining a first generator loss value based on the fourth feature vector, the second expanded intensity sequence and the third sequence of amplitude values;

[0036] inputting the fourth feature vector into the first discriminator, so that the first discriminator determines a first discriminator loss value through a cross-entropy loss function;

[0037] inputting the fourth feature vector into the first classifier, so that the first classifier determines a first classifier loss value through a cross-entropy loss function;

[0038] updating the first generative adversarial network model based on the first generator loss value, the first discriminator loss value and the first classifier loss value to obtain a trained generative adversarial network model, wherein the trained generative adversarial network model comprises the trained generator.

[0039] According to some embodiments of the present application, the first generator loss value is determined based on the fourth feature vector, the second expanded intensity sequence and the third sequence of amplitude values, comprising:

[0040] determining a mean square error loss value through a mean square error loss function based on the fourth feature vector and the second expanded intensity sequence;

[0041] filtering out a maximum third amplitude value in the third sequence of amplitude values;

[0042] determining an amplitude loss value based on the maximum third amplitude value, the fourth feature vector and the third sequence of amplitude values;

[0043] Determine the first generator loss value based on the mean square error loss value and the amplitude loss value.

[0044] In a second aspect, the application provides a transfusion medicine category identification system, which comprises:

[0045] A mixing module is configured to mix the millimeter wave transmission signal and the reflection signal to obtain an intermediate frequency signal.

[0046] A first data pair sequence determination module is configured to determine a first data pair sequence by fast Fourier transform based on the intermediate frequency signal, wherein the first data pair sequence comprises a first center frequency value sequence and a first amplitude value sequence of all windows, and the window is obtained by using a window function on the intermediate frequency signal.

[0047] An abnormal data deletion module is configured to delete abnormal data from the first data pair sequence of all windows by using a Euclidean distance function to obtain a second data pair sequence.

[0048] A first feature vector output module is configured to input the second data pair sequence into a trained generator to obtain a first feature vector output by the trained generator.

[0049] A category identification module is configured to input the first feature vector into a trained classifier to obtain a category identification result of the transfusion medicine to be identified output by the trained classifier.

[0050] The system first mixes the millimeter wave transmission signal and the reflection signal to obtain an intermediate frequency signal, which can provide more accurate and stable data for subsequent steps. Then, the first data pair sequence is determined by fast Fourier transform based on the intermediate frequency signal, which can extract frequency information and improve subsequent calculation efficiency. Then, the abnormal data of the first data pair sequence of all windows is deleted by using a Euclidean distance function, which improves the accuracy of the data. Finally, the second data pair sequence is input into the trained generator to obtain the first feature vector output by the trained generator. The first feature vector is input into the trained classifier to obtain the category identification result of the transfusion medicine to be identified output by the trained classifier, which realizes lossless identification and improves the efficiency and accuracy of transfusion medicine category identification.

[0051] In a third aspect, the application provides a transfusion medicine category identification electronic device, which comprises at least one control processor and a memory connected in communication with the at least one control processor. The memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the above-mentioned transfusion medicine category identification method.

[0052] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer executable instructions for causing a computer to execute the infusion medicine category identification method.

[0053] It should be noted that the beneficial effects between the second aspect to the fourth aspect of the present application and the prior art are the same as the beneficial effects between the above-mentioned infusion medicine category identification system and the prior art, which will not be described here.

[0054] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0055] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, wherein:

[0056] Figure 1 is a flowchart of an embodiment of the infusion medicine category identification method provided by the present application;

[0057] Figure 2 is a flowchart of a millimeter wave radar data acquisition system in an embodiment of the present application;

[0058] Figure 3 is a structural schematic diagram of an embodiment of the infusion medicine category identification system provided by the present application;

[0059] Figure 4 is a structural schematic diagram of an embodiment of the electronic device provided by the present application. DETAILED DESCRIPTION

[0060] Embodiments of the present application are described in detail below with reference to the accompanying drawings. Examples of the embodiments are shown in the drawings, in which the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0061] In the description of the present application, if there is a description of first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the sequence of indicated technical features.

[0062] In the description of the present application, it should be understood that the orientation description, such as the orientation or position relationship indicated by the above, below, etc. is based on the orientation or position relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0063] In the description of the present application, it should be noted that, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and the person skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical solution.

[0064] At present, since the solute component cannot be directly judged, only manual marking can be relied on, which not only easily causes confusion or drug deterioration, but also is difficult to realize nondestructive identification. The existing technology mostly adopts random sampling, and uses a vector network analyzer or a liquid chromatograph for destructive detection, which not only causes drug pollution and waste, but also has low recognition accuracy.

[0065] In order to solve the above technical defects, the embodiments of the present application provide a kind of infusion medicine class identification method, system, equipment and storage medium.

[0066] Please refer to Figure 1 It is a kind of infusion medicine class identification method provided by the embodiments of the present application, and the method is applied to electronic equipment, which can be a server, etc. As shown in Figure 1 The infusion medicine class identification method includes:

[0067] Step S101, in the case of obtaining the millimeter wave transmission signal and the reflection signal of the infusion medicine to be identified, the millimeter wave transmission signal and the reflection signal are mixed to obtain an intermediate frequency signal;

[0068] Please refer to Figure 2In the step S101, in the case of obtaining the millimeter wave transmission signal and the reflection signal of the infusion medicine to be identified, the millimeter wave transmission signal and the reflection signal are mixed to obtain the intermediate frequency signal. The millimeter wave transmission signal and the reflection signal of the infusion medicine to be identified are received, and the millimeter wave transmission signal and the reflection signal are mixed by the millimeter wave radar to obtain the intermediate frequency signal in the case of collecting and uploading the millimeter wave transmission signal and the reflection signal of the infusion medicine to be identified by the automatic sliding rail device and the millimeter wave radar data collection system constructed based on the millimeter wave radar sensor and the data collection board. The motion control part of the sliding rail of the automatic sliding rail device can be an Arduino Uno single-chip microcomputer and a DM542 step motor driver, and the driver controls the linear motion of the 57-type step motor to complete the predetermined track. The millimeter wave radar sensor can be an AWR1843 millimeter wave radar sensor, and the data collection board can be a DCA1000EVM data collection board.

[0069] The application realizes high-precision and repeatable positioning movement of the target to be measured by the automatic sliding rail device, avoids measurement errors caused by manual operation, and the millimeter wave radar data collection system is responsible for transmitting a frequency-modulated continuous wave (FMCW) signal and receiving a reflection signal from the target, completes the collection and digitization of the echo signal, and provides a raw data basis for subsequent signal processing and classification tasks.

[0070] In step S102, a first data pair sequence is determined based on the intermediate frequency signal by fast Fourier transform, wherein the first data pair sequence includes a first center frequency value sequence and a first amplitude value sequence of all windows, and the window is obtained by using a window function on the intermediate frequency signal.

[0071] The above-mentioned intermediate frequency signal can be four-dimensional data in the form of frame, chirp, receiving channel and sampling point.

[0072] Step S102 can include steps S1021 to S1027:

[0073] In step S1021, since the data collected at each position is the same for each frame, the intermediate frequency signal is divided into multiple three-dimensional data in the form of chirp, receiving channel and sampling point by frame.

[0074] In step S1022, a window sliding in the sampling point dimension is defined, and for the kth window, the starting index is defined as , and the ending index is defined as , so as to obtain the window size (win_size), wherein the window size is equal to the ending index minus the starting index.

[0075] In step S1023, the three-dimensional data (dimension of chirp number channel number Extract window data from the current window (number of sampling points);

[0076] Step S1023: Apply a Hanning window function (a three-dimensional matrix with a size of chirp number) to the window data. Number of channels (Window size), to obtain the data after windowing;

[0077] Step S1024: Multiply the windowed data by a scaling factor pre-set according to actual needs to obtain normalized data, in order to compensate for the energy loss caused by the window function and achieve normalization;

[0078] Step S1025: Perform a Fast Fourier Transform (FFT) along the sampling point dimension on the windowed and scaled data to transform the windowed and scaled data from the time domain to the frequency domain, and obtain frequency domain data;

[0079] Step S1026: Calculate the amplitude spectrum of the frequency domain data using the following formula:

[0080] ;

[0081] in, For the first One chirp The first channel The amplitude spectrum of each frequency sampling point For the first One chirp The first channel Frequency domain data of each frequency sampling point;

[0082] Step S1027: Pre-set a chirp according to actual needs. The first number is calculated using the following formula. The average amplitude spectrum of each chirp:

[0083] ;

[0084] in, For the first One chirp The average amplitude spectrum of each frequency sampling point This represents the total number of channels. For the first One chirp The first channel The amplitude spectrum of each frequency sampling point;

[0085] Step S1027: Calculate the first amplitude value of the current window using the following formula:

[0086] ;

[0087] wherein, is a first amplitude value;

[0088] Step S1028, a first center frequency value of the current window is calculated by the following formula:

[0089] ;

[0090] wherein, is a first center frequency value of the current window, is a total number of sampling points; is a value set in advance according to actual needs;

[0091] Step S1029, the window is slid forward by a fixed step set in advance according to actual needs, and and are updated, and steps S1022 to S1028 are repeated until all data are processed, to obtain a first center frequency value sequence and a first amplitude value sequence of all windows.

[0092] Step S103, the first data pair sequence of all windows is subjected to abnormal data deletion by a Euclidean distance function, to obtain a second data pair sequence;

[0093] The second data pair sequence includes a second center frequency value sequence and a second amplitude value sequence.

[0094] Step S104, the second data pair sequence is input into the trained generator, to obtain a first feature vector output by the trained generator;

[0095] Step S105, the first feature vector is input into the trained classifier, to obtain a class recognition result of the infusion medicine to be identified output by the trained classifier.

[0096] The method first mixes the millimeter wave transmission signal and the reflected signal to obtain an intermediate frequency signal, which can provide a data basis with higher measurement accuracy and stronger stability for subsequent steps; then based on the intermediate frequency signal, a first data pair sequence is determined by fast Fourier transform, which can extract frequency information and improve subsequent calculation efficiency; then the first data pair sequence of all windows is subjected to abnormal data deletion by a Euclidean distance function, which improves the accuracy of the data; finally, the second data pair sequence is input into the trained generator, to obtain a first feature vector output by the trained generator; the first feature vector is input into the trained classifier, to obtain a class recognition result of the infusion medicine to be identified output by the trained classifier, which realizes lossless identification and improves the efficiency and accuracy of infusion medicine class recognition.

[0097] In some embodiments, step S104 can include steps S201 to S202:

[0098] In step S201, the second data pair sequence is input into the trained generator, so that the trained generator determines the first signal intensity sequence according to the second center frequency value sequence; so that the trained generator normalizes the second amplitude value sequence to obtain the first normalized amplitude sequence; and so that the activation layer of the trained generator extracts the second feature vector based on the second center frequency value sequence and the second amplitude value sequence.

[0099] In step S201, the second feature vector extracted by the activation layer of the trained generator based on the second center frequency value sequence and the second amplitude value sequence can be a second feature vector of the second center frequency value sequence and the second amplitude value sequence extracted by the activation layer of the trained generator.

[0100] In step S202, the trained generator is used to determine the first feature vector based on the first signal intensity sequence, the first normalized amplitude sequence, and the second feature vector, and output the first feature vector.

[0101] The present application improves the robustness and interpretability of the features by fusing multi-source signal features, thereby improving the efficiency and accuracy of infusion drug class recognition.

[0102] In some embodiments, step S202 can include steps S301 to S303:

[0103] In step S301, the dimension of the second feature vector is obtained.

[0104] In step S302, the first signal intensity sequence is expanded based on the dimension of the second feature vector to obtain a first expanded intensity sequence.

[0105] In step S302, the first signal intensity sequence is expanded based on the dimension of the second feature vector to obtain a first expanded intensity sequence, which can be expanding the dimension of the first signal intensity sequence to be the same as the dimension of the second feature vector, and taking the expanded first signal intensity sequence as the first expanded intensity sequence.

[0106] In step S303, the second feature vector, the first normalized amplitude sequence, and the first expanded intensity sequence are fused to obtain the first feature vector.

[0107] In step S303, the expression of the first feature vector obtained by fusing the second feature vector, the first normalized amplitude sequence, and the first expanded intensity sequence can be:

[0108] ;

[0109] ;

[0110] in, The pre-set gate weight matrix in the trained generator according to actual needs. This is the second feature vector. This is the first extended intensity sequence. This is the amplitude sequence after the first normalization. To integrate weights, The first eigenvector, This indicates element-wise multiplication.

[0111] This application achieves feature fusion based on a second feature vector, a first normalized amplitude sequence, and a first expanded intensity sequence. By complementing each other, the robustness of the features is improved, providing more accurate data for subsequent identification of infusion drug categories, thereby improving the efficiency and accuracy of infusion drug category identification.

[0112] In some embodiments, steps S401 to S403 may be included before step S201:

[0113] Step S401: Obtain the relative permittivity of each dielectric layer, wherein the dielectric layer includes an air layer, a container layer, and a liquid layer;

[0114] In step S401 above, the relative permittivity of each dielectric layer can be obtained by experimental measurement.

[0115] Step S402: Determine the total signal strength function based on the relative permittivity of each dielectric layer, the preset propagation phase delay parameter, and the Fresnel reflection formula;

[0116] In step S402 above, based on the relative permittivity of each dielectric layer, the preset propagation phase delay parameter, and the Fresnel reflection formula, the total signal strength function can be determined as follows:

[0117] Let the relative permittivity of the air layer be... The container layer is The liquid layer is Introducing the refractive index of non-magnetic materials With relative permittivity Relationship The refractive indices of the air, container, and liquid parts are respectively To simplify the formula, the refractive index will be used directly in the following formulas. This indirectly reflects the influence of the relative permittivity of each material. Based on Fresnel's reflection formula, the corresponding expression can be obtained:

[0118] The expressions for the reflection coefficient and transmission coefficient at the air-container wall interface are:

[0119] ;

[0120] ;

[0121] where, is the reflection coefficient at the air-container wall interface, is the transmission coefficient at the air-container wall interface.

[0122] Similarly, the expressions for the reflection and transmission coefficients at the container wall-air interface are:

[0123] ;

[0124] ;

[0125] where, is the reflection coefficient at the container wall-air interface, is the transmission coefficient at the container wall-air interface.

[0126] And the reflection and transmission coefficients at the container wall-liquid interface are:

[0127] ;

[0128] ;

[0129] where, is the reflection coefficient at the container wall-liquid interface, is the transmission coefficient at the container wall-liquid interface.

[0130] Based on the Fresnel electromagnetic wave propagation theory, the complex amplitude relationships of the reflection and transmission components can be determined. For the established three-layer medium model, the calculation process of the electric field amplitude is as follows:

[0131] Propagation process first step: let the incident wave electric field amplitude be Then at the air-container wall interface:

[0132] The reflected wave amplitude at the air-container wall interface is where, is the reflected wave amplitude at the air-container wall interface;

[0133] The transmitted wave amplitude at the air-container wall interface is where, is the transmitted wave amplitude at the air-container wall interface.

[0134] Propagation process second step: the transmitted wave propagates in the container wall layer to the container wall-liquid interface:

[0135] The wave amplitude value of the container wall-liquid interface that occurs reflection is wherein, is the wave amplitude value of the container wall-liquid interface reflected back to the container wall-air interface, is the imaginary unit in the complex domain, is the phase constant of the electromagnetic wave in the container wall medium, in a non-loss or low-loss medium, the value is equal to the wave number, and the value is (physical formula), is the speed of light, is the frequency, is the container wall thickness;

[0136] The wave amplitude value of the container wall-liquid interface that occurs transmission is wherein, is the wave amplitude value of the container wall-liquid interface.

[0137] The third step of the propagation process: when the reflected wave propagating back to the air-container wall interface from the container wall-liquid interface in the container wall:

[0138] The wave amplitude value of the second reflection is wherein, is the wave amplitude value of the reflected wave propagating back to the container wall-liquid interface;

[0139] The wave amplitude value of the transmission back to the air is wherein, is the wave amplitude value of the transmission back to the air.

[0140] In the subsequent calculation, according to the actual needs, the following is set in advance , characterizes the phase delay of the electromagnetic wave in the container wall medium layer for a round trip, and is defined as a phase shift, therefore, the corresponding phase factor of the electromagnetic wave in the container wall for a round trip is , is the phase factor, which is used to quantify the phase change caused by the wave path in the process of multiple reflections in the expression of the comprehensive reflection coefficient. Based on the above analysis of the signal propagation process, the signal is reflected multiple times in the container wall cavity, and the total electric field of the reflected signal at the receiving end of the radar signal is composed of different phase electric field components reflected from the container wall, and the specific expression is as follows:

[0141] ;

[0142] wherein, is the initial electric field when the signal is incident, is the first electric field component, is the second electric field component, is the first one of the electric field components.

[0143] Therefore, the total electric field formed by the multiple reflections of the electromagnetic wave inside the layered medium can be represented as the coherent superposition of a series of reflection components. Simplifying it into a compact closed-form analytical solution, the final expression of the total electric field is obtained as follows:

[0144] ;

[0145] Based on the basic physical law between the electromagnetic wave power and the electric field strength, the total signal strength is proportional to the square of the electric field amplitude, therefore, the total signal strength can be represented as:

[0146] ;

[0147] wherein, is the complex conjugate of .

[0148] Finally, substituting the complex part into the Euler formula and substituting , the expression of the total signal strength is simplified as:

[0149] ;

[0150] The expression of the total signal strength is analyzed from the perspective of the reflection signal propagation, which contains the material information, i.e. certain properties of the container and the liquid. The total signal strength corresponds to the signal amplitude finally obtained by the data processing module, which provides a theoretical basis for the data used in the present application. The data itself can reflect the independence information of the liquid to be measured. The reflection coefficient is a supplementary explanation to perfect the derivation of the above reflection process, which shows that the signal reflection is not a simple one or two times of reflection, or the reflection process of the container wall being ignored as an approximate ultra-thin surface.

[0151] Based on the polynomial-Fourier mixed expansion principle of the electromagnetic wave propagation theory and the expression of the total signal strength , according to the even function property of the cosine function, the expression of can be simplified as follows:

[0152] ;

[0153] ;

[0154] wherein, is the first intermediate parameter, is the second intermediate parameter, is the third intermediate parameter, The fourth intermediate parameter, This is the fifth intermediate parameter.

[0155] After rational function decomposition, the following expression is obtained:

[0156] ;

[0157] Through Fourier series expansion, we obtain the following expression:

[0158] ;

[0159] in, , , The sixth intermediate parameter, The seventh intermediate parameter, Let be the harmonic order in the Fourier expansion. Based on the principle of physical energy attenuation, in a multilayer medium reflection model, one round trip ( (equal to 1) and two round trips ( The reflection at point 2) already accounts for the majority of the energy. The energy from reflections at higher round-trip times contributes less to the total signal strength and can be ignored. Therefore, here... The value is 2.

[0160] Based on the above analysis, we get The expansion is:

[0161] ;

[0162] Will The expansion of is used as the total signal strength function mentioned above.

[0163] Step S403: Construct a preset basis function based on the total signal strength function;

[0164] In step S403 above, the preset basis functions constructed based on the total signal strength function can be used to compensate for non-ideal characteristics in the actual system, such as frequency response, phase shift, and nonlinear effects, and to ensure the accuracy and robustness of the theoretical model in the real millimeter-wave radar system. In addition to the three basis functions included in the expansion above, four other basis functions are introduced here. The expression for constructing the preset basis functions in step S403 can be as follows:

[0165] ;

[0166] in, As a preset basis function, For the first The second center frequency value of each window.

[0167] Step S201 may include step S404:

[0168] Step S404: Input the second data pair sequence into the trained generator so that the trained generator determines the first signal strength sequence based on the second center frequency value sequence and a preset basis function.

[0169] In step S404 above, the second data pair sequence is input into the trained generator so that the trained generator can determine the expression of the first signal intensity sequence based on the second center frequency value sequence as follows:

[0170] ;

[0171] in, For the first The first signal strength corresponding to the second center frequency value of each window The first weight parameter is pre-set according to actual needs in the trained generator. The second weight parameter is a pre-set parameter in the trained generator according to actual needs. The third weight parameter is pre-set in the trained generator according to actual needs. The fourth weight parameter is pre-set in the trained generator according to actual needs. The fifth weight parameter is pre-set according to actual needs in the trained generator. The sixth weight parameter is pre-set in the trained generator according to actual needs. The seventh weight parameter is pre-set according to actual needs in the trained generator. For parameters, The dielectric constant of the container for the infusion drug to be identified. For container wall thickness, It is the speed of light.

[0172] The expansion derived from the polynomial-Fourier hybrid expansion principle:

[0173] ;

[0174] This application determines the first signal intensity sequence by using a preset basis function, which enhances the physical interpretability of the first signal intensity sequence. By combining the learning ability of the generator with the physical constraints of the preset basis function, the accuracy, generation efficiency, and dynamic adaptability of the first signal intensity sequence are improved while ensuring its physical interpretability. This provides more accurate data for subsequent identification of infusion drug categories, thereby improving the efficiency and accuracy of infusion drug category identification.

[0175] In some embodiments, step S103 can include steps S501-S504:

[0176] Step S501, calculating the Euclidean distance between each two first data pairs in the first data pair sequence by a Euclidean distance function;

[0177] Step S502, calculating the average value of the Euclidean distance between each first data pair and all other first data pairs in the first data pair sequence as the average Euclidean distance;

[0178] Step S503, determining the median value of all average Euclidean distances in the first data pair sequence, and taking the median value as the anomaly number threshold;

[0179] Step S504, deleting the first data pairs in the first data pair sequence whose average Euclidean distances are greater than the anomaly number threshold, to obtain the second data pair sequence.

[0180] The application can effectively identify and eliminate the abnormal data group with higher deviation by combining the Euclidean distance and the dynamic anomaly number threshold to determine the abnormal data, and provides more accurate data basis for subsequent infusion drug category identification, thereby improving the efficiency and accuracy of infusion drug category identification.

[0181] In some embodiments, the training process of the generator can include steps S601-S609:

[0182] Step S601, constructing a first adversarial generation network model and a training data set, wherein the first adversarial generation network model includes a first generator, a first discriminator and a first classifier, and the training data set is a labeled historical data pair sequence, and the historical data pair sequence includes a third center frequency value sequence and a third amplitude value sequence;

[0183] Step S602, inputting the training data set into the first generator, so that the trained generator determines a second signal intensity sequence according to the third center frequency value sequence and a preset basis function; so that the first generator normalizes the third amplitude value sequence to obtain a second normalized amplitude sequence; and so that the activation layer of the first generator extracts a third feature vector based on the third center frequency value sequence and the third amplitude value sequence;

[0184] In step S602, the data determination process of making the trained generator determine the second signal intensity sequence according to the third center frequency value sequence and the preset basis function is similar to the data determination process of making the trained generator determine the first signal intensity sequence according to the second center frequency value sequence, which is not repeated here.

[0185] In step S602, the feature extraction process of extracting the third feature vector based on the third sequence of center frequency values and the third sequence of amplitude values for the activated layer of the first generator is similar to the feature extraction process of extracting the second feature vector based on the second sequence of center frequency values and the second sequence of amplitude values for the activated layer of the trained generator, which is not repeated here.

[0186] In step S603, the dimension of the third feature vector is obtained.

[0187] In step S604, the second signal intensity sequence is expanded based on the dimension of the third feature vector to obtain a second expanded intensity sequence.

[0188] In step S604, the expansion process of expanding the second signal intensity sequence based on the dimension of the third feature vector to obtain a second expanded intensity sequence is similar to the expansion process of expanding the first signal intensity sequence based on the dimension of the second feature vector to obtain a first expanded intensity sequence, which is not repeated here.

[0189] In step S605, the fourth feature vector is obtained by feature fusion based on the third feature vector, the second normalized amplitude sequence, and the second expanded intensity sequence.

[0190] In step S605, the feature fusion process of obtaining the fourth feature vector by feature fusion based on the third feature vector, the second normalized amplitude sequence, and the second expanded intensity sequence is similar to the feature fusion process of obtaining the first feature vector by feature fusion based on the second feature vector, the first normalized amplitude sequence, and the first expanded intensity sequence, which is not repeated here.

[0191] In step S606, the first generator loss value is determined based on the fourth feature vector, the second expanded intensity sequence, and the third sequence of amplitude values.

[0192] In step S607, the fourth feature vector is input into the first discriminator to enable the first discriminator to determine the first discriminator loss value through the cross-entropy loss function.

[0193] In step S607, the fourth feature vector is input into the first discriminator to enable the first discriminator to determine the first discriminator loss value through the cross-entropy loss function, which can be that the fourth feature vector is input into the first discriminator to enable the first discriminator to calculate the cross-entropy loss value through the cross-entropy loss function, and the cross-entropy loss value calculated by the first discriminator is taken as the first discriminator loss value.

[0194] In step S608, the fourth feature vector is input into the first classifier to enable the first classifier to determine the first classifier loss value through the cross-entropy loss function.

[0195] In step S608, inputting the fourth feature vector into the first classifier so that the first classifier determines the first classifier loss value through the cross-entropy loss function can be inputting the fourth feature vector into the first classifier so that the first classifier calculates the cross-entropy loss value through the cross-entropy loss function, and taking the cross-entropy loss value calculated by the first classifier as the first classifier loss value.

[0196] In step S609, updating the first generative adversarial network model based on the first generator loss value, the first discriminator loss value and the first classifier loss value to obtain a trained generative adversarial network model, wherein the trained generative adversarial network model comprises a trained generator.

[0197] The trained generative adversarial network model further comprises a trained discriminator and a trained classifier.

[0198] In step S609, updating the first generative adversarial network model based on the first generator loss value, the first discriminator loss value and the first classifier loss value to obtain a trained generative adversarial network model can be that when the first generator loss value is less than a first preset loss threshold value set according to actual requirements, the first discriminator loss value is less than a second preset loss threshold value set according to actual requirements, and the first classifier loss value is less than a third preset loss threshold value set according to actual requirements, taking the first generative adversarial network model as the trained generative adversarial network model.

[0199] The present application updates the first generative adversarial network model by combining the first generator loss value, the first discriminator loss value and the first classifier value to obtain a trained generative adversarial network model, thereby improving the accuracy of model training.

[0200] In some embodiments, step S606 can comprise steps S701 to S704:

[0201] In step S701, determining the mean square error loss value through the mean square error loss function based on the fourth feature vector and the second expanded intensity sequence;

[0202] In step S701, the specific calculation formula of determining the mean square error loss value through the mean square error loss function is as follows:

[0203] ;

[0204] wherein, is the mean square error loss value, is the total number of the training data set, is the total number of windows, is the total number of feature dimensions, represents the first sample a fourth feature vector of a fourth feature dimension of the first window of the first sample, a fourth feature vector of a fourth feature dimension of the first window of the first sample, a fourth feature vector of a fourth feature dimension of the first window of the first sample, a fourth feature vector of a fourth feature dimension of the first window of the first sample, a fourth feature vector of a fourth feature dimension of the first window of the first sample, a fourth feature vector of a fourth feature dimension of the first window of the first sample,

[0205] Step S702, screening out a maximum third amplitude value in the third amplitude value sequence;

[0206] Step S703, determining an amplitude loss value based on the maximum third amplitude value, the fourth feature vector and the third amplitude value sequence;

[0207] In step S703, a specific calculation formula for determining the amplitude loss value based on the maximum third amplitude value, the fourth feature vector and the third amplitude value sequence is as follows:

[0208] ;

[0209] wherein, is the amplitude loss value, is the fourth feature vector of the first window of the first sample, is the fourth feature vector of the first window of the first sample, is the fourth feature vector of the first window of the first sample, is the third amplitude value of the first window of the first sample, is the third amplitude value of the first window of the first sample, is the maximum third amplitude value, is the maximum third amplitude value, is the maximum third amplitude value, is the maximum third amplitude value.

[0210] Step S704, determining a first generator loss value based on the mean square error loss value and the amplitude loss value.

[0211] In step S704, a specific calculation formula for determining the first generator loss value based on the mean square error loss value and the amplitude loss value is as follows:

[0212] ;

[0213] wherein, is the first generator loss value, is a first hyperparameter, is a first hyperparameter, is a minimum positive value set in advance according to actual needs.

[0214] The present application can improve the accuracy and efficiency of model training by determining the first generator loss value based on the mean square error loss value and the amplitude loss value.

[0215] In addition, with reference to Figure 3In an embodiment of the present application, a transfusion medicine category identification system is provided, comprising a mixing module 1100, a first data pair sequence determination module 1200, an abnormal data deletion module 1300, a first feature vector output module 1400, and a category identification module 1500, wherein:

[0216] The mixing module 1100 is configured to mix the millimeter wave transmission signal and the reflection signal to obtain an intermediate frequency signal when the millimeter wave transmission signal and the reflection signal of the transfusion medicine to be identified are obtained.

[0217] The first data pair sequence determination module 1200 is configured to determine a first data pair sequence by fast Fourier transform based on the intermediate frequency signal, wherein the first data pair sequence comprises a first center frequency value sequence and a first amplitude value sequence of all windows, and the window is obtained by using a window function on the intermediate frequency signal.

[0218] The abnormal data deletion module 1300 is configured to delete abnormal data of the first data pair sequence of all windows by using a Euclidean distance function to obtain a second data pair sequence.

[0219] The first feature vector output module 1400 is configured to input the second data pair sequence into the trained generator to obtain the first feature vector output by the trained generator.

[0220] The category identification module 1500 is configured to input the first feature vector into the trained classifier to obtain the category identification result of the transfusion medicine to be identified output by the trained classifier.

[0221] The system first mixes the millimeter wave transmission signal and the reflection signal to obtain an intermediate frequency signal, which can provide more accurate and stable data basis for subsequent steps; then determines a first data pair sequence by fast Fourier transform based on the intermediate frequency signal, which can extract frequency information and improve subsequent calculation efficiency; then deletes abnormal data of the first data pair sequence of all windows by using a Euclidean distance function, which improves the accuracy of the data; finally, inputs the second data pair sequence into the trained generator to obtain the first feature vector output by the trained generator; inputs the first feature vector into the trained classifier to obtain the category identification result of the transfusion medicine to be identified output by the trained classifier, realizes lossless identification, and improves the efficiency and accuracy of transfusion medicine category identification.

[0222] It should be noted that the system embodiment and the above-mentioned method embodiment are based on the same inventive concept, so the related content of the above-mentioned method embodiment is also applicable to the system embodiment, which will not be repeated here.

[0223] Figure 4 A hardware structure schematic diagram of the transfusion medicine category identification provided by the embodiment of the present application is shown.

[0224] The infusion drug category identification device may include a processor 301 and a memory 302 storing computer program instructions.

[0225] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0226] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 302 is non-volatile solid-state memory.

[0227] In some embodiments, memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. Thus, generally, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of this disclosure.

[0228] The processor 301 reads and executes computer program instructions stored in the memory 302 to implement any of the infusion drug category identification methods in the above embodiments.

[0229] In one example, the infusion drug category identification device may also include a communication interface 303 and a bus 310. For example, Figure 4 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 310 and complete communication with each other.

[0230] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0231] Bus 310 includes hardware, software, or both, to couple components of the infusion drug class identification device to each other and to couple components to other components, such as a printer, keyboard, mouse, or monitor (not shown). While bus 310 is shown for the sake of clarity as a single bus, bus 310 can include one or more buses operating together. Bus 310 can be implemented using any suitable type of bus or buses, including, but not limited to, an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association local (VLB) bus, or another suitable bus or a combination of two or more of these. Where appropriate, bus 310 can include one or more bus controllers. Although this application describes and shows a particular bus, this application contemplates any suitable bus or interconnect.

[0232] The infusion drug class identification device can perform the infusion drug class identification method in the embodiments of the application based on a three-dimensional design model, thereby realizing the infusion drug class identification method and system described in the embodiments of the application. Figure 1 and Figure 3 The infusion drug class identification method and system described in the embodiments of the application.

[0233] In addition, in combination with the infusion drug class identification method in the above embodiments, the embodiments of the application can provide a computer storage medium to realize. The computer storage medium has computer program instructions stored thereon; the computer program instructions are executed by a processor to realize any one of the infusion drug class identification methods in the above embodiments.

[0234] It needs to be clear that the application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the application is 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 understanding the spirit of the application.

[0235] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0236] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0237] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0238] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. An infusion medicine category recognition method characterized by, The infusion medicine category recognition method comprises: In the case of obtaining the millimeter wave transmission signal and the reflection signal of the infusion medicine to be recognized, the millimeter wave transmission signal and the reflection signal are mixed to obtain an intermediate frequency signal; Based on the intermediate frequency signal, a first data pair sequence is determined by fast Fourier transform, wherein the first data pair sequence comprises a first center frequency value sequence and a first amplitude value sequence of all windows, and the window is obtained by using a window function on the intermediate frequency signal; All first data pair sequences of the window are deleted by an Euclidean distance function to obtain a second data pair sequence; The second data pair sequence is input into a trained generator to obtain a first feature vector output by the trained generator, wherein the training process of the generator comprises: A first adversarial generation network model and a training data set are constructed, wherein the first adversarial generation network model comprises a first generator, a first discriminator and a first classifier, and the training data set is a labeled historical data pair sequence, and the historical data pair sequence comprises a third center frequency value sequence and a third amplitude value sequence; The training data set is input into the first generator, so that the trained generator determines a second signal intensity sequence according to the third center frequency value sequence and a preset base function; so that the first generator normalizes the third amplitude value sequence to obtain a second normalized amplitude sequence; and so that the activation layer of the first generator extracts a third feature vector based on the third center frequency value sequence and the third amplitude value sequence; The dimension of the third feature vector is obtained; Based on the dimension of the third feature vector, the second signal intensity sequence is expanded to obtain a second expanded intensity sequence; Based on the third feature vector, the second normalized amplitude sequence and the second expanded intensity sequence, feature fusion is performed to obtain a fourth feature vector; Based on the fourth feature vector, the second expanded intensity sequence and the third amplitude value sequence, a first generator loss value is determined; The fourth feature vector is input into the first discriminator, so that the first discriminator determines a first discriminator loss value by a cross-entropy loss function; The fourth feature vector is input into the first classifier, so that the first classifier determines a first classifier loss value by a cross-entropy loss function; Based on the first generator loss value, the first discriminator loss value and the first classifier loss value, the first adversarial generation network model is updated to obtain a trained adversarial generation network model, wherein the trained adversarial generation network model comprises the trained generator; The first feature vector is input into a trained classifier to obtain a category recognition result of the infusion medicine to be recognized output by the trained classifier.

2. The infusion medicine category recognition method according to claim 1, characterized by, The second data pair sequence comprises a second center frequency value sequence and a second amplitude value sequence, and the second data pair sequence is input into a trained generator to obtain a first feature vector output by the trained generator, which comprises: inputting the second data pair sequence into the trained generator, so that the trained generator determines a first signal intensity sequence according to the second center frequency value sequence; so that the trained generator normalizes the second amplitude value sequence to obtain a first normalized amplitude sequence; so that an activation layer of the trained generator extracts a second feature vector based on the second center frequency value sequence and the second amplitude value sequence; determining the first feature vector based on the first signal intensity sequence, the first normalized amplitude sequence and the second feature vector through the trained generator, and outputting the first feature vector.

3. The infusion medicine category recognition method according to claim 2, characterized by, The determination of the first feature vector based on the first signal intensity sequence, the first normalized amplitude sequence and the second feature vector comprises: obtaining the dimension of the second feature vector; expanding the first signal intensity sequence based on the dimension of the second feature vector to obtain a first expanded intensity sequence; performing feature fusion based on the second feature vector, the first normalized amplitude sequence and the first expanded intensity sequence to obtain the first feature vector.

4. The infusion medicine category recognition method of claim 2, wherein, Before the inputting of the second data pair sequence into the trained generator, so that the trained generator determines a first signal intensity sequence according to the second center frequency value sequence, the method comprises: obtaining the relative dielectric constant of each medium layer, wherein the medium layer comprises an air layer, a container layer and a liquid layer; determining a total signal intensity function based on the relative dielectric constant of each medium layer, a preset propagation phase delay parameter and a Fresnel reflection formula; constructing the preset base function based on the total signal intensity function; The inputting of the second data pair sequence into the trained generator, so that the trained generator determines a first signal intensity sequence according to the second center frequency value sequence, comprises: inputting the second data pair sequence into the trained generator, so that the trained generator determines the first signal intensity sequence based on the second center frequency value sequence through the preset base function.

5. The infusion medicine category recognition method according to claim 4, characterized by, The deletion of abnormal data from all the first data pair sequences of the windows through the Euclidean distance function to obtain the second data pair sequence comprises: calculating the Euclidean distance between each two first data pairs in the first data pair sequence through the Euclidean distance function; calculating the average value of the Euclidean distance between each first data pair and all other first data pairs in the first data pair sequence as an average Euclidean distance; determining the median value of all the average Euclidean distances in the first data pair sequence, and taking the median value as an abnormal number threshold value; deleting the first data pairs with the average Euclidean distance greater than the abnormal number threshold value in the first data pair sequence to obtain the second data pair sequence.

6. The infusion medicine category recognition method according to claim 5, wherein, The determination of the first generator loss value based on the fourth feature vector, the second expanded intensity sequence and the third amplitude value sequence comprises: determining a mean square error loss value through a mean square error loss function based on the fourth feature vector and the second expanded intensity sequence; screening a maximum third amplitude value in the third amplitude value sequence; determining an amplitude loss value based on the maximum third amplitude value, the fourth feature vector and the third amplitude value sequence; determining the first generator loss value based on the mean square error loss value and the amplitude loss value.

7. An infusion medicine class identification system, characterized by, The infusion medicine category identification system comprises: a mixing module configured to mix the millimeter wave transmission signal and the reflection signal to obtain an intermediate frequency signal when the millimeter wave transmission signal and the reflection signal of the infusion medicine to be identified are obtained; a first data pair sequence determination module configured to determine a first data pair sequence by fast Fourier transform based on the intermediate frequency signal, wherein the first data pair sequence comprises a first center frequency value sequence and a first amplitude value sequence of all windows, and the window is obtained by using a window function on the intermediate frequency signal; an abnormal data deletion module configured to delete abnormal data of the first data pair sequence of all the windows by using a Euclidean distance function to obtain a second data pair sequence; a first feature vector output module configured to input the second data pair sequence into a trained generator to obtain a first feature vector output by the trained generator, wherein the training process of the generator comprises: constructing a first adversarial generation network model and a training data set, wherein the first adversarial generation network model comprises a first generator, a first discriminator and a first classifier, and the training data set is a labeled historical data pair sequence, and the historical data pair sequence comprises a third center frequency value sequence and a third amplitude value sequence; inputting the training data set into the first generator, so that the trained generator determines a second signal intensity sequence according to the third center frequency value sequence and a preset basis function; so that the first generator normalizes the third amplitude value sequence to obtain a second normalized amplitude sequence; so that the activation layer of the first generator extracts a third feature vector based on the third center frequency value sequence and the third amplitude value sequence; obtaining the dimension of the third feature vector; based on the dimension of the third feature vector, expanding the second signal intensity sequence to obtain a second expanded intensity sequence; performing feature fusion based on the third feature vector, the second normalized amplitude sequence and the second expanded intensity sequence to obtain a fourth feature vector; determining a first generator loss value based on the fourth feature vector, the second expanded intensity sequence and the third amplitude value sequence; inputting the fourth feature vector into the first discriminator to make the first discriminator determine a first discriminator loss value by using a cross-entropy loss function; inputting the fourth feature vector into the first classifier to make the first classifier determine a first classifier loss value by using a cross-entropy loss function; updating the first adversarial generation network model based on the first generator loss value, the first discriminator loss value and the first classifier loss value to obtain a trained adversarial generation network model, wherein the trained adversarial generation network model comprises the trained generator. The category recognition module is configured to input the first feature vector into a trained classifier to obtain a category recognition result of the to-be-recognized infusion medicine output by the trained classifier.

8. An infusion medicine category identification apparatus characterized by, The computer readable storage medium stores computer executable instructions for causing a computer to perform the infusion medicine category recognition method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer readable storage medium stores computer executable instructions for causing a computer to perform the infusion medicine category recognition method according to any one of claims 1 to 6.

Images