User medicine taking behavior detection system and method based on passive radio frequency identification technology

Abstract

The application discloses a user medicine taking behavior detection system and method based on passive radio frequency identification technology. Passive RFID tags are arranged on the surfaces of a water cup and a medicine bottle, radio frequency signals of the water cup and the medicine bottle in static and different motion states are collected, a multi-classification neural network model is used to establish a mapping relationship between the radio frequency signals and object states, and the multi-classification neural network model is deployed to a processor. A reader / writer is used to read transmission time, RSSI and phase data of the passive RFID tag radio frequency signals, and current states of the medicine bottle and the water cup are identified. If motion states of the medicine bottle and the water cup are both identified as translation and then rotation within a set time threshold, it is considered that a medicine taking action occurs, and a medicine taking time is recorded. If a medicine taking behavior is not detected within a medicine taking time range set by a user, the user is informed to take medicine in time through a mobile phone APP. The method can realize non-invasive detection without feeling, and has wide application value and market potential.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of digital data processing, and relates to collection and processing identification of sensor signals, in particular to a user medication behavior detection system and method based on passive radio frequency identification technology. BACKGROUND

[0002] Standard medication is the premise of ensuring effectiveness and safety. However, in daily life, due to irregular life and other reasons, medication is often missed or taken incorrectly. In order to remind users to take medication on time, many types of intelligent medicine boxes have appeared on the market. Common ones include alarm reminder type medicine boxes, mobile phone controlled intelligent medicine boxes, APP controlled intelligent medicine boxes, etc. However, these products all need power modules for power supply, and once the battery is exhausted, all functions will be lost. The charging type product needs the user to pay attention to the power cycle and charge in time. The battery powered product may cause environmental pollution problems.

[0003] Passive radio frequency identification (RFID) is an electronic data carrier without external power supply. When the RFID reader interrogates it, it sends back its own global unique electronic product code (EPC) and other stored information through backscattering modulation of the incident wave. Since the data returned to the reader end in RFID read-write communication through the inquiry protocol is in digital form, but the radio frequency wave carrying the data is in analog form, the response of the RFID tag has both digital and analog characteristics. Using the digital characteristics of the RFID response, high-speed communication between the tag and the reader and the Internet of Things gateway can be achieved, the uniqueness of the tag and the corresponding target daily object can be determined, and the static characteristics of the observed target can be described. Using the analog characteristics of the RFID response, the dynamic characteristics of the observed target can be described by observing the change of the physical signal received by the RFID reader over time. The combination of the two lays the foundation for daily activity monitoring based on passive radio frequency identification. By installing RFID tags on non-intelligent objects, the communication, sensing and interaction capabilities of non-intelligent objects can be expanded, the Internet of Things access of non-intelligent daily objects can be realized, more continuous user activity information can be provided for the intelligent home platform, and more personalized and user habit conforming intelligent home applications become possible.

[0004] Prior art 1 (CN107832647A) discloses a phase gesture recognition method based on passive tags, which detects user actions by wearing RFID tags on the specified positions of fingers. Prior art 2 (CN110575174A) discloses a method and system for activity recognition and behavior analysis, which uses a radio frequency identification reader to read the data of passive RFID tags installed in a home environment or worn by a user, and detects the interaction activities between the user and non-intelligent objects. These solutions all require wearing devices, which will affect the user's life to some extent. Therefore, it is necessary to design a truly unobtrusive solution to achieve passive, wireless, and non-intrusive user activity detection without affecting the user's daily life and being completely perceived by the user. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application proposes a user medication behavior detection system and method based on passive radio frequency identification technology, which installs RFID tags on the user's water cup and medicine bottle, combines a neural network deep learning method to recognize the signal strength and phase change of the RFID tags on the water cup and medicine bottle under different user actions, establishes a mapping relationship, and thus judges whether there is a medication behavior, reminds the user who has not taken medicine to take medicine on time at the user's set time, and realizes passive, wireless, and non-intrusive user activity detection.

[0006] The user medication behavior detection system based on passive radio frequency identification technology comprises a state acquisition module, a state analysis module, an action detection module, and a warning module.

[0007] The state acquisition module comprises passive RFID tags fixed on the surfaces of the water cup and the medicine bottle, and an RFID reader / writer fixed in the home environment. The RFID reader / writer reads the reading time, RSSI (Received Signal Strength Indicator), and phase data of the passive RFID tags, and sends them to the state analysis module.

[0008] The state analysis module integrates a trained multi-classification neural network model, which judges the motion state of the water cup or the medicine bottle as static, translation, or rotation after translation according to the RSSI and phase data of the tags.

[0009] The action detection module judges whether a medication action occurs according to the motion state of the water cup and the medicine bottle output by the state analysis module, and records the time when the medication action occurs.

[0010] The warning module is used to save the detection results of the action detection module, record the medication time, and notify the user through a mobile phone APP when no medication behavior is detected or multiple medication behaviors are detected within a predetermined time range.

[0011] The user medicine taking behavior detection method based on passive radio frequency identification technology specifically comprises the following steps:

[0012] Step 1, design an RFID tag that fits the curved outer wall of a water cup and a medicine bottle, and fix it on the surface of the medicine bottle and the water cup respectively. Ensure that when the medicine bottle and the water cup are placed stationary on the horizontal plane, the radiation direction of the tag antenna is horizontal and omnidirectional, and the vertical direction is the radiation blind area.

[0013] Step 2, test the radio frequency signal reading time, RSSI and phase data of the tags attached to the medicine bottle and the water cup during the stationary, translation, rotation and translation-then-rotation activities through the RFID reader-writer.

[0014] Step 3, pre-process the data collected in step 2, divide by sliding time window, and label each data corresponding user behavior, including stationary, translation and translation-then-rotation, to construct a training data set.

[0015] Step 4, use the training data set obtained in step 3 to train a multi-classification neural network model, and deploy it to the processor as a state detection model.

[0016] Step 5, read the radio frequency signal reading time, RSSI and phase data of the RFID tags on the medicine bottle and the water cup through the RFID reader-writer, and input them into the state detection model to identify the motion state of the medicine bottle and the water cup.

[0017] Step 6, according to the detection result of step 5, judge whether the user takes medicine or not and record the time of taking medicine. If the motion state of the medicine bottle and the water cup is detected as translation-then-rotation within the set time threshold, it is considered that a medicine taking action has occurred, and the time of taking medicine is recorded. If no medicine taking behavior is detected within the time range set by the user in advance, the user will be notified in time to take medicine through the mobile phone APP. If multiple medicine taking behaviors are detected within the preset medicine metabolism time, an excessive medicine taking warning will be issued through the mobile phone APP.

[0018] The present application has the following beneficial effects:

[0019] 1. The user medicine taking behavior detection system based on radio frequency identification technology proposed in the present application has simple structure, only includes an RFID reader-writer and two RFID tags, does not need traditional video equipment or sensors, does not need complex wiring and processing, is easy to install, and is suitable for various application scenarios.

[0020] 2. Passive design, no battery power supply, long service life, no environmental pollution, and the effective working range of wireless transmission can be up to more than ten meters and is not limited by position and azimuth angle, and the same reader-writer can support multiple devices and multiple users for simultaneous detection.

[0021] 3. Compared with traditional activity detection equipment, it is low in cost, simple in testing process, does not require users to wear any equipment, does not interfere with users' daily life, does not require users to perform any additional operations, and can complete the detection process without the user's awareness; after completing machine learning training in a fixed complex home environment, the detection results are not affected by the multipath effect of complex environment, showing high accuracy and high robustness. Attached Figure Description

[0022] Figure 1 Schematic diagram of RFID tag fixing;

[0023] Figure 2 The radiation pattern of the RFID tag antenna under different conditions;

[0024] Figure 3 A schematic diagram of a symmetrical folded dipole tag fixing method;

[0025] Figure 4 This is the accuracy result for detecting the movement state of the water cup in the example;

[0026] Figure 5 The results show the accuracy of the detection of the movement state of the medicine bottle in the examples.

[0027] Figure 6 The results show the accuracy of user medication behavior detection in this example. Detailed Implementation

[0028] The present invention will be further explained below with reference to the accompanying drawings;

[0029] The RFID tag's radio frequency signal RSSI is known to be... r 4 Inversely proportional, r It is the distance between the reader antenna and the tag, so when the medicine bottle or water cup is moved, the RSSI will change accordingly. rThe change of the RSSI is changed accordingly, but the change has randomness, such as the user may move the cup farther or closer or up and down, so that the RSSI decreases or increases or does not change, and therefore the moving action cannot be judged by the RSSI; but when the cup or the medicine bottle is displaced, the distance between the reader antenna and the tag changes, causing the wave path of the electromagnetic wave to change, causing the signal phase of the tag on the displaced object to change significantly, and the phase change is quite sensitive to the millimeter-level slight movement, so the change of the phase can be used to reflect the displacement action of the medicine bottle or the cup. When the medicine bottle or the cup rotates, the orientation of the tag antenna attached to the medicine bottle or the cup changes; if the direction of the tag antenna pointing to the reader antenna before and after rotation corresponds to the maximum radiation direction and the radiation blind area respectively, the RSSI before and after rotation will change sharply, so the change of the RSSI can be used to reflect the rotation action of the medicine bottle or the cup. Based on this, the application provides a user medicine taking behavior detection method based on passive radio frequency identification technology, and the specific steps are as follows:

[0030] Step 1: Design an RFID tag conformal to the curved outer wall of the cup and the medicine bottle. The RFID tag adopts a dipole antenna structure, as shown in Figure 1 , when fixed on the surface of the medicine bottle and the cup, the direction is perpendicular to the horizontal plane, thereby forming an omnidirectional radiation field in the plane perpendicular to the tag, and a radiation blind area in the direction along the length of the tag. The antenna of the RFID reader and the RFID tag antenna are installed on the same horizontal plane, when the cup or the medicine bottle rotates, because the dipole antenna on the cup or the medicine bottle is in a radiation blind area in the direction along the tag, a 90-degree rotation makes the radiation blind area change from the direction perpendicular to the ground to the direction parallel to the ground, so that the signal strength read by the reader is greatly reduced, and therefore the sharp attenuation of the RSSI can be used to judge the rotation of the medicine bottle or the cup, as shown in Figure 2 , the radiation pattern when the cup or the medicine bottle is static and tilted is simulated respectively.

[0031] As shown in Figure 3 , for a medicine bottle with a small height, the RFID tag can be designed as a symmetrical folded dipole, which does not change the radiation pattern of the dipole.

[0032] Step 2: Test the radio frequency signal reading time, RSSI and phase data of the tag attached to the medicine bottle and the cup in the static, translation, rotation, translation and rotation activities by the RFID reader respectively.

[0033] Step 3: In order to better establish the relationship between the RSSI and the phase of the backscattered signal read by the RFID reader and the user's activities, the measured signal RSSI and phase need to be processed to obtain as many characteristics as possible as input data to make the machine learning more effective.

[0034] Since the unit of RSSI is dBm, it is necessary to convert its unit to W first, so as to more intuitively reflect the proportional relationship with the tag antenna radiation gain. In order to eliminate the influence of the RSSI caused by the direct distance and azimuth angle between the reader antenna and the tag antenna and the multipath effect in the complex environment, the RSSI data in W is further normalized and the standard deviation is solved. Since the measured phase data is a folded phase, that is, it is limited in the interval [0, 2π], it is necessary to judge the continuity of the measured phase first, and the phase is unfolded by 2π compensation for the jumping phase. The phase difference and the standard deviation of the phase difference of the unfolded phase are calculated. And mark the corresponding user behavior of each data, including static, translation and rotation after translation, Table 1 shows the data format before and after processing, wherein states 1~3 correspond to static, translation and rotation after translation respectively:

[0035] Table 1

[0036] Time / s RSSI / dbm Phase RSSI / w Phase difference RSSI standard deviation Standard deviation of phase difference Status 0 -38.5 0.83448555 0.39768076 0.50143678 0.23426514 0.014552641 1 0.007 -38 0.85289332 0.4462916 0.50574713 0.23426514 0.014591914 1 0.0138 -38 0.84675739 0.4462916 0.5 0.23426514 0.014557471 1 0.01954 -38 0.86516516 0.4462916 0.50574713 0.23488891 0.014621985 1 0.02674 -38 0.85902924 0.4462916 0.5 0.23549919 0.014600656 1 0.03415 -38 0.84062147 0.4462916 0.49712644 0.23549919 0.014613543 1 0.03956 -38 0.85289332 0.4462916 0.50431034 0.23549919 0.014578864 1 0.04624 -38 0.84675739 0.4462916 0.5 0.23549919 0.014600656 1 0.05431 -38.5 0.84675739 0.39768076 0.50143678 0.23609608 0.014613543 1 0.06069 -38.5 0.85902924 0.39768076 0.50431034 0.23609608 0.014617894 1 3.68252 -35.5 0.86516516 0.79418253 0.50143678 1.06225397 0.026740964 2 3.68919 -35.5 0.88970886 0.79418253 0.50718391 1.07869171 0.026732401 2 3.69436 -35.5 0.87130109 0.79418253 0.49712644 1.09467649 0.026712893 2 3.70229 -35.5 0.88970886 0.79418253 0.50431034 1.11568327 0.026687356 2 3.70919 -35.5 0.87130109 0.79418253 0.5 1.13072401 0.026687356 2 3.71532 -35.5 0.87130109 0.79418253 0.50143678 1.14537021 0.026661652 2 3.72265 -35.5 0.89584478 0.79418253 0.50143678 1.16479385 0.026662495 2 3.72865 -35.5 0.87743701 0.79418253 0.51005747 1.18366342 0.026833951 2 3.7361 -35.5 0.88970886 0.79418253 0.5 1.20200501 0.026851675 2 3.7421 -35.5 0.88357293 0.79418253 0.50431034 1.21984245 0.026877173 2 11.1876 -35.5 1.36217494 0.79418253 0.50143678 4.71081532 0.046034534 3 11.1965 -35.5 0.88357293 0.79418253 0.5 4.75222063 0.048850556 3 11.202 -35.5 0.88357293 0.79418253 0.50143678 4.79270947 0.051846959 3 11.2117 -35.5 0.89584478 0.79418253 0.50574713 4.83230489 0.05582738 3 11.2195 -35.5 0.88970886 0.79418253 0.50143678 4.87102867 0.068746729 3 11.2252 -35.5 0.88970886 0.79418253 0.49712644 4.90890143 0.091891604 3 11.234 -35.5 0.88357293 0.79418253 0.50574713 4.94982742 0.09189946 3 11.2411 -35 0.88357293 0.8911739 0.49856322 4.98984588 0.09212918 3 11.2469 -35.5 0.9019807 0.79418253 0.50574713 5.0236553 0.09241866 3 11.2549 -35.5 0.9019807 0.79418253 0.49712644 5.06587018 0.092537108 3

[0037] The length of the moving window is set to 400 and the step is 200. The processed data is divided into more groups as input samples by sliding the time window, and a training data set is constructed.

[0038] Step 4, using the training data set obtained in step 3 to train an LSTM neural network as a state detection model, and deploying it to the processor.

[0039] Step 5, reading the reading time, RSSI and phase data of the RFID tag radio frequency signal on the water cup and the medicine bottle through the RFID reader, and inputting them into the processor, and identifying the motion state of the water cup and the medicine bottle through the state detection model. Figure 4 、 Figure 5 The comparison of the motion state detection results of the state detection model for the water cup and the medicine bottle with the true value respectively shows that the state detection model trained by the method can establish a more accurate mapping relationship between the radio frequency signal RSSI, phase and the motion state of the object, so as to identify the state change of the object.

[0040] Step 6, the process of taking medicine by the user is divided into two actions of taking medicine and drinking water. Among them, taking medicine can be regarded as the medicine bottle passing through translation and then rotating, so the signal of the tag on the medicine bottle will show phase change and then a large decrease in RSSI in a short time. Similarly, drinking water can also be regarded as the water cup passing through translation and then rotating, and the signal of the tag on the water cup also shows phase change and a large decrease in RSSI in a short time. Therefore, both the medicine bottle and the water cup need to have translation and then rotation in a short time, so as to consider that the complete action sequence of taking medicine is completed.

[0041] According to the detection result of step 5, if the motion state of the medicine bottle is detected as translation followed by rotation within 5 seconds, and the motion state of the cup is detected as translation followed by rotation, it is considered that a medication action has occurred, and the time when the motion state of the cup is translation followed by rotation is recorded as the medication time, as shown in the following table. Figure 6 The prediction result is compared with the time point of the recorded medication video, and the accuracy rate can reach 91%.

[0042] If no medication behavior is detected within the time range set by the user in advance, the user will be notified in time to take medicine through the mobile phone APP. If multiple medication behaviors are detected within the preset drug metabolism time, an over-dose warning will be issued through the mobile phone APP.

[0043] The user medication behavior detection system based on passive radio frequency identification technology includes a state collection module, a state analysis module, an action detection module, and a warning module.

[0044] The state collection module includes passive RFID tags fixed on the surface of the cup and the medicine bottle, and an RFID reader fixed in the home environment. The RFID reader reads the reading time, RSSI, and phase data of the passive RFID tag radio frequency signal and sends them to the state analysis module.

[0045] The state analysis module integrates a trained multi-classification neural network model to determine the motion state of the cup or medicine bottle as stationary, translation, or translation followed by rotation based on the RSSI and phase data of the tags.

[0046] The action detection module determines whether a medication action has occurred based on the motion state of the cup and medicine bottle output by the state analysis module, and records the time when the medication action occurs.

[0047] The warning module is used to save the detection results of the action detection module, record the medication time, and notify the user through the mobile phone APP if no medication behavior is detected within the predetermined time range or multiple medication behaviors are detected.

Claims

1. A method for detecting user medication behavior based on passive radio frequency identification technology, characterized in that: Passive RFID tags are set on the surface of water cups and medicine bottles respectively. The RSSI and phase data of the passive RFID tags are received by an RFID reader. The movement status of water cups and medicine bottles is determined based on the time-series changes of the RSSI value and phase data of the radio frequency tags received by the reader. A multi-class neural network model is used to establish the mapping relationship between radio frequency signals and object states, and the multi-class neural network model is deployed to the processor; The transmission time, RSSI, and phase data of the radio frequency signals of the passive RFID tags on the medicine bottle and water cup are read by the RFID reader and written, and then input into the processor. The motion state sequence of the medicine bottle and water cup is identified by a multi-class neural network model. The system determines the associated user behavior sequence based on the motion sequence of the medicine bottle and water cup. If the motion sequence of the medicine bottle and water cup is detected to be translation followed by rotation within the set time threshold, the corresponding user behavior sequence is considered to be a medication action, and the medication time is recorded. If no user behavior sequence of medication is detected within the user's preset medication time range, or if multiple user behavior sequences of medication are detected, an alert is pushed through the mobile APP.

2. The method for detecting user medication behavior based on passive radio frequency identification technology as described in claim 1, characterized in that: The passive RFID tag conforms to the outer wall of the medicine bottle or water cup, allowing it to fit snugly against the surface.

3. The method for detecting user medication behavior based on passive radio frequency identification technology as described in claim 1, characterized in that: The antenna of the passive RFID tag is in the form of a dipole.

4. The method for detecting user medication behavior based on passive radio frequency identification technology as described in claim 1, characterized in that: When medicine bottles and water cups are placed at rest, the radiation direction of the passive RFID tag antenna is omnidirectional in the horizontal direction, and the vertical direction is a radiation blind zone.

5. The method for detecting user medication behavior based on passive radio frequency identification technology as described in claim 1, characterized in that: The RSSI and phase data of radio frequency signals from water cups and medicine bottles are collected by RFID readers when they are stationary, in translation, and rotated after translation.

6. The method for detecting user medication behavior based on passive radio frequency identification technology as described in claim 5, characterized in that: The RSSI units are converted to W, then normalized, and the standard deviation is calculated as an RSSI feature; phase expansion is performed on the phase data transitions, and the phase difference between two adjacent data points and the standard deviation of the phase difference are calculated as phase features. Using RSSI and phase features at the same time as input samples and corresponding object states as labels, a training dataset is constructed to train a multi-class neural network model. This model is able to establish a mapping relationship between radio frequency signals and object states, and determine the associated user interaction actions and their timing by judging the object's motion state. Finally, the combination of timing actions is used to determine whether medication has been taken.

7. The method for detecting user medication behavior based on passive radio frequency identification technology as described in claim 6, characterized in that: The multi-class neural network model is an LSTM neural network, which can identify the state of an object as stationary, translated, rotated, or translated and then rotated based on the input radio frequency signal data.

8. A user medication behavior detection system based on passive radio frequency identification technology, characterized in that: It includes a status acquisition module, a status analysis module, an action detection module, and an early warning module; The status acquisition module includes passive RFID tags fixed to the surface of water cups and medicine bottles, and RFID readers fixed in the home environment; the RFID readers read the reading time, RSSI and phase data of the radio frequency signals of the passive RFID tags and send them to the status analysis module. The state analysis module integrates a trained multi-class neural network model to determine the motion state of a water cup or medicine bottle based on the temporal changes of the label's RSSI and phase data. The action detection module determines the associated user interaction actions and their timing based on the motion state sequence of the water cup and medicine bottle output by the state analysis module, and finally determines whether the medication action has occurred by combining the timing actions, and records the time when the medication action occurs. The early warning module is used to save the detection results of the action detection module, record the medication time, and notify the user via mobile APP when no medication behavior is detected within a predetermined time range or when multiple medication behaviors are detected.

9. The user medication behavior detection system based on passive radio frequency identification technology as described in claim 8, characterized in that: The antenna of the passive RFID tag is in the form of a dipole; when the medicine bottle or water cup is placed at rest, the radiation direction of the passive RFID tag antenna is omnidirectional in the horizontal direction, and the vertical direction is a radiation dead zone.

Images