UHF-RFID-based access door warehouse-in and warehouse-out identification method and identification system

Abstract

The invention relates to the technical field of radio frequency identification, in particular to a UHF-RFID (Ultra High Frequency-Radio Frequency Identification)-based access door warehouse-in and warehouse-out identification method and identification system. The method comprises the following steps: acquiring activated and scattered signal data of an RFID tag array on goods in a recognition and reading area, acquiring position information and signal change characteristics of each tag in the RFID tag array, and establishing a multi-characteristic data model of the RFID tag array in combination with path information of the goods in the recognition and reading area; and calculating to obtain high-dimensional multi-source data features of the RFID tag array in the multi-antenna space, inputting the high-dimensional multi-source data features into a pre-trained tag classification model to identify a target tag, obtaining space-time dynamic feature vectors of the target tag passing through different antenna reading areas, and inputting the space-time dynamic feature vectors into a pre-trained warehouse-in and warehouse-out classification model to identify warehouse-in and warehouse-out states of goods in the reading areas. According to the invention, the target goods in the access door can be accurately identified, the warehouse-in and warehouse-out state of the target label on the target goods can be reliably judged, and the warehouse-in and warehouse-out process can be monitored and managed in real time.

Description

Technical Field

[0001] The present invention relates to the field of radio frequency identification technology, in particular to a UHF-RFID-based access door entry and exit identification method and identification system. Background Art

[0002] In current warehouse logistics environments, particularly those with high-density, dynamically changing aisle gates, existing ultra-high-frequency (UHF) radio frequency identification (RFID) systems face significant challenges in accurately identifying specific target goods and determining their inbound and outbound status. Their reliability and environmental adaptability remain limited. The key challenge lies in accurately identifying and tracking the target tags attached to the target goods as they pass through the aisle gates, while also effectively filtering out or ignoring the large number of interfering tags present in the environment, as well as those within the aisle gates that are not intended for tracking. These include stationary tags attached to shelves or other static equipment, as well as nomadic tags that accidentally enter the reading area but are not intended for tracking. Existing technologies often struggle to accurately and reliably identify the true target tags within this mixed tag population, resulting in misidentification or missed targets. This uncertainty in identification directly hinders subsequent determination of inbound and outbound status and ineffective monitoring of abnormal behavior.

[0003] In the prior art, the invention patent with publication number CN119538951A discloses an interference tag filtering method for an automatic entry and exit system for goods based on an RFID channel door. However, it only considers a single phase eigenvalue, has poor robustness, and only identifies the target tag without judging the entry and exit status, which cannot meet the needs of logistics warehouses and other scenarios with clear requirements; the invention patent with publication number CN114708678A discloses an RFID channel door system and an entry and exit method based on multiple sensors, but it needs to be equipped with a radar module and an infrared sensor module. Compared with an RFID system without an auxiliary sensor unit, the hardware configuration cost is greatly increased. At the same time, in order to ensure the synchronization and coordination of different sensors, , may also require additional software or hardware support, increasing costs and deployment difficulty; the invention patent with publication number CN110166995A discloses a warehouse access door integrating RFID technology, but due to its high degree of integration, it requires the use of a complete set of independent equipment, including a capacitive touch industrial tablet computer and infrared sensors, light sensors, etc., which will occupy additional space and increase costs and deployment difficulty; the invention patent with publication number CN104636773A discloses a directional monitoring system with radio frequency signal recognition, but it can only uniformly identify and judge the tags passing through the access door, and cannot distinguish between interference tags and target tags. It cannot meet the requirements in scenarios with clear needs such as logistics warehouses.

[0004] Obviously, the current existing technology in UHF RFID channel door applications still has obvious limitations and needs for improvement in terms of how to accurately identify target tags, effectively filter out interference, reliably determine the warehousing status, and take into account system complexity and cost control. Summary of the Invention

[0005] The technical problem to be solved by the embodiments of the present invention is to provide a channel door entry and exit identification method and identification system based on UHF-RFID to solve the problem that the existing technology cannot accurately identify the target tag and judge the entry and exit status of goods in the channel door application, and the cost is high.

[0006] The present invention discloses a UHF-RFID-based access door entry and exit identification method, comprising: Multiple antennas are deployed on the access door to form a reading area covering the access door. A reader / writer transmits radio frequency signals through the antennas and acquires signal data backscattered by the activated RFID tag arrays on the goods in the reading area. Acquire the position information and signal change characteristics of each tag in the RFID tag array, and establish a multi-feature data model of the RFID tag array in combination with the path information of the goods in the reading area; Obtaining high-dimensional multi-source data features of the RFID tag array in a multi-antenna space by calculation according to the multi-feature data model; Based on the RFID tag array on the goods in known motion, a tag classification model is established using machine learning. High-dimensional multi-source data features of the RFID tag array on the goods in the reading area are input into the tag classification model, and the RFID tag array with the target tag identified is output; The spatiotemporal dynamic feature vectors of the target tag passing through different antenna reading areas are obtained, and based on the RFID tag array on the goods with known movement directions, a warehouse entry and exit classification model is established using machine learning. The spatiotemporal dynamic feature vectors of the target tag are input into the warehouse entry and exit classification model, and the identification results representing the entry and exit of goods in the reading area are output.

[0007] Optionally, acquiring backscattered signal data after the RFID tag array on the goods in the reading area is activated includes: Side antennas are symmetrically arranged at preset heights on both sides of the passage door, and at least one group of the side antennas is arranged along the direction of goods entering and leaving the warehouse; A top antenna is arranged on the top of the passage door to cooperate with the side antenna, and together they form a multi-antenna spatial reading area covering the passage door, and the top antenna is horizontally offset from the corresponding side antenna in the direction of goods entering and exiting the warehouse; At least one tag is arranged on the top surface and each side surface of the cargo to form the RFID tag array; The reader / writer sends radio frequency signals through multiple antennas deployed on the top and both sides of the channel door, and obtains the signal data backscattered by each tag on the surface of the goods after being activated by the radio frequency signal within the time window when the goods completely pass through the reading area. The signal data at least includes the RSSI, phase, timestamp and antenna ID of each tag when it is read by each antenna.

[0008] Optionally, establishing the multi-feature data model of the RFID tag array includes: Establish a three-dimensional rectangular coordinate system with the intersection of the central plane of the access door as the origin, define the position coordinates of each antenna on the access door in the three-dimensional rectangular coordinate system, and determine the position relationship of each tag in the RFID tag array relative to each antenna based on the layout position of each tag on the surface of the cargo; Acquire path information of the automated guided vehicle, including speed and multiple movement paths, based on the entry or exit of the automated guided vehicle loaded with goods; The signal characteristics of the RSSI and phase of each tag on the surface of the cargo that change over time at each antenna are obtained based on the radio frequency propagation model, and the multi-feature data model is established in combination with the relative position of each tag in the RFID tag array and the path information of the automated guided vehicle.

[0009] Optionally, the establishing of the multi-feature data model of the RFID tag array further includes: Based on the position and signal characteristics of each tag in the RFID tag array and the path information of the automatic guided vehicle, a multi-feature data model is established to associate the RSSI and phase of each tag in the RFID tag array at each antenna with the distance between the tag and the corresponding antenna. The function expression of the multi-feature data model is:

[0010] Where, is the RSSI value of the tag at antenna i at time t, is the transmission power at time t, and are the transmitting antenna gain and receiving antenna gain at time t, is the signal wavelength, n is the path loss exponent, is the reference distance, is the distance from the tag to antenna i at time t, is the random noise at time t, is the initial phase, is the phase value of the tag at antenna i at time t, is the phase noise or error at time t.

[0011] Optionally, the calculating and obtaining high-dimensional multi-source data features of the RFID tag array in the multi-antenna space according to the multi-feature data model includes: The reader / writer obtains the time series of RSSI and phase of each tag on the cargo surface at each antenna; Inputting the acquired time series into the multi-feature data model to calculate basic time-domain statistical features and spatial features reflecting the RFID tag array, the basic time-domain statistical features include the maximum value, minimum value, mean, standard deviation, skewness, and kurtosis of RSSI and phase, and the spatial features include the RSSI sequence cross-correlation value between antennas, the variance of the phase difference after unwrapping between antennas, and the aggregated RSSI statistical features of different spatial position subsets in the RFID tag array; The basic time-domain statistical features and the spatial features are combined to form high-dimensional multi-source data features that characterize the RFID tag array in the multi-antenna space.

[0012] Optionally, inputting the acquired time series into the multi-feature data model to calculate basic time-domain statistical features and spatial features reflecting the RFID tag array includes: Based on the multi-feature data model, the RSSI time series of the tags synchronously recorded at multiple sampling points by the antennas on both sides of the channel gate is calculated and obtained. The function expression of the RSSI time series is:

[0013] Where, and are the RSSI time series recorded by the antennas on both sides of the channel gate, and n is the number of sampling points; According to the RSSI time series recorded by the antennas on both sides of the channel gate, the Pearson correlation coefficient representing the cross-correlation of the RSSI sequences between the antennas on both sides of the channel gate is calculated. The functional expression of the Pearson correlation coefficient is:

[0014] Where, is the Pearson correlation coefficient; Based on the multi-feature data model, the phase time series of the antennas on both sides of the channel gate after synchronous unwrapping is calculated, and the instantaneous phase difference series between the antennas on both sides of the channel gate is calculated. The function expression of the instantaneous phase difference series is:

[0015] Where, is the instantaneous phase difference sequence at time t, is the phase time series of the tag at one of the side antennas at time t, is the phase time series of the tag at the other side antenna at time t; According to the average value of the local variance of the instantaneous phase difference sequence in multiple time windows, the variance of the unwrapped phase difference between the antennas on both sides of the channel gate is calculated. The function expression of the variance is:

[0016] Where PVD is the variance of the unwrapped phase difference between the antennas on both sides of the channel gate, W is the window length, is the mean phase difference within the window, K is the total number of windows; According to the shape of the surface of the goods, the RFID tag array is divided into subsets corresponding to each surface of the goods; Based on the multi-feature data model, the RSSI values ​​of all tags in each subset at each antenna are calculated and obtained respectively, and the mean RSSI values ​​of all tags in each subset at each antenna are calculated and obtained to form an aggregated RSSI time series. The function expression of the aggregated RSSI time series is:

[0017] Where, The subset at time t The aggregate RSSI time series of all tags at antenna i, For subset The number of internal labels, k is the number of subsets, For subset RSSI value of inner tag j at antenna i; The basic time domain statistical features including maximum value, minimum value, mean value, standard deviation, skewness and kurtosis are extracted from the aggregated RSSI time series corresponding to each subset.

[0018] Optionally, the access door entry and exit identification method further includes using the tag classification model to identify the RFID tag array with the object as the target tag from goods in different motion states within the reading area, including: Standardizing the high-dimensional multi-source data features of the RFID tag array on each item in the reading area, and performing dimensionality reduction processing on the standardized high-dimensional multi-source data features to obtain low-dimensional multi-source data features; According to the low-dimensional multi-source data features of the RFID tag arrays on known stationary goods, nomadic goods, and target goods, a multi-layer perceptron classification model based on Bayesian optimization is input for training to obtain the tag classification model. The function expression of the tag classification model is:

[0019] Where P is the recognition result of the label classification model, is the type of target label, X is the input low-dimensional multi-source data feature, is the optimal hyperparameter for Bayesian optimization, W is the weight, b is the bias, is the activation function, is the classification function; The low-dimensional multi-source data features of the RFID tag array on each item in the reading area are input into the tag classification model, and the RFID tag array whose identification object is the target tag is output.

[0020] Optionally, obtaining the spatiotemporal dynamic feature vector of the target tag sequentially passing through different antenna reading areas includes: Based on the multi-feature data model, the RSSI time series of the target tag at each antenna is calculated and obtained, and each RSSI time series is subjected to a multi-layer discrete wavelet transform to obtain a high-frequency detail coefficient; Use the high-frequency detail coefficient to detect the target tag's signal mutation, and combine the peak and variance of the high-frequency detail coefficient to determine the effective interaction time window when the target tag enters and leaves the reading area of ​​different antennas; The antenna activation order, inter-antenna peak time offset, signal morphology statistical features, and reading frequency features are sequentially extracted from the effective interaction time window, and the extracted features are combined in a preset order to form a spatiotemporal dynamic feature vector representing the target tag passing through different antenna reading areas. The function expression of the spatiotemporal dynamic feature vector is:

[0021] Where F is the spatiotemporal dynamic feature vector, is the number of extracted features of D dimensions, and T represents the transposition operation.

[0022] Optionally, the access door entry and exit identification method further includes using the entry and exit classification model to identify the entry and exit status of goods in the reading area, including: A random forest classification model using the Gini coefficient is constructed, and the spatiotemporal dynamic feature vectors of the RFID tag array on the goods with known movement directions are input into the random forest classification model for training. The functional expression of the Gini coefficient is:

[0023] Where, is the Gini coefficient, is the proportion of samples in the i-th category; According to the prediction results of node splitting based on the Gini coefficient of the random forest classification model, the optimal splitting feature is selected, and the majority voting mechanism is used to establish the inbound and outbound classification model. The function expression of the majority voting mechanism is:

[0024] Where, is the classification result of incoming and outgoing warehouses, is the indicator function, is the prediction function of the t-th decision tree, T is the total number of decision trees, and argmax is the classification result with the most votes; Input the spatiotemporal dynamic feature vectors of the target tag through different antenna reading areas into the in-and-out classification model, and output the in-and-out status of the goods in the reading area; Based on the identification of the in-and-out status of the goods in the reading area, by comparing the duration of the target tag's effective interaction time window with the preset time threshold, it is determined whether the goods in the reading area have stayed abnormally, and by matching the target tag's spatiotemporal dynamic feature vector with the expected normal traffic pattern, it is determined whether the goods in the reading area have deviated from the path.

[0025] The present invention also discloses an identification system, which adopts the above-mentioned UHF-RFID-based access door entry and exit identification method, and the identification system includes: The radio frequency signal transmission module is used to form a reading area covering the passage door through multiple antennas deployed on the passage door, use a reader to send radio frequency signals through the antennas, and obtain signal data backscattered after the RFID tag array on the goods in the reading area is activated; A multi-feature data model building module is used to obtain the position information and signal change characteristics of each tag in the RFID tag array, and to build a multi-feature data model of the RFID tag array in combination with the path information of the goods in the reading area; A high-dimensional multi-source data feature acquisition module, configured to calculate and acquire high-dimensional multi-source data features of the RFID tag array in the multi-antenna space according to the multi-feature data model; A target tag identification module is configured to use machine learning to establish a tag classification model based on the RFID tag array on the goods in a known moving state, input the high-dimensional multi-source data features of the RFID tag array on the goods in the reading area into the tag classification model, and output the RFID tag array identified as the target tag; The entry and exit status recognition module is used to obtain the spatiotemporal dynamic feature vectors of the target tag passing through different antenna reading areas, and use machine learning to establish an entry and exit classification model based on the RFID tag array on the goods with known movement directions. The spatiotemporal dynamic feature vectors of the target tag are input into the entry and exit classification model, and the recognition results representing the entry and exit of goods in the reading area are output.

[0026] Compared with the prior art, the UHF-RFID-based access door entry and exit identification method and identification system provided by the embodiment of the present invention have the following advantages: By deploying multiple antennas on the access door to create a coverage area and combining them with radio frequency signals from a reader / writer, the signal data from the RFID tag arrays on goods within the reading area can be effectively activated and captured. This allows accurate differentiation of the RFID tag arrays on each item, even when multiple items in motion coexist within the reading area. By analyzing the signal variation characteristics of the RFID tag arrays and combining them with the path information of the goods, a multi-feature data model for the RFID tag arrays is established. This model accurately calculates the high-dimensional, multi-source data features of the RFID tag arrays in the multi-antenna space. These features can then be used to train a tag classification model, enabling accurate classification of RFID tag arrays on stationary goods, nomadic goods, and target goods, identifying RFID tag arrays with target tags, and reliably determining the entry and exit status of target tags. This approach has broad application prospects in intelligent warehousing, logistics automation, and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments, in which: Figure 1 A schematic block diagram of the overall steps of the UHF-RFID-based access door entry and exit identification method provided in an embodiment of the present invention; Figure 2 Schematic diagram of antenna distribution for identifying goods entering and leaving a warehouse through a passage door according to an embodiment of the present invention; Figure 3 A schematic diagram of a coordinate system for establishing a three-dimensional rectangular coordinate system with the intersection of the central plane of the channel door as the origin provided by an embodiment of the present invention; Figure 4 Time domain characteristic distribution diagram of the original phase and unwrapped phase during the tag movement process provided by the embodiment of the present invention; Figure 5 Schematic diagram of the low-dimensional multi-source data feature distribution of the labels provided in an embodiment of the present invention; Figure 6 A schematic block diagram of the process of establishing a label classification model provided by an embodiment of the present invention; Figure 7 A schematic block diagram of the process of establishing an inbound and outbound classification model provided by an embodiment of the present invention.

[0028] The symbols in the accompanying drawings represent the following: 1. Side antenna; 2. Top antenna; 3. Cargo; 4. Automatic guided vehicle; 5. Target tag; 6. Stationary tag; 7. Nomadic tag. DETAILED DESCRIPTION

[0029] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. Now, in conjunction with the accompanying drawings, the preferred embodiments of the present invention will be described in detail.

[0030] The present invention discloses a method for identifying access doors based on UHF-RFID. Figure 1 and Figure 2 Shown, including: S1. Multiple antennas are deployed on the access door to form a reading area covering the access door. A reader / writer transmits radio frequency signals through the antennas and obtains signal data backscattered by the activated RFID tag array on the goods 3 in the reading area. S2. Acquire the position information and signal change characteristics of each tag in the RFID tag array, and combine it with the path information of the goods 3 in the reading area to establish a multi-feature data model of the RFID tag array; S3, calculating and obtaining high-dimensional multi-source data features of the RFID tag array in the multi-antenna space according to the multi-feature data model; S4. Based on the RFID tag arrays on the goods in known motion, a tag classification model is established using machine learning. The high-dimensional multi-source data features of the RFID tag arrays on the goods 3 in the reading area are input into the tag classification model, and the RFID tag array is outputted as the target tag 5; S5. Obtain the spatiotemporal dynamic feature vectors of the target tag 5 as it passes through different antenna reading areas, and use machine learning to establish an in-and-out classification model based on the RFID tag array on the goods with known movement directions. Input the spatiotemporal dynamic feature vectors of the target tag 5 into the in-and-out classification model, and output the recognition results representing the in-and-out of the goods 3 in the reading area.

[0031] By implementing the above-described embodiment of the access door entry and exit identification method, the multi-antenna space deployed on the access door achieves comprehensive coverage of the RFID tag array within the reading area. The reader / writer uses these antennas to transmit radio frequency signals, effectively activating the RFID tag array and acquiring signal data scattered by the activated tags. This not only improves the tag activation rate but also enhances the received signal strength, providing a stable data foundation for subsequent target tag 5 identification. By analyzing the position information and signal variation characteristics of each tag in the RFID tag array, combined with the path information of the goods 3 within the reading area, a multi-feature data model is established. This model can fully reflect the physical properties and dynamic behavior of each tag in the RFID tag array. Based on the multi-feature data model, high-dimensional multi-source data features of the RFID tag array in the multi-antenna space are calculated and acquired. These features contain a variety of tag information, such as signal strength, frequency, and time, making the identification process more accurate and comprehensive. Using the RFID tag array on goods with known motion states, a tag classification model is established using machine learning techniques. By inputting the high-dimensional multi-source data features into the model, the target tag 5 can be accurately identified from the RFID tag array, significantly improving the accuracy and efficiency of identification. That is, each tag in the RFID tag array on the target cargo is a target tag 5, each tag in the RFID tag array on the stationary cargo is a stationary tag 6, and each tag in the RFID tag array on the nomadic cargo is a nomadic tag 7. After identifying the RFID tag array as target tag 5, the spatiotemporal dynamic feature vectors of the target tag 5 as it passes through different antenna reading zones are further acquired. These feature vectors contain the tag's motion trajectory and time information within the access door, providing an important basis for determining the entry and exit status. Based on the RFID tag array on cargo with known movement directions, a machine learning approach is used to establish an entry and exit classification model. By inputting the spatiotemporal dynamic feature vectors of the target tag 5 into the model, the entry and exit status of cargo 3 within the reading zone can be accurately identified, enabling real-time monitoring and management of the entry and exit process. Thus, by acquiring the multi-dimensional features of the RFID tag array across multiple antennas, it is possible to effectively distinguish the target cargo from other cargo in motion within the reading zone, and further determine the entry and exit status of the target cargo. This significantly improves the accuracy and reliability of identifying the entry and exit of cargo 3 within the access door, as well as its applicability in complex environments.

[0032] Furthermore, obtaining signal data of the activated scattering of the RFID tag array on the goods 3 in the reading area includes: Side antennas 1 are symmetrically arranged at preset heights on both sides of the passage door, and at least one group of side antennas 1 is arranged along the direction of goods 3 entering and leaving the warehouse; A top antenna 2 is arranged on the top of the passage door in coordination with the side antenna 1, and together they form a multi-antenna spatial reading area covering the passage door. The top antenna 2 is horizontally offset from the corresponding side antenna 1 in the direction of goods 3 entering and exiting the warehouse. At least one tag is arranged on the top surface and each side surface of the cargo 3 to form an RFID tag array; The reader / writer sends radio frequency signals through multiple antennas deployed on the top and both sides of the channel door, and obtains the signal data backscattered by each tag on the surface of the goods 3 after being activated by the radio frequency signal within the time window when the goods 3 completely pass through the reading area. The signal data includes at least the RSSI, phase, timestamp and antenna ID of each tag in the RFID tag array when it is read by each antenna.

[0033] Through the implementation of the above-mentioned channel door entry and exit identification method embodiment, the top antenna 2 has a certain horizontal offset relative to the two side antennas in the direction of the goods 3 entering and exiting the warehouse. Since the gain and polarization mode of each antenna are known, its installation position and orientation are adjusted to form a multi-antenna spatial reading area covering the channel door, thereby improving the recognition probability of the target tag 5 and the accuracy of the entry and exit status judgment. The RFID tag array is arranged on the side and top surface of the goods 3, forming a spatial layout customized according to the size and shape of the goods 3 to maximize the probability of multi-antenna reading of the RFID tag array and the reading range of the goods 3. Preferably, the two side antennas 1 constitute the door frame of the channel door, with a spacing of 150 cm and a height of 120 cm. The top antenna 2 is located above the center of the door frame, 200 cm high, and 50 cm away from the center line of the two side antennas 1 in the horizontal direction, forming a multi-antenna spatial reading area covering the channel door.

[0034] Further, combined with Figure 3 As shown, a multi-feature data model of the RFID tag array is established, including: A three-dimensional rectangular coordinate system is established with the intersection of the central plane of the access door as the origin. The position coordinates of each antenna on the access door in the three-dimensional rectangular coordinate system are defined. Combined with the layout position of each tag on the surface of the cargo 3, the position relationship of each tag in the RFID tag array relative to each antenna is determined; According to the entry or exit of the automated guided vehicle 4 loaded with goods 3, path information of the automated guided vehicle including speed and multiple moving paths is obtained; Based on the radio frequency propagation model, the signal characteristics of the RSSI and phase of each tag on the surface of the cargo 3 that change with time at each antenna are obtained, and a multi-feature data model is established by combining the relative position of each tag in the RFID tag array and the path information of the automated guided vehicle.

[0035] By implementing the above-mentioned embodiment of the channel door entry and exit identification method, a three-dimensional rectangular coordinate system is established with the intersection of the channel door center plane as the origin, wherein the X-axis is parallel to the entry and exit direction of the goods 3, the Y-axis is parallel to the width direction of the channel door, and the Z-axis is vertically upward. The precise position coordinates of each antenna in this coordinate system are defined, and combined with the layout of the RFID tag array, the positional relationship of each tag in the RFID tag array relative to each antenna is determined. The path of the automated guided vehicle includes but is not limited to entering and exiting the warehouse along the channel door; the direction of movement of the goods 3 is perpendicular to the direction of the main entry and exit channel (90° intersection) and maintains horizontal movement, and does not enter the channel door area; the direction of movement of the goods 3 is parallel to the direction of the main entry and exit channel, but the path is offset and does not pass through any channel door opening.

[0036] Furthermore, establishing a multi-feature data model of the RFID tag array also includes: Based on the signal characteristics and position of each tag in the RFID tag array and the path information of the automated guided vehicle, a multi-feature data model is established that associates the RSSI and phase of each tag in the RFID tag array at each antenna with the distance between the tag and the corresponding antenna. The function expression of the multi-feature data model is:

[0037] Where, is the RSSI value of the tag at antenna i at time t, is the transmission power at time t, and are the transmitting antenna gain and receiving antenna gain at time t, is the signal wavelength, n is the path loss exponent, is the reference distance, is the distance from the tag to antenna i at time t, is the random noise at time t, is the initial phase, is the phase value of the tag at antenna i at time t, is the phase noise or error at time t.

[0038] Through the implementation of the above-mentioned channel door entry and exit identification method embodiment, as shown in FIG Figure 3 As shown, the three antennas are antenna ,antenna and antennas , each tag returns , Phase , timestamp t and other information to the reader, For antenna The height from the horizontal plane where the cargo is located, and v is the speed of the AGV. Based on the above parameters, a multi-feature data model is established that associates the RSSI and phase of each tag in the RFID tag array at each antenna with the distance between the tag and the corresponding antenna.

[0039] Furthermore, high-dimensional multi-source data features of the RFID tag array in the multi-antenna space are calculated and obtained based on the multi-feature data model, including: The reader / writer obtains the time series of RSSI and phase of each tag on the surface of cargo 3 at each antenna; The acquired time series is input into a multi-feature data model to calculate basic time-domain statistical features and spatial features reflecting the RFID tag array. The basic time-domain statistical features include the maximum, minimum, mean, standard deviation, skewness, and kurtosis of RSSI and phase. The spatial features include the cross-correlation value of RSSI sequences between antennas, the variance of the phase difference after unwrapping between antennas, and the aggregated RSSI statistical features of different spatial position subsets in the RFID tag array. The basic time-domain statistical features and spatial features are combined to form high-dimensional multi-source data features that characterize the RFID tag array in the multi-antenna space.

[0040] Furthermore, the acquired time series is input into a multi-feature data model to calculate the basic time-domain statistical features and the spatial features reflecting the RFID tag array, including: Based on the multi-feature data model, the RSSI time series of the tags recorded synchronously at multiple sampling points by the antennas on both sides of the channel gate is calculated. The function expression of the RSSI time series is:

[0041] Where, and are the RSSI time series recorded by the antennas on both sides of the channel gate, and n is the number of sampling points; According to the RSSI time series recorded by the antennas on both sides of the channel gate, the Pearson correlation coefficient representing the cross-correlation of the RSSI sequences between the antennas on both sides of the channel gate is calculated. The functional expression of the Pearson correlation coefficient is:

[0042] Where, is the Pearson correlation coefficient; Based on the multi-feature data model, the phase time series of the antennas on both sides of the channel gate after synchronous unwrapping is calculated, and the instantaneous phase difference series between the antennas on both sides of the channel gate is calculated. The function expression of the instantaneous phase difference series is:

[0043] Where, is the instantaneous phase difference sequence at time t, is the phase time series of the tag at one of the side antennas 1 at time t, is the phase time series of the tag at the other side antenna 1 at time t; According to the average value of the local variance of the instantaneous phase difference sequence in multiple time windows, the variance of the unwrapped phase difference between the antennas on both sides of the channel gate is calculated. The function expression of the variance is:

[0044] Where PVD is the variance of the unwrapped phase difference between the antennas on both sides of the channel gate, W is the window length, is the mean phase difference within the window, K is the total number of windows; According to the shape of the surface of the goods 3, the RFID tag array is divided into subsets corresponding to each surface of the goods 3; Based on the multi-feature data model, the RSSI values ​​of all tags in each subset at each antenna are calculated and the mean RSSI values ​​of all tags in each subset at each antenna are calculated to form an aggregated RSSI time series. The function expression of the aggregated RSSI time series is:

[0045] Where, The subset at time t The aggregate RSSI time series of all tags at antenna i, For subset The number of internal labels, k is the number of subsets, For subset RSSI value of inner tag j at antenna i; The basic time domain statistical features including maximum value, minimum value, mean value, standard deviation, skewness and kurtosis are extracted from the aggregated RSSI time series corresponding to each subset.

[0046] Through the implementation of the above-mentioned channel gate entry and exit identification method embodiment, the Pearson correlation coefficient is used to quantify the linear similarity between the signal patterns of the antennas on both sides; the variance of the phase difference after unpacking between the antennas on both sides of the channel gate is used to characterize the degree of path deviation when the tag passes through the channel gate, thereby characterizing whether the tag passes through the channel gate accurately; the aggregated RSSI time series is used to characterize the differentiated RF responses of different tags in the RFID tag array when passing through the channel gate. Figure 4 As shown, Figure 4 Figure (a) shows the original phase time domain characteristic distribution of the RFID tag array on goods in different motion states. Figure 4 The middle (b) figure is for phase phase The periodic constraint of phase unwrapping is used to eliminate phase ambiguity and obtain the unwrapped phase sequence , and ensure the continuity of phase information based on the time domain characteristics of the original phase and unwrapped phase during the tag movement process.

[0047] Furthermore, the access door entry and exit identification method further includes using a tag classification model to identify an RFID tag array of a target tag 5 from goods 3 in different motion states within a reading area, including: The high-dimensional multi-source data features of the RFID tag array on each item 3 in the reading area are standardized respectively, and the standardized high-dimensional multi-source data features are subjected to dimensionality reduction processing to obtain low-dimensional multi-source data features; According to the low-dimensional multi-source data features of the RFID tag arrays on known stationary goods, nomadic goods, and target goods, a multi-layer perceptron classification model based on Bayesian optimization is input for training to obtain a label classification model. The function expression of the label classification model is:

[0048] Where P is the recognition result of the label classification model, is the type of target label 5, X is the input low-dimensional multi-source data feature, is the optimal hyperparameter for Bayesian optimization, W is the weight, b is the bias, is the activation function, is the classification function; The low-dimensional multi-source data features of the RFID tag arrays on each cargo in the reading area are input into the tag classification model, and the RFID tag array whose identification object is the target tag 5 is output.

[0049] Through the implementation of the above-mentioned channel door entry and exit identification method embodiment, as shown in FIG Figure 6 As shown, Figure 6 The two-dimensional distribution of low-dimensional features of labels based on PCA (Principal Component Analysis) dimensionality reduction. Based on the low-dimensional multi-source data features of the labels to be identified, the multi-layer perceptron classification model based on Bayesian optimization is used to identify the target label 5. Figure 7 As shown, the model determines the optimal hyperparameters through pre-training and Bayesian optimization: During initialization, set the MLP (Multilayer Perceptron) classification model parameters and the Bayesian optimizer parameters; define the objective function, such as cross-validation accuracy; use the Bayesian optimizer to generate the optimal hyperparameters that optimize the objective function value. . The optimal hyperparameters obtained are The combined retraining MLP classification model evaluates the model performance and propagates forward until the MLP classification model reaches the number of iterations or meets the stopping condition, and determines the predefined type of each tag in the RFID tag array (stationary tag 6, nomadic tag 7 or target tag 5), thereby identifying the target tag 5. That is, the recognition result is given the input low-dimensional multi-source data features, under the optimal hyperparameters Under the model configured and trained with weights W and bias b, the probability of predicting category k is obtained by applying weights layer by layer , bias and activation function Perform forward calculation and finally obtain it through the softmax function.

[0050] Furthermore, obtaining the spatiotemporal dynamic feature vectors of the target tag 5 passing through different antenna reading areas in sequence includes: The RSSI time series of the target tag 5 at each antenna is calculated based on the multi-feature data model, and each RSSI time series is subjected to a multi-layer discrete wavelet transform to obtain the high-frequency detail coefficient; Use the high-frequency detail coefficient to detect the signal mutation of the target tag 5, and combine the peak value and variance of the high-frequency detail coefficient to determine the effective interaction time window of the target tag 5 entering and leaving the reading area of ​​different antennas; The antenna activation order, inter-antenna peak time offset, signal morphology statistical features, and reading frequency features are sequentially extracted from the effective interaction time window. The extracted features are combined in a preset order to form a spatiotemporal dynamic feature vector representing the target tag 5 passing through different antenna reading areas. The function expression of the spatiotemporal dynamic feature vector is:

[0051] Where F is the spatiotemporal dynamic feature vector, is the number of extracted features of D dimensions, and T represents the transposition operation.

[0052] By implementing the above-mentioned embodiment of the access door entry and exit identification method, the RSSI time series of the target tag 5 is analyzed to determine the start and end time points of entering and leaving the effective interaction area of ​​the access door, and the predefined features of the time series pattern and relative timing relationship of the RFID tag array passing through different antenna reading areas in sequence are extracted within this time window. That is, the RSSI time series of the target tag 5 is first subjected to a multi-layer discrete wavelet transform, and the multi-layer discrete wavelet transform is analyzed by selecting "db4" as the wavelet basis function to decompose the high-frequency detail coefficients. (where j = 1, 2, 3, which is the number of decomposition layers, i is the antenna index, and k is the offline time point). To detect signal mutations and use their peak and variance To determine the time window in which target tag 5 effectively interacts with the channel gate Within this window, key features of the spatiotemporal dynamic characteristics of target tag 5 passing through the channel gate are extracted, including antenna activation order, inter-antenna peak time offset, signal morphology statistical features (such as the maximum, minimum, mean, and standard deviation of the RSSI value), and reading frequency features. These features are used to characterize the spatiotemporal characteristics of the signal characteristics of target tag 5 passing through the channel gate. The extracted features are then combined into a D-dimensional spatiotemporal dynamic feature vector, providing a data basis for subsequent entry and exit identification.

[0053] Furthermore, the access door entry and exit identification method further includes using an entry and exit classification model to identify the entry and exit status of the goods 3 in the reading area, including: A random forest classification model using the Gini coefficient is constructed. The spatiotemporal dynamic feature vectors of the RFID tag array on cargo 3 with known movement direction are input into the random forest classification model for training. The functional expression of the Gini coefficient is:

[0054] Where, is the Gini coefficient, is the proportion of samples in the i-th category; According to the prediction results of node splitting based on the Gini coefficient of the random forest classification model, the optimal splitting feature is selected, and the majority voting mechanism is used to establish the inbound and outbound classification model. The function expression of the majority voting mechanism is:

[0055] Where, is the classification result of incoming and outgoing warehouses, is the indicator function, is the prediction function of the t-th decision tree, T is the total number of decision trees, and argmax is the classification result with the most votes; The spatiotemporal dynamic feature vectors of the target tag 5 in the reading areas of different antennas are input into the warehouse entry and exit classification model, and the recognition result representing the entry and exit of the goods 3 in the reading area is output; Based on the identification of the in-and-out status of the goods 3 in the reading area, by comparing the duration of the effective interaction time window of the target tag 5 with the preset time threshold, it is determined whether the goods 3 in the reading area have stayed abnormally, and by matching the spatiotemporal dynamic feature vector of the target tag 5 with the expected normal traffic pattern, it is determined whether the goods 3 in the reading area have deviated from the path.

[0056] By implementing the above-mentioned embodiment of the access door entry and exit identification method, the spatiotemporal dynamic feature vectors of the target tag 5 passing through different antenna reading areas are input into a pre-trained random forest classification model. Figure 7 As shown in Figure 2, the random forest model first randomly extracts a certain number of samples from the entire input spatiotemporal dynamic feature vector dataset. These samples will be used for feature selection and decision tree construction. From the extracted samples, a portion of features are randomly selected as the candidate feature set, which will be used to build the decision tree. Using the Gini coefficient The Gini coefficient is used to assess the importance of each feature in the candidate feature set. It measures the purity of a dataset; lower values ​​indicate a purer dataset and easier classification. Within the candidate feature set, the Gini coefficient is used to select features that minimize the dataset's purity as test features. These features are then used to generate nodes in the decision tree. Using the selected test features, nodes are created in the dataset. Each node represents a decision rule used to partition the data into different subsets. For each node, the model determines whether it can become a node and whether to stop growing the decision tree. These criteria include the amount of data remaining in the node, a purity threshold, or a maximum depth limit. If the current node does not meet the criteria for continued growth, the model returns and selects another set of features as candidate features, repeating the process. When the required number of decision trees is met, the generation of new decision trees ceases, and all generated decision trees are stored. Ultimately, the resulting random forest model outputs an importance score for each feature, reflecting its contribution to the model's predictive power. This model integrates multiple decision trees. Each tree is trained using randomly sampled data and feature subsets. The prediction results of node splitting are based on criteria such as the Gini coefficient, and a majority voting mechanism is used to ultimately decide the direction of label movement (in or out of storage).

[0057] Finally, based on the judged entry and exit status, direction and path compliance checks are performed to identify abnormal entry and exit situations such as abnormal stops and significant deviations from the path. By analyzing the duration of the effective interaction time window, if the duration significantly exceeds the preset normal passage time threshold, it is determined to be an abnormal stop. At the same time, after the direction judgment, the degree of match between the extracted D-dimensional spatiotemporal feature vector and the expected normal passage pattern is evaluated, such as checking whether the antenna activation sequence meets expectations and whether the peak time offset is within a reasonable range. Any significant deviation from the normal pattern, such as confusion in the activation sequence, abnormal fluctuations in the signal peak, or abnormal time intervals, may indicate that the path has deviated significantly. Once an abnormal stop or significant path deviation is detected, the system will immediately trigger an early warning.

[0058] The present invention deploys multiple antennas on the top and both sides of the channel door to collect multiple eigenvalue signals such as RSSI and phase of the RFID tag array, and combines the layout of the RFID tag array, the motion information of the automatic guided vehicle, and the spatial relationship of the channel door equipment to construct a multi-feature data model. The model is used to calculate the multi-source data features of the RFID tag array in the multi-antenna space, and through multi-feature data fusion analysis and machine learning algorithms, accurate identification of stationary tags 6, nomadic tags 7 and target tags 5, as well as reliable judgment of the entry and exit status of the target tag 5, including compliance checks of direction and path, are achieved, thereby effectively identifying abnormal entry and exit situations such as abnormal stays and path deviations, and triggering early warnings. The present invention significantly improves the accuracy of target tag 5 identification and entry and exit status judgment through the combination of multi-antenna spatial layout, multi-eigenvalue fusion analysis and machine learning algorithms, and has broad application prospects in the fields of intelligent warehousing, logistics automation, etc.

[0059] The present invention also discloses an identification system, which adopts the above-mentioned UHF-RFID-based access door entry and exit identification method, and the identification system includes: The RF signal transmission module is used to form a reading area covering the passage door through multiple antennas deployed on the passage door, use the reader to send RF signals through the antenna, and obtain the signal data backscattered after the RFID tag array on the goods 3 in the reading area is activated; A multi-feature data model building module is used to obtain the position information and signal change characteristics of each tag in the RFID tag array, and to build a multi-feature data model of the RFID tag array in combination with the path information of the goods 3 in the reading area; A high-dimensional multi-source data feature acquisition module is used to calculate and acquire high-dimensional multi-source data features of RFID tag arrays in a multi-antenna space based on a multi-feature data model; The target tag 5 identification module is used to establish a tag classification model using machine learning based on the RFID tag array on the goods 3 in known motion. The high-dimensional multi-source data features of the RFID tag array on the goods 3 in the reading area are input into the tag classification model, and the RFID tag array of the target tag 5 is output; The entry-and-exit status recognition module is used to obtain the spatiotemporal dynamic feature vectors of the target tag 5 passing through different antenna reading areas, and use machine learning to establish an entry-and-exit classification model based on the RFID tag array on the goods with known movement directions. The spatiotemporal dynamic feature vectors of the target tag 5 are input into the entry-and-exit classification model, and the recognition results representing the entry and exit of the goods 3 in the reading area are output.

[0060] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned UHF-RFID-based channel door entry and exit identification method are implemented.

[0061] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, the steps of the above-mentioned UHF-RFID-based channel door entry and exit identification method are implemented.

[0062] The present invention is described based on flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to specific embodiments. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts 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, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0063] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0064] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0065] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Those skilled in the art may modify the technical solutions described in the above embodiments, or replace some of the technical features therein with equivalents; and all these modifications and replacements should fall within the scope of protection of the present invention.

Claims

1. A UHF-RFID-based access door entry and exit identification method, characterized in that: The UHF-RFID-based access door entry and exit identification method includes: Multiple antennas are deployed on the access door to form a reading area covering the access door. A reader / writer transmits radio frequency signals through the antennas and acquires signal data backscattered by the activated RFID tag arrays on the goods in the reading area. Acquire the position information and signal change characteristics of each tag in the RFID tag array, and establish a multi-feature data model of the RFID tag array in combination with the path information of the goods in the reading area; Obtaining high-dimensional multi-source data features of the RFID tag array in a multi-antenna space by calculation according to the multi-feature data model; Based on the RFID tag array on the goods in known motion, a tag classification model is established using machine learning. High-dimensional multi-source data features of the RFID tag array on the goods in the reading area are input into the tag classification model, and the RFID tag array with the target tag identified is output; The spatiotemporal dynamic feature vectors of the target tag passing through different antenna reading areas are obtained, and based on the RFID tag array on the goods with known movement directions, a warehouse entry and exit classification model is established using machine learning. The spatiotemporal dynamic feature vectors of the target tag are input into the warehouse entry and exit classification model, and the identification results representing the entry and exit of goods in the reading area are output.

2. The UHF-RFID-based access door entry and exit identification method according to claim 1 is characterized in that: The step of obtaining backscattered signal data after the RFID tag array on the goods in the reading area is activated includes: Side antennas are symmetrically arranged at preset heights on both sides of the passage door, and at least one group of the side antennas is arranged along the direction of goods entering and leaving the warehouse; A top antenna is arranged on the top of the passage door to cooperate with the side antenna, and together they form a multi-antenna spatial reading area covering the passage door, and the top antenna is horizontally offset from the corresponding side antenna in the direction of goods entering and exiting the warehouse; At least one tag is arranged on the top surface and each side surface of the cargo to form the RFID tag array; The reader / writer transmits radio frequency signals through multiple antennas deployed on the top and both sides of the passage door, and obtains the signal data backscattered by each tag on the surface of the goods after being activated by the radio frequency signal within the time window when the goods completely pass through the reading area. The signal data includes at least the RSSI, phase, timestamp and antenna ID of each tag in the RFID tag array when it is read by each antenna.

3. The UHF-RFID-based access door entry and exit identification method according to claim 2 is characterized in that: The establishing of the multi-feature data model of the RFID tag array includes: Establish a three-dimensional rectangular coordinate system with the intersection of the central plane of the access door as the origin, define the position coordinates of each antenna on the access door in the three-dimensional rectangular coordinate system, and determine the position relationship of each tag in the RFID tag array relative to each antenna based on the layout position of each tag on the surface of the cargo; Acquire path information of the automated guided vehicle, including speed and multiple movement paths, based on the entry or exit of the automated guided vehicle loaded with goods; The signal characteristics of the RSSI and phase of each tag on the surface of the cargo that change over time at each antenna are obtained based on the radio frequency propagation model, and the multi-feature data model is established in combination with the relative position of each tag in the RFID tag array and the path information of the automated guided vehicle.

4. The UHF-RFID-based access door entry and exit identification method according to claim 3 is characterized in that: The establishing of the multi-feature data model of the RFID tag array further includes: Based on the position and signal characteristics of each tag in the RFID tag array and the path information of the automatic guided vehicle, a multi-feature data model is established to associate the RSSI and phase of each tag in the RFID tag array at each antenna with the distance between the tag and the corresponding antenna. The function expression of the multi-feature data model is: Where, is the RSSI value of the tag at antenna i at time t, is the transmission power at time t, and are the transmitting antenna gain and receiving antenna gain at time t, is the signal wavelength, n is the path loss exponent, is the reference distance, is the distance from the tag to antenna i at time t, is the random noise at time t, is the initial phase, is the phase value of the tag at antenna i at time t, is the phase noise or error at time t.

5. The UHF-RFID-based access door entry and exit identification method according to claim 4 is characterized in that: The step of calculating and obtaining high-dimensional multi-source data features of the RFID tag array in the multi-antenna space according to the multi-feature data model includes: The reader / writer obtains the time series of RSSI and phase of each tag on the cargo surface at each antenna; Inputting the acquired time series into the multi-feature data model to calculate basic time-domain statistical features and spatial features reflecting the RFID tag array, the basic time-domain statistical features include the maximum value, minimum value, mean, standard deviation, skewness, and kurtosis of RSSI and phase, and the spatial features include the RSSI sequence cross-correlation value between antennas, the variance of the phase difference after unwrapping between antennas, and the aggregated RSSI statistical features of different spatial position subsets in the RFID tag array; The basic time-domain statistical features and the spatial features are combined to form high-dimensional multi-source data features that characterize the RFID tag array in the multi-antenna space.

6. The UHF-RFID-based access door entry and exit identification method according to claim 5 is characterized in that: The step of inputting the acquired time series into the multi-feature data model to calculate basic time-domain statistical features and spatial features reflecting the RFID tag array includes: Based on the multi-feature data model, the RSSI time series of the tags synchronously recorded at multiple sampling points by the antennas on both sides of the channel gate is calculated and obtained. The function expression of the RSSI time series is: Where, and are the RSSI time series recorded by the antennas on both sides of the channel gate, and n is the number of sampling points; According to the RSSI time series recorded by the antennas on both sides of the channel gate, the Pearson correlation coefficient representing the cross-correlation of the RSSI sequences between the antennas on both sides of the channel gate is calculated. The functional expression of the Pearson correlation coefficient is: Where, is the Pearson correlation coefficient; Based on the multi-feature data model, the phase time series of the antennas on both sides of the channel gate after synchronous unwrapping is calculated, and the instantaneous phase difference series between the antennas on both sides of the channel gate is calculated. The function expression of the instantaneous phase difference series is: Where, is the instantaneous phase difference sequence at time t, is the phase time series of the tag at one of the side antennas at time t, is the phase time series of the tag at the other side antenna at time t; According to the average value of the local variance of the instantaneous phase difference sequence in multiple time windows, the variance of the unwrapped phase difference between the antennas on both sides of the channel gate is calculated. The function expression of the variance is: Where PVD is the variance of the unwrapped phase difference between the antennas on both sides of the channel gate, W is the window length, is the mean phase difference within the window, K is the total number of windows; According to the shape of the surface of the goods, the RFID tag array is divided into subsets corresponding to each surface of the goods; Based on the multi-feature data model, the RSSI values ​​of all tags in each subset at each antenna are calculated and obtained respectively, and the mean RSSI values ​​of all tags in each subset at each antenna are calculated and obtained to form an aggregated RSSI time series. The function expression of the aggregated RSSI time series is: Where, The subset at time t The aggregate RSSI time series of all tags at antenna i, For subset The number of internal labels, k is the number of subsets, For subset RSSI value of inner tag j at antenna i; The basic time domain statistical features including maximum value, minimum value, mean value, standard deviation, skewness and kurtosis are extracted from the aggregated RSSI time series corresponding to each subset.

7. The UHF-RFID-based access door entry and exit identification method according to claim 6 is characterized in that: The access door entry and exit identification method further includes using the tag classification model to identify the RFID tag array having an object as a target tag from goods in different motion states within a reading area, including: Standardizing the high-dimensional multi-source data features of the RFID tag array on each item in the reading area, and performing dimensionality reduction processing on the standardized high-dimensional multi-source data features to obtain low-dimensional multi-source data features; According to the low-dimensional multi-source data features of the RFID tag arrays on known stationary goods, nomadic goods, and target goods, a multi-layer perceptron classification model based on Bayesian optimization is input for training to obtain the tag classification model. The function expression of the tag classification model is: Where P is the recognition result of the label classification model, To characterize the identification object of the target label, X is the input low-dimensional multi-source data feature, is the optimal hyperparameter for Bayesian optimization, W is the weight, b is the bias, is the activation function, is the classification function; The low-dimensional multi-source data features of the RFID tag array on each item in the reading area are input into the tag classification model, and the RFID tag array whose identification object is the target tag is output.

8. The UHF-RFID-based access door entry and exit identification method according to claim 7 is characterized in that: The obtaining of the spatiotemporal dynamic feature vector of the target tag passing through different antenna reading areas includes: Based on the multi-feature data model, the RSSI time series of the target tag at each antenna is calculated and obtained, and each RSSI time series is subjected to a multi-layer discrete wavelet transform to obtain a high-frequency detail coefficient; Use the high-frequency detail coefficient to detect the target tag's signal mutation, and combine the peak and variance of the high-frequency detail coefficient to determine the effective interaction time window when the target tag enters and leaves the reading area of ​​different antennas; The antenna activation order, inter-antenna peak time offset, signal morphology statistical features, and reading frequency features are sequentially extracted from the effective interaction time window, and the extracted features are combined in a preset order to form a spatiotemporal dynamic feature vector representing the target tag passing through different antenna reading areas. The function expression of the spatiotemporal dynamic feature vector is: Where F is the spatiotemporal dynamic feature vector, is the number of extracted features of D dimensions, and T represents the transposition operation.

9. The UHF-RFID-based access door entry and exit identification method according to claim 8, characterized in that: The access door entry and exit identification method further includes using the entry and exit classification model to identify the entry and exit status of goods in the reading area, including: A random forest classification model using the Gini coefficient is constructed, and the spatiotemporal dynamic feature vectors of the RFID tag array on the goods with known movement directions are input into the random forest classification model for training. The functional expression of the Gini coefficient is: Where, is the Gini coefficient, is the proportion of samples in the i-th category; According to the prediction results of node splitting based on the Gini coefficient of the random forest classification model, the optimal splitting feature is selected, and the majority voting mechanism is used to establish the inbound and outbound classification model. The function expression of the majority voting mechanism is: Where, is the classification result of incoming and outgoing warehouses, is the indicator function, is the prediction function of the t-th decision tree, T is the total number of decision trees, and argmax is the classification result with the most votes; Input the spatiotemporal dynamic feature vectors of the target tag through different antenna reading areas into the warehouse entry and exit classification model, and output the recognition results representing the entry and exit of goods in the reading area; Based on the identification of the in-and-out status of the goods in the reading area, by comparing the duration of the target tag's effective interaction time window with the preset time threshold, it is determined whether the goods in the reading area have stayed abnormally, and by matching the target tag's spatiotemporal dynamic feature vector with the expected normal traffic pattern, it is determined whether the goods in the reading area have deviated from the path.

10. An identification system, using the UHF-RFID-based access door entry and exit identification method according to any one of claims 1 to 9, characterized in that: The identification system comprises: The radio frequency signal transmission module is used to form a reading area covering the passage door through multiple antennas deployed on the passage door, use a reader to send radio frequency signals through the antennas, and obtain signal data backscattered after the RFID tag array on the goods in the reading area is activated; A multi-feature data model building module is used to obtain the position information and signal change characteristics of each tag in the RFID tag array, and to build a multi-feature data model of the RFID tag array in combination with the path information of the goods in the reading area; A high-dimensional multi-source data feature acquisition module, configured to calculate and acquire high-dimensional multi-source data features of the RFID tag array in the multi-antenna space according to the multi-feature data model; A target tag identification module is configured to use machine learning to establish a tag classification model based on the RFID tag array on the goods in a known moving state, input the high-dimensional multi-source data features of the RFID tag array on the goods in the reading area into the tag classification model, and output the RFID tag array identified as the target tag; The entry and exit status recognition module is used to obtain the spatiotemporal dynamic feature vectors of the target tag passing through different antenna reading areas, and use machine learning to establish an entry and exit classification model based on the RFID tag array on the goods with known movement directions. The spatiotemporal dynamic feature vectors of the target tag are input into the entry and exit classification model, and the recognition results representing the entry and exit of goods in the reading area are output.

Images