Intelligent library book retrieval method based on Internet of Things

Abstract

The invention relates to the technical field of book position retrieval, in particular to an intelligent library book retrieval method based on the Internet of Things, which comprises the following steps of: deploying an RFID (Radio Frequency Identification Device) dense read-write array on a bookshelf layer, arranging a UWB (Ultra Wideband) positioning base station in a library area channel, and embedding a composite label integrated with various sensors into a book cover; a dynamic position discrimination algorithm is introduced to identify the moving state of the book, an extended Kalman filter is utilized to fuse data to establish a motion trail prediction model, a hierarchical position display mechanism is adopted to provide detailed position information, and for temporarily placed books and overtime non-return books, the accuracy of positioning and tracking of the books is improved. And special processing and alarm prompting are respectively carried out, so that the problems of dynamic position updating lagging and inaccurate retrieval result in traditional library book retrieval are effectively solved, and the book retrieval efficiency of readers is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of book location retrieval, and in particular to an intelligent library book retrieval method based on the Internet of Things. Background Art

[0002] In intelligent library systems, efficient and accurate book retrieval is crucial to improving user experience and library management efficiency. However, due to the large number of books in the library, frequent borrowing and returning of books by readers, and the limited means of existing positioning and monitoring technologies, there is currently a widespread problem of inaccurate retrieval results caused by delayed updates of the dynamic location of books.

[0003] Traditional library retrieval systems mostly rely on static book location information records. When books are borrowed, returned, or temporarily moved by readers in the library, the system cannot obtain these dynamic changes in time. Readers may mistakenly put books in other locations when looking for them between shelves, or fail to follow the prescribed procedures during the borrowing process, resulting in the actual location of the books not matching the system records. Existing positioning technology uses RFID tags for identification, which results in it only being able to obtain book information at specific collection points and unable to track the movement of books in the library in real time. As a result, the book location information in the system often lags behind the actual situation. This not only makes readers spend a lot of time blindly searching for books between shelves, reducing reading and learning efficiency, but also brings great difficulties to the organization and management work of library staff, affecting the service quality and operational efficiency of the library. In order to solve this technical problem, we provide an intelligent library book retrieval method and system based on the Internet of Things. Summary of the Invention

[0004] The purpose of the present invention is to provide a smart library book retrieval method and system based on the Internet of Things to solve the problems raised in the above background technology.

[0005] 1. Due to the large number of books in the library and the frequent reader activity, the lag in updating the dynamic location of books leads to inaccurate search results. Therefore, this case deploys a dense RFID read / write array, UWB positioning base stations, and composite sensor tags. Using an extended Kalman filter to fuse data and track book locations, this method can obtain the dynamic location of books in real time and improve search accuracy.

[0006] 2. Because existing positioning technology cannot track the movement trajectory of books in real time, resulting in a lag in system location information, this case constructs a book movement trajectory prediction model and combines it with a heuristic search algorithm to predict the movement path of books. This can predict the trajectory of books, facilitate timely monitoring of book dynamics, and assist library management and reader retrieval.

[0007] To achieve the above objectives, one of the objectives of the present invention is to provide a smart library book retrieval method based on the Internet of Things, comprising the following steps: S1. Deploy dense RFID reader arrays on the bookshelf level, set up UWB positioning base stations in the library channels, and embed composite sensor tags on book covers. The tags integrate RFID chips, three-axis accelerometers, and pressure sensors. S2. A dynamic position determination algorithm is introduced. When the three-axis accelerometer detects that the book's continuous displacement is greater than 15 cm / s² and the pressure sensor reading disappears, the book is determined to be in a moving state, triggering positioning tracking by the UWB positioning base station. S3. Using an extended Kalman filter to fuse RFID phase ranging data obtained from a dense RFID reader array with UWB arrival time difference data, a book trajectory prediction model is established. When a book is stationary for 30 seconds, it is automatically registered as a temporary location point. S4. Use a hierarchical location display mechanism to simultaneously display the original shelf code and real-time location status of books, provide a heat map of the last disappearing area for moving books, and mark virtual coordinates with a countdown for temporarily placed books; S5. For books in temporary locations that have not been returned to the shelves for more than 48 hours, the positioning beacon autonomous alarm mode is activated, and a gradient-enhanced prompt tone is emitted through the built-in buzzer of the tag.

[0008] As a further improvement of this technical solution, the extended Kalman filter fusion process in S3 includes: A three-layer data quality assessment mechanism is established. The first layer evaluates the signal strength fluctuation coefficient of RFID phase ranging, the second layer evaluates the multipath effect of UWB arrival time difference, and the third layer evaluates the time stamp synchronization error between the two. The fusion weight is dynamically adjusted according to the evaluation results, and a self-calibration process is automatically performed every 100 fusion calculations. The system error is corrected by comparing the fusion results of known position points with the actual coordinates.

[0009] As a further improvement of this technical solution, the establishment of the book motion trajectory prediction model in S3 includes: A four-element state space is constructed, including the book position coordinates, movement speed, acceleration and direction angle. The physical space of the library is divided into three-dimensional grid units of 1m×1m×2m. A travel cost coefficient is assigned to each grid. The bookshelf area is set as the high-cost area, and the passage area is set as the low-cost area. The cost coefficient greater than or equal to 80 is considered a high-cost area, and the cost coefficient less than or equal to 20 is considered a low-cost area. A heuristic search algorithm is used to combine the current position, speed and travel cost coefficient to predict the movement path within the next 3 seconds, generating a probability distribution containing 5 candidate trajectories.

[0010] As a further improvement of this technical solution, the temporary location point registration in S3 includes: When a book is detected to be stationary, a three-level confirmation mechanism is activated. The first level confirms that the accelerometer's three-axis data are all less than 0.05g for 20 seconds. The second level confirms that the RFID signal strength fluctuation is less than 3dBm for 10 seconds. The third level verifies that the position change is less than 5cm for 5 seconds through UWB positioning. When registering a temporary location point, the environmental parameters are recorded synchronously, including the light intensity, temperature and humidity, and noise level of the area, and a multi-dimensional code containing spatiotemporal characteristics is generated for the temporary location point. The first 12 bits are the location coordinate hash value, the middle 8 bits are the timestamp code, and the last 4 bits are the environmental parameter classification code.

[0011] As a further improvement of this technical solution, S3 further includes an abnormal trajectory identification step: Five types of abnormal trajectory patterns are defined, and a sliding time window with a length of 5 minutes is established. The cosine similarity between the current trajectory and the normal pattern is calculated in real time. When the similarity is lower than the preset threshold, it is marked as abnormal. The secondary positioning enhancement is automatically triggered for books with abnormal trajectories, the UWB sampling frequency is increased from 10Hz to 50Hz, and three adjacent RFID readers are activated at the same time.

[0012] As a further improvement of this technical solution, S3 further includes a multi-label interference elimination step: When it is detected that there are three or more mobile tags within the signal range of the same RFID reader, the interference elimination process is started; A time division multiple access time slot allocation algorithm is used to dynamically allocate an exclusive communication time slot to each tag. The time slot width is dynamically adjusted according to the tag's moving speed. The phase fingerprint characteristics, Doppler frequency shift characteristics and signal strength change characteristics of each tag's RFID signal are extracted, and multi-tag data decoupling is achieved through feature clustering.

[0013] As a further improvement of this technical solution, S3 further includes a three-dimensional space mapping step: Build a three-dimensional digital twin model of the library, map the physical bookshelves, aisles, tables and chairs to the virtual space, and establish a two-way mapping table between physical coordinates and logical coordinates. The physical coordinates are accurate to the centimeter level, and the logical coordinates use a four-level coding system of "floor-area-bookshelf-layer number". When the position of a book changes, the virtual book position in the digital twin model is updated synchronously, and a visual path of the movement trajectory is generated in the model. The path color changes dynamically according to the movement speed.

[0014] As a further improvement of this technical solution, S3 further includes an energy consumption optimization step: A tag energy consumption prediction model is established to calculate the tag's motion energy consumption based on accelerometer data, and the data transmission energy consumption based on the communication frequency. A dynamic power adjustment strategy is implemented, and the mechanical energy generated by the three-axis accelerometer when the tag is in motion is converted into electrical energy through the piezoelectric effect to replenish the tag battery.

[0015] As a further improvement of this technical solution, S3 also includes a historical trajectory learning step: A book borrowing trajectory database is established, and the historical borrowing path, stay area and borrowing duration of each book are stored by ISBN classification. A clustering analysis algorithm is used to divide borrowing trajectories with similarity values ​​in a similar range into different pattern categories. When a new book movement trajectory appears, the most similar historical pattern is matched in real time, and the current trajectory prediction result is optimized based on the subsequent trajectory probability distribution of the historical pattern.

[0016] As a further improvement of this technical solution, S3 further includes a multimodal verification step: When a book is stationary for a timeout and registers a temporary location, a visual verification is triggered synchronously, and the surveillance camera above the location is called to confirm the actual status of the book through the image recognition algorithm; Compare the temperature, humidity and light intensity registered at the temporary location with the library environment monitoring system data. If the deviation exceeds the threshold, reposition.

[0017] Compared with the prior art, the present invention has the following beneficial effects: A smart library book retrieval method based on the Internet of Things deploys multiple sensors and positioning base stations, integrates different data to track the location of books, predicts movement trajectories, and accurately presents the real-time location of books, allowing readers to find books quickly, significantly reducing search time and improving reading and learning efficiency. The processing mechanism for temporarily placed and timed-out books, as well as functions such as abnormal trajectory recognition and multi-tag interference elimination, make it convenient for staff to organize bookshelves, inventory books, optimize collection management processes, and improve management efficiency. In addition, technologies such as three-dimensional space mapping, energy consumption optimization, historical trajectory learning, and multimodal verification not only realize digital management of library space, but also reduce equipment energy consumption, optimize trajectory prediction, and ensure accurate location information, providing comprehensive support for the efficient and intelligent operation of smart libraries. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the overall workflow diagram of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] See also Figure 1 As shown, this embodiment provides a smart library book retrieval method based on the Internet of Things, including the following steps: S1. Deploy RFID dense reading and writing arrays on the bookshelf level, set up UWB positioning base stations in the library channels, and embed composite sensor tags into book covers. The tags integrate RFID chips, three-axis accelerometers, and pressure sensors to eliminate positioning blind spots and achieve full coverage of the entire library. Multi-angle signal transmission improves tag recognition rate, and 3D positioning space construction provides the basis for precise positioning.

[0021] S2. Introducing a dynamic position discrimination algorithm, using a three-level threshold judgment mechanism. The first level detects that the three-axis composite acceleration of the accelerometer is greater than 0.8g and lasts longer than 0.3 seconds, preliminarily screening for movement events. The second level verifies that the pressure sensor reading is less than 50kPa and lasts longer than 0.5 seconds, eliminating slight vibration interference. When the three-axis accelerometer detects that the book's continuous displacement is greater than 15 cm / s2 and the pressure sensor reading disappears, it is determined that the book has entered a moving state. The final judgment of the moving state is completed and the positioning tracking of the UWB positioning base station is triggered, improving positioning efficiency and accuracy. S3. By using an extended Kalman filter to fuse the RFID phase ranging data obtained by the RFID dense read / write array and the UWB arrival time difference data, a book movement trajectory prediction model is established. When a book is stationary for more than 30 seconds, it is automatically registered as a temporary location point.

[0022] The extended Kalman filter fusion process in S3 includes: A three-layer data quality assessment mechanism is established. The first layer evaluates the signal strength fluctuation coefficient of RFID phase ranging, calculates the standard deviation of the signal strength in each RFID read and write cycle, defines the fluctuation coefficient = standard deviation / average signal strength, and sets three levels of fluctuation thresholds. Green (less than 0.15) indicates a stable signal, yellow (0.15-0.3) indicates moderate fluctuation, and red (greater than 0.3) indicates severe fluctuation. When the fluctuation coefficient is in the yellow area, the sampling frequency of the reader is automatically increased to 20Hz. In the red area, the adjacent reader joint verification mechanism is started. The second layer evaluates the impact of the multipath effect of the UWB arrival time difference, analyzes the pulse shape of the UWB signal, and The characteristic parameters of the multipath components are extracted, and a multipath effect impact model is constructed based on these parameters. The impact factor is calculated as the power of each multipath component × the sum of the squares of the delay time. A threshold is set. When the impact factor exceeds the threshold, the system automatically switches to the UWB multipath suppression mode and reduces multipath interference through a spatial smoothing algorithm. The third layer evaluates the timestamp synchronization error between the two. A two-way time transfer protocol is used to synchronize the RFID and UWB system clocks every 5 minutes. The sliding standard deviation of the timestamp difference between the two is calculated. When the standard deviation is greater than the preset threshold, the hardware clock calibration process is triggered. An adaptive timestamp compensation algorithm is developed to dynamically adjust the compensation value according to the data transmission path, achieving refined management of the data source quality. Dynamically adjust the fusion weight according to the evaluation results, establish a weight adjustment decision tree model, the input parameters are the three-layer evaluation results, and the output is the fusion weight of RFID and UWB. When the RFID signal fluctuation coefficient is green and the UWB multipath influence is green, the weight is assigned. When RFID is red and UWB is yellow, the weight is adjusted and the abnormal data marking process is started. A weight smooth transition algorithm is designed to avoid trajectory jumps caused by weight mutations. The adjustment amplitude is less than 15% each time, and the transition is completed in 3 cycles, so that the fusion result maintains the best accuracy in various scenarios. The smooth transition algorithm ensures trajectory continuity. 20 calibration points with known coordinates are preset in the library, distributed in the bookshelf area, aisles and lending desks, and the 3 nearest calibration points are automatically selected for self-calibration after completing 100 fusion calculations. A three-step calibration process is performed. The first step: calculate the absolute error between the fusion result and the true coordinate. When the error is greater than 20cm, fine calibration is triggered. The second step: collect 100 sets of fusion data at the point, and use the least squares method to calculate the accuracy of the fusion result. The error correction function is fitted by the method. The third step is to feed back the correction function coefficients to the state transfer matrix of the extended Kalman filter and set up a calibration quality assessment mechanism. When the calibration error is less than 5 cm for three consecutive times, the calibration cycle is extended to 200 calculations to achieve automatic correction of the system error. The adaptive calibration cycle balances accuracy and computational efficiency. By comparing the fusion results of known position points with the true coordinates, the system error is corrected and a three-dimensional space confidence ellipse model is constructed. The ellipse parameters are dynamically adjusted according to historical data. The ellipse parameters include the major axis, minor axis and deflection. When the new data point deviates from the ellipse center by more than 3 times the standard deviation, it is judged as abnormal data. Three levels of processing are performed on the abnormal data. The first level is marked as suspicious data and does not participate in the fusion temporarily, waiting for verification in the next sampling cycle. The second level is: if it is judged to be abnormal for two consecutive cycles, cross-validation of adjacent sensor data is started. The third level is: if the validity cannot be confirmed, the Kalman prediction value of the historical trajectory is called to replace the data point, effectively filtering out noise data and ensuring that the fusion process is not interfered with.

[0023] The establishment of the book movement trajectory prediction model in S3 includes: A quaternion state space is constructed, including the book's position coordinates, movement speed, acceleration, and azimuth. The Northeast Celestial coordinate system is used, with position coordinates accurate to the centimeter level, velocity resolution of 0.01m / s, acceleration resolution of 0.05m / s², and azimuth accuracy of ±2°. The state transition equation X(K+1)=F・X(K)+G・U(K)+W(K) is established, where X contains position (x, y, z), velocity (vx, vy, vz), acceleration (ax, ay, az), and azimuth. An adaptive noise covariance matrix is ​​designed and dynamically adjusted according to the book's motion state. The physical space of the library is divided into three-dimensional grid cells of 1m×1m×2m. A three-dimensional grid index table is established, with each grid storing the coordinate range, cost coefficient, and physical attributes. A multi-factor cost allocation algorithm is used to assign a travel cost coefficient to each grid. The bookshelf area is set as a high-cost area (80-100), the passage area is set as a low-cost area (10-20), and the table and chair area is set as (30-50). The cost coefficient is greater than or equal to 80 for the high-cost area, and the cost coefficient is less than or equal to 20 for the low-cost area. The dynamic cost is the real-time traffic density × 10. If there are more than 3 people per square meter, the cost increases by 30. The temporary cost is 150 for the construction area and 120 for the equipment failure area. This makes the predicted path more consistent with the actual traffic rules. The dynamic update mechanism adapts to the real-time changes of the library. A heuristic search algorithm is used to combine the current position, speed and travel cost coefficient to predict the movement path within the next 3 seconds. The A* algorithm is used to design a three-stage search strategy: Coarse search: Generate an initial path on a 1m resolution grid, with a heuristic function h(n) = 0.7 × Euclidean distance + 0.3 × cost accumulation.

[0024] Refine search: Optimize the path on a 0.2m resolution grid, adding velocity constraints (v≤1.5m / s) and acceleration constraints (a≤0.8m / s²).

[0025] Smoothing: The path is smoothed using Bezier curves with a curvature radius of ≥ 0.5m.

[0026] Develop a candidate trajectory generator that generates five differentiated paths based on the current state and generates a probability distribution containing the five candidate trajectories, as follows: Straight track: Maintain current direction and speed.

[0027] Deceleration trajectory: decelerates at -0.3m / s² until it stops.

[0028] Turn trajectory: Turn 30° left / right and maintain speed.

[0029] Acceleration trajectory: Accelerate at 0.2m / s² and maintain direction.

[0030] A probability value is assigned to each trajectory, and the weight is adjusted based on the frequency of human motion patterns in historical data statistics. A probability distribution function based on the Gaussian mixture model is constructed. The initial parameters are obtained through training with historical trajectory data and updated in real time. When a book is detected entering a specific area, the weight of the corresponding motion pattern is adjusted. An attention mechanism is introduced to assign weights to historical trajectories near the current state. The formula is: attention weight = exp(-d² / 2p²), where d is the trajectory similarity distance and p is the adaptive bandwidth parameter. Online learning is performed every 10 minutes using newly collected trajectory data to update the Gaussian mixture model parameters and establish a prediction-observation comparison window. The actual position is compared with the predicted trajectory every 500ms, and the mean square error is calculated. When the mean square error is greater than 0.25m², the trajectory correction process is triggered: Reinitialize the state space, use the current observation as the starting point, backtrack to the last five time steps, perform state re-estimation through particle filtering, regenerate candidate trajectories, optimize the path in the deviation direction, set the prediction confidence index: C = exp(-MSE / p²), when C < 0.6, reduce the display priority of the predicted trajectory to ensure that the prediction error is controllable and the re-estimation process effectively corrects the accumulated deviation.

[0031] Temporary location registration in S3 includes: When the book is detected to be stationary, a three-level confirmation mechanism is activated. The first level confirms that the accelerometer's three-axis data are all less than 0.05g for 20 seconds. A sliding window variance analysis method is used to calculate the variance of the three-axis acceleration data within the 20-second window. When the variances of the three axes (X, Y, and Z) are all less than 0.0025g² and the mean is less than 0.05g, it is determined to be in a preliminary stationary state. Sudden changes with a rate of change greater than 0.1g / s are smoothed to avoid misjudgments caused by slight shaking of the book. The second level confirms that the RFID signal strength fluctuation is less than 3dBm for 10 seconds. The multi-reader joint verification mechanism is activated, and the three nearest RFID reader signals are monitored simultaneously. The sliding standard deviation of the signal strength is calculated. If the standard deviation is less than 1dBm for 10 consecutive seconds and the signal strength fluctuation of each reader is less than 3dBm, the third level verification is entered. The Kalman filter is used to predict the trend of signal strength changes. If the deviation between the actual value and the predicted value is greater than 5dBm, the timing is reset. The third level verifies that the position change is less than 5cm for 5 seconds through UWB positioning. Switch UWB positioning to high-precision mode. High precision increases the sampling frequency to 50Hz, with a measurement accuracy of 3cm. The Euclidean distance of adjacent positions within 5 seconds in three-dimensional space is calculated. When the distance between 10 consecutive sampling points is less than 5cm, it is confirmed to be in a stationary state. If the position change trends measured by more than three UWB base stations are inconsistent, the verification time is extended to 15 seconds. When registering a temporary location point, the environmental parameters are recorded synchronously, including the light intensity, temperature and humidity, and noise level of the area. A multi-dimensional code containing spatiotemporal characteristics is generated for the temporary location point. The first 12 bits are the location coordinate hash value, the middle 8 bits are the timestamp code, and the last 4 bits are the environmental parameter classification code. The three-dimensional coordinates (x, y, z) are converted into a one-dimensional integer through coding, and then a hash operation is performed. The first 12 bits of the hash value are intercepted as the location identifier. The mapping accuracy is 10cm×10cm×20cm. When a conflict occurs, a 1-bit random number is appended to the original code and the conflict log is recorded. The timestamp is used to convert the millisecond time The stamp is compressed into 8-bit Base64 encoding, with a time accuracy of 1 minute. The encoding format is that the first 4 digits represent the date offset, and the last 4 digits represent the minutes of the day. A distributed hash table storage architecture is established, and the temporary location point data is sharded and stored on 3 nodes according to the encoded hash value. The space partitioning algorithm is used to recursively divide the library space into 8 sub-nodes. Each node has a maximum capacity and automatically splits when the capacity is exceeded. It supports spatial range queries, such as querying "all temporary location points within 2 meters around the coordinates (10,5,2)", which balances access efficiency and storage costs. The spatial index greatly improves the regional retrieval speed.

[0032] S3 also includes the abnormal trajectory identification step: Five types of abnormal trajectory patterns are defined, including trajectories passing through physically inaccessible bookshelf-dense areas, staying in non-borrowing areas for more than 15 minutes, traveling back and forth between two locations more than 5 times / hour, path similarity > 80%, instantaneous speed > 2m / s (normal human walking speed 1.1-1.5m / s), and vertical upward movement of more than 2 meters. A sliding time window is established with a length of 5 minutes and a step size of 10 seconds to ensure the integrity of trajectory information. The trajectory within the window is resampled and unified into an equally spaced sequence of 100 points. The trajectory feature vector is calculated and updated every 10 seconds. The cosine similarity between the current trajectory and the normal pattern is calculated in real time, and the maximum value is taken as the matching degree. When the similarity is When the accuracy is lower than the preset threshold, it is marked as abnormal. For books with abnormal tracks, the secondary positioning enhancement is automatically triggered, the UWB sampling frequency is increased from 10Hz to 50Hz, and three adjacent RFID readers are activated simultaneously based on the current position. The reader working parameters are adjusted, the transmission power is increased to 30dBm, and the receiving sensitivity is adjusted to -85dBm. Phase differential positioning technology is used, and the three readers form a triangulated positioning. A Kalman filter is established to fuse UWB and RFID data. The state vector contains position, velocity and acceleration. The data quality weight is set, the UWB position weight is 0.7, and the RFID phase difference weight is 0.3. When the deviation between the two is greater than 30cm, the third level verification is started.

[0033] S3 also includes a multi-label interference elimination step: Deploy the RFID signal feature monitoring module to analyze the time domain and frequency domain features of the received signal in real time. The time domain analysis is to calculate the signal duration and the slope of the rising / falling edge. When the signal duration is detected to be abnormally prolonged (>15% of the standard value), it is marked as potential interference. The signal spectrum purity is detected by fast Fourier transform. When the harmonic component accounts for >20%, it is determined that there is multi-tag interference. The number of tags is estimated based on the variance of the signal strength fluctuation. The formula is: N≈M² / 0.76 (M is the standard deviation of the signal strength). Set three levels of warning thresholds: green (N<3), yellow (3≤N<5), and red (N≥5). When the yellow warning is received for three consecutive sampling periods, the interference elimination process is automatically started, that is, the establishment of A speed-to-slot mapping model was established. Low-speed tags (v < 0.5 m / s) were assigned a basic slot width of 200 μs. For medium-speed tags (0.5 ≤ v < 1.5 m / s), the slot width was 200 μs + (v - 0.5) × 100 μs. For high-speed tags (v ≥ 1.5 m / s), the slot width was 300 μs + (v - 1.5) × 50 μs. A slot allocation optimization algorithm was designed that applied the maximum-minimum fairness principle, prioritizing high-speed tags. The slot allocation scheme was recalculated every 50 ms and dynamically adjusted based on the latest speed data. When a sudden speed increase was detected for a tag, 20% of the time resources were borrowed from the slots of low-speed tags to ensure efficient resource utilization. The phase fingerprint feature, Doppler frequency shift feature and signal strength change feature of each tag RFID signal are extracted. Among them, the phase fingerprint feature is to extract the change curve of the signal phase over time, the Doppler frequency shift feature is to detect the signal frequency offset through a mixer, and the signal strength change feature is to calculate the change rate of the received signal strength indication within a 10ms window. The neighborhood radius ε=0.3 (feature space distance) and the minimum number of points MinPts=5 are set. The feature similarity measurement function is constructed, d=0.5d phase + 0.3d frequency shift + 0.2d intensity, where d is the Euclidean distance. After each round of clustering, the cluster center is recalculated and the feature weight is adjusted. When a signal collision is detected, a preliminary classification is first performed based on the phase fingerprint feature, and the classification result is verified by Doppler frequency shift to eliminate abnormal points. A secondary confirmation is performed through the signal strength change feature to ensure that each cluster corresponds to a unique tag, and multi-label data decoupling is achieved through feature clustering.

[0034] S3 also includes three-dimensional space mapping steps: Build a 3D digital twin model of the library, map the physical bookshelves, aisles, and tables and chairs into the virtual space, and establish a bidirectional mapping table between physical coordinates and logical coordinates. The physical coordinates are accurate to the centimeter level, and the logical coordinates use a four-level coding system of "floor-area-bookshelf-layer number", as follows: The northeast celestial coordinate system is used, with the origin set at the center of the main entrance of the library, the X / Y axis ±1cm, and the Z axis ±2cm. Floors are represented by numbers (1-5 floors), areas are represented by AZ, bookshelves are represented by two digits (01-99), and layer numbers are represented by 1-6 (there are 6 bookshelves in total). A complete coding example: 2-B-15-3 represents the 3rd layer of bookshelf No. 15 in Area B on the 2nd floor. The mapping relationship is stored in a B+ tree index structure. The key is the physical coordinate and the value is the logical code. A reverse index table is established to support fast query of the physical coordinate range through logical coding. When the bookshelf layout is adjusted, only the affected mapping relationship is updated. Incremental updates reduce maintenance costs. When the book position changes, the virtual book position in the digital twin model is updated synchronously, and a movement trajectory visualization path is generated in the model. The path color changes dynamically according to the movement speed, as shown below: The WebSocket protocol is used to establish a real-time communication channel between the digital twin server and the positioning system, and a three-level cache mechanism is implemented, including memory cache (the last 1,000 updates), Redis cache (the last 24 hours), and database persistence. When the position of the book changes by more than 5 cm, an update is triggered to avoid frequent refreshes caused by small fluctuations. The particle system is used to render the movement trajectory, and each particle represents a 0.1-second position record to achieve trajectory color mapping, where low speed (<0.5m / s) is displayed in green, medium speed (0.5-1.5m / s) is displayed in yellow, and high speed (>1.5m / s) is displayed in red. The trajectory is retained for 30 minutes after generation, after which the transparency gradually decreases until it disappears. The kernel density estimation algorithm is used to calculate the space For hotspots, the bandwidth parameter is set to 1.5m. Hotspot changes are counted by time period. For example, the area around the borrowing desk is a hotspot during the morning rush hour (8:00-9:00), and the leisure area becomes more popular during the lunch break (12:00-13:00). A heat map visualization is generated, and a gradient color (blue-green-yellow-red) is used to represent the heat level. Based on historical trajectory data, the Dijkstra algorithm is used to calculate the optimal borrowing path. The recommended path is highlighted in the digital twin interface, along with the estimated travel time. The space utilization rate of each area is calculated, that is, actual used area / total area × 100%. The distribution of space idle time is analyzed. For example, the idle rate of some bookshelves at night reaches 80%. Based on the analysis results, it is recommended to adjust the bookshelf layout or open new areas to reduce operating costs.

[0035] S3 also includes energy consumption optimization steps: Establish a tag energy consumption prediction model, calculate the tag motion state energy consumption based on accelerometer data, and calculate the data transmission energy consumption based on the communication frequency, as follows: A model for the relationship between three-axis acceleration and energy consumption was established: Emotion = k1・|ax|+k2・|ay|+k3・|az|+C, where k1, k2, and k3 are calibration coefficients and C is the static power consumption. The parameters were calibrated experimentally: C = 0.5μW in the static state, k1 = k2 = 2.3μW / (m / s²) for horizontal movement, and k3 = 3.1μW / (m / s²) for vertical movement. A sliding window integral calculation was used, with energy consumption calculated every 100ms, and the window length was set to 5 seconds. The relationship between communication frequency and energy consumption was analyzed: Ecommunication = f・(a・f+b), where f is the communication frequency and a = 0.05nJ / Hz². , b=20nJ / Hz, each byte of data transmitted consumes an additional 0.1μJ of energy, and the energy consumption is calculated according to different communication modes (positioning, data upload, heartbeat). The total energy consumption model is Etotal=Emotion+Ecommunication+Esensing, where Esensing is the sensor sampling energy consumption. Kalman filtering is used to predict the energy consumption trend in the next 5 minutes. The state vector contains the current energy consumption and the rate of change. The energy consumption warning threshold is set. When the predicted remaining power is less than 20%, the low power mode is triggered and a dynamic power adjustment strategy is implemented. The mechanical energy generated by the three-axis accelerometer when the tag moves is converted into electrical energy through the piezoelectric effect and replenished to the tag battery. The details are as follows: Using PZT-5H piezoelectric ceramic material, measuring 6mm × 4mm × 0.5mm and with a resonant frequency of 180Hz, it converts AC power into DC power with a conversion efficiency of >85%. A 10mF supercapacitor with a withstand voltage of 5V is then used to store the converted energy. A DC-DC converter adjusts the load impedance to achieve maximum power point tracking (MPPT). An energy conversion model is established: Ecollected = k・a²・t, where a is the acceleration amplitude, t is the vibration duration, and k = 0.2μJ / (m²・s³). The charging circuit is activated when the capacitor voltage exceeds 2.5V and stops supplying power when it is less than 1.8V. A dual-power switching circuit is designed to prioritize the use of harvested energy, switching to battery power when insufficient. An energy allocation strategy is implemented: 60% of the harvested energy is used to maintain basic functions and 40% is used for charging. An energy log is established to record the daily energy collected, consumed, and battery status.

[0036] S3 also includes historical trajectory learning steps: A book borrowing trajectory database was established, which was classified by ISBN and stored with the historical borrowing path, stay area, and borrowing duration of each book. A cluster analysis algorithm was used to classify borrowing trajectories with similarity values ​​into different pattern categories, as follows: The geometric features of the trajectory are extracted, including the total length, average speed, number of turns, and maximum dwell time. The regional visit features are calculated, namely the number of areas passed, the sequence of high-frequency dwelling areas, and the proportion of dwellings in different functional areas. The temporal feature vector is generated, including the distribution of borrowing periods and the probability density of borrowing duration. A dynamic density threshold is introduced to adjust the clustering density according to the popularity of the book. The threshold for popular books is reduced by 20%, and the threshold for unpopular books is increased by 30%. The weighted Euclidean distance is fused with multi-dimensional distance metrics, with a geometric feature weight of 0.4, a regional feature weight of 0.3, and a temporal feature weight of 0.3. Trajectories with a density lower than 1 / 3 of the global average density are marked as outliers and do not participate in pattern classification. Five typical borrowing modes are divided, including fast borrowing mode, research-based borrowing mode, casual browsing mode, return mode, and abnormal mode. When a new book movement trajectory appears, the most similar historical mode is matched in real time, and the current trajectory prediction result is optimized based on the subsequent trajectory probability distribution of the historical mode, as follows: Construct a historical pattern index tree, hierarchically index by book category, borrowing period, and pattern type, reduce the matching search space, use the approximate nearest neighbor search algorithm to prioritize matching the first three most similar historical patterns, calculate the dynamic time-warped distance between the current trajectory and the historical pattern, count the common path branches of the subsequent trajectory of each historical pattern, calculate the probability of each branch, such as 60% going to the study area, 30% going to the book return, and 10% continuing to browse, and adjust the probability distribution based on the current book position and moving direction. If the current location is close to the book return, the probability of the book return path will increase by 20%, and integrate the prediction model to weight the probability prediction results based on the historical pattern and the extended Kalman filter prediction results. When the newly collected trajectory data is combined with the When the prediction deviation exceeds the threshold, the historical pattern is re-matched and the prediction is updated. New trajectory data from the previous 24 hours is automatically imported at midnight every day, and the pattern category is updated using an online clustering algorithm. When the new data causes the existing pattern cluster center to shift beyond the threshold, the pattern is triggered to be re-divided. If the frequency of a certain pattern is less than 0.5% for 7 consecutive days, it will be removed from the historical pattern library. The prediction results based on the historical pattern are compared with the actual trajectory every month. The parameters of the pattern with an accuracy rate of less than 70% are adjusted or retrained. The library activity calendar and holiday schedule information are connected to analyze the impact of special events on borrowing patterns, dynamically adjust the pattern weight, and establish exclusive pattern libraries for different time periods according to seasonal changes to improve the targeted nature of the prediction.

[0037] S3 also includes multimodal verification steps: When a book is stationary for a timeout and registers at a temporary location, visual verification is triggered synchronously, and the surveillance camera above the location is called to confirm the actual status of the book through an image recognition algorithm. A wide-angle camera is deployed above each bookshelf area to establish a spatial coordinate mapping relationship with the RFID reader array. When a book is stationary for a timeout and triggers temporary location registration, the system automatically retrieves the real-time image of the corresponding camera, and the retrieval response time is controlled within 200ms. The target detection model is used to perform transfer learning on the morphological characteristics of books. The training data set contains 200,000 book images at different angles and lighting conditions. During detection, the book outline, cover features and placement status are identified, and a confidence assessment system is established. When the confidence score of the detected book is greater than 0.8, the recognition is confirmed to be valid. If the score is in the range of 0.5-0.8, the adjacent camera is automatically called for cross-verification. If the score is less than 0.5, the manual review process is triggered to reduce the impact of environmental interference. The dynamic judgment mechanism improves verification reliability. Compare the temperature, humidity and light intensity registered at the temporary location with the library environment monitoring system data. If the deviation exceeds the threshold, reposition the location. Deploy environmental monitoring nodes at intervals of 10m×10m in the library. Each node integrates temperature and humidity sensors and light sensors. The monitoring data is uploaded to the central server in a 1-minute cycle. The threshold is updated daily based on historical data. When the deviation between the environmental parameters recorded at the temporary location and the monitoring system data exceeds the threshold, triple verification is initiated. First, the sensor data is reread to eliminate instantaneous errors. Secondly, the data of adjacent monitoring nodes are called for spatial interpolation verification. If there is still a deviation, the UWB and RFID joint repositioning is triggered, and the repositioning sampling frequency is increased to 50Hz. Different weights are assigned to visual verification (weight 0.6), environmental parameter comparison (weight 0.3), and basic sensor data (weight 0.1), and the comprehensive confidence is calculated: Z=0.6×Zvision+0.3×Zenvironment+0.1×Zbase, where Z is the comprehensive confidence and Zx is the confidence score of each modality. When the comprehensive confidence is greater than 0.9, the temporary position is directly confirmed. Secondary verification is initiated in the range of 0.7-0.9. A full relocation is triggered when the confidence is less than 0.7, and an early warning message is pushed to the administrator. An abnormal case library is established to record the process data and final results of each multimodal verification. Through the reinforcement learning algorithm, the weight distribution and threshold setting of each modality are optimized according to historical cases, and the model parameters are automatically updated every week.

[0038] S4. Multi-dimensional location information display can help administrators quickly understand the status of books. Heat maps and virtual coordinates can improve the recognition of books in special status. A hierarchical location display mechanism is used to simultaneously display the original shelf code and real-time location status of books. A heat map of the last disappearance area is provided for moving books. Virtual coordinates with a countdown are marked for temporarily placed books. Layer rendering improves interface performance. Heat maps can intuitively display loss patterns. Dynamic labels can enhance the recognition of temporary books. S5. For books in temporary locations that have not been returned to the shelves for more than 48 hours, the positioning beacon autonomous alarm mode is activated, and a gradient-enhanced prompt tone is emitted through the built-in buzzer of the tag.

[0039] The above-mentioned smart library book retrieval method based on the Internet of Things achieves accurate positioning and tracking of books by deploying RFID dense read-write arrays on the bookshelf level, setting up UWB positioning base stations in the library channels, and embedding composite tags integrating multiple sensors on the book covers. It introduces a dynamic position discrimination algorithm to identify the movement status of books, uses the extended Kalman filter to fuse data to establish a motion trajectory prediction model, and adopts a hierarchical position display mechanism to provide detailed location information. For books that are temporarily placed or have not been returned after a timeout, special processing and alarm prompts are respectively performed. This effectively solves the problems of delayed dynamic position updates and inaccurate retrieval results in traditional library book retrieval, and significantly improves the efficiency of readers in retrieving books.

[0040] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A book retrieval method for an intelligent library based on the Internet of Things, characterized in that: The following steps are involved: S1. Deploy dense RFID reader arrays on the bookshelf level, set up UWB positioning base stations in the library channels, and embed composite sensor tags on book covers. The tags integrate RFID chips, three-axis accelerometers, and pressure sensors. S2. Introducing a dynamic position discrimination algorithm: When the three-axis accelerometer detects that the book's continuous displacement is greater than 15 cm / s² and the pressure sensor reading disappears, it is determined that the book has entered a moving state and triggers positioning tracking by the UWB positioning base station; S3. Using an extended Kalman filter to fuse RFID phase ranging data obtained from a dense RFID reader array with UWB arrival time difference data, a book trajectory prediction model is established. When a book is stationary for 30 seconds, a temporary location point is automatically registered. S4. Use a hierarchical location display mechanism to simultaneously display the original shelf code and real-time location status of books, provide a heat map of the last disappearing area for moving books, and mark virtual coordinates with a countdown for temporarily placed books; S5. For books in temporary locations that have not been returned to the shelves for more than 48 hours, the positioning beacon autonomous alarm mode is activated, and a gradient-enhanced prompt tone is emitted through the built-in buzzer of the tag.

2. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, characterized in that: The extended Kalman filter fusion process in S3 includes: A three-layer data quality assessment mechanism is established. The first layer evaluates the signal strength fluctuation coefficient of RFID phase ranging, the second layer evaluates the multipath effect of UWB arrival time difference, and the third layer evaluates the time stamp synchronization error between the two. The fusion weight is dynamically adjusted according to the evaluation results, and a self-calibration process is automatically performed every 100 fusion calculations. The system error is corrected by comparing the fusion results of known position points with the actual coordinates.

3. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, wherein: The establishment of the book movement trajectory prediction model in S3 includes: A four-element state space is constructed, including the book position coordinates, movement speed, acceleration and direction angle. The physical space of the library is divided into three-dimensional grid units of 1m×1m×2m. A travel cost coefficient is assigned to each grid. The bookshelf area is set as the high-cost area, and the passage area is set as the low-cost area. The cost coefficient greater than or equal to 80 is considered a high-cost area, and the cost coefficient less than or equal to 20 is considered a low-cost area. A heuristic search algorithm is used to combine the current position, speed and travel cost coefficient to predict the movement path within the next 3 seconds, generating a probability distribution containing 5 candidate trajectories.

4. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, wherein: The temporary location point registration in S3 includes: When a book is detected to be stationary, a three-level confirmation mechanism is activated. The first level confirms that the accelerometer's three-axis data are all less than 0.05g for 20 seconds. The second level confirms that the RFID signal strength fluctuation is less than 3dBm for 10 seconds. The third level verifies that the position change is less than 5cm for 5 seconds through UWB positioning. When registering a temporary location point, the environmental parameters are recorded synchronously, including the light intensity, temperature and humidity, and noise level of the area, and a multi-dimensional code containing spatiotemporal characteristics is generated for the temporary location point. The first 12 bits are the location coordinate hash value, the middle 8 bits are the timestamp code, and the last 4 bits are the environmental parameter classification code.

5. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, characterized in that: The S3 also includes the abnormal trajectory identification step: Five types of abnormal trajectory patterns are defined, and a sliding time window with a length of 5 minutes is established. The cosine similarity between the current trajectory and the normal pattern is calculated in real time. When the similarity is lower than the preset threshold, it is marked as abnormal. The secondary positioning enhancement is automatically triggered for books with abnormal trajectories, the UWB sampling frequency is increased from 10Hz to 50Hz, and three adjacent RFID readers are activated at the same time.

6. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, characterized in that: The S3 also includes a multi-label interference elimination step: When it is detected that there are three or more mobile tags within the signal range of the same RFID reader, the interference elimination process is started; A time division multiple access time slot allocation algorithm is used to dynamically allocate an exclusive communication time slot to each tag. The time slot width is dynamically adjusted according to the tag's moving speed. The phase fingerprint characteristics, Doppler frequency shift characteristics and signal strength change characteristics of each tag's RFID signal are extracted, and multi-tag data decoupling is achieved through feature clustering.

7. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, characterized in that: The S3 also includes a three-dimensional space mapping step: A three-dimensional digital twin model of the library is constructed, and the physical bookshelves, aisles, tables and chairs are mapped to the virtual space. A two-way mapping table between physical coordinates and logical coordinates is established. The physical coordinates are accurate to the centimeter level, and the logical coordinates adopt a four-level coding system of "floor-area-bookshelf-layer number". When the position of a book changes, the virtual book position in the digital twin model is updated synchronously, and a visual path of the movement trajectory is generated in the model. The path color changes dynamically according to the movement speed.

8. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, characterized in that: The S3 also includes an energy consumption optimization step: A tag energy consumption prediction model is established to calculate the tag's motion energy consumption based on accelerometer data, and the data transmission energy consumption based on the communication frequency. A dynamic power adjustment strategy is implemented, and the mechanical energy generated by the three-axis accelerometer when the tag is in motion is converted into electrical energy through the piezoelectric effect to replenish the tag battery.

9. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, wherein: The S3 also includes historical trajectory learning steps: A book borrowing trajectory database is established, and the historical borrowing path, stay area and borrowing duration of each book are stored by ISBN classification. A clustering analysis algorithm is used to divide borrowing trajectories with similarity values ​​in a similar range into different pattern categories. When a new book movement trajectory appears, the most similar historical pattern is matched in real time, and the current trajectory prediction result is optimized based on the subsequent trajectory probability distribution of the historical pattern.

10. The method for searching books in an intelligent library based on the Internet of Things according to claim 1, characterized in that: The S3 also includes a multimodal verification step: When a book is stationary for a timeout and registers a temporary location, a visual verification is triggered synchronously, and the surveillance camera above the location is called to confirm the actual status of the book through the image recognition algorithm; Compare the temperature, humidity and light intensity registered at the temporary location with the library environment monitoring system data. If the deviation exceeds the threshold, reposition.

Images