Intelligent Live Animal Detection and Customs Clearance Management System Based on Millimeter-Wave Radar

Abstract

This invention relates to the field of intelligent detection and customs clearance management technology, particularly to a live animal intelligent detection and customs clearance management system based on millimeter-wave radar. The system includes a millimeter-wave radar array module, a data processing and analysis module, a baggage sorting control module, a quarantine process management module, an alarm response module, and a system management platform. The millimeter-wave radar array module uses multi-band fusion detection technology to acquire live biological characteristics. The data processing and analysis module uses algorithms such as an improved dual-stream convolutional neural network to identify live animals and output confidence scores. Based on the detection results, the system management platform links the various modules to achieve dynamic baggage sorting, tiered alarms, intelligent allocation of quarantine resources, and closed-loop control throughout the entire process. This invention can effectively identify concealed live animals, prevent biological invasion and disease transmission, and significantly improve customs clearance efficiency, reduce manual intervention costs, and ensure the traceability and security of customs clearance data through automated and intelligent management.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection and customs clearance management technology, specifically to a live animal intelligent detection and customs clearance management system based on millimeter-wave radar. Background Technology

[0002] Traditional customs baggage inspection mainly relies on manual inspection and single-technology detection, which has many limitations.

[0003] Manual inspection relies on the experience and visual observation of customs officers, which is not only inefficient and unable to handle large-scale traffic, but also prone to missing live animals hidden in baggage compartments and hidden pockets. Furthermore, frequent opening of bags increases the risk of biological contamination, impacting the passenger experience. Existing detection technologies, such as X-ray imaging, while providing images of the internal structure of baggage, struggle to effectively distinguish live animals from similarly shaped non-living items, resulting in a high false positive rate. Infrared thermal imaging technology is significantly affected by ambient temperature, leading to insufficient accuracy in complex environments.

[0004] With the continuous growth of global customs business volume, traditional detection methods can no longer meet the dual requirements of efficient customs clearance and precise supervision. Therefore, a live animal intelligent detection and customs clearance management system based on millimeter-wave radar is proposed to address the above issues. Summary of the Invention

[0005] The purpose of this invention is to provide a live animal intelligent detection and customs clearance management system based on millimeter-wave radar to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar includes a millimeter-wave radar array module, a data processing and analysis module, a baggage sorting control module, a quarantine process management module, an alarm response module, and a system management platform.

[0008] Millimeter-wave radar array modules are deployed at key nodes of the customs baggage conveyor belt to scan the live biological characteristics inside the baggage in real time.

[0009] The data processing and analysis module identifies the presence of live animals through millimeter-wave feature extraction algorithms and outputs the detection results to the system management platform;

[0010] The system management platform dynamically triggers the following collaborative processes based on the test results:

[0011] The baggage sorting control module adjusts the sorting path in real time, guiding baggage containing live animals to the quarantine channel;

[0012] The alarm response module generates tiered alarm signals and associates them with quarantine priorities.

[0013] The quarantine process management module automatically allocates quarantine resources and generates electronic quarantine task orders.

[0014] As a preferred option, the millimeter-wave radar array module employs multi-band fusion detection technology, specifically including:

[0015] Low-frequency bands are used to penetrate luggage materials to obtain the outline of a living person; the frequency of the low-frequency band is between 24-33 GHz.

[0016] High-frequency bands are used to capture micro-motion characteristics, with frequencies between 60-90 GHz.

[0017] A live biological feature model is established based on dual-band data fusion to distinguish live animals from static interference objects.

[0018] As a preferred option, the data processing and analysis module includes:

[0019] The millimeter-wave signal preprocessing unit performs motion compensation and noise filtering on the radar echo;

[0020] The liveness feature recognition unit extracts biological micro-motion spectral features through an improved two-stream convolutional neural network model;

[0021] The dynamic threshold determination unit outputs a liveness confidence score based on an adaptive confidence calibration mechanism.

[0022] As a preferred option, in the improved two-stream convolutional neural network model:

[0023] The probability of a living organism is obtained by weighting the outputs of each convolutional layer by applying the Sigmoid function.

[0024] The output of each convolutional layer is generated by processing low-frequency and high-frequency band data separately through convolution kernels, summing them, and then passing them through the ReLU activation function.

[0025] The dynamic frequency band weighting coefficients are determined based on the exponentially normalized values ​​of the signal-to-noise ratios (SNRs) of each layer, where the SNR weights are adjusted by the ambient noise energy value.

[0026] The environmental disturbance adjustment factor approaches 1 as the environmental noise energy increases, and the empirical constant for material attenuation ranges from 0.1 to 0.5.

[0027] As a preferred approach, the adaptive confidence calibration mechanism includes:

[0028] The final confidence score is a weighted sum of the probability of the presence of a living organism and the Shannon entropy of the micromotion feature;

[0029] The weighting coefficients are dynamically adjusted according to changes in ambient temperature; for every 1 degree Celsius increase in temperature, the probability value weight increases by 0.05.

[0030] The threshold for determining liveness is determined by adding a compensation amount to the square of the material density, which is based on a base threshold of 0.6. The material compensation coefficient is fixed at 0.02.

[0031] When the final confidence score exceeds the dynamic threshold, it is determined that the luggage contains a live animal.

[0032] As a preferred solution, the baggage sorting control module performs dynamic path optimization, including:

[0033] The sorting priority index is calculated by multiplying the confidence score by a negative exponential decay function of the quarantine channel load;

[0034] When the real-time load of the channel reaches the baseline load value, the priority index decays to 1 / e of the original value;

[0035] By using radio frequency identification (RFID) technology to link baggage with inspection data, the sorting machine is controlled to transfer high-priority baggage to the quarantine area.

[0036] As a preferred solution, the alarm response module adopts a three-level collaborative handling mechanism, including:

[0037] Level 1 alarm: When the confidence score is >0.9, an audible and visual alarm is triggered and the conveyor belt is frozen;

[0038] Level 2 alarm: When the confidence score is 0.7 ≤ confidence score ≤ 0.9, a "to be reviewed" label is generated and a manual inspection is notified;

[0039] Level 3 alarm: When the confidence score is 0.6 ≤ confidence score < 0.7, only a low-risk alert is sent to the quarantine management module;

[0040] Alarm signals are pushed to customs personnel's mobile terminals in real time.

[0041] As a preferred option, the quarantine process management module includes:

[0042] The quarantine plan generation unit matches a pre-set treatment plan library based on the type of living organism.

[0043] The resource scheduling unit allocates quarantine resources based on the spatial location coordinates of baggage and confidence scores;

[0044] The electronic customs clearance document generation unit automatically generates electronic release instructions and updates the customs clearance system after quarantine is completed.

[0045] As a preferred option, the resource scheduling unit executes an optimization algorithm, including:

[0046] The scheduling cost function minimizes the weighted response time of all baggage awaiting quarantine;

[0047] The weighting factor is the ratio of the confidence score to the distance from the quarantine officer to the luggage;

[0048] The response time is calculated based on the average time taken for similar quarantine procedures in the past.

[0049] As a preferred solution, the system management platform achieves closed-loop control throughout the entire process, including:

[0050] Millimeter-wave detection data is integrated with the customs clearance system via an application programming interface (API).

[0051] An electronic traceability chain is constructed from detection to release, with data at each stage stored using blockchain technology;

[0052] A heatmap of customs clearance efficiency is generated based on confidence scores and quarantine time, visually displaying process bottlenecks.

[0053] As can be seen from the technical solution provided by the present invention above, the intelligent detection and customs clearance management system for live animals based on millimeter-wave radar provided by the present invention has the following beneficial effects:

[0054] I. Precise detection to strengthen biosafety control:

[0055] The millimeter-wave radar array module employs multi-band fusion detection technology, combined with advanced algorithms from the data processing and analysis module. It can effectively penetrate various luggage materials and accurately capture the minute physiological characteristics of live animals, achieving highly reliable identification of concealed live animals. This technological breakthrough greatly reduces the risk of misjudgment and missed detection by traditional detection methods, and can effectively intercept illegally imported live animals.

[0056] II. Intelligent linkage improves customs clearance efficiency:

[0057] The system management platform coordinates various functional modules to achieve fully automated and intelligent management of the entire process from detection, sorting, alarms to quarantine. The baggage sorting control module dynamically plans routes based on detection results and channel load, the quarantine process management module intelligently allocates resources, and the alarm response module accurately handles cases according to their classification. All links work closely together and operate efficiently. Compared with traditional manual operation processes, it significantly shortens the processing time for baggage containing live animals, greatly improves the overall customs clearance efficiency of the port, reduces passenger waiting time, and optimizes the customs clearance experience.

[0058] III. Scientific Decision-Making to Enhance Emergency Management Capabilities:

[0059] The alarm response module's tiered handling mechanism, combined with the system management platform's intelligent analysis function, enables rapid and accurate responses to anomalies and provides scientific decision-making support for customs management personnel through data analysis. This proactive management model allows customs to plan resource allocation in advance, respond calmly to various emergencies, and effectively improve the scientific nature and proactivity of emergency management.

[0060] IV. Full traceability to ensure regulatory compliance and security:

[0061] The electronic traceability chain built using blockchain technology achieves the immutability and full traceability of data throughout the entire customs clearance process, strictly complying with customs regulations and audit requirements. It provides strong data support for law enforcement and reduces enforcement risks. At the same time, the system's comprehensive security protection and access control system effectively prevents data leakage and illegal operations, ensuring the security of customs business information.

[0062] V. Optimize configuration and reduce system operating costs:

[0063] Automated and intelligent process design reduces manual intervention and lowers labor costs; intelligent algorithm-driven precise resource scheduling avoids equipment idleness and resource waste, improves the work efficiency of equipment and personnel, optimizes the allocation of customs operation resources, and effectively reduces long-term operating costs.

[0064] VI. Flexible adaptation, expanding the scope of application scenarios:

[0065] The millimeter-wave radar array module has environmental adaptability and can dynamically adjust detection parameters according to different luggage materials and complex environments; the data processing and analysis module adapts to the differences in various port environments through online learning; this makes the invention widely applicable to different types of customs ports, international logistics hubs and other scenarios, demonstrating strong versatility and adaptability. Attached Figure Description

[0066] Figure 1 This is a schematic diagram of the overall structure of the intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0068] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0069] like Figure 1As shown, this embodiment of the invention provides a live animal intelligent detection and customs clearance management system based on millimeter-wave radar, including a millimeter-wave radar array module, a data processing and analysis module, a baggage sorting control module, a quarantine process management module, an alarm response module, and a system management platform;

[0070] Millimeter-wave radar array modules are deployed at key nodes of the customs baggage conveyor belt to scan the live biological characteristics inside the baggage in real time.

[0071] The data processing and analysis module identifies the presence of live animals through millimeter-wave feature extraction algorithms and outputs the detection results to the system management platform;

[0072] The system management platform dynamically triggers the following collaborative processes based on the test results:

[0073] The baggage sorting control module adjusts the sorting path in real time, guiding baggage containing live animals to the quarantine channel;

[0074] The alarm response module generates tiered alarm signals and associates them with quarantine priorities.

[0075] The quarantine process management module automatically allocates quarantine resources and generates electronic quarantine task orders.

[0076] In this embodiment, the millimeter-wave radar array module employs multi-band fusion detection technology, specifically including:

[0077] Low-frequency bands are used to penetrate luggage materials to obtain the outline of a living person; the frequency of the low-frequency band is between 24-33 GHz.

[0078] High-frequency bands are used to capture micro-motion characteristics, with frequencies between 60-90 GHz.

[0079] A live biological feature model was established based on dual-band data fusion to distinguish live animals from static interference objects;

[0080] Furthermore, the millimeter-wave radar array module is a key component of the system for achieving liveness detection, and its performance directly affects the detection accuracy and efficiency. The following section elaborates on this module in terms of its function, composition, and technical principles:

[0081] I. Overall Function Overview:

[0082] The millimeter-wave radar array module is deployed at key nodes of the customs baggage conveyor belt. By emitting millimeter-wave signals and receiving reflected echoes, it enables non-contact scanning of live animals inside baggage. Its core function is to capture the spatial contours and micro-motion characteristics (such as minute displacements caused by breathing and heartbeat) of objects inside baggage in real time, and transmit the raw radar data to the data processing and analysis module. At the same time, the module has environmental adaptability and can dynamically adjust the detection parameters according to the baggage material and environmental interference to ensure stable acquisition of high-quality detection data in complex scenarios.

[0083] II. Submodule Composition and Functions:

[0084] (a) Multi-band radar transmitting and receiving unit:

[0085] Dual-band signal transmission: The device integrates millimeter-wave transmitters in both low-frequency (24–33 GHz) and high-frequency (60–90 GHz) bands. The low-frequency transmitter utilizes its long wavelength to penetrate common luggage materials such as fabric and plastic to obtain the outline and location information of objects. The high-frequency transmitter, with its high resolution advantage, captures Doppler frequency shift signals generated by the minute physiological movements of live animals (such as respiratory rate of about 0.5–2 Hz and heart rate of about 1–3 Hz).

[0086] Array antenna design: The phased array antenna adopts a linear or area array layout and achieves ±45° field of view coverage through electronic scanning technology to ensure that there are no blind spots in the detection when luggage moves on the conveyor belt; each antenna unit independently receives echo signals from different angles, providing multi-dimensional information for subsequent data fusion;

[0087] Echo signal acquisition: Equipped with a high-sensitivity receiving circuit, the reflected echo is amplified, mixed, and demodulated with low noise to convert the analog signal into a digital baseband signal; at the same time, pulse compression technology is used to improve distance resolution, which can distinguish different objects with a distance of less than 5cm.

[0088] (II) Data Preprocessing and Synchronization Unit:

[0089] Time synchronization module: Through the Global Positioning System (GPS) or a high-precision clock chip, nanosecond-level time synchronization of multi-band radar transmission and reception is achieved, ensuring the consistency of high and low frequency data in the time dimension and laying the foundation for subsequent dual-band data fusion;

[0090] Clutter suppression and filtering: Adaptive filtering algorithms (such as Kalman filtering) are used to process the raw echo data to remove environmental noise (such as electromagnetic interference and clutter generated by conveyor belt vibration), and motion compensation technology is used to eliminate Doppler frequency shift interference caused by luggage movement to extract pure target micro-motion signals;

[0091] Data formatting: The preprocessed radar data is converted into a range-Doppler map or tensor form, packaged according to time series, and transmitted to the data processing and analysis module. The data format includes parameters such as timestamps and amplitudes of low-frequency contour information and high-frequency micro-motion features.

[0092] (III) Environmental Adaptive Adjustment Unit:

[0093] Material recognition module: Based on the attenuation characteristics and reflection coefficient of radar echoes, a luggage material classification model is established; the spectral characteristics of echo signals are analyzed through machine learning algorithms (such as support vector machines) to identify materials such as metal, plastic, and fabric in real time, and automatically adjust radar transmission power and frequency band switching strategies; for example, when metal materials are detected, the high-frequency signal strength is reduced to avoid signal saturation, while the penetration capability of low-frequency signals is enhanced.

[0094] Interference detection and response: Continuously monitor the intensity of environmental electromagnetic interference. When interference signals (such as electromagnetic waves of the same frequency generated by nearby communication equipment) are detected, activate the frequency band switching mechanism to dynamically switch sub-frequency bands within the range of 24–33 GHz or 60–90 GHz to ensure the stability of the detected signal.

[0095] III. Key Technology Principles:

[0096] (I) Multi-band fusion detection principle:

[0097] Based on the physical characteristics of millimeter waves at different frequency bands, the low-frequency band (24–33 GHz) utilizes penetration to construct a three-dimensional contour model of the target, while the high-frequency band (60–90 GHz) captures the physiological movements of tiny animals through the Doppler effect. A data-level fusion strategy is adopted to fuse low-frequency contour data with high-frequency micro-motion features in the feature space, forming a multidimensional dataset containing location, shape, and life characteristics, which significantly improves the ability to distinguish between live animals and static objects (such as clothing and electronic products).

[0098] (II) Phased array antenna scanning principle:

[0099] By controlling the phase and amplitude of each element of the array antenna, rapid electronic scanning of millimeter-wave beams is achieved; by utilizing the principle of wave interference, beams pointing in different directions are synthesized in space, covering the entire area of ​​the conveyor belt without mechanical rotation; this technology enables the radar to have millisecond-level frame rate scanning capability, ensuring that fast-moving luggage can be completely detected.

[0100] (III) Environmental Adaptive Adjustment Principle:

[0101] By combining material identification algorithms and interference detection mechanisms, a closed-loop feedback adjustment system is constructed. When environmental changes are detected (such as changes in material or interference intensity), the radar transmission parameters (frequency band, power, pulse repetition frequency) are automatically adjusted, and the data preprocessing filter coefficients are updated to ensure the stability of radar performance and detection accuracy under different operating conditions.

[0102] In this embodiment, the data processing and analysis module includes:

[0103] The millimeter-wave signal preprocessing unit performs motion compensation and noise filtering on the radar echo;

[0104] The liveness feature recognition unit extracts biological micro-motion spectral features through an improved two-stream convolutional neural network model;

[0105] The dynamic threshold determination unit outputs a liveness confidence score based on an adaptive confidence calibration mechanism.

[0106] In the improved two-stream convolutional neural network model:

[0107] The probability of a living organism is obtained by weighting the outputs of each convolutional layer by applying the Sigmoid function.

[0108] The output of each convolutional layer is generated by processing low-frequency and high-frequency band data separately through convolution kernels, summing them, and then passing them through the ReLU activation function.

[0109] The dynamic frequency band weighting coefficients are determined based on the exponentially normalized values ​​of the signal-to-noise ratios (SNRs) of each layer, where the SNR weights are adjusted by the ambient noise energy value.

[0110] The environmental disturbance adjustment factor approaches 1 as the environmental noise energy increases, and the empirical constant for material attenuation ranges from 0.1 to 0.5.

[0111] Among them, the adaptive confidence calibration mechanism includes:

[0112] The final confidence score is a weighted sum of the probability of the presence of a living organism and the Shannon entropy of the micromotion feature;

[0113] The weighting coefficients are dynamically adjusted according to changes in ambient temperature; for every 1 degree Celsius increase in temperature, the probability value weight increases by 0.05.

[0114] The threshold for determining liveness is determined by adding a compensation amount to the square of the material density, which is based on a base threshold of 0.6. The material compensation coefficient is fixed at 0.02.

[0115] When the final confidence score exceeds the dynamic threshold, it is determined that the luggage contains a live animal;

[0116] Furthermore, the data processing and analysis module acts as the "brain" of the system, undertaking the crucial task of transforming the raw data collected by the millimeter-wave radar array module into accurate detection results. The following section provides a comprehensive and in-depth explanation of this module from multiple aspects, including its functions, structure, principles, and processes.

[0117] I. Overall Function Overview:

[0118] The data processing and analysis module is primarily responsible for receiving raw radar data transmitted from the millimeter-wave radar array module. After a series of complex data processing and intelligent analysis operations, it accurately determines whether there are live animals in the luggage and outputs the detection results to the system management platform in a timely manner. Its core functions include preprocessing millimeter-wave radar echo signals, extracting biological micro-movement characteristics using advanced deep learning algorithms, and providing a confidence score for the presence of live animals using an adaptive threshold mechanism. In addition, this module can continuously improve the accuracy and stability of detection by continuously optimizing algorithms and calibration parameters, providing solid data support and reliable decision-making basis for customs to carry out efficient customs clearance management.

[0119] II. Submodule Composition and Functions:

[0120] (a) Millimeter-wave signal preprocessing unit:

[0121] Motion Compensation Module: When luggage moves on the conveyor belt, it generates Doppler frequency shift interference, affecting the accuracy of detection. The motion compensation module constructs an accurate motion model by deeply analyzing the speed and direction of the luggage. Based on this model, the time axis of the original radar echo signal is precisely translated and stretched, effectively eliminating false Doppler components caused by the motion of objects, making micro-motion features such as breathing and heartbeat clearer and more prominent, laying a good foundation for subsequent feature extraction and analysis.

[0122] Noise filtering module: To remove noise from the signal, this module combines median filtering and wavelet transform techniques. Median filtering can effectively remove salt-and-pepper noise while protecting the edge information of the signal and avoiding edge blurring. Wavelet transform decomposes the signal into different frequency scales and, by setting appropriate thresholds, effectively removes high-frequency environmental noise, such as electromagnetic interference and noise generated by conveyor belt mechanical vibration, while fully preserving the 0.1-5Hz characteristic frequency band where the micro-motion signal is located, ensuring the purity of the input data.

[0123] Data normalization module: Since the original radar data has differences in amplitude and frequency range, it will affect the performance of subsequent feature extraction algorithms. The data normalization module normalizes the preprocessed radar data, mapping data of different scales to a unified range, such as [0,1]. This not only eliminates the adverse effects of data scale differences, but also significantly improves the convergence speed and stability of the algorithm, making subsequent feature extraction and analysis more efficient and accurate.

[0124] (ii) Liveness detection unit:

[0125] Improved dual-stream convolutional neural network model: This network uses a dual-input structure to process low-frequency and high-frequency data from millimeter-wave radar, and achieves accurate identification of live animals by fusing multi-band features through dynamic weights.

[0126] Additive frequency band weighting fusion: (in, For the first The weighting coefficients for fusing low-frequency and high-frequency features in convolutional layers. ; For the first The signal-to-noise ratio of a layer feature is calculated by the ratio of signal power to noise power; As an environmental disturbance modulator, ; This is an empirical constant for material attenuation, with a value range of [value range missing]. The unit is c / g is used to quantify the attenuation characteristics of different materials for millimeter waves; the environmental noise energy value is the energy integral of the environmental noise spectrum collected in real time, calculated by Fourier transform.

[0127] Forward propagation computation: (in, The output value represents the probability of a living organism, and its range is [value missing]. ; The Sigmoid activation function maps the linear output to a probability distribution. The dynamic frequency band weighting coefficient is calculated using the formula above. , These are the low-frequency and high-frequency band feature convolution kernel parameter matrices, with dimensions of [missing information]. , , These represent the number of input and output channels, respectively. This is a convolution operator, using a two-dimensional convolution operation with a stride of 1; , These are the preprocessed low-frequency and high-frequency radar data tensors, respectively, with dimensions of... , , The feature map height and width; For the first Layer bias terms are used to adjust the threshold of the activation function; ReLU is the modified linear unit activation function, expressed as follows: , used to introduce nonlinear characteristics; The total number of convolutional layers in the network is [number] in this model. ;

[0128] Network structure optimization:

[0129] Batch normalization layer is used: (in, Input data; , These are the mean and variance of the mini-batch data, respectively. To prevent constants with a denominator of zero, the value is taken as follows: ; , These are learnable scaling and translation parameters; (The data is after normalization).

[0130] Dropout layer uses probability Randomly discard neurons to prevent overfitting;

[0131] Transfer learning optimization:

[0132] Based on a pre-trained 3DCNN model (such as ResNet3D), the parameters are fine-tuned using the following formula:

[0133] (in, These are the fine-tuned model parameters; These are the parameters for the pre-trained model; The gradient is calculated based on the customs dataset; The learning rate is initially set to 0.001 and uses an exponential decay strategy, decreasing to 0.9 times its original value every 10 training epochs.

[0134] (III) Dynamic Threshold Determination Unit:

[0135] Adaptive confidence calibration mechanism:

[0136] Confidence score calculation: using the formula (in, The final confidence score has a range of values. The higher the value, the stronger the reliability of detecting live animals; The probability-feature entropy balance coefficient. ; This indicates the ambient temperature value, in °C. It is the probability of a living organism output by the two-stream network; For micro-motion feature vectors, ; It is a micro-motion feature vector The Shannon entropy is calculated using the following formula: It is used to measure the uncertainty of micro-motion characteristics; The final confidence score is calculated by comprehensively considering the probability of energy distribution in the characteristic frequency domain. In this way, the model output probability and feature stability are balanced, effectively avoiding misjudgment that may occur if a single indicator is used for judgment.

[0137] Dynamic threshold setting: via formula (in, This is the threshold for determining liveness; The base threshold is set at 0.6. This is the material compensation coefficient, with a value of 0.02. This is the density of the luggage material obtained from radar inversion, in units of... Calculate the liveness detection threshold This formula allows the judgment threshold to be dynamically adjusted according to the different materials of the luggage, thus better adapting to various complex detection scenarios.

[0138] Results Output and Feedback: When the calculated confidence score is obtained... When the system determines that the luggage contains live animals, it promptly sends detailed detection results, including confidence scores and location information of suspected live animals, to the system management platform to trigger subsequent luggage sorting, alarm responses, and other operations. At the same time, the unit also compares and analyzes historical detection data with actual quarantine results, feeding the comparison results back to the algorithm to continuously optimize algorithm parameters and improve the accuracy and reliability of detection.

[0139] (I) The principle of fusion in two-stream convolutional neural networks:

[0140] Based on the multi-band characteristics of millimeter-wave radar, the two branches of a dual-stream convolutional neural network extract spatial features such as the contour and position of objects from low-frequency data, and capture micro-motion spectral features from high-frequency data; through dynamic weights... It achieves adaptive fusion at the feature level; this innovative design breaks the limitations of traditional single-stream networks, fully combines the advantages of data from different frequency bands, and greatly improves the accuracy of identifying live animals, enabling more precise detection of live animals from complex radar data;

[0141] (II) Adaptive Confidence Calibration Principle:

[0142] Considering that factors such as ambient temperature and luggage material can affect the test results, this module introduces a temperature adjustment coefficient. and material compensation threshold This enables dynamic adjustment of confidence scores and judgment criteria; simultaneously, it utilizes feature entropy. Quantifying the stability of micro-motion features effectively avoids the model from over-relying on low-quality data and making incorrect judgments, ensuring that the detection results have high reliability and accuracy in various complex environments;

[0143] (III) Principles of Real-Time Online Learning:

[0144] This module establishes an efficient feedback closed-loop mechanism, using the actual quarantine results returned by the system management platform, such as live or non-live tags confirmed by manual re-inspection, as a monitoring signal. Based on this feedback information, the weight parameters and confidence calibration coefficients of the dual-flow network are updated online. Through this real-time online learning method, the algorithm can automatically adapt to the differences in the detection environment and baggage types at different ports, continuously improve its detection performance, and always maintain its optimal working state.

[0145] IV. Module Workflow:

[0146] (a) Data Reception and Preprocessing Stage:

[0147] The data processing and analysis module receives raw detection data transmitted from the millimeter-wave radar array module. This data includes radar echo signals in both low and high frequency bands, which form the basis for subsequent analysis.

[0148] After the data enters the millimeter-wave signal preprocessing unit, motion compensation, noise filtering, and data normalization are performed sequentially. Through this series of preprocessing steps, interference and noise in the data are removed, the data scale is unified, and standardized data is generated to prepare for subsequent liveness feature extraction.

[0149] (II) Liveness Feature Extraction Stage:

[0150] The preprocessed standardized data is input into the low-frequency and high-frequency branches of the improved dual-stream convolutional neural network. Inside the network, spatial features and micro-motion features are extracted from the data through multiple convolution and pooling operations.

[0151] Using dynamic frequency band weighting coefficients The extracted high- and low-frequency features are fused to enable the network to fully utilize information from different frequency bands; finally, the probability of liveness is output through the sigmoid activation function. This completes the extraction and preliminary assessment of liveness features;

[0152] (III) Confidence Assessment and Decision-Making Stage:

[0153] The dynamic threshold determination unit determines the threshold based on the current ambient temperature. and luggage material density Calculate the probability-feature entropy balance coefficient and liveness detection threshold The calculation of these parameters fully considers the influence of the actual testing environment, ensuring the rationality of the judgment criteria.

[0154] The probability of combining the output of the two-stream network Entropy of micro-motion features Calculate the final confidence score. ,Will With the judgment threshold Compare the results and output a clear determination of whether there are live animals in the luggage;

[0155] (iv) Results Output and Feedback Stage:

[0156] Once the detection results are confirmed, the module will send detailed information, including confidence score and suspected live animal location, to the system management platform. Based on this information, the system management platform will promptly trigger subsequent processes such as baggage sorting and alarm response to achieve efficient handling of baggage containing live animals.

[0157] The module collects actual quarantine results and compares and analyzes them with historical testing data. Based on the analysis results, the parameters and confidence calibration mechanism of the dual-flow network are iteratively optimized to continuously improve the detection performance and accuracy of the module, making it better suited to actual application needs.

[0158] (V) Conclusion:

[0159] When the system stops running or receives a stop command, the data processing and analysis module will save important information such as the current model parameters, calibration coefficients, and historical detection records, then release the occupied computing resources, enter standby mode, and wait for the next task to start.

[0160] In this embodiment, the baggage sorting control module performs dynamic path optimization, including:

[0161] The sorting priority index is calculated by multiplying the confidence score by a negative exponential decay function of the quarantine channel load;

[0162] When the real-time load of the channel reaches the baseline load value, the priority index decays to 1 / e of the original value;

[0163] By using radio frequency identification (RFID) technology to link baggage with inspection data, the sorting machine is controlled to transfer high-priority baggage to the quarantine area.

[0164] Furthermore, the baggage sorting control module, as the core execution unit for the system to achieve intelligent customs clearance management, undertakes the crucial task of converting inspection results into precise sorting actions. The following will elaborate on the technical solution in the claims from the dimensions of functional architecture, core algorithms, and workflow:

[0165] I. Overall Function Overview:

[0166] The baggage sorting control module receives the live animal detection results (confidence score) output by the data processing and analysis module. By combining the real-time load status of the quarantine channel, dynamic path optimization is implemented for baggage on the conveyor belt. Its core functions include: calculating sorting priority based on confidence level and channel load, binding baggage and inspection data through radio frequency identification (RFID) technology, accurately controlling the sorting machine to guide baggage containing live animals to the quarantine channel, and simultaneously updating the baggage status information on the system management platform to ensure the efficient and orderly operation of the customs clearance process.

[0167] II. Submodule Composition and Functions:

[0168] (a) Priority Calculation Unit:

[0169] Data access and integration: Real-time acquisition of baggage liveness detection confidence scores output by the data processing and analysis module. (range of values) ), and the real-time load of each quarantine channel as reported by the quarantine process management module. (Unit: pieces, representing the number of bags awaiting inspection at the current channel), Channel baseline load value (Preset normal channel capacity);

[0170] Priority algorithm execution: using the formula priority index Calculate the sorting priority for each piece of luggage; this formula uses an exponential function to adjust the channel load. Attenuation processing is applied to ensure that high-confidence baggage (those close to the presence of live animals) or baggage in low-load aisles receive higher priority; for example, when a baggage... The passage , When the priority index is ;

[0171] Priority sorting and output: The calculation results are sorted in descending order to generate a real-time sorting queue, and the priority information is bound to the baggage ID (obtained via RFID) and transmitted to the sorting execution unit;

[0172] (ii) RFID Data Management Unit:

[0173] Tag Reading and Writing Control: Before baggage enters the inspection area, a unique electronic tag is assigned to each piece of baggage using a fixed RFID reader, and basic baggage information (such as flight number and passenger ID) is written into it; after inspection, additional inspection data (confidence score) is written in. (Detection time);

[0174] Data association and tracking: Establish a mapping relationship between baggage ID and detection results and sorting path, and realize dynamic tracking of baggage from detection to sorting by updating tag data in real time; for example, when baggage needs to change its path due to priority adjustment, the system automatically updates the target channel information in the tag;

[0175] Collision avoidance: The Time Division Multiple Access (TDMA) algorithm is used to solve the signal collision problem when multiple tags are read and written at the same time, ensuring the accuracy of data reading and writing in the scenario of dense baggage transportation;

[0176] (III) Sorting Execution Unit:

[0177] Path planning engine: Based on the priority queue and the real-time status of the quarantine channel, the optimal sorting path is generated using Dijkstra's shortest path algorithm; for example, when multiple quarantine channels are available, the channel closest to the current baggage location is selected first.

[0178] Equipment collaborative control: Control commands are sent to sorting machines (such as cross belt sorting machines and swing arm sorting machines) via industrial Ethernet (such as Profinet protocol) to precisely adjust the sorting machine's motion parameters (such as sorting angle and trigger time); at the same time, the conveyor belt speed controller is linked to ensure that the luggage enters the target channel smoothly.

[0179] Anomaly tolerance mechanism: When a mechanical failure or path blockage is detected in the sorting machine, an emergency plan is automatically triggered: the affected baggage is temporarily transferred to a backup buffer zone, and maintenance personnel are notified through the alarm response module, while the fault log of the system management platform is updated.

[0180] III. Key Technology Principles:

[0181] (I) Principle of Dynamic Priority Calculation:

[0182] Based on Bayesian decision theory, confidence scores will be used. As a probability basis for the existence of living organisms, combined with channel load The impact on sorting timeliness is assessed by constructing a priority model using an exponential decay function. This model can dynamically balance quarantine urgency with channel resource utilization. For example, when there is high confidence (close to the presence of live animals) but the channel is busy, the priority can be appropriately reduced to avoid excessive channel congestion.

[0183] (II) RFID end-to-end tracking principle:

[0184] By leveraging the contactless identification capabilities of RFID, a unique binding between baggage and inspection data is achieved; through real-time updates of tag data, a full-process electronic traceability chain is constructed from inspection, sorting to quarantine, ensuring data traceability and operational auditability, and meeting customs supervision and compliance requirements;

[0185] (III) Sorting Path Optimization Principles:

[0186] Heuristic search algorithms (such as Dijkstra's algorithm) are combined with real-time path status (channel occupancy, equipment operation status) to dynamically plan the shortest sorting path; by introducing path weight coefficients (such as distance weight and time weight), the transmission efficiency of luggage in complex conveyor network is optimized and sorting delay is reduced.

[0187] IV. Module Workflow:

[0188] (a) Initialization phase:

[0189] Start the RFID reader, sorter and other hardware equipment, and complete the equipment self-test and parameter calibration (such as sorter angle calibration and conveyor belt speed calibration).

[0190] Load preset configuration parameters, including the baseline load value of the quarantine channel. Set threshold values ​​for sorting machine action parameters (such as maximum sorting angle and minimum trigger interval) and establish a communication connection with the system management platform.

[0191] (II) Data Acquisition and Calculation Stage:

[0192] Real-time reception of detection results (confidence score) from the data processing and analysis module. Channel load data of the quarantine process management module ( , );

[0193] Priority calculation unit according to the formula priority index Calculate the sorting priority of each piece of luggage and generate a sorting queue;

[0194] (III) Path Planning and Execution Phase:

[0195] The path planning engine calculates the optimal sorting path based on the priority queue and channel status, and generates control instructions that include the target channel ID and the sorting trigger time.

[0196] The sorting execution unit obtains the baggage electronic tag information through the RFID data management unit. After verifying the data integrity, it sends the control command to the sorting machine and drives the baggage to the corresponding quarantine channel.

[0197] (iv) Status Feedback and Update Phase:

[0198] After sorting is completed, the arrival of the luggage is confirmed by the RFID reader installed at the entrance of the passage, and the actual sorting result (success / failure) is fed back to the system management platform;

[0199] Update the status of the baggage electronic tag (e.g., mark it as "sorted"), and record the sorting time and route information in the system database for subsequent traceability and statistical analysis;

[0200] (V) Conclusion:

[0201] When the system stops running or receives a stop command, the baggage sorting control module stops data processing and equipment control operations, saves the current priority queue, sorting logs and other data, turns off the RFID reader and sorting machine power, and releases system resources.

[0202] In this embodiment, the quarantine process management module includes:

[0203] The quarantine plan generation unit matches a pre-set treatment plan library based on the type of living organism.

[0204] The resource scheduling unit allocates quarantine resources based on the spatial location coordinates of baggage and confidence scores;

[0205] The electronic customs clearance document generation unit automatically generates an electronic release instruction and updates the customs clearance system after quarantine is completed.

[0206] Furthermore, the quarantine process management module is the core hub of the customs clearance management system, ensuring the scientific and efficient operation of quarantine work. Driven by test results, it achieves intelligent allocation of quarantine resources and closed-loop management of the process. The following will elaborate on its functional positioning, module composition, technical principles, and workflow:

[0207] I. Overall Function Overview:

[0208] The quarantine process management module uses the baggage liveness detection results (confidence score) output by the data processing and analysis module. The system automatically matches quarantine plans, dynamically schedules resources, and generates electronic customs clearance documents based on the information of live animals and their locations. Its core functions include: intelligently matching quarantine and disposal plans according to the type of live animal; rationally allocating quarantine personnel, equipment, and other resources based on optimization algorithms; and automatically generating electronic release instructions and updating the customs clearance system after quarantine is completed. This ensures that the entire quarantine process is traceable and highly efficient, providing a solid guarantee for customs to prevent and control biosafety risks.

[0209] II. Submodule Composition and Functions:

[0210] (a) Quarantine Plan Generation Unit:

[0211] Live animal identification and classification: Receives suspected live animal characteristic information from the data processing and analysis module, combines it with historical quarantine data and species database (including physiological characteristics, risk levels, etc. of common smuggled live animals), and uses pattern recognition algorithms to preliminarily determine the live animal type (such as mammals, reptiles, birds, etc.) and risk level (high risk, medium risk, low risk).

[0212] Intelligent matching of quarantine protocols: Establish a quarantine protocol database covering standardized procedures for different types of live organisms and risk levels (such as isolation and observation, sampling and testing, disinfection, etc.); based on the identification results, automatically retrieve and match the optimal quarantine protocol, and generate an electronic task sheet containing operation steps and a list of required resources; for example, if a high-risk alien species is detected, automatically match a strict isolation and quarantine protocol with a professional laboratory testing protocol.

[0213] Dynamic adjustment mechanism for the plan: Supports manual intervention and plan update functions; quarantine personnel can manually adjust the quarantine plan according to the actual situation (such as abnormal health status of live animals); at the same time, the system regularly optimizes the plan library content based on the latest quarantine standards and case data to ensure the scientific and compliant nature of the quarantine process;

[0214] (ii) Resource Scheduling Unit:

[0215] Real-time monitoring of resource status: Connect to various resource management systems at the quarantine site to obtain real-time resource data such as the work status of quarantine personnel (busy, idle, on standby), equipment availability (testing instruments, sampling tools, disinfection equipment, etc.), and site occupancy information (quarantine room, isolation area), and build a dynamic resource pool;

[0216] Optimize scheduling algorithm execution: Use formulas to calculate scheduling costs. (in, The number of bags awaiting quarantine; For the first Confidence rating for each piece of luggage; From the current location of quarantine personnel to their luggage The distance is expressed in meters (m). The algorithm calculates the optimal scheduling scheme based on the historical average quarantine time for similar cases (in minutes). This algorithm comprehensively considers the baggage risk level (confidence score). Resource accessibility (distance) ) and processing efficiency (time consumption) Prioritize assigning high-confidence baggage to nearby, experienced quarantine personnel to maximize resource utilization efficiency;

[0217] Dispatch instruction issuance and feedback: The generated resource dispatch instructions (such as specifying quarantine personnel to go to a certain baggage location or allocating specific equipment to the quarantine area) are sent to the relevant execution terminals (such as quarantine personnel handheld terminals and equipment management systems), and the execution status of the instructions is monitored in real time; if resource conflicts occur (such as equipment failure or personnel emergencies), the re-dispatch mechanism is automatically triggered to ensure uninterrupted quarantine work;

[0218] (III) Electronic Customs Clearance Form Generation Unit:

[0219] Quarantine result data integration: Receive quarantine process records (sampling data, test reports, handling measures, etc.), on-site monitoring video clips, and relevant data from other modules of the system (such as test time and sorting path) uploaded by quarantine personnel through handheld terminals to form a complete quarantine file;

[0220] Automated customs clearance decision-making: Based on a pre-set customs clearance rule base (including laws and regulations, health standards, risk thresholds, etc.), the quarantine results are automatically reviewed; if the quarantine is qualified, an electronic customs clearance form is automatically generated, which includes information such as release instructions, quarantine conclusions, and validity period; if the quarantine is unqualified, a return or destruction notice is generated, and the baggage status is marked as "pending processing" simultaneously.

[0221] Data synchronization and system integration: Electronic customs clearance data is pushed to the customs clearance system in real time through the application programming interface (API) to update the baggage clearance status; at the same time, quarantine files are encrypted and stored on the blockchain to ensure that the data is tamper-proof and traceable throughout the process, meeting the requirements of customs supervision and auditing.

[0222] III. Key Technology Principles:

[0223] (I) Principle of Intelligent Matching of Quarantine Plans:

[0224] A knowledge graph-based network is constructed to link live animals with quarantine protocols, structurally associating knowledge nodes such as species characteristics, risk levels, and quarantine standards. Natural language processing (NLP) technology is used to analyze the feature descriptions in the test results, and matching paths are quickly retrieved in the knowledge graph to achieve accurate recommendations for quarantine protocols, thereby improving the accuracy and efficiency of protocol matching.

[0225] (II) Principles of Resource Optimization Scheduling:

[0226] A multi-objective optimization theory is adopted to construct a mathematical model with the goal of minimizing scheduling costs (taking into account both the timeliness of risk handling and resource consumption). The model is solved by heuristic algorithms (such as genetic algorithms and simulated annealing algorithms) to quickly search for the global optimal solution in complex resource combination and task allocation scenarios, ensuring that quarantine resources are always efficiently configured in a dynamically changing environment.

[0227] (III) Principles of Electronic Customs Clearance Automation:

[0228] By using smart contract technology, customs clearance rules are transformed into automatically executable code logic. When the quarantine result data meets the preset conditions, the smart contract automatically triggers the generation of electronic customs clearance documents and the system data update process, realizing unmanned operation from quarantine to customs clearance and reducing errors and delays caused by human intervention.

[0229] IV. Module Workflow:

[0230] (a) Initialization phase:

[0231] After the quarantine process management module is started, it loads the latest basic data such as the quarantine scheme library, customs clearance rule library, and species database, and completes the system parameter configuration (such as risk level threshold and scheduling algorithm weight coefficient).

[0232] Establish communication connections with other modules of the system (data processing and analysis module, baggage sorting control module, alarm response module) to obtain real-time information on baggage awaiting quarantine and system status data;

[0233] (II) Task Reception and Analysis Phase:

[0234] Receive baggage information containing live animals from the baggage sorting control module, including baggage ID and location coordinates. Confidence score and preliminary results of the live organism type determination;

[0235] The quarantine plan generation unit quickly matches quarantine plans based on the received information and generates electronic task orders; the resource scheduling unit simultaneously analyzes the number, distribution and resource status of luggage to be quarantined and initiates resource scheduling calculations.

[0236] (III) Resource Allocation and Quarantine Implementation Phase:

[0237] The resource scheduling unit issues task instructions to quarantine personnel and equipment management system based on the scheduling scheme determined by the optimization algorithm; the quarantine personnel receive the tasks through handheld terminals and go to the designated location to perform quarantine operations.

[0238] During the quarantine process, quarantine personnel upload sampling data, test results, and other information to the system in real time; the resource scheduling unit continuously monitors resource usage and quarantine progress, and dynamically adjusts scheduling strategies (such as replenishing consumed test reagents and allocating reinforcements).

[0239] (iv) Customs clearance form generation and feedback stage:

[0240] The electronic customs clearance document generation unit collects complete quarantine result data, performs automated review according to customs clearance rules, generates electronic customs clearance documents or processing notices, and pushes them to the customs clearance system and relevant personnel terminals.

[0241] The entire quarantine process data (including plan implementation records, resource usage details, and customs clearance results) is encrypted and stored on the blockchain. At the same time, the baggage status dashboard on the system management platform is updated to show the customs clearance progress to customs management personnel.

[0242] (V) Conclusion:

[0243] When all luggage awaiting quarantine has been processed or the system receives a stop command, the quarantine process management module stops task scheduling and data processing, backs up quarantine files and system logs, releases the occupied computing resources, and enters standby mode.

[0244] In this embodiment, the alarm response module adopts a three-level collaborative handling mechanism, including:

[0245] Level 1 alarm: When the confidence score is >0.9, an audible and visual alarm is triggered and the conveyor belt is frozen;

[0246] Level 2 alarm: When the confidence score is 0.7 ≤ confidence score ≤ 0.9, a "to be reviewed" label is generated and a manual inspection is notified;

[0247] Level 3 alarm: When the confidence score is 0.6 ≤ confidence score < 0.7, only a low-risk alert is sent to the quarantine management module;

[0248] Alarm signals are pushed to customs personnel's mobile terminals in real time;

[0249] Furthermore, the alarm response module acts as a security sentinel in the customs clearance management system, bearing the crucial responsibility of real-time early warning and rapid response to abnormal situations. This module triggers a tiered alarm mechanism through multi-dimensional risk assessment, combined with intelligent scheduling and visualized command, constructing an efficient and precise emergency response system to ensure that customs personnel can respond promptly to various emergencies. The following details its functional architecture, technical implementation, and workflow:

[0250] I. Overall Function Overview:

[0251] The alarm response module is based on the detection results (confidence score) of the data processing and analysis module. The system monitors system operation status data and real-time abnormal situations during customs clearance. Its core functions include: determining alarm levels through risk threshold algorithms, triggering audible and visual alarms and multi-channel notifications (SMS, email, APP push), linking with on-site monitoring systems to lock target locations, generating task work orders containing handling plans, and recording alarm process data for post-event analysis and system optimization.

[0252] II. Submodule Composition and Functions:

[0253] (a) Risk Assessment Unit:

[0254] Multi-dimensional threshold detection:

[0255] Risks of liveness detection: Based on confidence scores Set three threshold levels ( , , ),when When an alert is triggered, The alarm level is upgraded to medium level. Advanced alarms will be activated at any time;

[0256] Equipment anomaly monitoring: Real-time collection of operating parameters (such as temperature, current, vibration frequency) of equipment such as sorting machines and RFID readers, and establishment of equipment health models through machine learning algorithms. When the parameters deviate from the normal range and exceed the preset threshold (such as temperature exceeding 75℃, vibration frequency fluctuation exceeding ±15%), it is determined that the equipment has potential failure risk.

[0257] System operating status: Monitor system throughput, response time, data transmission success rate and other indicators. When the data packet loss rate exceeds 3% or the processing delay exceeds 5 seconds, trigger a system performance abnormality alarm.

[0258] Comprehensive risk level calculation: using the formula risk index (in, The severity of the equipment malfunction is scored, with a value ranging from 0 to 10. The system's operational status is scored, with a value ranging from 0 to 10. , , These are the weighting coefficients, and The risk index is calculated using a comprehensive approach, with default values ​​of 0.6, 0.25, and 0.15 respectively. This formula ensures that the true risk level is accurately reflected in different scenarios through dynamic weight allocation.

[0259] (ii) Alarm triggering unit:

[0260] Tiered alarm mechanism:

[0261] Low-level alarm (risk index < 6): Triggers a yellow warning signal, and the alarm information is only displayed on the system management platform to notify on-duty personnel to pay attention;

[0262] Intermediate alarm (6≤risk index<8): Activate the audible and visual alarm device (such as flashing warning lights and sounding buzzer), and simultaneously send a text message and APP push to designated personnel, along with brief alarm details (such as suspected location of a living person and risk level).

[0263] Advanced Alarm (Risk Index ≥ 8): Based on the intermediate alarm, it automatically dials the person in charge and sends an email containing the handling plan, and links the on-site monitoring system to focus the camera on the alarm location and start recording.

[0264] Alarm filtering and suppression: A time window mechanism (e.g., within 10 minutes) is used to merge repeated alarms of the same location and type to avoid wasting resources; for known temporary anomalies (e.g., parameter fluctuations during equipment maintenance), a preset whitelist is used for filtering to reduce false alarm interference;

[0265] (III) Emergency Response Unit:

[0266] Emergency response plan generation: Establish an emergency response knowledge base and pre-set standardized response procedures for different types of alarms (such as live animal alarms and equipment failure alarms); when an alarm occurs, automatically match the corresponding response plan and generate a task work order that includes operation steps, responsible personnel, and required resources; for example, for high-risk live animal alarms, automatically generate a three-level response procedure of "immediately isolate the luggage - notify quarantine experts - activate biosafety protection measures";

[0267] Resource scheduling and coordination: It interacts in real time with the quarantine process management module and the baggage sorting control module, and adjusts resource allocation synchronously when an alarm is triggered; for example, when a live animal alarm is detected, quarantine personnel and testing equipment are prioritized, and the baggage sorting process in the relevant area is suspended to ensure that the handling work is not disturbed.

[0268] Process tracking: A unique identifier is generated for each alarm task using QR code technology. When the personnel perform each step of the operation, they scan the code to confirm. The system updates the task status in real time (such as received, being processed, or completed) and records the operation time and the personnel who performed the operation, forming a complete closed loop for the handling process.

[0269] III. Key Technology Principles:

[0270] (a) Intelligent threshold adaptive mechanism:

[0271] Based on historical alarm data and response results, reinforcement learning algorithms are used to dynamically adjust risk thresholds (such as...). , , When the system detects that the false alarm rate of a certain type of alarm is too high, it automatically raises the corresponding threshold; if the missed alarms occur frequently, it appropriately lowers the threshold and continuously learns to optimize the accuracy of alarm triggering.

[0272] (II) Multimodal alarm fusion technology:

[0273] The DS evidence theory is used to fuse alarm information from different sources (such as liveness detection, equipment monitoring, and system logs) for decision-making. By calculating the confidence and likelihood of each piece of evidence, the system comprehensively judges whether to trigger an alarm and determines the final risk level, effectively reducing the possibility of misjudgment from a single data source.

[0274] (III) Emergency Response Knowledge Graph:

[0275] A knowledge graph centered on "alarm type - handling plan - resource requirements" is constructed to link and store structured knowledge such as customs emergency handling specifications, historical cases, and expert experience. When an alarm occurs, the optimal handling path is quickly inferred through a graph neural network, and a handling plan that complies with the latest regulatory requirements is dynamically generated.

[0276] IV. Module Workflow:

[0277] (a) Initialization phase:

[0278] After the alarm response module is activated, it loads the preset risk threshold parameters (such as...). , , (Equipment parameters within normal range), alarm classification rules, and emergency response knowledge base; complete system configuration initialization.

[0279] Establish real-time data channels with the data processing and analysis module, baggage sorting control module, and quarantine process management module, and subscribe to key monitoring indicators (such as confidence scores and equipment operating status).

[0280] (II) Monitoring and Evaluation Phase:

[0281] The risk assessment unit continuously receives data transmitted from each module and performs real-time analysis of the liveness detection results, equipment operating parameters, and system performance indicators to calculate the risk index.

[0282] When the risk index exceeds the preset threshold, an alarm event is generated, and key information such as alarm time, location, type, and risk level is recorded. The event is then pushed to the alarm triggering unit.

[0283] (III) Alarm and Response Phase:

[0284] The alarm triggering unit activates the corresponding alarm mechanism (audible and visual alarm, multi-channel notification) according to the risk level, and sends the alarm details and emergency response plan to the designated personnel's terminal at the same time;

[0285] The emergency response unit generates task orders, dispatches relevant resources (such as quarantine personnel and testing equipment) to execute the response plan, and tracks the response progress in real time through QR code scanning; if the risk level changes during the response process, the alarm level and response measures are automatically adjusted.

[0286] (iv) Feedback and Archiving Stage:

[0287] Once the disposal is completed, the system automatically collects the disposal results (such as the live animals have been properly handled and the equipment malfunction has been repaired), updates the alarm status to "resolved", and stores the disposal process data (including operation records, on-site photos, and video clips) in the blockchain for evidence storage.

[0288] Regularly perform statistical analysis on alarm data to evaluate the rationality of alarm thresholds and the effectiveness of handling plans, providing a basis for system parameter optimization and knowledge base updates;

[0289] (V) Conclusion:

[0290] When the system stops running or receives a stop command, the alarm response module stops data acquisition and analysis, saves the current alarm status and system configuration, shuts down the audible and visual alarm devices and communication channels, and releases system resources.

[0291] In this embodiment, the system management platform implements closed-loop control throughout the entire process, including:

[0292] Millimeter-wave detection data is integrated with the customs clearance system via an application programming interface (API).

[0293] An electronic traceability chain is constructed from detection to release, with data at each stage stored using blockchain technology;

[0294] A heatmap of customs clearance efficiency is generated based on confidence scores and quarantine time to visually display process bottlenecks.

[0295] Furthermore, the system management platform, as the "central nervous system" of the millimeter-wave radar-based intelligent detection and customs clearance management system for live animals, undertakes the core functions of cross-module collaboration, full-process control, and data-driven intelligent decision-making. It integrates data and operations from various functional modules to achieve closed-loop management from detection to release, while leveraging information and intelligent technologies to improve the transparency and efficiency of customs supervision. The following will elaborate on its functional architecture, core modules, technical principles, and workflow:

[0296] I. Overall Function Overview:

[0297] The system management platform is data-driven, connecting millimeter-wave radar array modules, data processing and analysis modules, baggage sorting control modules, quarantine process management modules, and alarm response modules to achieve information sharing and collaborative operation among multiple modules. Its core functions include: real-time monitoring of system operation status and dynamic scheduling of tasks by each module; building an electronic traceability chain to ensure data traceability throughout the customs clearance process; generating visualized analysis reports based on test results and quarantine time to provide decision support for customs management; and ensuring system data security and legal compliance through access control and security protection mechanisms.

[0298] II. Submodule Composition and Functions:

[0299] (a) Data Integration and Interaction Module:

[0300] Multi-source data access: Communication is established with various functional modules through standardized interfaces (such as RESTful API, WebSocket) to receive raw data collected by millimeter-wave radar and detection results (confidence scores) output by the data processing module in real time. Multi-source heterogeneous data, including baggage sorting status, quarantine process records, and alarm information;

[0301] Data cleaning and storage: The received data is format-converted, missing values ​​are filled, and outliers are filtered. Structured data (such as test results and equipment status) is stored in a relational database (such as MySQL), and unstructured data (such as radar images and surveillance videos) is stored in a distributed file system (such as Ceph). Indexes are also created to support fast queries.

[0302] Data sharing and distribution: Based on the needs of each module, the integrated data is pushed to the corresponding module in real time; for example, the confidence score and location information of luggage containing live animals are sent to the baggage sorting control module, and the quarantine task order and related resource requirements are transmitted to the quarantine process management module.

[0303] (II) Process Collaboration Control Module:

[0304] Task scheduling engine: Dynamically triggers collaborative operations of various modules based on detection results; when the data processing and analysis module determines that the luggage contains live animals, it sequentially calls the luggage sorting control module to adjust the sorting path, the alarm response module to trigger tiered alarms, and the quarantine process management module to allocate quarantine resources, forming an automated workflow of "detection-sorting-alarm-quarantine";

[0305] Conflict detection and resolution: Real-time monitoring of the task execution status of each module, and by establishing a task dependency graph (e.g., quarantine tasks need to be started after baggage sorting is completed), resource conflicts are avoided when multiple tasks are executed concurrently; if a conflict occurs (e.g., multiple quarantine tasks compete for the same equipment), the task priority is automatically adjusted or resources are reallocated.

[0306] Anomaly Handling Mechanism: When a module malfunctions (such as a sorting machine jam or a quarantine equipment failure), the relevant tasks are immediately suspended, an anomaly message is sent to the alarm response module, and an emergency plan is activated (such as activating a backup sorting path or deploying backup equipment) to ensure that the overall operation of the system is not seriously affected.

[0307] (III) Electronic Traceability and Supervision Module:

[0308] Blockchain Evidence Storage System: Utilizing consortium blockchain technology, the entire process of data from millimeter-wave radar detection to final release (detection records, sorting logs, quarantine reports, customs clearance documents, etc.) is encrypted and stored on the blockchain; each data node contains information such as timestamp, operator, and data hash value, ensuring that the data is tamper-proof and traceable;

[0309] Traceability service: Provides customs supervisors with a visual traceability interface, supporting quick query of the customs clearance history of specific baggage by baggage ID, time range, operator, etc.; for example, it can trace the inspection time, sorting path, quarantine results and handling personnel of a piece of baggage, providing a basis for law enforcement audit and accountability.

[0310] Compliance check: Built-in customs regulations and business rules library (such as the "Law of the People's Republic of China on Entry and Exit Animal and Plant Quarantine"), automatically compares customs clearance process data with regulatory requirements, and if any violations are found (such as releasing without quarantine as required), an early warning is immediately triggered and a rectification work order is generated;

[0311] (iv) Data Analysis and Decision-Making Module:

[0312] Visualized monitoring dashboard: Utilizing visualization tools such as ECharts and D3.js, it displays the system's operational status in real time (e.g., current baggage inspection volume, load of each quarantine channel, alarm statistics) and a heatmap of clearance efficiency (based on confidence scores). (Based on quarantine time generation) and the distribution of abnormal events, it helps managers quickly grasp the overall situation;

[0313] Intelligent analysis model: Utilizes machine learning algorithms (such as time series analysis and cluster analysis) to mine historical data and predict future business trends (such as changes in the detection rate of live animals during peak periods) and equipment failure probability (based on abnormal trends in operating parameters), providing decision support for resource allocation and preventive maintenance;

[0314] Report generation and export: Regularly generate various statistical reports (such as monthly live animal detection reports and comparison tables of customs clearance efficiency at various ports), and support export in Excel, PDF and other formats to meet the needs of customs business statistics and reporting to higher-level departments.

[0315] (v) System Security and Access Control Module:

[0316] Identity authentication and authorization: Multi-factor authentication (username / password + dynamic token) is used to ensure the legitimacy of user identity. Role-based access control (RBAC) model is used to assign differentiated permissions to different users (such as inspectors, quarantine officers, and administrators) (e.g., inspectors can only view inspection data, while administrators can modify system configuration).

[0317] Data encryption and protection: SSL / TLS encryption protocol is used for transmitted data, and AES-256 encryption algorithm is used for stored data; Intrusion Detection System (IDS) and firewall are deployed to monitor network attack behavior in real time and prevent data leakage or malicious tampering;

[0318] Log auditing and backup: Record all user operation logs (login time, data access, configuration modification) and conduct regular security audits; establish an off-site disaster recovery backup mechanism, perform incremental backups of core data daily and full backups weekly to ensure data security and system reliability;

[0319] III. Key Technology Principles:

[0320] (I) Microservice Architecture Collaboration Principles:

[0321] The system management platform adopts a microservice architecture, which breaks down each functional module into independently running services (such as data integration services and process control services) and achieves efficient interaction between services through lightweight communication protocols (such as gRPC). This architecture improves the scalability and fault tolerance of the system. The failure of a certain service will not affect the overall operation, and it is easy to add functional modules or upgrade existing services.

[0322] (II) Principles of Blockchain Electronic Traceability:

[0323] Based on the distributed ledger and consensus mechanism of blockchain, customs clearance data is packaged into blocks in chronological order and linked into a chain using a hash algorithm; each block contains the hash value of the previous block, ensuring data integrity and immutability; the consortium blockchain model restricts data to be readable and writable only by authorized nodes (such as various departments of customs), meeting the privacy and compliance requirements of regulatory data.

[0324] (III) Principles of Intelligent Decision Analysis:

[0325] By utilizing big data analytics, massive amounts of historical data are cleaned, aggregated, and modeled. By training predictive models (such as LSTM-based equipment failure prediction models and random forest-based live animal risk assessment models) and combining them with real-time data input, intelligent analysis and trend prediction of the system's operating status can be achieved, assisting customs management personnel in making scientific decisions.

[0326] IV. Module Workflow:

[0327] (a) Initialization phase:

[0328] After the system management platform starts, it loads configuration files (database connection parameters, service interface addresses, permission policies, etc.) and completes the initialization of each sub-module.

[0329] Establish connections with external modules such as millimeter-wave radar array modules and data processing and analysis modules, and subscribe to key data topics (such as detection results and equipment status changes).

[0330] (II) Data Processing and Interaction Stage:

[0331] The data integration and interaction module continuously receives data sent by various functional modules, cleans and transforms it, stores it in the corresponding database, and distributes it to other modules as needed.

[0332] If a test result containing live animals is received, the process coordination control module immediately triggers baggage sorting, alarm and quarantine tasks, generates task scheduling instructions and sends them to the relevant modules;

[0333] (III) Process Monitoring and Management Phase:

[0334] Real-time monitoring of the task execution status of each module; ensuring that tasks are executed in sequence through the process collaboration control module; handling task conflicts and abnormal situations.

[0335] The electronic traceability and supervision module stores data from each stage on the blockchain in real time and updates baggage traceability information; at the same time, it checks the compliance of the process based on the legal database and issues timely warnings when problems are found.

[0336] (iv) Data Analysis and Decision-Making Stage:

[0337] The data analysis and decision-making module regularly extracts data from the database to generate visual reports and analysis reports, showcasing system operation and business trends.

[0338] Based on intelligent analysis models, historical data is mined to predict potential risks (such as equipment failure and reduced customs clearance efficiency), and early warning information and optimization suggestions are pushed to managers.

[0339] (v) Safety and Maintenance Phase:

[0340] The system security and access control module continuously monitors user operations and network security, records logs and performs audits; it also regularly backs up data to ensure the system operates securely and stably.

[0341] Based on business needs or system upgrade requirements, adjust the configuration, update functions, or expand services of each sub-module, such as adding data interfaces or optimizing task scheduling algorithms.

[0342] (vi) Closing phase:

[0343] When the system stops running or receives a stop command, the system management platform closes the connection with external modules, saves the current running status and configuration parameters, releases resources, and stops all services.

[0344] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A live animal intelligent detection and customs clearance management system based on millimeter-wave radar, characterized in that: It includes a millimeter-wave radar array module, a data processing and analysis module, a baggage sorting control module, a quarantine process management module, an alarm response module, and a system management platform; The millimeter-wave radar array module is deployed at key nodes of the customs baggage conveyor belt to scan for live biological characteristics inside the baggage in real time. The millimeter-wave radar array module adopts multi-band fusion detection technology, specifically including: Low-frequency bands are used to penetrate luggage materials to obtain the outline of a living person; the frequency of the low-frequency band is between 24-33 GHz. High-frequency bands are used to capture micro-motion characteristics, with frequencies between 60-90 GHz. A live biological feature model was established based on dual-band data fusion to distinguish live animals from static interference objects; The data processing and analysis module identifies the presence of live animals using a millimeter-wave feature extraction algorithm and outputs the detection results to the system management platform. The data processing and analysis module includes: The millimeter-wave signal preprocessing unit performs motion compensation and noise filtering on the radar echo; The liveness feature recognition unit extracts biological micro-motion spectral features through an improved two-stream convolutional neural network model; The dynamic threshold determination unit outputs a liveness confidence score based on an adaptive confidence calibration mechanism. In this adaptive confidence calibration mechanism: The final confidence score is a weighted sum of the probability of the presence of a living organism and the Shannon entropy of the micromotion feature; The weighting coefficients are dynamically adjusted according to changes in ambient temperature; for every 1 degree Celsius increase in temperature, the probability value weight increases by 0.

05. The threshold for determining liveness is determined by adding a compensation amount to the square of the material density, which is based on a base threshold of 0.

6. The material compensation coefficient is fixed at 0.

02. When the final confidence score exceeds the dynamic threshold, it is determined that the luggage contains a live animal; The system management platform dynamically triggers the following collaborative processes based on the detection results: The baggage sorting control module adjusts the sorting path in real time, guiding baggage containing live animals to the quarantine channel; The alarm response module generates tiered alarm signals and associates them with quarantine priorities. The quarantine process management module automatically allocates quarantine resources and generates electronic quarantine task orders.

2. The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to claim 1, characterized in that: In the improved two-stream convolutional neural network model: The probability of a living organism is obtained by weighting the outputs of each convolutional layer by applying the Sigmoid function. The output of each convolutional layer is generated by processing low-frequency and high-frequency band data separately through convolution kernels, summing them, and then passing them through the ReLU activation function. The dynamic frequency band weighting coefficients are determined based on the exponentially normalized values ​​of the signal-to-noise ratios (SNRs) of each layer, where the SNR weights are adjusted by the ambient noise energy value. The environmental disturbance adjustment factor approaches 1 as the environmental noise energy increases, and the empirical constant for material attenuation ranges from 0.1 to 0.

5.

3. The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to claim 1, characterized in that: The baggage sorting control module performs dynamic path optimization, including: The sorting priority index is calculated by multiplying the confidence score by a negative exponential decay function of the quarantine channel load; When the real-time load of the channel reaches the baseline load value, the priority index decays to 1 / e of the original value; By using radio frequency identification (RFID) technology to link baggage with inspection data, the sorting machine is controlled to transfer high-priority baggage to the quarantine area.

4. The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to claim 1, characterized in that: The alarm response module adopts a three-level collaborative handling mechanism, including: Level 1 alarm: When the confidence score is >0.9, an audible and visual alarm is triggered and the conveyor belt is frozen; Level 2 alarm: When the confidence score is 0.7 ≤ confidence score ≤ 0.9, a "to be reviewed" label is generated and a manual inspection is notified; Level 3 alarm: When the confidence score is 0.6 ≤ confidence score < 0.7, only a low-risk alert is sent to the quarantine management module; Alarm signals are pushed to customs personnel's mobile terminals in real time.

5. The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to claim 1, characterized in that: The quarantine process management module includes: The quarantine plan generation unit matches a pre-set treatment plan library based on the type of living organism. The resource scheduling unit allocates quarantine resources based on the spatial location coordinates of baggage and confidence scores; The electronic customs clearance document generation unit automatically generates electronic release instructions and updates the customs clearance system after quarantine is completed.

6. The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to claim 5, characterized in that: The resource scheduling unit executes an optimization algorithm, including: The scheduling cost function minimizes the weighted response time of all baggage awaiting quarantine; The weighting factor is the ratio of the confidence score to the distance from the quarantine officer to the luggage; The response time is calculated based on the average time taken for similar quarantine procedures in the past.

7. The intelligent detection and customs clearance management system for live animals based on millimeter-wave radar according to claim 1, characterized in that: The system management platform implements closed-loop control throughout the entire process, including: Millimeter-wave detection data is integrated with the customs clearance system via an application programming interface (API). An electronic traceability chain is constructed from detection to release, with data at each stage stored using blockchain technology; A heatmap of customs clearance efficiency is generated based on confidence scores and quarantine time, visually displaying process bottlenecks.

Images