A luggage intelligent detection method and system based on multi-source data fusion
By fusing multi-source data and updating component parameters using graph neural networks, the problem of insufficient modeling of complex component relationships in luggage inspection is solved, enabling accurate identification of early minor defects and stable detection results, and providing specific anomaly location information.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGXI PROD QUALITY SUPERVISION & TESTING INST (JIANGXI DEFECTIVE PROD RECALL CENT)
- Filing Date
- 2025-08-25
- Publication Date
- 2026-06-02
AI Technical Summary
Existing methods for detecting luggage and bags have shortcomings in terms of multi-source data collaborative utilization, complex component relationship modeling, and detection result stability. They are difficult to accurately identify early minor defects, and data fusion strategies based on simple weighting cannot effectively reflect the physical connections between components, resulting in low detection accuracy, insufficient adaptability, and imprecise anomaly localization.
An intelligent detection method based on multi-source data fusion is adopted. By acquiring basic detection data of luggage, a forward model of component relationships and physical laws is established. Uncertainty-driven additional data acquisition is used, and the component parameter vector is updated by combining graph neural network to generate counterfactual synthetic signals and perform comparative detection.
It improves the accuracy and stability of detection, enabling more accurate identification of anomalies in complex structures, providing specific location information of abnormal components, and enhancing the reliability of detection results and their engineering application value.
Smart Images
Figure CN121388909B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent detection and data processing technology, and in particular to an intelligent detection method and system for travel bags based on multi-source data fusion. Background Technology
[0002] With the increasing demand for travel, the frequency of use of luggage in air, rail, and daily transportation has risen significantly. The complexity of their structure and the diversity of their usage environments have led to increasingly prominent problems such as damage, loosening, cracking, and material aging. Traditional luggage inspection methods often rely on manual inspection or localized testing based on a single signal, such as visual inspection, manual tapping, or single-channel acoustic testing, to determine the integrity of the case and its components. While these methods are simple to operate, they often suffer from limitations such as strong subjectivity in the inspection results, reliance on experience, and insufficient ability to identify minor defects, making it difficult to meet the dual requirements of inspection efficiency and accuracy in large-scale travel scenarios.
[0003] In recent years, advancements in signal processing and sensing technologies have propelled the automation and intelligentization of luggage inspection. Some studies have attempted to extract material density, damping characteristics, and structural response using guided wave signals, resonance spectrum analysis, or multipath scanning. However, single-source data is prone to information loss and misjudgment when dealing with complex component relationships and diverse stress states. For example, guided wave signals are susceptible to propagation path interference, resonance spectrum peak drift cannot accurately reflect local damage, and multipath scanning has limited stability in noisy environments. This results in low accuracy for single-signal-based detection methods when identifying subtle anomalies such as early microcracks and loose assembly.
[0004] To improve the comprehensiveness of detection, some methods have introduced multi-source signal fusion, attempting to simultaneously acquire guided wave and resonance spectrum data and use statistical analysis methods to compare the correlation between different data sources. However, most existing methods employ weighted superposition or simple matching strategies, lacking systematic modeling of complex component relationships and physical laws, making it difficult to achieve a global analysis of the overall structure of the bag. Furthermore, traditional data fusion methods fail to effectively utilize residual information and fail to drive data acquisition optimization through uncertainty, resulting in "over-detection" or "missed detection" issues in some key areas. Summary of the Invention
[0005] In view of the aforementioned existing problems, the present invention is proposed.
[0006] Therefore, this invention provides an intelligent detection method for luggage based on multi-source data fusion to address the significant shortcomings of existing luggage detection methods in terms of collaborative utilization of multi-source data, modeling of complex component relationships, and stability of detection results. Traditional manual or single-signal detection methods struggle to accurately identify early, minute defects. Data fusion strategies based on simple weighting cannot effectively reflect the physical connections between components. Existing machine learning methods lack joint analysis of guided wave propagation, resonance spectrum characteristics, and multi-path scanning results, leading to problems such as low detection accuracy, insufficient adaptability, and imprecise anomaly localization.
[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0008] In a first aspect, the present invention provides an intelligent detection method for suitcases based on multi-source data fusion, which includes: acquiring basic detection data of suitcases and inputting the basic detection data into a forward model based on component relationships and physical laws to obtain component parameter vectors;
[0009] Additional data acquisition is driven by the uncertainty of the component parameter vector to obtain supplementary observation data;
[0010] The component parameter vector, basic detection data, and supplementary observation data are input into the graph neural network to update the component parameter vector and generate a counterfactual synthetic signal.
[0011] The counterfactual synthetic signal is compared with the basic detection data and the supplementary observation data, and the luggage is detected based on the comparison results, and the detection results of the luggage are output.
[0012] As a preferred embodiment of the intelligent detection method for luggage based on multi-source data fusion described in this invention, the basic detection data includes density and damping information formed by guided wave signals and resonance spectra, as well as multi-path scanning observation results.
[0013] The forward model is established by creating a component graph based on the relationships between the components of the suitcase, where the nodes of the component graph represent the components of the suitcase and the edges represent the assembly and connection relationships between the components; combining multipath guided wave propagation and resonance spectrum analysis, a forward model is established based on the structural relationships of the component graph to obtain the parameters of each component, and then spliced into a component parameter vector in a set order.
[0014] As a preferred embodiment of the intelligent detection method for travel bags based on multi-source data fusion described in this invention, the step of driving additional data acquisition based on the uncertainty of the component parameter vector includes: when the uncertainty of the component parameter vector exceeds a preset threshold, determining the corresponding component as a high-uncertainty component and triggering additional data acquisition; the uncertainty includes the parameter dispersion of the component parameter vector, the residual between the predicted observation of the forward model and the basic detection data, and the main peak drift of the resonance spectrum;
[0015] The acquisition of supplementary observation data includes selecting additional propagation paths on the edges of the component graph corresponding to the high uncertainty component, performing directional acquisition, and acquiring guided wave data and resonance spectrum data of the propagation paths.
[0016] As a preferred embodiment of the intelligent detection method for travel bags based on multi-source data fusion described in this invention, the updated component parameter vector includes: using a graph neural network to perform multi-layer propagation of the node features of each component on the component graph, mapping the basic detection data and the supplementary observation data into node feature vectors, using the residual between the observations predicted by the forward model and the basic detection data and supplementary observation data as edge feature input, and using the component parameter vector as global condition input, and obtaining the updated component parameter vector through graph convolution propagation and aggregation operations.
[0017] As a preferred embodiment of the intelligent detection method for travel bags based on multi-source data fusion described in this invention, the generation of the counterfactual synthetic signal includes: calling the forward model based on the updated component parameter vector, generating a predicted observation signal under the constraints of the propagation path and resonance characteristics of the forward model, and correcting the predicted observation signal by combining the uncertainty weight based on uncertainty calculation to obtain the counterfactual synthetic signal.
[0018] As a preferred embodiment of the intelligent detection method for travel bags based on multi-source data fusion described in this invention, the step of detecting travel bags based on comparison results includes: performing time synchronization and frequency domain feature alignment on the counterfactual synthetic signal, the basic detection data, and the supplementary observation data; calculating the residual vector; constructing a detection index based on the mean square value of the residual vector; comparing the detection index with a preset threshold and completing the detection judgment; outputting a qualified judgment when the detection index is less than the preset threshold, and outputting an abnormal judgment when the detection index is greater than or equal to the preset threshold.
[0019] As a preferred embodiment of the intelligent detection method for travel bags based on multi-source data fusion described in this invention, the output of the detection result of the travel bag includes outputting a qualified judgment or an abnormal judgment; when an abnormal judgment is output, the position of the component corresponding to the abnormality is output as the detection result, and the detection result is provided to the user interface or the upper management system.
[0020] Secondly, the present invention provides an intelligent detection system for suitcases and bags based on multi-source data fusion, comprising: a data module, which acquires basic detection data of suitcases and bags, and inputs the basic detection data into a forward model based on component relationships and physical laws to obtain component parameter vectors;
[0021] The supplementary module acquires additional observation data based on the uncertainty of the component parameter vector.
[0022] The update module inputs the component parameter vector, basic detection data, and supplementary observation data into the graph neural network to update the component parameter vector and generate a counterfactual synthetic signal.
[0023] The detection module compares the counterfactual synthetic signal with the measured data, detects the luggage based on the comparison results, and outputs the detection results of the luggage.
[0024] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the intelligent detection method for luggage based on multi-source data fusion as described in the first aspect of the present invention.
[0025] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent detection method for luggage based on multi-source data fusion as described in the first aspect of the present invention.
[0026] The beneficial effects of this invention are as follows: By inputting basic detection data into a forward model constructed based on component relationships and physical laws, the physical characteristics of luggage components can be accurately characterized in multiple dimensions, reducing the bias caused by isolated data in traditional methods; by utilizing an uncertainty-driven supplementary acquisition mechanism, guided wave signals and resonance spectrum data are dynamically supplemented at key locations, making the detection range more targeted and improving the effectiveness of observation data; by updating and propagating component parameter vectors based on graph neural networks, multi-level information aggregation can be achieved on the component diagram, thereby more realistically reflecting the mutual influence of complex assembly structures; by generating counterfactual synthetic signals and comparing them with measured data in the time and frequency domains, detection errors can be effectively reduced, and accurate location of abnormal parts can be achieved; the final output detection results include both overall judgment and specific location of abnormal components, providing a directly usable diagnostic basis for the user interface and management system. Attached Figure Description
[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0028] Figure 1 This is a flowchart of an intelligent detection method for luggage based on multi-source data fusion. Detailed Implementation
[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0030] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0031] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0032] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a method for intelligent detection of luggage based on multi-source data fusion, including the following steps:
[0033] S1: Obtain basic detection data of the luggage and input the basic detection data into the forward model based on component relationships and physical laws to obtain the component parameter vector.
[0034] First, a comprehensive inspection of the luggage was conducted to collect basic data for subsequent analysis. This basic data mainly includes two categories:
[0035] Guided wave signal: Sensors are placed at different locations in the bag to transmit and receive guided wave signals. The time delay, energy attenuation and dispersion characteristics are extracted from the propagation results of multiple paths.
[0036] Resonance spectrum: The frequency domain response obtained under frequency sweep or impact excitation is used to extract the position and variation of spectral lines that reflect the material density and damping characteristics.
[0037] After acquiring basic detection data, a forward model that reflects physical laws and component relationships is constructed.
[0038] First, abstract the suitcase / bag into a component diagram. , where nodes This indicates the various components (such as handle base, inner frame, shell plate, fasteners, etc.), and their edges. This indicates the assembly and connection relationships between components.
[0039] Based on the component diagram, and combining multipath guided wave propagation and resonance spectrum analysis, a differentiable forward model is established:
[0040]
[0041] in, This represents the actual detection data acquired from the guided wave signal and resonance spectrum, including time delay, energy attenuation, dispersion, and resonance spectrum peak value, etc. Represents the guided wave propagation function; Represents a component parameter vector; This represents the set of excitation parameters, including the settings for the center frequency, bandwidth, and propagation path of the guided wave excitation. Represents the resonance spectrum function; This represents the set of information on the resonant side, including the density distribution of the contents and the damping factor; The noise term represents random errors, environmental noise, and non-ideal factors present in the actual measurement.
[0042] The parameters of each component are obtained through model calculation, such as adhesion integrity, damping attenuation coefficient, interface reflectivity, and density offset, and are then assembled in a preset order to form a component parameter vector.
[0043]
[0044] in, Represents the total component parameter vector; These represent the parameters of different components in the suitcase; (No. (Number of parameters).
[0045] The initial acquisition phase is no longer limited to a single physical field, but simultaneously incorporates guided wave signals, density and damping information from the resonant spectrum, and multipath scanning results to ensure the complete capture of the structure's volumetric properties and local responses. This multimodal input avoids the shortcomings of traditional methods where "acoustic echoes only reflect the overall thickness" or "a single scan only covers a local area," allowing the multidimensional physical characteristics of each component to serve as mutual references. By uniformly transforming different signals to a consistent detection coordinate system, the problems of amplitude scale mismatch, phase reference drift, and time window asynchrony between cross-source data are resolved, forming a complete dataset with engineering comparability and providing a more stable foundation for subsequent modeling.
[0046] In the modeling phase, the method no longer relies on global regression but instead establishes a component graph driven by component relationships. Nodes correspond to specific components of the suitcase, and edges represent the assembly connections between components. Waveguide propagation characteristics and resonant peak offsets are directly mapped to the structural relationships in this graph, making the parameters not just a set of numerical values, but a state representation strongly bound to the component topology and physical connections. This graph structure modeling overcomes the drawbacks of traditional "parameter isolation," avoids spurious correlations caused by ignoring assembly relationships, and enables the prediction model to distinguish anomalies from different sources such as loose connections, material delamination, and structural defects, improving the targeting of localization and diagnosis.
[0047] After constructing the component graph, a forward model is built using the graph structure and physical laws. The parameters of each component are concatenated into a component parameter vector in a predetermined order. This sequential concatenation ensures the consistency of the input vector during training and inference, avoiding instability caused by disordered arrangement of data from different batches. Simultaneously, each component of the parameter vector corresponds to a specific component and physical characteristic, possessing traceability and interpretability, and is no longer an abstract black-box variable. This approach addresses the shortcomings of existing methods in feature interpretation and model portability, enabling detection results to support automated judgment and provide clear evidence for manual review, thus enhancing its practical application value.
[0048] S2: Based on the uncertainty of the component parameter vector, additional data acquisition is performed to obtain supplementary observation data; the component parameter vector, basic detection data and supplementary observation data are input into the graph neural network to update the component parameter vector, and a counterfactual synthetic signal is generated based on the updated parameters.
[0049] The system has obtained the component parameter vectors and their confidence intervals. To avoid performing full measurements every time, this step involves targeted additional acquisition for components with insufficient information. The criteria for determining insufficient information include three aspects: first, the observed parameter dispersion, i.e., the variance or width of the confidence interval for each parameter in the component parameter vector; second, the fit of the forward model to the observations, manifested as the residual between the forward model's predicted observations and the basic detection data along a specific path or frequency band; and third, the shift of the main peak of the resonance spectrum, especially when the spectral lines exhibit abnormal drift and existing guided wave evidence is insufficient to support a definitive conclusion, or when the number of effective propagation paths for a particular component in the component diagram is small, these should be considered as information gaps. Any of the above situations indicates that additional acquisition is needed to supplement the observations.
[0050] To facilitate unified judgment, the above three types of evidence are converted into comparable indicators: the variance or confidence interval width of the component parameter vector is used as a measure of parameter dispersion; the residual magnitude of the forward model's predicted observations and the basic detection data on each path is used as a residual indicator; and the expected reduction in parameter entropy after obtaining new path observations is used as a mutual information gain indicator. These indicators are normalized and accumulated according to predetermined weights to obtain a comprehensive information debt score for subsequent ranking and scheduling.
[0051] The triggering rules consist of two parts: a single threshold and a combined threshold. If the variance of the component's parameter vector exceeds the corresponding threshold, or the residual of a critical propagation path exceeds the corresponding threshold, or the relative drift of the main peak of the resonance spectrum exceeds the corresponding threshold, then additional acquisition of that component is triggered. Furthermore, additional acquisition should also be triggered when the information debt of a component exceeds the total threshold set by the system. To ensure detection cycle time and efficiency, the system establishes a priority queue according to information debt from high to low, and initiates additional acquisition for several components at the front of the queue; the same component must not be triggered repeatedly in two consecutive rounds to avoid oversampling.
[0052] After triggering additional data collection, the first step is to determine the appropriate additional data collection method. The optimization objective is to select the method that yields the maximum mutual information gain from the candidate method set.
[0053]
[0054] in, This represents the optimal set of additional acquisition system parameters (including propagation path, center frequency, bandwidth, etc.). This indicates taking parameters that maximize the objective function; This represents the set of additional acquisition system parameters for the candidate; Represents the candidate space for all possible sets of institutional parameters; This represents an estimate of the conditional mutual information; Represents a component parameter vector; Represents observation data; This represents the density and damping information obtained from the resonance spectrum.
[0055] After determining the target system selection, the first step is to optimize the frequency band to maximize the identifiability of the parameters:
[0056]
[0057]
[0058] in, Indicates the optimal frequency; This indicates the search for the frequency that maximizes the objective function; Indicates candidate frequency; The search interval representing the frequency; Represents the trace operation of a matrix; Indicates frequency Based on estimated parameters With density damping information The constructed Fisher information matrix; Indicates frequency Fisher's information matrix below; Indicates the predicted mean For component parameter vectors The partial derivatives; Indicates the transpose operation; Indicates frequency The noise or uncertainty covariance matrix is shown below. This represents the inverse matrix of the covariance matrix; Indicates the forward model at frequency The predicted mean.
[0059] After the frequency band is determined, the candidate paths are sorted, and the mutual information score and component graph coverage are combined:
[0060]
[0061] in, Indicates in path Mutual information estimation between the component parameter vector and the observed data; Representing a path The rating; Weighting coefficients representing mutual information scores; The weighting coefficients representing the coverage score; Representing a path In the component diagram Coverage score; This represents a component diagram.
[0062] Based on this, the optimal path combination is selected to ensure both distinctiveness and coverage of key components.
[0063] After determining the frequency band and path, the acquisition parameters are refined. By optimizing the pulse width, a balance is achieved between time delay resolution and signal-to-noise ratio.
[0064]
[0065] At the same time, an energy judgment threshold is set to ensure robust detection:
[0066]
[0067] in, Indicates the optimal pulse width; This indicates the search for parameters that minimize the objective function; Indicates the candidate pulse width; Indicates the pulse width The latency resolution is as follows; Indicates the target latency resolution; Indicates the balance factor; Indicates the pulse width Signal-to-noise ratio at the following levels; Indicates within the time window Internal signal energy; Indicates the window Sum all time points within the range; Indicates time The amplitude of the observed signal; This indicates the energy threshold.
[0068] These refined configurations ensure the stability and robustness of additional data acquisition.
[0069] After data collection is complete, the results will be organized into a supplementary observation dataset:
[0070]
[0071] in, This indicates a supplementary observation dataset; Indicates in path and frequency Additional guided wave observation data acquired below; Indicates in path and frequency Additional resonance spectrum observation data collected below; This represents the corresponding set of institutional parameters; Indicates all A set of paths; Indicates the first Path; This indicates the optimal frequency.
[0072] In traditional detection, supplementary observations often rely on manual experience or fixed rules, which can easily lead to data redundancy or omission of key areas. This paper introduces an uncertainty-driven acquisition mechanism that dynamically identifies high-uncertainty components by monitoring the dispersion of component parameter vectors, the difference between forward model predictions and basic detection data, and the shift of the main peak of the resonance spectrum. This approach avoids blind acquisition, focusing supplementary resources on the most needed information gaps, significantly improving the efficiency and relevance of the detection.
[0073] After identifying components with high uncertainty, additional propagation paths are selected based on the topology of the component graph. Instead of relying on uniform scanning or global encryption, directional data acquisition is performed using a path-guided approach. The selection of propagation paths considers not only node locations but also edge connectivity to ensure that the acquired guided wave data and resonance spectrum data are most valuable for correcting existing models. This design allows the supplementary data to directly address missing information, reducing unnecessary repeated testing and demonstrating high engineering applicability.
[0074] The introduction of supplementary observation data is not a one-time operation, but rather forms a dynamic feedback loop with uncertainty assessment. After each directional acquisition, the parameter uncertainty is updated, and the new distribution determines whether further data is needed. This closed-loop optimization mechanism overcomes the limitations of traditional "sampling stops" methods, enabling the acquisition strategy to be adjusted in real time, gradually compressing the uncertainty interval, achieving refined capture and layer-by-layer convergence of the luggage's condition, and improving the overall stability and reliability of the detection.
[0075] S3: Input the component parameter vector, basic detection data and supplementary observation data into the graph neural network, update the component parameter vector, and generate a counterfactual synthetic signal.
[0076] The basic detection data and supplementary observation data obtained in the preceding steps are mapped to nodes and edges in the component graph, forming node features corresponding to each component. The residuals between the forward model's predicted observations and the actual observations are introduced as edge features. Simultaneously, the existing component parameter vectors are used as global conditional inputs to ensure consistency between the update process and the preceding modeling.
[0077] Building upon this, a graph neural network is used to perform multi-layer propagation and aggregation operations on the component graph, achieving joint updates of node features and global conditions, thereby obtaining a new component parameter vector. The update process follows the computational relationships below:
[0078]
[0079] in, Indicates the first Layer Time The feature vector of each node; Represents a nonlinear activation function; Represents a node Belongs to node Neighbor set ; This represents the message passing and feature update functions; Indicates the first Layer Time The feature vector of each node; Indicates the first Layer-time neighbor nodes eigenvectors; This represents the residual between the observations predicted by the forward model and the observed observations, corresponding to the node. and Edge features between them; Indicates the first The component parameter vector during round iteration.
[0080] Finally, the updated component parameter vector is obtained through aggregation:
[0081]
[0082] in, Represents the updated component parameter vector; AGG( ) represents an aggregation function used to integrate node features; Indicates the first All nodes in the layer The feature vector set, with a total number of nodes. .
[0083] Subsequently, the updated component parameter vectors are input into the forward model to generate a counterfactual synthesized signal under the constraints of the component graph structure. The generated result incorporates uncertainty correction based on the fundamental forward computation to enhance signal stability.
[0084]
[0085] in, Indicates a counterfactual synthetic signal; Represents the forward model function; This represents the updated component parameter vector; Represents a component diagram; This represents the uncertainty correction function; This represents the uncertainty covariance matrix of the component parameters.
[0086] Through the above processing, not only were the component parameters updated, but counterfactual synthetic signals that could reflect different possible detection scenarios were also obtained, providing a basis for subsequent detection and judgment.
[0087] During the process of updating the component parameter vectors, the basic detection data and supplementary observation data are first mapped to the nodes of the component graph, forming corresponding node feature vectors. Simultaneously, the residuals between the observations predicted by the forward model and the measured data are introduced into the edges of the component graph, participating in propagation as edge features. In this way, node features reflect the direct detection information of the component, while edge features provide cross-component error information, enabling the data to be fully integrated at both spatial and relational levels.
[0088] After input mapping, a graph neural network is used to perform multi-layer propagation and aggregation operations on the component graph. During propagation, node features continuously interact with information from neighboring nodes, while edge features adjust the direction and intensity of the interaction. The component parameter vector, which is a global conditional input, provides background constraints for propagation. Through multi-layer propagation, each node can absorb comprehensive information from surrounding nodes and edges while maintaining its own attributes, achieving deep coupling and layer-by-layer correction between data.
[0089] After propagation and aggregation, an updated component parameter vector is obtained. This parameter vector, through the combined effects of basic detection data, supplementary observation data, and residual information, better reflects the actual state of the luggage components. The updated result not only enhances the model's sensitivity to local anomalies but also improves the stability and accuracy of the overall parameter representation, providing a reliable input foundation for subsequent counterfactual synthetic signal generation.
[0090] S4: Compare the counterfactual synthetic signal with the basic detection data and the supplementary observation data, and detect the suitcases based on the comparison results, and output the detection results of the suitcases.
[0091] First, the counterfactual synthesized signal is aligned with the measured data to ensure the rationality of subsequent comparisons. Time alignment uses cross-correlation to determine the optimal shift amount, using the following formula:
[0092]
[0093] in, Indicates the optimal time alignment amount; Indicates a time-shifted variable; This represents the variable that makes the subsequent summation terms reach their maximum value; This represents the amplitude of the measured signal at time t; This represents the amplitude of the counterfactual synthesized signal at time t; This indicates that the measured signal has been shifted in time. The sampling points after that.
[0094] Frequency domain alignment is achieved by calculating the optimal frequency shift through spectral cross-correlation, expressed by the formula:
[0095]
[0096] in, Indicates the optimal frequency alignment amount; Indicates the frequency shift variable; This indicates summing over frequency f; This represents the amplitude spectrum of the measured signal at frequency f; This represents the amplitude spectrum of the counterfactual synthesized signal at frequency f; This indicates that the measured signal is in translation. The frequency point after that.
[0097] After completing the time and frequency domain alignment, the aligned measured signal is denoted as... and synthesize signals with counterfactual information. By subtracting, we obtain the residual vector. The residual vector reflects the deviation between actual observations and model predictions, and serves as the basis for subsequent judgments.
[0098] With residual vector The square mean of the values is used as the detection index. This indicator characterizes the overall degree of deviation and is used to quantitatively determine whether there are structural abnormalities in luggage.
[0099] detection indicators With preset threshold Comparison: When When, output "Pass"; when When this happens, output "Abnormal Detection".
[0100] threshold Two setting methods can be used: fixed threshold ( ), obtained based on experimental calibration; adaptive threshold ( The data is updated in real time based on the mean and variance of historical detection indicators.
[0101] The final test results include: pass / fail judgment: output "normal"; anomaly judgment: output "abnormal". Combined with the residual energy distribution of the path in the component graph, the location of the component that may have problems is further located, and the diagnostic information is output to the user interface and the upper management system.
[0102] After the counterfactual synthetic signal is generated, it needs to be compared with the measured data. Before comparison, time synchronization processing is performed to ensure that the two types of signals correspond consistently in the time dimension. Simultaneously, characteristic peaks are aligned in the frequency domain to maintain comparability of the different signals in key frequency distributions. This preprocessing step avoids spurious differences caused by time drift or frequency shift, thus ensuring that the comparison results accurately reflect the true state of the luggage components.
[0103] After signal alignment, the counterfactual synthesized signal is compared point-by-point with the measured data to obtain a residual vector representing the difference between the two. A detection index is then calculated based on the residual vector, and the overall difference is quantified using mean square processing. This detection index comprehensively reflects the signal deviation across multiple observation points, exhibiting greater robustness compared to single-point differences and effectively avoiding the impact of local noise interference on the overall judgment.
[0104] After the detection indicators are constructed, they are compared with preset thresholds to arrive at the final detection conclusion. When the detection indicators are below the threshold, a judgment that the suitcase is in a qualified state is output; when the detection indicators are above the threshold, a judgment that the suitcase has an anomaly is output, and the location of the component corresponding to the anomaly is further located. The detection conclusion is finally presented through the user interface or the upper-level management system, providing an intuitive and clear reference for on-site inspection and subsequent maintenance.
[0105] This embodiment also provides an intelligent detection system for luggage based on multi-source data fusion, including:
[0106] The data module acquires basic inspection data of the luggage and inputs it into a forward model based on component relationships and physical laws to obtain component parameter vectors.
[0107] The supplementary module acquires additional observation data based on the uncertainty of the component parameter vector.
[0108] The update module inputs the component parameter vector, basic detection data, and supplementary observation data into the graph neural network, updates the component parameter vector, and generates a counterfactual synthetic signal.
[0109] The detection module compares the counterfactual synthetic signal with the measured data, detects the luggage based on the comparison results, and outputs the detection results of the luggage.
[0110] Example 2 is an embodiment of the present invention, which provides a method and system for intelligent detection of luggage based on multi-source data fusion. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.
[0111] In a testing experiment on a certain type of travel bag, 20 samples with similar appearance and consistent dimensions were selected. Ten of these were intact bags, while the other ten had varying degrees of structural defects pre-programmed in concealed locations, such as loose pull rod connections, delamination of sidewall materials, and minor cracks in the bottom load-bearing frame. The testing system consisted of a guided wave signal acquisition module, a resonance spectrum analysis module, a multi-path scanning platform, and a graph neural network computing unit. At the start of the experiment, 12 measuring points were set up on the surface of the bag using guided wave sensors, with an excitation frequency range of 1.0 kHz to 5.0 kHz and a step interval of 50 Hz. The basic testing data obtained through the initial excitation included material density distribution, damping coefficient information, and the time-domain response of the three main propagation paths. This data was input into a forward model established based on component relationships and physical laws to obtain an initial component parameter vector, covering the dynamic characteristics of major components such as the pull rod, frame, zipper, and fabric.
[0112] During the parameter calculation process, the variance of the component parameters of some samples exceeded 0.15, indicating a high uncertainty. The system automatically triggered additional acquisitions, added new propagation paths around the corresponding components, and directionally collected new guided wave and resonance spectrum data. The supplementary observation data volume was approximately 480 pieces for each sample. Subsequently, the system input the component parameter vector, basic detection data, and supplementary observation data into the graph neural network for iterative propagation. The updated parameter vector formed a stronger correlation between the features of adjacent nodes. Based on the updated parameters, the forward model was called to generate predicted observation signals, which were corrected by combining the uncertainty weights to obtain counterfactual synthetic signals. Finally, the synthetic signals and the measured data were aligned in the time and frequency domains, and the residual vector and detection indicators were calculated to output the detection conclusions for each travel suitcase. The entire process was fully automated, and the average detection time for a single sample was 12.35 seconds.
[0113] Among the 20 samples, the mean value of the detection indicators for 10 intact samples was 0.034, the standard deviation was 0.008, and the maximum value was 0.046. Among the 10 defective samples, the mean value of the detection indicators was 0.127, the standard deviation was 0.021, the minimum value was 0.101, and the maximum value was 0.168. The threshold was set at 0.080. The comparison of the judgment results showed that 19 out of 20 samples were judged to be consistent with the true state, and the detection accuracy was 95.00%. Among them, one slightly cracked sample had a crack depth less than 0.3 mm, the detection indicator was 0.078, slightly lower than the threshold, and was judged to be qualified, constituting a missed detection. The judgment accuracy of the intact samples was 100.00%, and the judgment accuracy of the defective samples was 90.00%. In addition, among the 9 samples with abnormal judgments, the system could accurately locate the positions of the abnormal components, and the positioning accuracy reached 88.89%.
[0114] From the experimental data, it can be seen that the detection indicators formed an obvious boundary between intact samples and defective samples. The detection indicators of intact samples were concentrated below 0.046, and the variance was small, indicating that the signal comparison results were stable and reliable. The detection indicators of defective samples were significantly higher than 0.100, reflecting the sensitivity of the abnormal state to the signal residuals. Compared with traditional single-frequency domain analysis, this method effectively improved the discrimination of complex structure states through multi-source data fusion and graph neural network updates, avoiding misjudgments caused by noise from a single data source. In addition, through the uncertainty-driven additional acquisition mechanism, more abundant observation paths could be obtained for components with high uncertainty, reducing the situation where local defects were ignored. In the experiment, the positioning results of the abnormal components showed that the method could provide specific diagnostic information, rather than just a binary judgment of overall pass or fail, which provided more targeted references for maintenance and repair in practical applications. Overall, this experiment demonstrated the effectiveness and advantages of the method in the detection of complex travel suitcase structures, especially in terms of detection accuracy and abnormal positioning ability, which were superior to traditional detection methods.
[0115] This embodiment also provides a computer device applicable to the intelligent detection method for luggage based on multi-source data fusion, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent detection method for luggage based on multi-source data fusion as proposed in the above embodiment.
[0116] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.
[0117] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the intelligent detection method for travel bags based on multi-source data fusion as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.
[0118] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A smart luggage detection method based on multi-source data fusion, characterized in that, include: Acquire basic test data of the luggage and input the basic test data into a forward model based on component relationships and physical laws to obtain component parameter vectors; the forward model includes a component graph based on the component relationships of the luggage, where the nodes of the component graph represent the components of the luggage and the edges represent the assembly and connection relationships between the components; combining multipath guided wave propagation and resonance spectrum analysis, the forward model is established based on the structural relationships of the component graph to obtain the parameters of each component, and then spliced into a component parameter vector in a set order; Additional data acquisition is driven by the uncertainty of the component parameter vector to obtain supplementary observation data; The component parameter vector, basic detection data, and supplementary observation data are input into the graph neural network to update the component parameter vector and generate a counterfactual synthetic signal. The generation of the counterfactual synthetic signal includes calling the forward model based on the updated component parameter vector, generating a predicted observation signal under the constraints of the propagation path and resonance characteristics of the forward model, and correcting the predicted observation signal by combining the uncertainty weight based on uncertainty calculation to obtain the counterfactual synthetic signal. The counterfactual synthetic signal is compared with the basic detection data and the supplementary observation data, and the luggage is detected based on the comparison results, and the detection results of the luggage are output. The step of detecting luggage based on comparison results includes performing time synchronization and frequency domain feature alignment on the counterfactual synthetic signal, the basic detection data, and the supplementary observation data, respectively, calculating the residual vector, constructing a detection index based on the mean square value of the residual vector, and comparing the detection index with a preset threshold to complete the detection judgment; when the detection index is less than the preset threshold, a qualified judgment is output, and when the detection index is greater than or equal to the preset threshold, an abnormal judgment is output.
2. The intelligent detection method for travel bags based on multi-source data fusion as described in claim 1, characterized in that: The basic detection data includes density and damping information formed by guided wave signals and resonance spectra, as well as multipath scanning observation results.
3. The intelligent detection method for travel bags based on multi-source data fusion as described in claim 2, characterized in that: The additional acquisition driven by the uncertainty of the component parameter vector includes determining the corresponding component as a high uncertainty component and triggering additional acquisition when the uncertainty of the component parameter vector exceeds a preset threshold; the uncertainty is determined by the parameter dispersion of the component parameter vector, the residual between the predicted observation of the forward model and the basic detection data, and the main peak drift of the resonance spectrum. The acquisition of supplementary observation data includes selecting additional propagation paths on the edges of the component graph corresponding to the high uncertainty component, performing directional acquisition, and acquiring guided wave data and resonance spectrum data of the propagation paths.
4. The intelligent detection method for travel bags based on multi-source data fusion as described in claim 3, characterized in that: The updated component parameter vector includes using a graph neural network to propagate the node features of each component in multiple layers on the component graph, mapping the basic detection data and the supplementary observation data into node feature vectors, using the residual between the observations predicted by the forward model and the basic detection data and supplementary observation data as edge feature input, and using the component parameter vector as global condition input. After graph convolution propagation and aggregation operations, the updated component parameter vector is obtained.
5. The intelligent detection method for travel bags based on multi-source data fusion as described in claim 4, characterized in that: The output test results for the travel bags include either a pass / fail determination or an anomaly determination. When an anomaly is detected, the location of the component corresponding to the anomaly is output as the detection result, and the detection result is provided to the user interface or the upper-level management system.
6. A smart detection system for luggage based on multi-source data fusion, based on the smart detection method for luggage based on multi-source data fusion as described in any one of claims 1 to 5, characterized in that: The data module acquires basic test data of the luggage and inputs it into a forward model based on component relationships and physical laws to obtain component parameter vectors. The forward model includes a component graph based on the component relationships of the luggage, where nodes represent the components of the luggage and edges represent the assembly and connection relationships between the components. Combining multipath waveguide propagation and resonance spectrum analysis, the forward model is established based on the structural relationships of the component graph to obtain the parameters of each component, and these parameters are then assembled into a component parameter vector in a predetermined order. The supplementary module acquires additional observation data based on the uncertainty of the component parameter vector. The update module inputs the component parameter vector, basic detection data, and supplementary observation data into the graph neural network to update the component parameter vector and generate a counterfactual synthetic signal. The generation of the counterfactual synthetic signal includes calling the forward model based on the updated component parameter vector, generating a predicted observation signal under the constraints of the propagation path and resonance characteristics of the forward model, and correcting the predicted observation signal by combining the uncertainty weight based on uncertainty calculation to obtain the counterfactual synthetic signal. The detection module compares the counterfactual synthetic signal with the basic detection data and the supplementary observation data, and detects the suitcases based on the comparison results, and outputs the detection results of the suitcases. The step of detecting luggage based on comparison results includes performing time synchronization and frequency domain feature alignment on the counterfactual synthetic signal, the basic detection data, and the supplementary observation data, respectively, calculating the residual vector, constructing a detection index based on the mean square value of the residual vector, and comparing the detection index with a preset threshold to complete the detection judgment; when the detection index is less than the preset threshold, a qualified judgment is output, and when the detection index is greater than or equal to the preset threshold, an abnormal judgment is output.