An object-carrying safety detection management system and method based on the Internet of Things

Abstract

The application discloses a kind of based on Internet of Things's carrying thing security detection management system and method, it is related to carrying thing safety detection technical field, the present application is based on the sensor combination of carrying thing type label selection, collects the multimodal sensing data of carrying thing, vehicle and driver;Uniform timestamp and space coordinates are applied to multimodal data Generation spatiotemporal correlation dataset;Filter spatiotemporal correlation dataset constitutes core feature set, the operation characteristic analysis fusion weight of vehicle is combined, and optimization feature is generated based on fusion weight;Optimized features and corresponding category labels are used to train SVM classifier, and the local data of each vehicle is used to train LSTM model;Analysis source domain and target domain MMD distance;Fusion multi-source evidence analysis joint trust degree, based on dynamic safety evaluation triggers abnormal early warning.Adaptive fusion multi-source feature, dynamically adjust threshold, adapt to different time periods risk fluctuation, improve early warning sensitivity.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carrier safety detection, and particularly relates to a carrier safety detection management system and method based on Internet of Things. BACKGROUND

[0002] The logistics transportation industry is facing challenges, especially in terms of driver monitoring and transportation safety. Intelligent logistics has the advantages of strong connectivity, wide integration, low economic cost, and high operation efficiency. Transportation is an important foundation for the construction of the logistics system, and promoting the development of intelligent logistics has various application scenarios, including Internet road freight. Logistics enterprises need to timely grasp vehicle information and conduct real-time monitoring and scheduling.

[0003] In the real-time detection of existing carrier safety, traditional sensor networks use fixed combinations such as temperature and humidity and acceleration sensors, which cannot dynamically adjust according to the characteristics of the carrier such as liquid shaking and fragile product vibration, resulting in missed detection or redundant collection of key parameters such as pressure distribution and vibration spectrum, and traditional analysis methods rely on single signals or fixed threshold values to judge risks, ignoring the synergistic effect of multiple evidences, and have a high false positive rate.

[0004] Therefore, the present application discloses a carrier safety detection management system and method based on Internet of Things to solve the above problems. SUMMARY

[0005] The present application aims to provide a carrier safety detection management system and method based on Internet of Things to solve the problems in the prior art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a carrier safety detection management method based on Internet of Things, which comprises the following steps:

[0007] S1: selecting a sensor combination based on a carrier type label, collecting multi-modal sensor data of the carrier, vehicle and driver, and applying a unified timestamp and spatial coordinate to the multi-modal data to generate a spatio-temporal correlation data set;

[0008] S2: screening the spatio-temporal correlation data set to form a core feature set, combining the running characteristics of the vehicle to analyze the fusion weight, and generating an optimized feature based on the fusion weight;

[0009] S3: training an SVM classifier based on the optimized feature and the corresponding category label, training an LSTM model using local data of each vehicle, and analyzing the MMD distance between the source domain and the target domain;

[0010] S4: fusing multi-source evidence analysis to analyze the joint trust degree, and triggering an abnormal early warning based on dynamic safety evaluation.

[0011] According to the above scheme, in S1, the following contents are included:

[0012] S101: According to the cargo type label L type Select sensor combination; if L type =Liquid, enable pressure sensor array S p Acquisition of pressure distribution ;where p n This represents the normalized measurement value of the nth pressure sensor, where n ∈ [1, N], N represents the total number of pressure sensors, and n is a positive integer; if L type =Fragile item, activate vibration sensor D h Collect vibration spectrum D h (t);

[0013] S102: Analyze the signal-to-noise ratio (SNR) of the i-th sensor. If the SNR of the i-th sensor is less than the SNR threshold, switch to the backup sensor. After switching, compare the spectral correlation between the backup sensor and the main sensor in the overlapping frequency band. Apply a unified timestamp t to the multimodal data. j and spatial coordinates (x, y, z) j Generate a spatiotemporal correlation dataset X ST ={(t) j x j y j , z j , data j )};data j Let represent the set of sensor measurements corresponding to the j-th spatiotemporal point.

[0014] This application intelligently selects pressure or vibration sensors based on the type of transported goods (liquid / fragile goods) to avoid redundant data acquisition and improve the monitoring accuracy of key parameters; it calculates the signal-to-noise ratio in real time and switches to backup sensors to dynamically eliminate noise interference data and improve the reliability of subsequent feature extraction; it unifies timestamps and spatial coordinates to eliminate spatiotemporal misalignment and constructs a spatiotemporally correlated dataset to provide data support for multi-sensor collaborative analysis.

[0015] According to the above scheme, S2 includes the following:

[0016] S201: For the spatiotemporal correlation dataset X ST Calculating the importance of random forest features using physical data I RF d and mutual information I MI d Where d represents the feature dimension index; retain those that satisfy I RF d > α and I MI d The features of >β constitute the core feature set F core ={f d}, where α represents the feature importance threshold of the random forest, α=1 / d 1 / 2 β represents the mutual information threshold, and β is equal to the median of mutual information in the dataset;

[0017] S202: Vehicle acceleration function a c (t) Convolution is performed with window lengths τ equal to τ1, τ2, and τ3, respectively, where τ1, τ2, and τ3 are preset constants, to obtain the local mutation amount M. local =Conv1D(a c , τ=S1), trend quantity M trend =Conv1D(a c , τ=S2) and event quantity M global =Conv1D(a c ,τ=S3);

[0018] S203: Merge feature X=[F core M local M trend M global Project onto the query matrix Q, the key matrix K, and the value matrix V; Q=W q X, K=W K X, V=W V X; where W q W K and W V Represents the trainable parameter matrix; the query matrix Q, key matrix K, and value matrix V are split into A1, A2, ..., A1 according to the total number of attention heads M. M Subvectors; calculate the matching score (Score) of the m-th head with all keys. m =A m ⊺ K; where ⊺ represents the transpose of the vector; analyze the fusion weight w m =exp(A m ⊺ K) / exp(A r ⊺ K); where exp(∙) represents an exponential function with the base of natural numbers, r∈[1,M], and r is a positive integer; Optimization feature F is generated based on fusion weights. opt = w m ⋅[F core M local M trend M global ].

[0019] This application combines the importance and mutual information of random forests to screen core features, balancing feature prediction ability and independence, and suppressing overfitting; it extracts local mutations, trends, and global events of acceleration signals through different convolutional windows to comprehensively characterize vehicle motion patterns; and it adaptively fuses multi-source features to enhance the contribution of key features (such as sudden vibrations) and improve the discriminative power of the feature space.

[0020] According to the above scheme, S3 includes the following:

[0021] S301: Using optimized feature F opt And train an SVM classifier with corresponding class labels b∈{-1, +1}, where the radial basis kernel function is:

[0022] ;

[0023] Where, x s Let f(x) represent the s-th support vector, x represent the feature vector of the new sample to be classified, exp(∙) represent the exponential function with base to natural number, γ represent the kernel function parameter, and D represent the optimized feature F. opt Feature dimensions, Var(F) opt () represents the sum of the variances of all features;

[0024] Decision function:

[0025] ;

[0026] Where, λ s Represents support vector x s The weight, b s Represents support vector x s The category label is denoted by c, which represents the bias term, and sign (∙) represents the sign function. The output is +1 or -1 to indicate the classification result.

[0027] When the new sample F new Upon arrival, support vectors are filtered using KKT conditions, retaining the original support vectors and newly added boundary samples to construct a reduced set S. ’ =S old ∪S new In the reduced set S ’ Solve the dual problem above and update λ. s and c;

[0028] S302: Train an LSTM model using the optimized feature matrix stored in the in-vehicle edge devices of each vehicle, and update the parameter set W. localThe parameter set includes the weight matrices of the input gate, forget gate, output gate, and candidate memory units in the LSTM model. The locally trained parameters are encrypted and uploaded to the cloud server. Gaussian noise is added to the aggregation result on the cloud, with the standard deviation σ of the Gaussian noise satisfying relaxed difference privacy. The cloud-aggregated parameters W... global = W e local +N(0,σ 2 E represents the total number of vehicles; W represents the total number of vehicles. e local Represents the parameters of the e-th vehicle; N(0,σ) 2 ) indicates that the center is 0 and the variance σ 2 Gaussian normal distribution sampling; cloud-aggregated parameters are distributed to corresponding vehicles to replace the parameter set;

[0029] S303: Denote the data distribution used during model training as the source domain and the new data distribution generated during actual model deployment as the target domain. Calculate the MMD distance between the source domain and the target domain. If the MMD distance is greater than the corresponding threshold, trigger model retraining.

[0030] Support vectors are selected using KKT conditions, and only boundary sample parameters are updated, reducing computational complexity and enabling rapid model iteration; Gaussian noise is added to aggregated parameters in the cloud to balance model performance and data security; data distribution shifts are identified in a timely manner to trigger model retraining to maintain prediction accuracy;

[0031] According to the above scheme, S4 includes the following:

[0032] S401: For each source of evidence g∈{L obj P state F drift Let L be the basic probability assignment function; where L is the probability assignment function. obj P represents the classification label output by the SVM classifier. state F represents the probability of the vehicle's state predicted by the LSTM model. drift The detected distribution offset markers are represented; Eye represents the attention index collected by the eye tracker; the joint confidence level is calculated using the DS evidence theory.

[0033] S402: Obtain the time interval t and corresponding accident rate R from historical accident data, and construct an exponential decay curve: R(t) = εexp(−κt) + μ, where ε represents the initial risk amplitude, κ represents the risk decay rate, and μ represents the baseline risk lower limit; fit the parameters ε, κ, and μ using the nonlinear least squares method; calculate the real-time risk threshold Risk. th(t)=R(t)+η∙std(R); η represents the system's preset sensitivity coefficient, and std(R) represents the standard deviation of the historical accident rate; if the real-time joint trust level is greater than the real-time risk threshold, an early warning is triggered, and the warning information is sent to the driver and the system administrator.

[0034] Calculate the joint trust level to reduce the risk of misjudgment based on single evidence; combine real-time standard deviation to dynamically adjust the threshold to adapt to risk fluctuations at different times (such as increased risk of nighttime transportation) and improve the sensitivity of early warning.

[0035] Another aspect of this application provides an Internet of Things (IoT)-based cargo safety detection and management system, which is applied to the above-mentioned IoT-based cargo safety detection and management method. The system includes a data acquisition and processing module, a feature extraction module, a model training module, and a dynamic early warning module.

[0036] The data acquisition and processing module is used to select sensor combinations based on the type label of the transport object, and collect multimodal sensing data of the transport object, vehicle and driver; and to apply a unified timestamp and spatial coordinates to the multimodal data to generate a spatiotemporal correlated dataset;

[0037] The feature extraction module is used to filter spatiotemporally related datasets to form a core feature set, combine vehicle operation characteristics to analyze and fuse weights, and generate optimized features based on the fusion weights.

[0038] The model training module trains an SVM classifier based on optimized features and corresponding category labels, trains an LSTM model using local data from each vehicle, and analyzes the MMD distance between the source and target domains.

[0039] The dynamic early warning module is used to integrate multi-source evidence analysis to jointly establish trust levels and trigger anomaly warnings based on dynamic security assessments.

[0040] According to the above scheme, the data acquisition and processing module includes a sensor selection unit and a data spatiotemporal correlation unit;

[0041] The sensor selection unit is used to select a sensor combination based on the type label of the transported goods. If the type label is liquid, the pressure sensor array is activated to collect the pressure distribution; if the type label is fragile, the vibration sensor is activated to collect the vibration spectrum.

[0042] The data spatiotemporal correlation unit applies a unified timestamp and spatial coordinates to multimodal data to generate a spatiotemporal correlation dataset.

[0043] According to the above scheme, the feature extraction module includes a feature analysis unit and a feature fusion unit;

[0044] The feature analysis unit is used to calculate the importance and mutual information of random forest features on the physical data in the spatiotemporally correlated dataset, and retains features that satisfy the random forest feature importance and mutual information greater than the corresponding threshold to form the core feature set. The vehicle acceleration function is convolved with different window lengths to generate acceleration feature quantities.

[0045] The feature fusion unit analyzes the fusion weights based on the fusion features and generates optimized features based on the fusion weights.

[0046] According to the above scheme, the model training module includes a data training unit and a offset analysis unit;

[0047] The data training unit is used to train an SVM classifier using optimized features and corresponding category labels; and to train an LSTM model using local data from each vehicle.

[0048] The offset analysis unit is used to denote the data distribution used in the model training phase as the source domain, the new data distribution generated when the model is actually deployed as the target domain, and to calculate the MMD distance between the source domain and the target domain.

[0049] According to the above scheme, the dynamic early warning module includes a risk analysis unit and a dynamic threshold early warning unit;

[0050] The risk analysis unit is used to define a basic probability allocation function for each source of evidence and to calculate the joint confidence level using DS evidence theory.

[0051] The dynamic threshold early warning unit is used to obtain the time interval and corresponding accident rate in historical accident data, construct an exponential decay curve, calculate the real-time risk threshold based on the exponential decay curve, and trigger an early warning if the real-time joint trust degree is greater than the real-time risk threshold, sending the early warning information to the driver and system administrator.

[0052] Compared with existing technologies, the beneficial effects of this invention are as follows: This application intelligently selects pressure or vibration sensors based on the type of transport vehicle, avoiding redundant data acquisition and improving the monitoring accuracy of key parameters; it calculates the signal-to-noise ratio in real time and switches to backup sensors, dynamically eliminating noise interference data and improving the reliability of subsequent feature extraction; it unifies timestamps and spatial coordinates to eliminate spatiotemporal misalignment, constructing a spatiotemporally correlated dataset to provide data support for multi-sensor collaborative analysis; this application combines random forest importance and mutual information to screen core features, balancing feature prediction ability and independence, and suppressing overfitting; it extracts acceleration signals through different convolutional windows. Local mutations, trends, and global events comprehensively characterize vehicle motion patterns; adaptive fusion of multi-source features enhances the contribution of key features and improves the discriminative power of the feature space; KKT conditional selection of support vectors reduces computational complexity and enables rapid model iteration by updating only boundary sample parameters; cloud-aggregated parameters are amplified and Gaussian noise is added to balance model performance and data security; timely identification of data distribution shifts triggers model retraining to maintain prediction accuracy; calculation of joint confidence reduces the risk of misjudgment based on single evidence; and dynamic adjustment of thresholds based on real-time standard deviation adapts to risk fluctuations at different times, improving early warning sensitivity. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 This is a flowchart illustrating an Internet of Things-based method for the safety detection and management of transport vehicles according to the present invention.

[0055] Figure 2 This is a schematic diagram of the structure of a vehicle safety detection and management system based on the Internet of Things according to the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figure 1 The present invention provides a technical solution: a method for safety detection and management of transport vehicles based on the Internet of Things, the method comprising the following steps:

[0058] S1: Select sensor combinations based on the type label of the transport vehicle to collect multimodal sensing data of the transport vehicle, the vehicle and the driver; apply a unified timestamp and spatial coordinates to the multimodal data to generate a spatiotemporal correlated dataset;

[0059] S1 includes the following:

[0060] S101: According to the cargo type label L type Select sensor combination; if L type =Liquid, enable pressure sensor array S p Collect pressure distribution {p n} EQ \o(\s\up 6(N ),\s\do 2(n=1)); where p n This represents the normalized measurement value of the nth pressure sensor, where n ∈ [1, N], N represents the total number of pressure sensors, and n is a positive integer; if L type =Fragile item, activate vibration sensor D h Collect vibration spectrum D h (t);

[0061] S102: Analyze the signal-to-noise ratio (SNR) of the i-th sensor. If the SNR of the i-th sensor is less than the SNR threshold, switch to the backup sensor. After switching, compare the spectral correlation between the backup sensor and the main sensor in the overlapping frequency band. Apply a unified timestamp t to the multimodal data. j and spatial coordinates (x, y, z) j Generate a spatiotemporal correlation dataset X ST ={(t) j x j y j , z j , data j )};data j Let represent the set of sensor measurements corresponding to the j-th spatiotemporal point.

[0062] S2: Select spatiotemporal related datasets to form a core feature set, combine vehicle operation characteristics to analyze and fuse weights, and generate optimized features based on the fusion weights;

[0063] S2 contains the following:

[0064] S201: For the spatiotemporal correlation dataset X ST Calculating the importance of random forest features using physical data I RF d and mutual information I MI d Where d represents the feature dimension index; retain those that satisfy I RF d > α and I MI dThe features of >β constitute the core feature set F core ={f d}, where α represents the feature importance threshold of the random forest, α=1 / d 1 / 2 β represents the mutual information threshold, and β is equal to the median of mutual information in the dataset;

[0065] S202: Vehicle acceleration function a c (t) Convolution is performed with window lengths τ equal to τ1, τ2, and τ3, respectively, where τ1, τ2, and τ3 are preset constants, to obtain the local mutation amount M. local =Conv1D(a c , τ=S1), trend quantity M trend =Conv1D(a c , τ=S2) and event quantity M global =Conv1D(a c ,τ=S3);

[0066] S203: Merge feature X=[F core M local M trend M global Project onto the query matrix Q, the key matrix K, and the value matrix V; Q=W q X, K=W K X, V=W V X; where W q W K and W V Represents the trainable parameter matrix; the query matrix Q, key matrix K, and value matrix V are split into A1, A2, ..., A1 according to the total number of attention heads M. M Subvectors; calculate the matching score (Score) of the m-th head with all keys. m =A m ⊺ K; where ⊺ represents the transpose of the vector; analyze the fusion weight w m =exp(A m ⊺ K) / exp(A r ⊺ K); where exp(∙) represents an exponential function with the base of natural numbers, r∈[1,M], and r is a positive integer; Optimization feature F is generated based on fusion weights. opt = w m ⋅[F core M local M trend M global ].

[0067] S3: Train an SVM classifier based on optimized features and corresponding category labels, train an LSTM model using local data from each vehicle, and analyze the MMD distance between the source and target domains.

[0068] S3 includes the following:

[0069] S301: Using optimized feature F opt And train an SVM classifier with corresponding class labels b∈{-1, +1}, where the radial basis kernel function is:

[0070] ;

[0071] Where, x s Let f(x) represent the s-th support vector, x represent the feature vector of the new sample to be classified, exp(∙) represent the exponential function with base to natural number, γ represent the kernel function parameter, and D represent the optimized feature F. opt Feature dimensions, Var(F) opt () represents the sum of the variances of all features;

[0072] Decision function:

[0073] ;

[0074] Where, λ s Represents support vector x s The weight, b s Represents support vector x s The category label is denoted by c, which represents the bias term, and sign (∙) represents the sign function. The output is +1 or -1 to indicate the classification result.

[0075] When the new sample F new Upon arrival, support vectors are filtered using KKT conditions, retaining the original support vectors and newly added boundary samples to construct a reduced set S. ’ =S old ∪S new In the reduced set S ’ Solve the dual problem above and update λ. s and c;

[0076] S302: Train an LSTM model using the optimized feature matrix stored in the in-vehicle edge devices of each vehicle, and update the parameter set W. local The parameter set includes the weight matrices of the input gate, forget gate, output gate, and candidate memory units in the LSTM model. The locally trained parameters are encrypted and uploaded to the cloud server. Gaussian noise is added to the aggregation result on the cloud, with the standard deviation σ of the Gaussian noise satisfying relaxed difference privacy. The cloud-aggregated parameters W... global = W elocal +N(0,σ 2 E represents the total number of vehicles; W represents the total number of vehicles. e local Represents the parameters of the e-th vehicle; N(0,σ) 2 ) indicates that the center is 0 and the variance σ 2 Gaussian normal distribution sampling; cloud-aggregated parameters are distributed to corresponding vehicles to replace the parameter set;

[0077] S303: Denote the data distribution used during model training as the source domain and the new data distribution generated during actual model deployment as the target domain. Calculate the MMD distance between the source domain and the target domain. If the MMD distance is greater than the corresponding threshold, trigger model retraining.

[0078] S4: Integrates multi-source evidence analysis to jointly establish trust levels and triggers anomaly warnings based on dynamic security assessments.

[0079] S4 includes the following:

[0080] S401: For each source of evidence g∈{L obj P state F drift Let L be the basic probability assignment function; where L is the probability assignment function. obj P represents the classification label output by the SVM classifier. state F represents the probability of the vehicle's state predicted by the LSTM model. drift The detected distribution offset markers are represented; Eye represents the attention index collected by the eye tracker; the joint confidence level is calculated using the DS evidence theory.

[0081] Example 1: In this example, the category label L is defined. obj Basic probability assignment function: Proposition H is high-risk.

[0082] Distribution offset marker F drift The generation rules are as follows: ;D MMD ξ represents the MMD distance between the source domain and the target domain, and ξ represents the MMD distance threshold.

[0083] Attention index is equal to the normalized weighted average of fixation duration and blink frequency;

[0084] S402: Obtain the time interval t and corresponding accident rate R from historical accident data, and construct an exponential decay curve: R(t) = εexp(−κt) + μ, where ε represents the initial risk amplitude, κ represents the risk decay rate, and μ represents the baseline risk lower limit; fit the parameters ε, κ, and μ using the nonlinear least squares method; calculate the real-time risk threshold Risk. th(t)=R(t)+η∙std(R); η represents the system's preset sensitivity coefficient, and std(R) represents the standard deviation of the historical accident rate; if the real-time joint trust level is greater than the real-time risk threshold, an early warning is triggered, and the warning information is sent to the driver and the system administrator.

[0085] Example 2: Fit ε=0.5, κ=0.1, μ=0.05 based on historical data; then R(t)=0.5exp(−0.1t)+0.05; when t=10 days, R(10)=0.5exp(−0.1×10)+0.05≈0.234;

[0086] In this embodiment, the standard deviation of the historical accident rate std(R) = 0.05; η = 2; then Risk th (t)=0.5exp(−0.1t)+0.05+2×0.05;

[0087] When t=5 days, Risk th (5)=0.5exp(−0.1×5)+0.05+0.1≈0.453;

[0088] Please see Figure 2 The present invention provides a technical solution: a carrier safety detection and management system based on the Internet of Things, which includes a data acquisition and processing module, a feature extraction module, a model training module and a dynamic early warning module;

[0089] The data acquisition and processing module is used to select sensor combinations based on the type label of the transport vehicle, collect multimodal sensing data of the transport vehicle, the vehicle and the driver; and apply a unified timestamp and spatial coordinates to the multimodal data to generate a spatiotemporal correlated dataset.

[0090] The feature extraction module is used to filter spatiotemporally related datasets to form a core feature set, combine vehicle operation characteristics to analyze and fuse weights, and generate optimized features based on the fusion weights;

[0091] The model training module trains an SVM classifier based on optimized features and corresponding category labels, trains an LSTM model using local data from each vehicle, and analyzes the MMD distance between the source and target domains.

[0092] The dynamic early warning module is used to integrate multi-source evidence analysis to jointly determine trust levels and trigger anomaly warnings based on dynamic security assessments.

[0093] The data acquisition and processing module includes a sensor selection unit and a data spatiotemporal correlation unit;

[0094] The sensor selection unit is used to select the sensor combination according to the type label of the transported goods. If the type label is liquid, the pressure sensor array is activated to collect the pressure distribution; if the type label is fragile, the vibration sensor is activated to collect the vibration spectrum.

[0095] The data spatiotemporal correlation unit applies a unified timestamp and spatial coordinates to multimodal data to generate a spatiotemporal correlation dataset.

[0096] The feature extraction module includes a feature analysis unit and a feature fusion unit;

[0097] The feature analysis unit calculates the importance and mutual information of random forest features on the physical data in the spatiotemporally correlated dataset, retaining features whose random forest feature importance and mutual information are greater than the corresponding thresholds to form the core feature set. The vehicle acceleration function is convolved with different window lengths to generate acceleration feature quantities.

[0098] The feature fusion unit analyzes the fusion weights based on the fusion features and generates optimized features based on the fusion weights.

[0099] The model training module includes a data training unit and a offset analysis unit;

[0100] The data training unit is used to train an SVM classifier using optimized features and corresponding category labels; and to train an LSTM model using local data from each vehicle.

[0101] The offset analysis unit is used to denote the data distribution used during the model training phase as the source domain and the new data distribution generated during the actual deployment of the model as the target domain, and to calculate the MMD distance between the source domain and the target domain.

[0102] The dynamic early warning module includes a risk analysis unit and a dynamic threshold early warning unit;

[0103] The risk analysis unit is used to define a basic probability allocation function for each source of evidence and to calculate the joint confidence level using DS evidence theory.

[0104] The dynamic threshold warning unit is used to obtain the time interval and corresponding accident rate from historical accident data, construct an exponential decay curve, calculate the real-time risk threshold based on the exponential decay curve, and trigger a warning if the real-time joint trust degree is greater than the real-time risk threshold, sending the warning information to the driver and system administrator.

[0105] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0106] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for safety detection and management of transported goods based on the Internet of Things, characterized in that, The method includes the following steps: S1: Select sensor combinations based on the type label of the transport vehicle to collect multimodal sensing data of the transport vehicle, the vehicle and the driver; apply a unified timestamp and spatial coordinates to the multimodal data to generate a spatiotemporal correlated dataset; S2: Select spatiotemporal related datasets to form a core feature set, combine vehicle operation characteristics to analyze and fuse weights, and generate optimized features based on the fusion weights; S3: Train an SVM classifier based on optimized features and corresponding category labels, train an LSTM model using local data from each vehicle, and analyze the MMD distance between the source and target domains. S4: Integrate multi-source evidence analysis to jointly establish trust levels and trigger anomaly warnings based on dynamic security assessments; S1 includes the following: S101: According to the cargo type label L type Select sensor combination; if L type =Liquid, enable pressure sensor array S p Collect pressure distribution ;where p n This represents the normalized measurement value of the nth pressure sensor, where n ∈ [1, N], N represents the total number of pressure sensors, and n is a positive integer; if L type =Fragile item, activate vibration sensor D h Collect vibration spectrum D h (t); S102: Analyze the signal-to-noise ratio (SNR) of the i-th sensor. If the SNR of the i-th sensor is less than the SNR threshold, switch to the backup sensor. After switching, compare the spectral correlation between the backup sensor and the main sensor in the overlapping frequency band. Apply a unified timestamp t to the multimodal data. j and spatial coordinates (x, y, z) j Generate a spatiotemporal correlation dataset X ST ={(t) j x j y j , z j , data j )};data j This represents the set of sensor measurements corresponding to the j-th spatiotemporal point; S4 includes the following: S401: For each source of evidence g∈{L obj P state F drift Let L be the basic probability assignment function; where L is the probability assignment function. obj P represents the classification label output by the SVM classifier. state F represents the probability of the vehicle's state predicted by the LSTM model. drift The detected distribution offset markers are represented; Eye represents the attention index collected by the eye tracker; the joint confidence level is calculated using the DS evidence theory. S402: Obtain the time interval t and corresponding accident rate R from historical accident data, and construct an exponential decay curve: R(t) = εexp(−κt) + μ, where ε represents the initial risk amplitude, κ represents the risk decay rate, and μ represents the baseline risk lower limit; fit the parameters ε, κ, and μ using the nonlinear least squares method; calculate the real-time risk threshold Risk. th (t)=R(t)+η∙std(R); η represents the system's preset sensitivity coefficient, and std(R) represents the standard deviation of the historical accident rate; if the real-time joint trust level is greater than the real-time risk threshold, an early warning is triggered, and the warning information is sent to the driver and the system administrator.

2. The method for safety detection and management of transport vehicles based on the Internet of Things according to claim 1, characterized in that: S2 contains the following: S201: For the spatiotemporal correlation dataset X ST Calculating the importance of random forest features using physical data I RF d and mutual information I MI d Where d represents the feature dimension number; Retain the condition that I RF d > α and I MI d The features of >β constitute the core feature set F core ={f d }, where α represents the feature importance threshold of the random forest, α=1 / d 1 / 2 β represents the mutual information threshold, and β is equal to the median of mutual information in the dataset; S202: Vehicle acceleration function a c (t) Convolution is performed with window lengths τ equal to τ1, τ2, and τ3, respectively, where τ1, τ2, and τ3 are preset constants, to obtain the local mutation amount M. local =Conv1D(a c , τ=S1), trend quantity M trend =Conv1D(a c , τ=S2) and event quantity M global =Conv1D(a c ,τ=S3); S203: Merge feature X=[F core M local M trend M global Project onto the query matrix Q, the key matrix K, and the value matrix V; Q=W q X, K=W K X, V=W V X; where W q W K and W V Represents the trainable parameter matrix; The query matrix Q, key matrix K, and value matrix V are split into A1, A2, ..., A1 based on the total number of attention heads M. M Subvectors; Calculate the matching score between the m-th header and all keys. m =A m ⊺ K; where ⊺ represents the transpose of the vector; analyze the fusion weight w m =exp(A m ⊺ K) / exp(A r ⊺ K); where exp(∙) represents an exponential function with the base of natural numbers, r∈[1,M], and r is a positive integer; Optimization feature F is generated based on fusion weights. opt = w m ⋅[F core M local M trend M global ].

3. The method for safety detection and management of transported goods based on the Internet of Things according to claim 2, characterized in that: S3 includes the following: S301: Using optimized feature F opt And train an SVM classifier with the corresponding category label b∈{-1,+1}; When the new sample F new Upon arrival, support vectors are filtered using KKT conditions, retaining the original support vectors and newly added boundary samples to construct a reduced set S'=S old ∪S new Solve the dual problem on the reduced set S' and update λ. s and c; S302: Train an LSTM model using the optimized feature matrix stored in the in-vehicle edge devices of each vehicle, and update the parameter set W. local The parameter set includes the weight matrices of the input gate, forget gate, output gate, and candidate memory units in the LSTM model. The locally trained parameters are encrypted and uploaded to the cloud server. Gaussian noise is added to the aggregation result on the cloud, with the standard deviation σ of the Gaussian noise satisfying relaxed difference privacy. The cloud-aggregated parameters W... global = W e local +N(0,σ 2 E represents the total number of vehicles; W represents the total number of vehicles. e local Represents the parameters of the e-th vehicle; N(0,σ) 2 ) indicates that the center is 0 and the variance σ 2 Gaussian normal distribution sampling; cloud-aggregated parameters are distributed to corresponding vehicles to replace the parameter set; S303: Denote the data distribution used during model training as the source domain and the new data distribution generated during actual model deployment as the target domain. Calculate the MMD distance between the source domain and the target domain. If the MMD distance is greater than the corresponding threshold, trigger model retraining.

4. An Internet of Things (IoT)-based cargo safety detection and management system, wherein the system is applied to the IoT-based cargo safety detection and management method described in any one of claims 1-3, characterized in that, The system includes a data acquisition and processing module, a feature extraction module, a model training module, and a dynamic early warning module; The data acquisition and processing module is used to select sensor combinations based on the type label of the transport object, and to collect multimodal sensing data of the transport object, vehicle and driver; Apply a uniform timestamp and spatial coordinates to multimodal data to generate a spatiotemporally correlated dataset; The feature extraction module is used to filter spatiotemporally related datasets to form a core feature set, combine vehicle operation characteristics to analyze and fuse weights, and generate optimized features based on the fusion weights. The model training module trains an SVM classifier based on optimized features and corresponding category labels, trains an LSTM model using local data from each vehicle, and analyzes the MMD distance between the source and target domains. The dynamic early warning module is used to integrate multi-source evidence analysis to jointly establish trust levels and trigger anomaly warnings based on dynamic security assessments.

5. The Internet of Things-based cargo safety detection and management system according to claim 4, characterized in that: The data acquisition and processing module includes a sensor selection unit and a data spatiotemporal correlation unit; The sensor selection unit is used to select a sensor combination based on the type label of the transported goods. If the type label is liquid, the pressure sensor array is activated to collect the pressure distribution; if the type label is fragile, the vibration sensor is activated to collect the vibration spectrum. The data spatiotemporal correlation unit applies a unified timestamp and spatial coordinates to multimodal data to generate a spatiotemporal correlation dataset.

6. The Internet of Things-based cargo safety detection and management system according to claim 4, characterized in that: The feature extraction module includes a feature analysis unit and a feature fusion unit; The feature analysis unit is used to calculate the importance and mutual information of random forest features on the physical data in the spatiotemporal correlation dataset, and retains features that satisfy the random forest feature importance and mutual information greater than the corresponding threshold to form the core feature set; the vehicle acceleration function is convolved with different window lengths to generate acceleration feature quantities. The feature fusion unit analyzes the fusion weights based on the fusion features and generates optimized features based on the fusion weights.

7. The Internet of Things-based cargo safety detection and management system according to claim 4, characterized in that: The model training module includes a data training unit and a offset analysis unit; The data training unit is used to train an SVM classifier using optimized features and corresponding category labels; and to train an LSTM model using local data from each vehicle. The offset analysis unit is used to denote the data distribution used in the model training phase as the source domain, the new data distribution generated when the model is actually deployed as the target domain, and to calculate the MMD distance between the source domain and the target domain.

8. The Internet of Things-based cargo safety detection and management system according to claim 4, characterized in that: The dynamic early warning module includes a risk analysis unit and a dynamic threshold early warning unit; The risk analysis unit is used to define a basic probability allocation function for each source of evidence and to calculate the joint confidence level using DS evidence theory. The dynamic threshold early warning unit is used to obtain the time interval and corresponding accident rate in historical accident data, construct an exponential decay curve, calculate the real-time risk threshold based on the exponential decay curve, and trigger an early warning if the real-time joint trust degree is greater than the real-time risk threshold, sending the early warning information to the driver and system administrator.

Images