A mine rescue multi-information life detection system and method

Abstract

The present invention relates to the field of mine rescue detection technology, and discloses a mine rescue multi-information life detection system and method. A mine rescue multi-information life detection method comprises: deploying a heterogeneous sensor network and applying a belief propagation algorithm to construct a self-organizing network to achieve dynamic topology optimization; constructing an environmental adaptability calibration system to achieve dynamic calibration of sensors in complex environmental changes; quantifying sensor performance through a functional space model and performing degradation evaluation to achieve intelligent reallocation of sensing tasks; utilizing a belief propagation algorithm to construct a collaborative detection framework to fuse multi-sensor data through a distributed decision algorithm; performing multi-dimensional feature extraction and analysis on life signals, and performing reliability evaluation on the results to output highly reliable detection results. The present invention solves the technical problems faced by traditional mine rescue detection systems in complex environments, such as sensor performance drift, functional degradation, and insufficient topological structure adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of mine rescue detection technology, and more particularly, to a mine rescue multi-information life detection system and method. Background Art

[0002] During mine rescue operations, accurate life detection of trapped personnel in complex and ever-changing environments is essential to inform effective decision-making. Existing mine rescue detection methods primarily rely on fixed sensor networks. These networks often face three core challenges in post-disaster environments: sensor performance drifts due to environmental factors (such as temperature, humidity, and air pressure); sensor functionality partially degrades after prolonged use or under extreme conditions; and the network topology struggles to dynamically adapt to post-disaster terrain changes. These issues severely limit the accuracy, reliability, and sustainability of life detection, hindering rescue efficiency.

[0003] In the existing technology, a single sensor array is often used for life signal detection, such as using seismic wave sensors to detect knocking signals, using infrared sensors to detect heat sources, and using gas sensors to monitor changes in carbon dioxide concentrations. However, these systems are usually based on preset fixed algorithms and threshold judgments, and cannot adapt to the complex and changing environment after a mine disaster. They also lack the ability to adaptively adjust to sensor performance degradation. For example, in high temperature and high humidity environments, sensor accuracy will drop significantly. Even in the case of most sensors failing, the system often cannot continue to provide effective detection support. In addition, the sensor network in the existing technology usually adopts a fixed topology structure, which is difficult to cope with situations such as terrain changes and node failures after a mine disaster. Once sensors in some areas are damaged or communications are interrupted, the detection capability of the entire network will be greatly reduced, and even lead to detection blind spots, making it impossible for trapped people to be discovered in time.

[0004] Therefore, there is a need for a multi-information life detection system and method that can adapt to the environment, self-reconstruct functions, and self-optimize topology in complex and changeable mine rescue environments, so as to improve the accuracy, reliability, and sustainability of life detection and provide more reliable technical support for rescue decisions. Summary of the Invention

[0005] The present invention provides a mine rescue multi-information life detection system and method, which solves the technical problems of sensor performance drift, function degradation and insufficient topological structure adaptability in related technologies.

[0006] The present invention provides a mine rescue multi-information life detection method, comprising the following steps:

[0007] Deploy heterogeneous sensor networks and apply belief propagation algorithms to build self-organizing networks and achieve dynamic topology optimization;

[0008] Based on the optimized network topology, an environmental adaptability calibration system is constructed to achieve dynamic calibration of sensors in complex environmental changes;

[0009] For calibrated sensors, the sensor performance is quantified and degradation assessment is performed using a functional space model to achieve intelligent reallocation of sensing tasks.

[0010] Based on the reallocated sensing tasks, a collaborative detection framework is constructed using the belief propagation algorithm to fuse multi-sensor data through a distributed decision-making algorithm.

[0011] The fused data is used to extract and analyze multi-dimensional features of life signals, and the results are evaluated for reliability to output highly reliable detection results.

[0012] In a preferred embodiment, in the step of deploying a heterogeneous sensor network and applying a belief propagation algorithm to construct a self-organizing network, the belief propagation algorithm is implemented as follows:

[0013] Each node sends information to its potential neighbor nodes, indicating its confidence in the neighbor node's state;

[0014] The node updates its belief based on all the neighbor information it receives;

[0015] By iteratively transmitting confidence information, the network gradually converges to a stable state and forms an optimal communication topology.

[0016] In a preferred embodiment, the step of constructing an environmental adaptability calibration system includes:

[0017] Deploy an environmental parameter monitoring subsystem to collect data on key environmental factors that affect sensor performance;

[0018] Construct a model of the impact of environmental parameters and sensor performance, and establish a mapping relationship between environmental factors and sensor performance drift;

[0019] Implement a distributed calibration algorithm to perform inter-sensor calibration using spatial correlation;

[0020] The calibration reference node group is dynamically selected based on environmental similarity to adapt to local environmental changes.

[0021] In a preferred embodiment, in the step of quantifying the sensor performance by a functional space model, the functional space model is defined as:

[0022] Quantitative evaluation models for the four core performance dimensions of measurement range, sensitivity, accuracy, and precision;

[0023] Each dimension receives a performance score between 0 and 1 through standardized testing;

[0024] The comprehensive health status index of the sensor is calculated through a weighted function.

[0025] In a preferred embodiment, the step of realizing intelligent reallocation of sensing tasks includes:

[0026] Establish a sensing task model and define the demand vectors for each functional dimension;

[0027] Calculate the matching degree between the remaining functions of the sensor and the mission requirements;

[0028] Use the matching function to evaluate the suitability of the task and the remaining functions of the sensor;

[0029] Construct a bipartite graph matching model and use the Hungarian algorithm to solve the optimal task allocation solution.

[0030] In a preferred embodiment, the step of constructing a collaborative detection framework using a belief propagation algorithm includes:

[0031] The detection area is divided into multiple sub-areas, each of which is monitored by multiple sensor nodes;

[0032] Define the node state vector to represent the node's detection status for different types of information;

[0033] Construct an inter-node compatibility function to quantify the consistency of the detection status of two nodes;

[0034] By iteratively updating information, the detection status of each node is optimized and the overall detection accuracy is improved.

[0035] In a preferred embodiment, the step of fusing the multi-sensor data through a distributed decision algorithm uses weighted Dempster-Shafer evidence theory to fuse multimodal features, including:

[0036] Extract features from each sensor data and convert them into support for each hypothesis set;

[0037] Define a trust function for each data source to measure the total support for the hypothesis;

[0038] Introduce weight vectors and apply weighted combination rules to perform feature fusion;

[0039] Generate fused detection results and their confidence scores.

[0040] In a preferred embodiment, the step of evaluating the reliability of the result includes:

[0041] Define a reliability evaluation index system, including signal strength, signal consistency, feature stability, and environmental interference;

[0042] Establish a reliability evaluation model based on evidence theory to calculate the overall reliability of the test results;

[0043] Implement a contextualized data credibility assessment model to calculate the credibility of the data source in the current environment;

[0044] Generate trapped personnel status information and reliability assessment report.

[0045] In a preferred embodiment, the mine rescue multi-information life detection method further includes an adaptive working mode switching step based on energy status:

[0046] Establish node energy status model to monitor battery charge, energy consumption rate and life expectancy;

[0047] Define multiple working modes, including high performance mode, standard mode, energy saving mode and sleep mode;

[0048] Design energy efficiency optimization functions to balance performance and energy consumption;

[0049] Based on the current energy status and detection requirements, the working mode of each node is dynamically adjusted.

[0050] In a preferred embodiment, a mine rescue multi-information life detection system is used to execute the above-mentioned mine rescue multi-information life detection method, including:

[0051] Heterogeneous sensor network self-organizing unit for dynamic topology optimization and node location self-awareness;

[0052] Environmental adaptability calibration unit, used to achieve dynamic parameter optimization of sensors in complex environmental changes;

[0053] Functional degradation perception and reconstruction unit, used to quantify sensor performance and achieve intelligent task reallocation;

[0054] Multi-point collaborative detection unit, used to achieve collaborative detection and distributed decision-making of multiple sensor nodes;

[0055] Life signal analysis and reliability assessment unit, used to generate highly reliable life detection results and reliability reports.

[0056] The beneficial effects of the present invention are:

[0057] Improved detection accuracy and reliability: Through a multi-point collaborative detection framework and multimodal signal fusion technology, combined with an environmental adaptability calibration system, the accuracy of life detection is improved and the false alarm rate is significantly reduced, providing more reliable data support for rescue decisions.

[0058] Enhanced environmental adaptability: By adopting self-organizing network construction methods and topology optimization technology, the system can adapt to the complex and changing environment after a mine disaster, improve detection coverage, and maintain effective life detection functions even when some nodes fail.

[0059] Extending effective working time: Through adaptive working mode switching technology based on energy status and intelligent reallocation of tasks after functional degradation, the system's energy efficiency is improved, and the continuous working time can be extended to meet the needs of long-term rescue.

[0060] Optimizing resource utilization efficiency: Sensor performance quantification technology and task adaptability evaluation models based on functional space models enable the system to accurately perceive sensor function degradation and fully utilize remaining functions, thereby improving system resource utilization efficiency.

[0061] Provide decision reliability assessment: Through the reliability assessment model based on evidence theory and the contextualized data credibility assessment model, the system quantitatively evaluates the reliability of the detection results and generates a detailed reliability assessment report to help rescue decision makers understand the credibility of the data and reduce decision risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of a mine rescue multi-information life detection method of the present invention; DETAILED DESCRIPTION

[0063] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0064] At least one embodiment of the present invention discloses a mine rescue multi-information life detection method, such as Figure 1 As shown, the following steps are included:

[0065] Step 1: deploy a heterogeneous sensor network and apply the belief propagation algorithm to build a self-organizing network to achieve dynamic topology optimization;

[0066] The specific steps include:

[0067] Step 1.1: Deploy a heterogeneous sensor network consisting of acoustic sensors, thermal imaging sensors, gas sensors, and electromagnetic sensors to collect multi-dimensional life signal data.

[0068] Among them, the acoustic sensor is used to collect sound wave signals and detect weak sounds of human activities;

[0069] Thermal imaging sensors collect thermal radiation data and locate areas with abnormal temperatures;

[0070] Gas sensors collect data on gas concentrations such as carbon dioxide and oxygen and detect respiratory activity;

[0071] Electromagnetic sensors collect data on changes in electromagnetic fields and detect human movement and heartbeats.

[0072] Step 1.2: A node communication system based on the belief propagation algorithm is used to enable each sensor node to build a self-organizing network by exchanging information.

[0073] The specific implementation of the belief propagation algorithm is as follows: Each node To its potential neighbor nodes Send a message , representing a node For Node state confidence level;

[0074] node Update its own beliefs based on all neighbor information received :

[0075] ;

[0076] in, Representation node About your own status The belief value of node Consider yourself to be in a state The probability estimate of For nodes The local evidence function represents the node's state based on its own sensor data Initial assessment; For nodes The neighbor set of node Nodes that communicate directly; is a normalization constant to ensure that the sum of belief probabilities is 1; For all neighbor nodes The information is multiplied.

[0077] By iteratively transmitting confidence information, the network gradually converges to a stable state and forms an optimal communication topology.

[0078] Step 1.3: Realize node location self-awareness and self-establishment of neighbor relationships;

[0079] The specific method is as follows:

[0080] Calculate the relative position of nodes based on the signal strength, propagation delay and multi-point triangulation algorithm between nodes;

[0081] Construct an initial neighbor relationship graph based on relative position information;

[0082] Dynamically adjust neighbor relationships based on communication quality and energy consumption.

[0083] Step 1.4: Use the topology optimization function to automatically adjust the network structure. The optimization function is expressed as: ;

[0084] in, Represents finding the objective function The topological structure that achieves the maximum value , is the optimal topology, indicating the network connection method finally selected by the system; is the set of all possible topological structures; For topology The coverage quality function is used to evaluate the network coverage effect on the target area; For topology Energy consumption function; Indicates the importance of balancing coverage quality and energy consumption. Coverage quality function By calculating the sensor coverage density and sensor data quality of key areas, the energy consumption function Then consider the node communication frequency, transmission distance and data volume.

[0085] Through the above steps, the system forms a heterogeneous sensor network with an adaptive topology, realizing multi-dimensional information collection and efficient transmission of complex environments, laying the foundation for subsequent sensor calibration and life signal detection.

[0086] Step 2: Based on the optimized network topology, an environmental adaptability calibration system is constructed to achieve dynamic calibration of sensors in complex environmental changes.

[0087] The specific steps include:

[0088] Step 2.1, deploy the environmental parameter monitoring subsystem;

[0089] Collect data on key environmental factors that affect sensor performance:

[0090] The temperature sensor collects ambient temperature data;

[0091] The humidity sensor collects environmental humidity data;

[0092] The air pressure sensor collects ambient air pressure data;

[0093] The gas concentration sensor collects gas concentration data such as methane and carbon monoxide in the environment.

[0094] Step 2.2: Construct a model of environmental parameters and sensor performance impact, and establish a mapping relationship between environmental factors and sensor performance drift;

[0095] The model of environmental parameters and sensor performance influence adopts a hierarchical Bayesian network structure, which can be expressed as:

[0096] ;

[0097] in, Indicates that the environmental parameters are known Sensor performance parameters under the conditions The posterior probability distribution of ; Indicates that the sensor performance parameters are known Environmental parameters were observed under the conditions of The likelihood probability of ; Indicates sensor performance parameters The prior probability distribution of , which is the initial estimate of sensor performance before observing environmental parameters; Indicates environmental parameters The marginal probability distribution of , acts as a normalization factor to ensure that the posterior probabilities sum to 1.

[0098] The specific implementation steps are as follows:

[0099] Collect a large amount of standard performance data of sensors under different environmental conditions and build a training data set;

[0100] Learning conditional probability distributions using maximum likelihood estimation , that is, the probability of observing specific environmental parameters when the sensor performance parameters are known;

[0101] Determine the prior distribution of sensor performance parameters by combining prior knowledge , reflecting the estimation of the initial performance state of the sensor;

[0102] The posterior distribution of sensor performance parameters under given environmental conditions is calculated by Bayesian inference, that is, , as the theoretical basis for sensor calibration.

[0103] The detailed implementation of the hierarchical Bayesian network structure is as follows:

[0104] The first layer (environmental parameter layer): contains environmental variable nodes such as temperature, humidity, and air pressure. Each node represents an environmental parameter.

[0105] The second layer (intermediate association layer): contains nodes for the combined effects of environmental factors, capturing the interactive effects of multiple environmental factors;

[0106] The third layer (sensor parameter layer): contains sensor performance parameter nodes such as sensitivity, accuracy, and precision.

[0107] The conditional probability table in the network is constructed in the following way:

[0108] Use expert knowledge to initialize the dependencies between nodes and determine the network topology;

[0109] Map the collected training data into the network structure and calculate the conditional probability between each node;

[0110] Apply the expectation maximization (EM) algorithm to optimize the conditional probability parameters;

[0111] The network performance is evaluated through cross-validation and the network structure is optimized iteratively.

[0112] In the mine rescue scenario, the following example shows the application of the hierarchical Bayesian network:

[0113] When a sudden temperature change (from 25°C to 40°C) and an increase in humidity (from 50% to 85%) are detected in a specific area, the system automatically infers that the acoustic sensor sensitivity may decrease by 18% and the transmission delay may increase by 250ms;

[0114] Based on the inference results, the system pre-adjusts the acoustic sensor's amplification and signal processing parameters to maintain data acquisition quality;

[0115] At the same time, the system increases the sampling frequency of the thermal imaging sensor to supplement the acoustic data that may be affected and ensure detection performance.

[0116] Step 2.3: Implement a distributed calibration algorithm to perform inter-sensor calibration using spatial correlation.

[0117] The sensor network is divided into multiple calibration groups, each group contains multiple spatially adjacent sensors;

[0118] In each calibration group, the sensor with the most stable environmental conditions is selected as the reference node;

[0119] Other nodes calculate calibration parameters by comparing data with the reference node;

[0120] The calibration parameters are optimized via Bayesian estimation:

[0121] ;

[0122] in, Represents the sensor calibration parameter vector, including parameters that need to be adjusted, such as gain and offset; represents the observation data set containing the reference node and the node to be calibrated, is the likelihood function, which means that given the calibration parameters The data were observed under the condition probability; is the parameter prior distribution, which represents the initial belief distribution of the calibration parameters before observing the data; is the posterior probability, which means that when the data is observed Post-calibration parameters Renewal of faith; Indicates a direct proportional relationship.

[0123] Step 2.4, dynamically select the calibration reference node group based on environmental similarity to adapt to local environmental changes;

[0124] Calculate the similarity between current environment parameters and historical environment parameters:

[0125] ;

[0126] in, Indicates the similarity between the current environment parameters and the historical environment parameters, is the current environment parameter vector, is the historical environment parameter vector, represents the square of the Euclidean distance between two environmental parameter vectors, that is, the sum of the squares of the differences between the parameters. It is the scaling parameter for the environment similarity calculation, which controls the sensitivity of the similarity function. The value makes the similarity more sensitive to environmental changes. is an exponential function;

[0127] Select the node group that is most suitable as a calibration reference based on the environmental similarity and sensor status score;

[0128] When the environmental change exceeds the threshold, the process of reselecting the reference node group is automatically triggered.

[0129] Through the above steps, the system can monitor changes in environmental parameters in real time, accurately establish a relationship model between environmental factors and sensor performance drift, and dynamically optimize sensor parameters through a distributed calibration algorithm, effectively overcoming the problem of sensor performance drift caused by environmental changes and improving the accuracy of data collection.

[0130] Step 3: For the calibrated sensors, the sensor performance is quantified and degradation assessment is performed using the functional space model to achieve intelligent reallocation of sensing tasks.

[0131] The specific steps include:

[0132] Step 3.1: Build a sensor functional space model to comprehensively quantify sensor performance;

[0133] Define the function space model:

[0134] ;

[0135] in, represents the functional space model, Indicates the measurement range, which indicates the range of physical quantities that the sensor can measure; It represents sensitivity, which indicates how responsive the sensor is to changes in input; It represents accuracy, which indicates how close the measured value is to the true value; Indicates accuracy and the repeatability of measurement results.

[0136] For different types of sensors, corresponding functional space parameter evaluation methods are established.

[0137] The detailed implementation of the functional space model is as follows:

[0138] Constructing a multi-dimensional performance feature vector: For each sensor type, establish a feature vector containing 4-8 core performance dimensions based on its operating principle and characteristics. For example, acoustic sensors can add frequency response range and signal-to-noise ratio dimensions.

[0139] Establish performance quantification standards: standardize the performance parameters of each dimension to a value between 0 and 1, where 1 represents an ideal state and 0 represents complete failure;

[0140] Implement inter-dimensional correlation analysis: Calculate the mutual influence between performance dimensions through the covariance matrix to identify key performance bottlenecks;

[0141] Construct a comprehensive performance scoring function:

[0142] ;

[0143] in, The weight of each dimension indicates the importance of each performance dimension in the overall score. The value range is 0-1, and the sum of all weights is 1. It is dynamically adjusted according to the task type. For the The standardized score of each performance dimension indicates the performance level of that dimension, with a value range of 0-1, where 1 indicates the best performance and 0 indicates complete failure. This is the overall performance score of the sensor, with a value range of 0-1. The closer to 1, the better the overall performance of the sensor.

[0144] Example of application of functional space model in mine rescue environment:

[0145] For gas sensors, the system automatically identifies its sensitivity in high humidity environments Reduced to 0.72 (standard value 0.95), while the measurement range and precision remained essentially normal (0.92 and 0.88, respectively);

[0146] The system determines that the sensor is still suitable for detecting gas concentration trends, but is not suitable for accurately measuring absolute concentration values;

[0147] Based on this, the system adjusted the sensor's task to monitor breathing patterns instead of measuring absolute carbon dioxide concentration, while increasing its sampling frequency to compensate for the decreased sensitivity;

[0148] The system also configures adjacent thermal imaging sensors as auxiliary verification means to form a complementary detection strategy.

[0149] Step 3.2: Realize real-time monitoring and degradation assessment of sensor performance;

[0150] Perform self-tests regularly to collect sensor performance data;

[0151] Compare with the baseline performance and calculate the performance degradation vector:

[0152] ;

[0153] in, Represents the current performance vector of the sensor, including current performance parameters such as measurement range, sensitivity, accuracy and precision; Represents the sensor's benchmark performance vector, which includes the standard performance parameters after factory or calibration. represents the performance degradation vector, which indicates the degree of degradation of each performance dimension relative to the baseline state;

[0154] The principal component analysis (PCA) method is used to map the multidimensional degradation vector to a significant feature space and extract key degradation features. This method identifies the main performance dimensions that contribute most to the degradation state by calculating the eigenvalues ​​and eigenvectors of the covariance matrix of the degradation data.

[0155] Calculate the sensor's health status index based on the degradation characteristics:

[0156] ;

[0157] in The Euclidean norm of the degradation vector, which is the square root of the sum of the squares of the components of the degradation vector, is used to quantify the overall degradation degree. represents the Euclidean norm of the benchmark performance vector, that is, the square root of the sum of the squares of the components of the benchmark performance vector, which is used to standardize the degree of degradation. Indicates the sensor health status index, with a value range of 0-1. Values ​​closer to 1 indicate healthier sensors, and values ​​closer to 0 indicate more severe sensor performance degradation.

[0158] Step 3.3, accurately locate the dimension and degree of functional degradation;

[0159] For degenerate vector Normalize the components of to obtain the normalized degradation vector:

[0160] ;

[0161] in, represents the normalized degradation vector, 、 、 、 Represents the normalized degradation values ​​of the measurement range dimension, sensitivity dimension, accuracy dimension, and precision dimension respectively. The value range is 0-1. The larger the value, the more serious the dimension degradation.

[0162] Set the degradation threshold vector for each dimension:

[0163] ;

[0164] in, represents the degradation threshold vector of each dimension, 、 、 、 They represent the maximum acceptable degradation thresholds of the measurement range dimension, sensitivity dimension, accuracy dimension, and precision dimension, respectively. Exceeding this value will affect the normal function of the sensor;

[0165] Will and Perform component comparison and identify dimensions exceeding a threshold as the main degraded dimensions;

[0166] Calculate the degradation percentage of each dimension , represents the degradation percentage of each performance dimension relative to its threshold, which is used to intuitively evaluate the severity of degradation, where Indicates the percentage of degradation of each dimension, represents the normalized degradation vector, Represents the degradation threshold vector for each dimension.

[0167] Step 3.4: Dynamically reallocate sensing tasks using the task adaptability evaluation model.

[0168] Building a sensing task model ,in, 、 、 Respectively represent 、 、 A sensing task, represents the number of sensing tasks;

[0169] Each task Contains the demand vector for each functional dimension ,in, 、 、 They represent the minimum required values ​​of the sensor's measurement range, sensitivity, accuracy, and precision, respectively;

[0170] Calculating sensor residual capacity ,in, The sensor's benchmark performance vector includes the standard performance parameters after factory or calibration. represents the performance degradation vector, which indicates the degree of degradation of each performance dimension relative to the baseline state; A vector representing the sensor's remaining functional performance;

[0171] The compatibility function is used to evaluate the suitability of the task with the remaining functions of the sensor:

[0172] ;

[0173] in, Indicates a task With sensor The matching degree is in the range of 0-1, and the closer the value is to 1, the higher the matching degree is; is the weight coefficient of each functional dimension, which indicates the importance of different performance dimensions to task completion. ; Indicates a task Functional Dimension Demand value; Indicates sensor In the functional dimension The remaining performance value on It means taking the smaller value between the mission requirement and the remaining performance of the sensor to ensure that the mission requirement is not exceeded; Indicates the proportion of the sensor's remaining performance that meets mission requirements;

[0174] Build a bipartite graph matching model and use the Hungarian algorithm to solve the optimal task allocation solution to maximize the overall matching degree ,in, It is a 0-1 variable. When the value is 1, it means that the task Assigned to sensor , when the value is 0, it means no allocation; It represents the overall matching degree of the entire system and is the optimization objective function.

[0175] Through the above steps, the system can perceive the degradation of sensor functions in real time, accurately locate the degradation dimension and degree, and realize intelligent redistribution of sensing tasks based on the task adaptability evaluation model, making full use of the remaining functions of the sensors and ensuring the stability and reliability of the system during long-term operation.

[0176] Step 4: Based on the reallocated sensing tasks, a collaborative detection framework is constructed using the belief propagation algorithm to fuse the multi-sensor data through a distributed decision-making algorithm.

[0177] The specific steps include:

[0178] Step 4.1: Build a multi-node collaborative detection framework and use the belief propagation algorithm to achieve information sharing and collaboration;

[0179] The detection area is divided into multiple sub-areas, each of which is monitored by multiple sensor nodes;

[0180] Define the node state vector ,in, Represents a sensor node The state vector contains the node's detection status for various types of information; 、 、 Represents nodes respectively For the first 、 、 The detection status of the class information; Indicates the total number of information categories that the system can detect;

[0181] Constructing an inter-node collaboration model based on the belief propagation algorithm:

[0182] ;

[0183] in, Represents a slave node Pass to node Message containing node For Node Inferred information about the state; Indicates the node The sum operation of all possible states; Representation node and nodes The compatibility function between them quantifies the consistency of the detection states of two nodes; Representation node The set of neighbor nodes of Representation node Exclude nodes from the neighbor node set The collection after Representation node From the The product of the messages received by all neighbor nodes outside the node is a comprehensive inference of state;

[0184] By iteratively updating information, the detection status of each node is optimized and the overall detection accuracy is improved.

[0185] The detailed implementation of the multi-node collaborative detection framework is as follows:

[0186] Constructing factor graph representation: Modeling the sensor network as a bipartite graph structure, where variable nodes represent sensor states and factor nodes represent the interaction relationships between sensors;

[0187] Define the message passing protocol: design two types of message formats, variable-to-factor messages and factor-to-variable messages, which contain state estimates and confidence information;

[0188] Implement adaptive message scheduling: dynamically adjust the message delivery order and frequency based on node importance, message novelty, and energy status;

[0189] Design convergence judgment criteria: define a confidence change threshold, and the algorithm is considered to have converged when the confidence change in consecutive iterations is less than the threshold.

[0190] Application examples of collaborative detection framework based on belief propagation algorithm:

[0191] In the mine collapse area, a set of acoustic sensors detected a faint knocking sound (confidence level 0.65), but this information alone could not determine whether it was made by trapped people;

[0192] The system initiated a collaborative detection process, and the adjacent thermal imaging sensor reported the detection of a possible human heat source (confidence level 0.58), but there was also uncertainty;

[0193] The gas sensor detected periodic changes in carbon dioxide concentration in the area (confidence level 0.72), which is consistent with human breathing patterns;

[0194] By using the belief propagation algorithm to perform state fusion, the three sensor data mutually verified each other, and the final comprehensive confidence level was increased to 0.91, and the system confirmed that there was a trapped person at the location;

[0195] At the same time, the system recognizes the synchronization between the acoustic signal and the breathing pattern, infers that the trapped person is awake, and provides the precise location coordinates to the rescuers.

[0196] Step 4.2: realize life signal feature extraction and multimodal fusion;

[0197] Apply time-frequency analysis and pattern recognition algorithms to acoustic signals to extract the sound characteristics of human activities;

[0198] Apply thermal anomaly detection and morphological analysis to thermal imaging data to identify human thermal radiation characteristics;

[0199] Apply fluctuation analysis and trend identification to gas concentration data to extract respiratory activity characteristics;

[0200] Apply spectrum analysis and pattern matching to electromagnetic field data to extract heartbeat and motion features;

[0201] The weighted Dempster-Shafer evidence theory is used to fuse multimodal features: ;

[0202] in, Represents the hypothesis set after fusion The basic probability distribution of Indicates the sensor data pair hypothesis set The basic probability distribution function of ; Represents all hypothesis sets The intersection of is equal to the hypothesis set ; Represents all hypothesis sets The intersection of is the empty set; represents the product of the basic probability distribution of all sensors; Represents the Dempster combination rule operator.

[0203] The detailed implementation of weighted Dempster-Shafer evidence theory multimodal fusion is as follows:

[0204] Identification and judgment framework construction:

[0205] Define a complete set of mutually exclusive hypotheses ,For example , and the corresponding power set ,in, represents a complete set of mutually exclusive hypotheses, 、 、 Respectively represent 、 、 Different assumptions; represents the total number of hypotheses;

[0206] Basic probability distribution calculation: For each type of sensor data, a specific feature probability mapping function is designed to convert the extracted features into the support for each hypothesis set;

[0207] Trust function construction: For each data source , define the trust function , measure the hypothesis The total support of Represents a data source For the hypothesis set The trust function value of ; Represents a data source For the hypothetical subset The basic probability distribution value of ; Represents the hypothesis set All subsets of Perform sum operation;

[0208] Weighted fusion implementation:

[0209] Introducing weight vector ,in, 、 、 Respectively represent 、 The weight of each sensor data source, Indicates the number of sensor data sources;

[0210] Apply the weighted combination rule:

[0211] ;

[0212] in, Represents the weighted fusion of the hypothesis set The basic probability distribution of Indicates the The weight of each sensor data source; Indicates that the The basic probability distribution value of each sensor is weighted exponentially calculated; Represents all hypothesis sets The intersection of is equal to the hypothesis set ; Represents all hypothesis sets The intersection of is the empty set; It represents the product of the weighted basic probability distribution of all sensors, reflecting the degree to which multiple sensors jointly support a hypothesis.

[0213] Examples of multimodal fusion applications in mine rescue environments:

[0214] At a mining accident site, the system simultaneously collected data from four types of sensors for life signal detection:

[0215] The acoustic sensor detects rhythmic tapping sounds, and its basic probability distribution is: , , ;

[0216] The thermal imaging sensor detects a weak heat source, and its basic probability distribution is: , , ;

[0217] The gas sensor detects periodic carbon dioxide fluctuations, and its basic probability distribution is: , , ;

[0218] The electromagnetic sensor does not detect any significant signal, and the basic probability distribution is: , , .

[0219] The system assigns weight vectors according to the current environmental conditions and the status of each sensor ;

[0220] Apply the weighted Dempster-Shafer fusion algorithm to obtain the basic probability distribution after fusion: , , ;

[0221] Based on this, the system confirms that there is a high possibility that there are trapped people at that location and initiates a directional rescue procedure.

[0222] Step 4.3: Implement a distributed decision-making algorithm and adaptively adjust the detection strategy;

[0223] Establish a hierarchical decision-making structure, including node level, regional level and global level;

[0224] Node-level decisions are based on preliminary judgments based on local sensor data and confidence levels;

[0225] Regional-level decision-making integrates the judgment results of multiple nodes in the same area;

[0226] The global layer decision integrates the decision results of multiple regions to form the final life signal detection conclusion;

[0227] Adopting an adaptive strategy optimization model based on Markov decision process (MDP):

[0228] ;

[0229] in, Indicates status The optimal decision strategy under Indicates the current state of the system; Indicates possible choices of action; Indicates execution of an action The next state to which the system may transition; Indicates that the status Take action After transfer to state probability; Indicates the slave state Take action Transfer to state Instant rewards received; is a discount factor (ranging from 0 to 1) that balances the importance of immediate rewards and long-term gains; Indicates status The optimal value function represents the optimal value function from the state The maximum cumulative reward that can be obtained at the beginning; Indicates selecting an action that maximizes the following expression .

[0230] Step 4.4, realize adaptive working mode switching based on energy status;

[0231] Establish node energy status model to monitor battery charge, energy consumption rate and life expectancy;

[0232] Define multiple working modes, including high performance mode, standard mode, energy saving mode and sleep mode;

[0233] Design energy efficiency optimization function:

[0234] ;

[0235] in, is the optimal working mode, that is, the operating mode with the highest energy efficiency that the system should choose; is the set of all possible working modes; Representing a collection A specific working mode in For mode The performance function quantifies the detection performance and response capability of the system in this mode; For mode The energy consumption function represents the energy consumption rate of the system in this mode; It is a trade-off coefficient used to balance the weight relationship between performance and energy consumption, reflecting the current task's emphasis on performance and endurance. Indicates that the working mode is selected so that the following expression can obtain the maximum value .

[0236] Based on the current energy status and detection requirements, the working mode of each node is dynamically adjusted to maximize energy efficiency.

[0237] Through the above steps, the system realizes the collaborative detection and distributed decision-making of multiple sensor nodes, can effectively integrate multiple life signal characteristics, adaptively adjust the detection strategy, and optimize the working mode according to the energy status, thereby improving the detection accuracy while extending the system working time and enhancing the system's adaptability in complex environments.

[0238] Step 5: Extract and analyze the multi-dimensional features of life signals from the fused data, conduct reliability assessment on the results, and output a highly reliable detection result.

[0239] The specific steps include:

[0240] Step 5.1: Extract and analyze multi-dimensional features of life signals:

[0241] Apply multi-scale time-frequency analysis to the raw sensor data to extract time domain, frequency domain and time-frequency domain features;

[0242] Principal component analysis and independent component analysis are used on acoustic features to separate possible acoustic features of human activities;

[0243] A thermal target tracking algorithm is used for thermal imaging features to identify thermal radiation sources with human characteristics;

[0244] Apply periodicity analysis and fluctuation pattern recognition to gas concentration characteristics to extract breathing characteristics;

[0245] Spectrum analysis and wavelet transform are used to extract the characteristics of the heartbeat signal from the electromagnetic field characteristics;

[0246] Constructing feature vectors ,in, A comprehensive feature vector representing a vital signal; 、 、 Respectively represent 、 、 Life signal characteristics; represents the total number of features;

[0247] Each feature All of them have been standardized to facilitate subsequent pattern recognition and classification analysis.

[0248] Step 5.2, apply the reliability assessment model to quantify the confidence of the test results;

[0249] Define a reliability evaluation index system, including signal strength, signal consistency, feature stability, environmental interference, and other dimensions;

[0250] Establish a reliability assessment model based on evidence theory:

[0251] ;

[0252] in, Indicates test results Overall reliability rating of Indicates the vital signal detection result to be evaluated; Indicates the total number of evaluation indicators; Indicates the Detection results under evaluation indicators The reliability score ranges from 0 to 1; Indicates the The weight coefficient of each evaluation indicator;

[0253] Use historical data and expert knowledge to determine the scoring criteria and weight coefficients for each indicator;

[0254] The test results are divided into three levels according to the reliability score: high confidence, medium confidence and low confidence.

[0255] The detailed implementation of the reliability assessment model based on evidence theory is as follows:

[0256] Constructing a multi-layer evaluation indicator system:

[0257] First layer: overall reliability;

[0258] The second layer includes four dimensions: data quality, model applicability, environmental impact, and result consistency;

[0259] The third layer: specific evaluation indicators under each dimension, a total of 12-15 detailed indicators;

[0260] Design scoring function mapping: Design a specific scoring function for each refined indicator to convert the observed value into a standardized score in the range of 0-1;

[0261] Implement AHP (Analytical Hierarchy Process) weight determination: through expert judgment and pairwise comparison, construct a judgment matrix and calculate the weight of indicators at each level;

[0262] Establish an evidence accumulation process: introduce the DS theoretical framework, use the scores of each indicator as supporting evidence, and accumulate them through combination rules to form an overall reliability assessment.

[0263] Application examples of reliability assessment models based on evidence theory:

[0264] After a suspected life signal was detected at a mining accident site, the system initiated a reliability assessment process:

[0265] Data quality dimension assessment (weight 0.30):

[0266] Signal strength index: The average signal-to-noise ratio of multiple points is 8.3dB, corresponding to a score of 0.76;

[0267] Data integrity indicator: sensor coverage rate is 92%, corresponding to a score of 0.85;

[0268] Sampling adequacy index: the sample size is sufficient, corresponding to a score of 0.90;

[0269] Data quality dimension weighted score: 0.83;

[0270] Model applicability dimension evaluation (weight 0.25):

[0271] Model environment matching: The current dust concentration affects the model performance, corresponding to a score of 0.72;

[0272] Parameter fitness: Automatically adjust parameters to achieve the optimal value, corresponding to a score of 0.88;

[0273] Weighted score of model applicability dimension: 0.80;

[0274] Environmental impact dimension assessment (weight 0.25):

[0275] Interference source strength: Equipment vibration interference exists, corresponding to a score of 0.65;

[0276] Environmental stability: large temperature fluctuations, corresponding to a score of 0.70;

[0277] Weighted score of environmental impact dimension: 0.68;

[0278] Result consistency dimension evaluation (weight 0.20):

[0279] Multi-source consistency: The four sensor data support each other, with a corresponding score of 0.92;

[0280] Temporal consistency: The signal is continuously and stably detected for 20 minutes, corresponding to a score of 0.88;

[0281] The weighted score of the result consistency dimension was 0.90;

[0282] Overall reliability calculation:

[0283] ;

[0284] The system classifies this detection result as "high confidence" (threshold is 0.75) and prioritizes the allocation of rescue resources.

[0285] Step 5.3: Implement a contextualized data credibility assessment model to calculate the credibility of the data source in the current environment;

[0286] Building an environmental scenario model , including environmental parameters such as temperature, humidity, air pressure, and noise level;

[0287] Collect credibility annotation data of data sources in different environmental scenarios and build training sets ;

[0288] The context similarity weighted algorithm is used to calculate the credibility of the data source in the current environment:

[0289] ;

[0290] in, Represents the environment Download data source credibility; Indicates the data sources; Represents the current environmental situation model; Indicates the first an environmental situation; represents the total number of historical environmental situation samples; Indicates environment = and historical environment The similarity ranges from 0 to 1, and the larger the value, the more similar the two environments are; In historical context Download data source known credibility; is the normalized weight coefficient, which is used to balance the importance of different historical environment samples;

[0291] Comprehensively evaluate the credibility of multiple data sources to form a final credibility report of the test results.

[0292] Step 5.4, generating trapped personnel status information and reliability assessment report;

[0293] Determine the location coordinates, number and physiological status of trapped people based on multi-dimensional life signal characteristics;

[0294] The location information is calculated using triangulation and signal strength weighted average algorithm, and the positioning error estimate is given;

[0295] The number of people estimated was determined through cluster analysis and multi-source information verification, and confidence intervals for the estimates were provided;

[0296] Physiological status analysis includes assessment of activity status (resting / moving), consciousness status (awake / comatose), and vital signs (normal / abnormal);

[0297] Generate a structured reliability assessment report containing the following:

[0298] The overall reliability score and confidence level of the test results;

[0299] An assessment of the credibility of each data source and its applicability in the current context;

[0300] Analysis of uncertainties that may affect test results;

[0301] Suggestions for improving the results, such as the need for additional sensor data or adjustment of detection parameters.

[0302] Through the above steps, the system can conduct a comprehensive analysis of the collected multi-dimensional life signal data, accurately locate the position of trapped people, assess their number and physiological status, and conduct reliability assessments on the detection results, providing highly reliable life detection results and detailed reliability reports, providing a scientific basis for rescue decisions.

[0303] Application examples of this embodiment:

[0304] This implementation was applied during the rescue process of a coal mine collapse accident in Shanxi Province. The mine was approximately 580 meters deep, and the collapsed area covered approximately 1,200 square meters. The geological structure was complex, and the collapse created multiple enclosed spaces. At the time of the accident, 23 miners were trapped. The collapse disrupted the communication system, partially damaged the ventilation system, and caused flooding in some areas, creating an extremely complex rescue environment. The main technical challenges faced by the rescue included:

[0305] Harsh environmental conditions: On-site temperature fluctuations (20-42°C), extremely high humidity (85-98%), and the presence of large amounts of dust and harmful gases severely limited the performance of traditional sensing equipment;

[0306] Changes in terrain structure: The channels formed after the collapse are narrow and unstable, with some areas only 40-60 cm wide, making it difficult to deploy conventional fixed networks;

[0307] Long-term rescue requirements: The rescue time is estimated to be 72-96 hours, and the sensing equipment needs to work stably for a long time in harsh environments.

[0308] In this rescue mission, a multi-information life detection system of this embodiment was deployed, which includes 85 heterogeneous sensor nodes, covering the collapsed area and surrounding areas where there may be living space, in order to accurately locate the trapped people and assess their physiological status, providing a decision-making basis for scientific rescue.

[0309] Implementation example:

[0310] Multi-dimensional information collection and network self-organization implementation process:

[0311] Within 2 hours of the start of the rescue operation, the rescue team deployed the heterogeneous sensor network shown in Table 1. Sensors were deployed to different locations in the collapse area through rescue channels, drill holes, and robots.

[0312] Table 1: Composition of heterogeneous sensor nodes deployed at the rescue site;

[0313]

[0314] After deployment, the system automatically initiated network self-organization. Initially, due to terrain constraints and signal attenuation, only 63% of nodes successfully established communication connections. The system then optimized the network topology using a belief propagation algorithm. After 27 minutes and 139 iterations, the network topology performance was achieved, as shown in Table 2.

[0315] Table 2: Performance comparison before and after network topology optimization;

[0316]

[0317] During the topology optimization process, the system also automatically completed node position perception, with an average error of 3.8 meters and a maximum error of no more than 7.2 meters, meeting the needs of life detection and positioning.

[0318] In practical applications, the self-organizing network demonstrated excellent adaptability. During the 28th hour of the rescue operation, a small secondary collapse caused nine sensor nodes to fail and 17 nodes to relocate. The system automatically reconfigured the network within eight minutes, reestablishing a stable communication topology and ensuring continuous data collection, avoiding the widespread paralysis that would have occurred in traditional fixed networks under such circumstances.

[0319] Environmental parameter monitoring and sensor dynamic calibration implementation process:

[0320] Environmental parameters at the rescue site fluctuated significantly, particularly in temperature, humidity, and gas concentrations. For example, the changes in environmental parameters over a 48-hour period in Area 12 (one of the most important potential survival areas) are shown in Table 3.

[0321] Table 3: Changes in environmental parameters in area 12 within 48 hours;

[0322]

[0323] These environmental changes cause significant drift in sensor performance. For example, acoustic sensors experience an average 23% decrease in sensitivity in high-temperature and high-humidity environments (temperature > 38°C, humidity > 92%), while electromagnetic sensors experience a 17% decrease in penetration in environments with high methane concentrations (> 1.8%).

[0324] The system effectively addressed these challenges through a dynamic calibration process. A hierarchical Bayesian network calculated the impact of environmental factors on sensor performance in real time and optimized its parameters. Upon first detecting an area of ​​high temperature and humidity, the system automatically adjusted the acoustic sensor parameters, including increasing gain and adjusting filter parameters. Table 4 shows a comparison of acoustic sensor performance in the same area before and after calibration.

[0325] Table 4: Comparison of acoustic sensor performance before and after calibration under environmental changes;

[0326]

[0327] The system dynamically selects calibration reference nodes based on environmental similarity. During the rescue, the calibration reference node group in Area 12 was dynamically changed eight times to ensure calibration accuracy. This approach enabled the system to maintain over 85% sensing performance even in extreme environmental conditions, significantly exceeding the approximately 60% level of conventional systems.

[0328] Sensor function degradation perception and capability reconstruction implementation process:

[0329] During the long rescue process, multiple sensor nodes experienced varying degrees of functional degradation. The system uses a functional space model to assess sensor health. Table 5 shows the statistical degradation of various sensor functions during the rescue process.

[0330] Table 5: Statistics of sensor function degradation during the 72-hour rescue process;

[0331]

[0332] The system uses a functional space model to accurately identify the functional degradation dimension and degree of each sensor. Taking thermal imaging sensor No. 5 as an example, its functional degradation dimension analysis is shown in Table 6.

[0333] Table 6: Analysis of the functional degradation dimension of thermal imaging sensor No. 5 (after 72 hours);

[0334]

[0335] Based on functional degradation assessments, the system adjusted its sensor task allocation strategy in real time. For thermal imaging sensor No. 5, due to a significant decrease in sensitivity while maintaining a relatively stable measurement range, the system reassigned it to screening for wide-area thermal anomalies, removing its use for precise temperature measurement. Simultaneously, the system increased the sampling frequency of other high-performing sensors in the same area to ensure data quality.

[0336] Through this intelligent task allocation strategy, the system still maintained an overall detection capability of 87% even when more than 25% of the nodes were severely degraded or failed, ensuring the continuation of the rescue operation.

[0337] Multi-point collaborative detection and distributed decision-making implementation process:

[0338] During the rescue operation, multi-point collaborative detection and a distributed decision-making framework played a key role in accurately identifying life signals. During the 18th hour of the rescue, acoustic sensors in area C3 detected a faint knocking sound, but the signal-to-noise ratio was only 4.2 dB, making it difficult to determine whether this signal alone was human activity. The system initiated a collaborative detection process based on the belief propagation algorithm, integrating information from multiple surrounding sensors. The detailed process is shown in Table 7.

[0339] Table 7: Co-detection process and confidence changes in the C3 region;

[0340]

[0341] Based on multimodal fusion of the Dempster-Shafer evidence theory, the system successfully integrated multiple low-confidence single-point detections (average confidence of approximately 0.56) into a high-confidence comprehensive judgment (0.94), and accurately located the position of the trapped person with an error of less than 4.8 meters.

[0342] In terms of energy management, the system dynamically adjusts its operating mode based on the node power status and task priority. Table 8 shows the system's energy consumption and operating time during the entire rescue process.

[0343] Table 8: Energy consumption and working time in different working modes;

[0344]

[0345] By switching to an adaptive working mode, the system's overall energy efficiency has increased by approximately 280%, allowing most sensors to continue working until the rescue is completed, while the traditional fixed working mode system is expected to only work for 25-30 hours.

[0346] Implementation process of comprehensive analysis and reliability assessment of life signals:

[0347] Throughout the rescue process, the system detected 18 possible life signals, 14 of which were confirmed to be trapped personnel, 3 were false positives (caused by equipment vibration or animal activity), and 1 was unconfirmed. The system performed a reliability assessment on each detection result. Table 9 shows the reliability scores and final results for each detection location.

[0348] Table 9: Life signal detection location and reliability evaluation (partial);

[0349]

[0350] Using a reliability assessment model based on evidence theory, the system successfully categorized all false positive detections as "low confidence" and all true life signals as "medium confidence" or "high confidence." This enabled the rescue team to allocate resources rationally, prioritizing high-confidence areas.

[0351] The system also uses a contextualized data credibility assessment model to dynamically evaluate the credibility of each data source based on current environmental conditions. For example, in areas of high temperature and humidity, the system automatically reduces the credibility weight of thermal imaging sensors and increases the weight of acoustic and gas sensors, resulting in more accurate detection results.

[0352] Ultimately, the system generated a comprehensive assessment report encompassing location, number of people, and physiological status, providing precise guidance for the rescue operation. All 23 trapped individuals were successfully rescued, with 21 of them matching the system's predicted locations (with an error of less than 5 meters), and the accuracy of the physiological status assessment reached 92%.

[0353] Technical effect verification:

[0354] In this mine rescue example, the two most important technical effects of this implementation method have been fully verified: improved detection accuracy and reliability, and enhanced environmental adaptability.

[0355] Improved detection accuracy and reliability:

[0356] This implementation significantly improves the accuracy and reliability of life detection through environmental adaptability calibration and multimodal information fusion. Table 10 shows the performance comparison between this system and traditional life detection systems under the same rescue conditions.

[0357] Table 10: Comparison of detection accuracy and reliability between this embodiment and the traditional system;

[0358]

[0359] As shown in Table 10, this implementation significantly outperforms the traditional system across all key metrics. In particular, the false alarm rate decreased from 21.4% to 4.3%, a 79.9% reduction. Positioning accuracy increased by 66.1%, which is crucial for accurately guiding rescuers. The overall reliability score reached 0.89, providing a highly reliable basis for rescue decisions.

[0360] Effect of enhancing environmental adaptability:

[0361] This implementation greatly enhances the system's adaptability in complex and changing environments through self-organizing network topology and function self-reconfiguration technology. Table 11 shows a comparison of the system's performance retention rates under different environmental conditions.

[0362] Table 11: Comparison of system performance retention rates under different environmental conditions;

[0363]

[0364] The data in Table 11 shows that this implementation maintains high system performance under various harsh environmental conditions, with an average performance retention rate of 85.0%, 61.3% higher than that of traditional systems. This implementation's advantages are particularly pronounced in extreme situations such as secondary collapse and large-scale network node loss, with performance retention rates increasing by 119.3% and 102.6%, respectively.

[0365] During the 72-hour continuous rescue process, the system responded to eight environmental structure changes through dynamic adjustments of the self-organizing network topology, and successfully handled 73 sensor nodes that showed varying degrees of degradation through functional self-reconstruction technology, ensuring the continuity and stability of the system functions and the smooth completion of the rescue operation.

[0366] In summary, this implementation has achieved excellent application results in actual mine rescue scenarios, breaking through multiple technical limitations of traditional systems in complex environments, and providing more reliable and efficient life detection technical support for mine rescue.

[0367] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A mine rescue multi-information life detection method, characterized in that: The following steps are involved: Deploy heterogeneous sensor networks and apply belief propagation algorithms to build self-organizing networks and achieve dynamic topology optimization; Based on the optimized network topology, an environmental adaptability calibration system is constructed to achieve dynamic calibration of sensors in complex environmental changes; For calibrated sensors, the sensor performance is quantified and degradation is evaluated using a functional space model to achieve intelligent reallocation of sensing tasks. The functional space model is defined as: Quantitative evaluation models for the four core performance dimensions of measurement range, sensitivity, accuracy, and precision; Each dimension receives a performance score between 0 and 1 through standardized testing; Calculate the comprehensive health status index of the sensor through a weighted function; Based on the reallocated sensing tasks, a collaborative detection framework is constructed using a belief propagation algorithm, and multi-sensor data is integrated through a distributed decision-making algorithm. In the steps of deploying a heterogeneous sensor network and applying the belief propagation algorithm to construct a self-organizing network, the belief propagation algorithm is implemented as follows: Each node sends information to its potential neighbor nodes, indicating its confidence in the neighbor node's state; The node updates its belief based on all the neighbor information it receives; By iteratively transmitting trust information, the network gradually converges to a stable state and forms an optimal communication topology; The steps to build a collaborative detection framework using the belief propagation algorithm include: The detection area is divided into multiple sub-areas, each of which is monitored by multiple sensor nodes; Define the node state vector to represent the node's detection status for different types of information; Construct an inter-node compatibility function to quantify the consistency of the detection status of two nodes; By iteratively updating information, the detection status of each node is optimized to improve the overall detection accuracy The fused data is used to extract and analyze multi-dimensional features of life signals, and the results are evaluated for reliability to output highly reliable detection results.

2. A mine rescue multi-information life detection method according to claim 1, characterized in that: The steps of constructing the environmental adaptability calibration system include: Deploy an environmental parameter monitoring subsystem to collect data on key environmental factors that affect sensor performance; Construct a model of the impact of environmental parameters and sensor performance, and establish a mapping relationship between environmental factors and sensor performance drift; Implement a distributed calibration algorithm to perform inter-sensor calibration using spatial correlation; The calibration reference node group is dynamically selected based on environmental similarity to adapt to local environmental changes.

3. The mine rescue multi-information life detection method according to claim 1, characterized in that: The steps of realizing intelligent reallocation of sensing tasks include: Establish a sensing task model and define the demand vectors for each functional dimension; Calculate the matching degree between the remaining functions of the sensor and the mission requirements; Use the matching function to evaluate the suitability of the task and the remaining functions of the sensor; Construct a bipartite graph matching model and use the Hungarian algorithm to solve the optimal task allocation solution.

4. The mine rescue multi-information life detection method according to claim 1, characterized in that: In the step of fusing the multi-sensor data through a distributed decision algorithm, weighted Dempster-Shafer evidence theory is used to fuse multimodal features, including: Extract features from each sensor data and convert them into support for each hypothesis set; Define a trust function for each data source to measure the total support for the hypothesis; Introduce weight vectors and apply weighted combination rules to perform feature fusion; Generate fused detection results and their confidence scores.

5. The mine rescue multi-information life detection method according to claim 1, characterized in that: The steps of evaluating the reliability of the results include: Define a reliability evaluation index system, including signal strength, signal consistency, feature stability, and environmental interference; Establish a reliability evaluation model based on evidence theory to calculate the overall reliability of the test results; Implement a contextualized data credibility assessment model to calculate the credibility of the data source in the current environment; Generate trapped personnel status information and reliability assessment report.

6. The mine rescue multi-information life detection method according to claim 1, characterized in that: It also includes adaptive working mode switching steps based on energy status: Establish node energy status model to monitor battery charge, energy consumption rate and life expectancy; Define multiple working modes, including high performance mode, standard mode, energy saving mode and sleep mode; Design energy efficiency optimization functions to balance performance and energy consumption; Based on the current energy status and detection requirements, the working mode of each node is dynamically adjusted.

7. A mine rescue multi-information life detection system, used to execute a mine rescue multi-information life detection method according to any one of claims 1 to 6, characterized in that: include: Heterogeneous sensor network self-organizing unit for dynamic topology optimization and node location self-awareness; Environmental adaptability calibration unit, used to achieve dynamic parameter optimization of sensors in complex environmental changes; Functional degradation perception and reconstruction unit, used to quantify sensor performance and achieve intelligent task reallocation; Multi-point collaborative detection unit, used to achieve collaborative detection and distributed decision-making of multiple sensor nodes; Life signal analysis and reliability assessment unit, used to generate highly reliable life detection results and reliability reports.

Images