Method and system for positioning and safety control of underground personnel in coal mine based on iris recognition

Abstract

The present application provides a kind of coal mine personnel positioning and safety control method and system based on iris recognition, it is related to mine safety monitoring technical field, first acquisition coal mine multiple iris acquisition module uploaded with module identifier and acquisition time stamp original iris image stream, it is dynamically scheduled and handled to generate the iris image set to be handled.For the eye image of miner in the iris image set to be handled Multi-modal iris feature collaborative extraction, generate multidimensional iris feature vector and input miner identity depth correlation network to carry out identity matching identification, obtain miner identity code and individual space-time trajectory original point set.Individual space-time trajectory original point set is projected to underground three-dimensional space coordinate system to generate the moving trajectory line of miner, after triggering safety control instruction generation operation after dynamic dangerous area boundary analysis.The present application improves the positioning accuracy and safety control level of coal mine personnel.

Description

Technical Field

[0001] This invention relates to the field of mine safety monitoring technology, and more specifically, to a method and system for locating and managing personnel underground in coal mines based on iris recognition. Background Technology

[0002] In the underground coal mine working environment, personnel positioning and safety management are crucial links in ensuring safe coal mine production and the safety of miners' lives. Traditional methods for personnel positioning in underground coal mines mainly rely on technologies such as radio frequency identification (RFID), wireless local area network (WLAN) positioning, and sensor-based positioning.

[0003] RFID technology uses electronic tags worn by miners and readers deployed at specific locations underground to locate personnel. However, this technology has low positioning accuracy, typically only determining the general area where the miner is located, and cannot achieve precise individual positioning. Moreover, electronic tags are susceptible to the complex underground environment, such as humidity, high temperatures, and electromagnetic interference, which can cause tag damage or signal loss, affecting the accuracy of positioning.

[0004] Wireless LAN positioning technology utilizes existing underground wireless LAN infrastructure to determine location by measuring parameters such as signal strength and time of arrival between the miner's equipment and the wireless access point. However, the underground environment is complex, with numerous metal devices and obstacles that cause severe reflection, refraction, and attenuation of wireless signals, resulting in significant positioning errors and making it difficult to meet the high-precision positioning requirements for coal mine safety management.

[0005] Sensor-based positioning technologies, such as accelerometers and gyroscopes, can track personnel movement to some extent, but they are susceptible to the cumulative effects of sensor errors. Over time, positioning errors increase, and they cannot accurately determine the absolute location of personnel underground. Furthermore, most existing coal mine safety management methods rely on alarms to alert personnel upon entering hazardous areas, lacking real-time analysis and prediction of personnel movement trajectories. This makes it impossible to detect potential safety risks in advance and implement effective control measures. Summary of the Invention

[0006] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for iris recognition-based personnel location and safety management in coal mines, the method comprising:

[0007] The system acquires raw iris image streams uploaded in real time by multiple iris acquisition modules deployed in the intersection area of ​​underground roadways, mining face area, and near key equipment in coal mines. The raw iris image streams contain images of the miner's eyes captured by each iris acquisition module according to a fixed acquisition cycle, each image having a module identifier and acquisition timestamp.

[0008] The original iris image stream is subjected to dynamic scheduling processing of iris images. The acquisition frequency and acquisition resolution of the multiple iris acquisition modules in the next acquisition cycle are dynamically adjusted according to the personnel density distribution parameters of different areas in the coal mine, and a set of iris images to be processed after dynamic scheduling is generated.

[0009] Multimodal iris feature collaborative extraction is performed on the miner's eye images in the set of iris images to be processed. At the same time, the micro-features of iris texture, the macro-features of iris gland distribution, and the topological features of iris vascular network are extracted to generate a multi-dimensional iris feature vector.

[0010] The multi-dimensional iris feature vector is input into a pre-constructed deep association network for miner identity to perform identity matching and recognition, outputting a miner identity code that is uniquely corresponding to each miner's eye image, and sorting the miner's eye images carrying the miner identity code according to their module identifier and acquisition timestamp to generate an individual spatiotemporal trajectory original point set corresponding to each miner identity code;

[0011] The original point set of the individual spatiotemporal trajectory is projected onto the three-dimensional spatial coordinate system of the coal mine. Combined with the fixed installation position coordinates of the iris acquisition module, a continuous movement trajectory line of the miner is generated underground. The spatial position relationship between the continuous movement trajectory line of the miner and the boundary of the pre-defined underground dynamic danger zone is analyzed. When the analysis shows that there is a spatial intersection or entry trend between the continuous movement trajectory line of the miner and the boundary of the underground dynamic danger zone, a safety control command generation operation bound to the underground dynamic danger zone is triggered.

[0012] Furthermore, embodiments of the present invention also provide an iris recognition-based underground personnel positioning and safety management system for coal mines, comprising:

[0013] A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned iris recognition-based method for locating and managing personnel underground in coal mines via executing the machine-executable instructions.

[0014] Based on the above, by acquiring raw iris image streams uploaded by multiple iris acquisition modules deployed in different areas of the coal mine, and dynamically scheduling these raw iris image streams, the acquisition frequency and resolution are dynamically adjusted according to the personnel density distribution parameters of different areas in the coal mine. This ensures both the efficiency of image acquisition and improves image quality, effectively utilizing system resources. A multimodal iris feature collaborative extraction operation is employed, simultaneously extracting microscopic features of iris texture, macroscopic features of iris gland distribution, and topological features of the iris vascular network to generate a multidimensional iris feature vector, significantly improving the accuracy and reliability of identity recognition. The multidimensional iris feature vector is input into a deep association network for miner identity matching and recognition, uniquely determining the identity code of each miner and generating an initial set of individual spatiotemporal trajectory points. Projecting this initial set of individual spatiotemporal trajectory points onto the three-dimensional spatial coordinate system of the coal mine generates a continuous movement trajectory line for the miner underground, and analyzing its spatial relationship with the pre-defined boundaries of dynamic hazardous areas underground. When a miner's movement trajectory is detected to intersect with or enter a dangerous area boundary, a safety control command is generated in a timely manner, enabling real-time location tracking and safety early warning for the miner, effectively preventing safety accidents. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the execution flow of the method for locating and managing personnel underground in coal mines based on iris recognition provided in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of exemplary hardware and software components of the coal mine underground personnel positioning and safety management system based on iris recognition provided in an embodiment of the present invention. Detailed Implementation

[0017] Figure 1 This is a flowchart illustrating a method for locating and managing personnel underground in coal mines based on iris recognition, provided in one embodiment of the present invention. A detailed description follows.

[0018] Step S110: Obtain the raw iris image stream uploaded in real time by multiple iris acquisition modules deployed in the intersection area of ​​underground roadways, the mining face area and near key equipment in the coal mine. The raw iris image stream contains the miner's eye image captured by each iris acquisition module according to a fixed acquisition cycle, with module identifier and acquisition timestamp.

[0019] In this embodiment, data uploaded in real time by iris recognition modules deployed at various key locations underground in the coal mine is first acquired. At the intersection of underground roadways, such as the junction of the main pedestrian passage and auxiliary work roadway responsible for transportation, a first iris recognition module with module identifier MZ-001 is installed. In the mining face area, such as a safe area five meters behind a roadheader undergoing tunneling operations, a second iris recognition module with module identifier MZ-002 is installed. Near key equipment, such as at the entrance of the underground central substation, a third iris recognition module with module identifier MZ-003 is installed. Each of these iris recognition modules captures an image according to a pre-set fixed acquisition cycle, for example, every 500ms. When a miner passes through the acquisition range of the module, the module captures an image containing the miner's eye area. Each iris recognition module attaches a unique module identifier to the image when generating image data. Meanwhile, the real-time clock inside the module marks each frame of image with a precise acquisition timestamp in the format YYYY-MM-DDHH:MM:SS.fff. The miner's eye images carrying the module identifier and acquisition timestamp are pushed to the processing center in real time and continuously through the underground industrial ring network to form the raw iris image stream.

[0020] Step S120: Perform dynamic scheduling processing on the original iris image stream. Based on the personnel density distribution parameters of different areas in the coal mine, dynamically adjust the acquisition frequency and acquisition resolution of the multiple iris acquisition modules in the next acquisition cycle to generate a set of iris images to be processed after dynamic scheduling.

[0021] Next, the received raw iris image stream is dynamically scheduled, specifically adjusting the operating parameters of the corresponding iris acquisition module based on the population density of each area. The system receives raw iris image streams uploaded in real-time from all iris acquisition modules (MZ-001, MZ-002, MZ-003, etc.) during the current acquisition cycle (e.g., within the past 500ms). Each module identifier is parsed from this image stream, and the number of miner eye images actually uploaded by each module during the current cycle is counted. For example, if the module identifier MZ-001 uploads Count_001 images, MZ-002 uploads Count_002, and MZ-003 uploads Count_003 during the current cycle, these image counts serve as a preliminary estimate of the population density in the area where each module is located.

[0022] Step S121: Receive the original iris image stream uploaded by the multiple iris acquisition modules in real time during the current acquisition cycle, and extract the module identifier corresponding to each iris acquisition module and the number of miner eye images actually acquired by each iris acquisition module in the current acquisition cycle from the original iris image stream as a preliminary estimate of the population density in the area where the iris acquisition module is located.

[0023] In the specific implementation, a data listening port is opened to continuously receive data packets forwarded from the underlying network devices. Each data packet header contains the identifier of the module sending the data, while the data packet payload contains one or more frames of binary data of the miner's eye image and their acquisition timestamps. After unpacking, the image frames are classified and counted according to the module identifier. For example, at the end of a processing cycle T_current (1000ms in length), the internally maintained counters show that the number of image frames from MZ-001 is Count_001, from MZ-002 is Count_002, and from MZ-003 is Count_003. These count values ​​Count_001, Count_002, and Count_003 are preliminary estimates of the population density in the area where each corresponding module is located.

[0024] Step S122: Query the pre-stored iris acquisition module deployment location mapping table according to the module identifier to obtain the specific installation location coordinates of each iris acquisition module in the underground roadway of the coal mine and its corresponding area type label. The area type label includes the main pedestrian passage area, mining equipment operation area, auxiliary operation area and restricted access area.

[0025] This embodiment maintains a statically configured iris recognition module deployment location mapping table, with the module identifier as the primary key. Upon obtaining the module identifier, the mapping table is immediately queried. For example, using MZ-001 as the index, the specific installation location coordinates are found to be a point (X_001, Y_001, Z_001) in the three-dimensional spatial coordinate system of an underground coal mine, with the area type label "Main Pedestrian Passage Area". Using MZ-002 as the index, its coordinates are found to be (X_002, Y_002, Z_002), with the area type label "Mining Equipment Operation Area". Using MZ-003 as the index, its coordinates are found to be (X_003, Y_003, Z_003), with the area type label "Restricted Access Area".

[0026] Step S123: Sum the preliminary estimates of the population density of multiple iris acquisition modules under the same area type label within the current acquisition period to generate the population density distribution parameters corresponding to each area type label.

[0027] Based on the area type labels obtained from the query, the preliminary estimates of personnel density for modules under the same label are merged. For example, under the label "Main Pedestrian Passage Area," there may be multiple modules such as MZ-001, MZ-004, and MZ-005. The corresponding Count_001, Count_004, and Count_005 values ​​for these modules are summed to obtain the personnel density distribution parameter Density_passageway for this area. Under the label "Mining Equipment Operation Area," there are modules such as MZ-002 and MZ-006. Their Count values ​​are summed to obtain Density_operation. Under the label "Restricted Access Area," there are modules such as MZ-003 and MZ-007. Their Count values ​​are summed to obtain Density_restricted.

[0028] Step S124: Compare the personnel density distribution parameters corresponding to each area type label with the preset high density threshold, medium density threshold and low density threshold for that area type label, and determine the personnel monitoring precision level required for each area type label in the next collection cycle.

[0029] For each area type label, three density thresholds are preset to classify the level of monitoring granularity. For example, for "Main Pedestrian Path Area," three preset density thresholds are Thresh_H_pass, Thresh_M_pass, and Thresh_L_pass. The calculated Density_passageway is compared to these three thresholds sequentially. If Density_passageway is greater than or equal to Thresh_H_pass, the area is determined to require "high-granularity monitoring" in the next cycle; if Density_passageway is between Thresh_M_pass and Thresh_H_pass, it is determined to be "medium-granularity monitoring"; if Density_passageway is between Thresh_L_pass and Thresh_M_pass, it is determined to be "low-granularity monitoring"; and if Density_passageway is less than Thresh_L_pass, it is determined to be "lowest-granularity monitoring". Similarly, the same comparison and judgment are performed on other area type labels to obtain their respective monitoring granularity levels: Level_passageway, Level_operation, and Level_restricted.

[0030] Step S125: Based on the personnel monitoring precision level, assign the acquisition frequency adjustment coefficient and acquisition resolution adjustment coefficient for the next acquisition cycle to the iris acquisition module within each area type label. The acquisition frequency adjustment coefficient is positively correlated with the personnel monitoring precision level, and the acquisition resolution adjustment coefficient is positively correlated with the personnel monitoring precision level.

[0031] A set of adjustment coefficients is predefined for each monitoring granularity level. For example, for the "high-granularity monitoring" level, the corresponding acquisition frequency adjustment coefficient K_freq_high and acquisition resolution adjustment coefficient K_res_high are both greater than 1; for the "medium-granularity monitoring" level, the corresponding adjustment coefficients K_freq_med and K_res_med are slightly greater than or equal to 1; for the "low-granularity monitoring" level, the corresponding adjustment coefficients K_freq_low and K_res_low are equal to 1 or slightly less than 1; for the "lowest-granularity monitoring" level, the corresponding adjustment coefficients K_freq_min and K_res_min are less than 1. Based on the Level_passageway determined in step S124, the same adjustment coefficients, namely K_freq_passageway and K_res_passageway, are assigned to all modules (such as MZ-001, MZ-004, and MZ-005) within the "main pedestrian passage area". Similarly, assign K_freq_operation and K_res_operation to modules within the "mining equipment operation area", and assign K_freq_restricted and K_res_restricted to modules within the "restricted access area".

[0032] Step S126: Multiply the acquisition frequency adjustment coefficient by the reference acquisition frequency of each iris acquisition module to obtain the actual acquisition frequency of each iris acquisition module in the next acquisition cycle.

[0033] Each iris recognition module has a factory-set or initially configured reference acquisition frequency F_base, for example, 500ms per acquisition corresponds to a frequency of 2Hz. An adjustment factor is applied to the reference frequency. For example, for the MZ-001 module, its actual acquisition frequency F_actual_001 = K_freq_passageway × F_base. For the MZ-002 module, its actual acquisition frequency F_actual_002 = K_freq_operation × F_base. For the MZ-003 module, its actual acquisition frequency F_actual_003 = K_freq_restricted × F_base.

[0034] Step S127: Multiply the acquisition resolution adjustment coefficient by the reference acquisition resolution of each iris acquisition module to obtain the actual acquisition resolution of each iris acquisition module in the next acquisition cycle.

[0035] Each iris recognition module also has a baseline recognition resolution R_base, for example, 640 pixels × 480 pixels. An adjustment factor is applied to the baseline resolution. For the MZ-001 module, its actual recognition resolution R_actual_001 = K_res_passageway × R_base. This multiplication operation means multiplying the length and width of the resolution by the square root of the adjustment factor, or more simply, directly applying the adjustment factor to the enumerated value that determines the resolution level. For example, if R_base corresponds to 720P and K_res_passageway is 1.5, then R_actual_001 might be adjusted to 1080P. For the MZ-002 module, R_actual_002 = K_res_operation × R_base. For the MZ-003 module, R_actual_003 = K_res_restricted × R_base.

[0036] Step S128: Send an adjustment instruction for acquisition parameters, including the actual acquisition frequency and the actual acquisition resolution, to each iris acquisition module, and receive the miner's eye images newly acquired and uploaded by each iris acquisition module in the next acquisition cycle according to the adjusted acquisition parameters. Combine the newly acquired miner's eye images with the miner's eye images uploaded but not yet processed in the current acquisition cycle to form the set of iris images to be processed.

[0037] The calculated actual acquisition frequency F_actual_001 and actual acquisition resolution R_actual_001 are encapsulated into a single instruction and sent to the MZ-001 module via the downhole network. Similarly, F_actual_002 and R_actual_002 are sent to MZ-002, and F_actual_003 and R_actual_003 are sent to MZ-003. Upon receiving the instruction, each module acquires and uploads images according to the new parameters within the next acquisition cycle (e.g., the next 1000ms). Simultaneously, the miner's eye images that have already been uploaded in the current acquisition cycle but have not yet undergone subsequent processing such as feature extraction are merged with the newly received images acquired according to the adjusted parameters to form a large pool of images to be processed, namely the set of iris images to be processed, Set_images_pending.

[0038] Step S129: In the set of iris images to be processed, add the module identifier of the source iris acquisition module, the acquisition timestamp of the acquisition time, and the parameter record of the actual acquisition resolution used for the miner's eye image to each frame of the miner's eye image.

[0039] For each frame (Image_frame_k) in the set of iris images to be processed (Set_images_pending), its metadata information needs to be bound to the image itself. The metadata information includes: the source module identifier (ID_module_k), the image's acquisition timestamp (TimeStamp_k), and the actual acquisition resolution (Resolution_k) used during acquisition. This information is organized into the image header or associated data records to ensure that the source and acquisition parameters of each frame can be traced at any time during subsequent processing.

[0040] Step S1210: Sort all miner eye images in the set of iris images to be processed globally according to the order of their acquisition timestamps, and divide the sorted miner eye images into multiple batches to be processed. The number of miner eye images in each batch to be processed does not exceed the preset batch processing capacity limit.

[0041] To ensure efficiency and orderliness in subsequent processing, all images in the set of iris images to be processed, `Set_images_pending`, are sorted in ascending order according to their associated acquisition timestamps, `TimeStamp_k`, resulting in a globally ordered image queue. Then, based on a preset batch size limit, such as `Batch_size_max` of 64 frames, this ordered queue is divided into multiple batches. The first batch contains images with the earliest acquisition timestamps, frames 1 to 64; the second batch contains images 65 to 128; and so on. Each batch, `Batch_n`, contains no more than 64 images of the miner's eyes, and the images within a batch are temporally consecutive. This completes the dynamic scheduling and initial organization of the iris images, generating an ordered and batched set of specially processed iris images that can be directly used for subsequent feature extraction operations.

[0042] Step S130: Perform multimodal iris feature collaborative extraction operation on the miner's eye image in the set of iris images to be processed, and simultaneously extract the microscopic features of iris texture, the macroscopic features of iris gland distribution, and the topological features of iris vascular network to generate a multidimensional iris feature vector.

[0043] Next, multimodal iris feature extraction is performed on each batch of miner eye images segmented in step S1210. This step aims to extract three different levels of complementary iris biometric information from the same eye image and fuse them into a unified feature vector. For each miner eye image Image_frame taken from batch_n, it is first preprocessed to locate the specific region where the iris is located.

[0044] Step S131: Sequentially extract each frame of the miner's eye image from the set of iris images to be processed, accurately locate the iris region of the miner's eye image, and determine the pixel coordinate range of the pupil region surrounded by the inner edge of the iris, the main annular region of the iris between the inner edge and the outer edge of the iris, and the sclera region outside the outer edge of the iris in the miner's eye image.

[0045] For each frame of the miner's eye image (Image_frame), iris localization is performed using a method based on Hough transform or an active shape model. First, the approximate center and inner edge of the pupil are detected in the image, obtaining the pupil center coordinates (X_pupil, Y_pupil) and the pupil radius R_pupil. The area enclosed by this inner edge is the pupil region. Next, the outer edge of the iris is searched outwards, at the junction of the iris and sclera, obtaining the outer edge radius R_iris. The annular region located between the inner edge (a circle with radius R_pupil) and the outer edge (a circle with radius R_iris) is identified as the main annular region of the iris. The region beyond the outer edge, extending to the image boundary, is marked as the sclera region. The boundaries of all these regions are recorded and stored as a set of pixel coordinates.

[0046] Step S132: Perform polar coordinate transformation on the main annular region of the iris, and expand the main annular region of the iris from the original image coordinate system into a normalized rectangular iris image with fixed angular resolution and fixed radius resolution. The rows of the normalized rectangular iris image correspond to the angular dimension in the polar coordinate system, and the columns correspond to the radius dimension in the polar coordinate system.

[0047] To eliminate the effects of pupil dilation and image rotation, the located annular region of the iris is normalized. Specifically, a polar coordinate transformation is used to map each point on the ring from Cartesian coordinates (X, Y) to polar coordinates (R_theta, Angle_alpha). The angle Alpha is set with the pupil center as the vertex, typically starting from the horizontal rightward direction (0 degrees) and increasing counter-clockwise, ranging from 0 to 360 degrees. The angle resolution is set to one sampling point per degree, for a total of 360 angle samples. The radius R_theta is linearly sampled from the inner edge R_pupil to the outer edge R_iris, with a radius resolution of 64 sampling points. Through this transformation, the original annular iris region is expanded into a normalized rectangular iris image (Normalized_iris) with dimensions of 360 rows (corresponding to the angular dimension) multiplied by 64 columns (corresponding to the radius dimension). The row coordinates i (0 to 359) of the normalized rectangular iris image correspond to the angle in polar coordinates, and the column coordinates j (0 to 63) correspond to the radius position from the inside to the outside in polar coordinates.

[0048] Step S133: Perform multi-scale two-dimensional Gabor filtering on the normalized rectangular iris image. Use a set of two-dimensional Gabor filters with different scales and directions to perform convolution operation on the normalized rectangular iris image to obtain the complex filter response value of each pixel position at multiple scales and multiple directions.

[0049] To extract the microscopic details of the iris texture, a two-dimensional Gabor filter bank is constructed. This filter bank contains S different scales (e.g., scale parameters s=1, 2, 3, corresponding to different Gaussian envelope sizes) and D different directions (e.g., directions d=0°, 45°, 90°, 135°). For each pixel in the normalized rectangular iris image (Normalized_iris), these S×D Gabor filters are convolved with the image patch in the neighborhood of that pixel. The convolution result of each filter is a complex number, where the real part represents the even-symmetric filtering response and the imaginary part represents the odd-symmetric filtering response. Therefore, for each pixel location, a complex response vector of length S×D is obtained, where each element of the vector is a complex number, denoted as Response_complex_{i, j, s, d}.

[0050] Step S134: Perform phase quantization encoding on the complex filter response value, quantize the phase information of the complex filter response value at each pixel position into binary code, and concatenate the binary codes of all pixel positions in a preset order to generate an iris texture micro-feature encoding string.

[0051] The amplitude information of complex numbers is ignored; only their phase information is extracted for encoding because phase information is more robust to changes in illumination and contrast. For the complex response Response_complex at each pixel location and for each filter (corresponding to a specific scale s and direction d), its phase angle Phase_angle = arctan(imaginary part / real part) is calculated. Based on the quadrant in which the phase angle lies, it is quantized into a 2-bit binary code. For example, a phase angle between 0° and 90° is encoded as "00", between 90° and 180° as "01", between 180° and 270° as "10", and between 270° and 360° as "11". The binary codes corresponding to all S×D filters at a single pixel location are concatenated to form a local binary code of length 2×S×D bits. Then, following the order from left to right and from top to bottom, the local binary codes of all 360×64 pixel positions on the normalized rectangular image Normalized_iris are concatenated end to end to form a final iris texture micro-feature encoding string Feature_texture, with a total length of 360×64×2×S×D bits.

[0052] Step S135: Perform local binary pattern texture analysis on the main annular region of the iris, divide the main annular region of the iris into multiple non-overlapping annular sub-bands, calculate the local binary pattern histogram in each annular sub-band, and concatenate the local binary pattern histograms of all annular sub-bands to generate a macroscopic feature vector of iris gland distribution.

[0053] Beyond the microscopic Gabor texture, the iris also contains macroscopic distribution patterns formed by iris glands (such as crypts and constriction grooves). First, the original annular region of the iris (before polar coordinate transformation) or the normalized rectangular image `Normalized_iris` is divided into N non-overlapping annular sub-bands along the radial direction. For example, N=8, with each sub-band corresponding to a width of 8 pixels in the radial direction. For each annular sub-band, a Local Binary Pattern (LBP) operator is used. For each pixel within the sub-band, its grayscale value is compared with the surrounding P sampling points in a circular neighborhood, centered on the pixel. Neighbors with a value greater than the center pixel are marked as 1, otherwise 0, forming a P-bit binary number, i.e., the LBP value of the pixel. Then, the frequency of LBP values ​​for all pixels within the sub-band is calculated, generating a histogram of dimension 2^P. The histogram vectors of N subbands (e.g., 8) are concatenated end to end to form a final macroscopic feature vector of iris gland distribution, Feature_gland, with a dimension of N×2 to the power of P.

[0054] Step S136: Perform vascular network enhancement processing on the main annular region of the iris, use Hessian matrix eigenvalue analysis to detect linear structures in the main annular region of the iris, extract the morphological skeleton of the detected linear structures, and generate a binary skeleton image of the iris vascular network.

[0055] The iris also contains a fine network of blood vessels, especially in light-colored irises. First, the original image of the main annular region of the iris (before polar coordinate transformation) is grayscaled and smoothed. Then, the eigenvalues ​​of the Hessian matrix in the image are calculated. The Hessian matrix is ​​a 2x2 matrix composed of the second-order partial derivatives of the image. For each pixel in the image, two eigenvalues ​​λ1 and λ2 of this matrix are calculated (assuming |λ1|≥|λ2|). For linear structures (blood vessels), their eigenvalues ​​typically satisfy |λ1| is much greater than |λ2|, and λ1 has a large absolute value. By setting a response function with respect to λ1 and λ2, the linear structures in the image can be enhanced while suppressing blocky and speckled noise. After obtaining the enhanced image, a morphological skeleton extraction algorithm, such as a thinning algorithm based on hit-and-miss transform, is used to iteratively peel away the edge pixels of the linear structures until a centerline of a single pixel width is obtained, ultimately generating a binarized skeleton image of the iris blood vessel network, Skeleton_vessel, where the blood vessel pixel value is 1 and the background is 0.

[0056] Step S137: Perform graph theory analysis on the bipathic skeleton image of the iris vascular network to extract the coordinates of branch nodes, intersection nodes, and the length and angle of the connecting edges between adjacent nodes in the vascular network, and construct the topology diagram of the iris vascular network.

[0057] The binarized skeleton image Skeleton_vessel is treated as a graph structure. First, the image is scanned to identify the 8-neighborhood connections of all blood vessel pixels. Based on the number of neighboring pixels, key nodes can be identified: if a blood vessel pixel has exactly two blood vessel pixels in its 8-neighborhood, it is a normal path point; if it has three or four blood vessel pixels, it is a branch node (fork point) or intersection node; if it has only one blood vessel pixel, it is an endpoint. The coordinates of all branch and intersection nodes are recorded, for example, the coordinates of node_A are (X_A, Y_A), and the coordinates of node_B are (X_B, Y_B). Then, along the blood vessel path, all path points connecting two adjacent nodes are traced, and the Euclidean length of the path is calculated as the length of the connecting edge Edge_AB. Simultaneously, the angle between the line connecting node A and node B and the horizontal axis is calculated as the angle of the connecting edge Angle_AB. Thus, a graph Graph_vessel, representing the topology of the iris vascular network, is constructed, consisting of a set of nodes V and a set of weighted edges E (length, angle).

[0058] Step S138: Statistically analyze the degree of all branch nodes and intersection nodes in the iris vascular network topology diagram to generate a node degree distribution histogram that reflects the complexity of the vascular network, and use the node degree distribution histogram as the topological feature vector of the iris vascular network.

[0059] In the topological graph Graph_vessel, the degree of each node refers to the number of edges connected to that node. The degree distribution of all branch nodes (usually degree 3) and intersection nodes (usually degree 4) is calculated separately. For example, the total number of nodes with degree 3 (N_deg3), the total number of nodes with degree 4 (N_deg4), and the number of nodes with other degree values ​​(such as nodes with endpoint degree 1, possibly nodes on a ring with degree 2) (N_deg1, N_deg2) can be calculated. These statistics are combined into a one-dimensional histogram vector. Alternatively, node degrees can be combined with the length and angle of connecting edges to form more complex distribution features. Here, the number of nodes corresponding to each degree value is directly arranged in ascending order of degree value to form an iris vascular network topological feature vector Feature_vessel=[N_deg1, N_deg2, N_deg3, N_deg4, ...].

[0060] Step S139: The microscopic feature encoding string of iris texture, the macroscopic feature vector of iris gland distribution, and the topological feature vector of iris vascular network are spliced ​​and fused end to end in the feature dimension to generate the multidimensional iris feature vector.

[0061] Features from three different modalities are fused. The microscopic feature encoding string for iris texture, Feature_texture, is a binary string, while the macroscopic feature vectors for iris gland distribution, Feature_gland, and iris vascular network topology, Feature_vessel, are numerical vectors. First, Feature_gland and Feature_vessel are normalized to ensure their numerical range aligns with or is standardized to match the numerical range of Feature_texture (0 or 1). Then, the three vectors are concatenated end-to-end. For example, if Feature_texture, Feature_gland, and Feature_vessel have lengths of L_texture, L_gland, and L_vessel, the fused multidimensional iris feature vector Feature_multi has dimensions of L_texture + L_gland + L_vessel. This fused vector Feature_multi represents the multimodal iris features of a single frame of a miner's eye image.

[0062] Step S1310: Add the module identifier and acquisition timestamp of the source miner's eye image to the multi-dimensional iris feature vector to generate a complete multi-dimensional iris feature data unit carrying spatiotemporal information.

[0063] Finally, the generated multi-dimensional iris feature vector (Feature_multi) is bound to the metadata of the image frame. A data structure (Unit_feature_data) is created, containing three fields: a feature vector field (Feature_vector) with the value of Feature_multi; a module identifier field (Module_ID) with the value of the source module identifier of the image (e.g., MZ-001); and a timestamp field (Timestamp) with the value of the image acquisition timestamp (e.g., YYYY-MM-DDHH:MM:SS.fff). This complete data unit (Unit_feature_data) serves as the direct input for subsequent identity matching and recognition. Steps S131 to S1310 are repeated for all image frames in the current batch (Batch_n) to obtain a batch of multi-dimensional iris feature data units carrying spatiotemporal information.

[0064] Step S140: Input the multi-dimensional iris feature vector into the pre-constructed deep association network for miner identity to perform identity matching and recognition, output the miner identity code that is uniquely corresponding to each miner's eye image, and sort the miner's eye images carrying the miner identity code according to their module identifier and acquisition timestamp to generate the original point set of individual spatiotemporal trajectory corresponding to each miner identity code.

[0065] Next, the multi-dimensional iris feature data unit carrying spatiotemporal information generated in step S1310 is input into a pre-trained deep neural network for identity recognition. This deep neural network is responsible for mapping high-dimensional iris features to specific miner identities.

[0066] Step S141: Obtain a pre-constructed deep association network for miner identities. The deep association network for miner identities includes an input layer, multiple hidden layers, and an output layer. The dimension of the input layer is consistent with the dimension of the multi-dimensional iris feature vector, and the dimension of the output layer is consistent with the total number of miner identity codes.

[0067] First, a pre-trained deep association network for miner identities is loaded. The number of nodes in the input layer of this deep association network is set to Input_dim, which must be equal to the dimension of the multi-dimensional iris feature vector Feature_multi generated in step S139, i.e., Input_dim = L_texture + L_gland + L_vessel. The network contains multiple hidden layers. For example, the first hidden layer is a fully connected layer with H1 nodes and the activation function is ReLU; the second hidden layer is also a fully connected layer with H2 nodes and the activation function is ReLU. The output layer of the network is a fully connected layer with the number of nodes Output_dim equal to the total number of registered miner identity codes N_miner. For example, if there are currently N_miner = 500 miners in the mine, then the number of nodes in the output layer is 500.

[0068] Step S142: Input the multi-dimensional iris feature vector into the input layer of the miner identity deep association network. After forward propagation calculation through the multiple hidden layers, perform linear transformation and nonlinear activation processing on the input data in each hidden layer to extract higher-order iris feature representations layer by layer.

[0069] The Feature_multi vector from a specific multidimensional iris feature data unit (Unit_feature_data) is extracted and used as the network input. This vector is first fed into the input layer. In the first hidden layer, the input vector is multiplied by the weight matrix W1, and a bias vector b1 is added to obtain the intermediate result Z1 = Feature_multi × W1 + b1. Then, Z1 is passed through the ReLU activation function to obtain the output of this layer, A1 = ReLU(Z1). A1 is then used as the input of the second hidden layer, and the above process is repeated: Z2 = A1 × W2 + b2, A2 = ReLU(Z2). Through the above layer-by-layer transmission and transformation, the original low-level iris features are gradually abstracted into a higher-level, more discriminative feature representation A_last_hidden (i.e., the output of the last hidden layer).

[0070] Step S143: In the output layer, the softmax activation function is used to perform a normalized exponential transformation on the output of the last hidden layer to generate a probability distribution vector. Each element value in the probability distribution vector represents the probability value that the current miner's eye image belongs to the corresponding miner identity code.

[0071] The output A_last_hidden of the last hidden layer is fed into the output layer. The output layer is also calculated using a linear transformation Z_out = A_last_hidden × W_out + b_out, resulting in a vector of length N_miner (e.g., 500), called logits. Then, the logits vector is normalized using the softmax function. The softmax function is calculated as follows: for each element z_i in logits, e raised to the power of z_i is calculated, and then divided by the sum of all elements raised to the power of z_j. That is, the probability of the i-th miner's identity P_i = exp(z_i) / (exp(z_1) + exp(z_2) + ... + exp(z_N_miner)). Finally, a probability distribution vector Probability_vector = [P_1, P_2, ..., P_N_miner] is obtained, where the sum of all P_i is 1, and each P_i represents the probability that the current image belongs to the i-th miner's identity code ID_miner_i.

[0072] Step S144: Select the miner identity code corresponding to the maximum probability value in the probability distribution vector as the preliminary identity matching result of the current miner's eye image, and record the maximum probability value as the matching confidence score.

[0073] From the probability distribution vector `Probability_vector`, the `argmax` function is used to find the index `Index_max` where the maximum value is located, i.e., `Index_max = argmax(Probability_vector)`. The miner identity code corresponding to this index `Index_max` is denoted as `ID_miner_candidate`, which is the miner identity that the network believes the current image most likely belongs to. At the same time, the maximum probability value is denoted as `Confidence_score`, i.e., `Confidence_score = Probability_vector[Index_max]`.

[0074] Step S145: Compare the matching confidence score with a preset acceptance threshold. If the matching confidence score is greater than or equal to the preset acceptance threshold, then determine the preliminary identity matching result as the miner identity code that uniquely corresponds to the current miner's eye image.

[0075] A preset confidence threshold for identity acceptance is established, for example, Accept_threshold=0.95. The Confidence_score obtained in step S144 is compared with this threshold. If Confidence_score ≥ 0.95, it indicates that the network is highly confident in its determination of the identity, and therefore the preliminary identity matching result ID_miner_candidate is accepted as the final, uniquely corresponding miner identity code, denoted as ID_miner_final.

[0076] Step S146: If the matching confidence score is less than the preset acceptance threshold, a manual review process is triggered. The miner's eye image and its multi-dimensional iris feature vector are pushed to the remote monitoring terminal. The manual confirmation miner identity code returned by the remote monitoring terminal is received, and the manual confirmation miner identity code is determined as the miner identity code that uniquely corresponds to the current miner's eye image.

[0077] If Confidence_score < 0.95, it indicates insufficient reliability of the automatic identification result, requiring manual intervention. In this case, a manual review process is triggered. The miner's eye image (Image_frame) itself, along with its multi-dimensional iris feature vector (Feature_multi), the automatically identified candidate identity (ID_miner_candidate), and the confidence score (Confidence_score), are packaged into a single review request and pushed to the remote monitoring terminal screen at the surface or underground monitoring center. After reviewing the image, the monitoring personnel, based on experience or other auxiliary information, manually select a correct miner identity code from the miner list, denoted as ID_miner_manual. The terminal returns this ID_miner_manual as a response. Upon receiving this response, it is determined as the final miner identity code: ID_miner_final = ID_miner_manual.

[0078] Step S147: Associate each frame of the miner's eye image in the set of iris images to be processed with the finally determined miner identity code, and generate a complete identification result data record containing the miner identity code, the miner's eye image itself, the module identifier and the acquisition timestamp.

[0079] After step S145 or S146, each frame of the miner's eye image obtains a final identity ID_miner_final. This identity code is combined with the corresponding image, module identifier, and acquisition timestamp to form a complete identification result data record. For example, for an image from module MZ-001 with timestamp T1, if the final identification code is determined to be miner M-123, then the record Record={ID_miner:M-123,Image:Image_frame_T1,Module_ID:MZ-001,Timestamp:T1} is generated. This process is repeated for all images in the current batch Batch_n to obtain a batch of complete identification result data records.

[0080] Step S148: Group all complete identification result data records according to the miner identity code, and group complete identification result data records with the same miner identity code into the same group.

[0081] All complete identification result data records obtained in the previous step are grouped according to the value of the ID_miner field. For example, all records with ID_miner of M-123 are grouped into one group, Group_M-123, and all records with ID_miner of M-456 are grouped into another group, Group_M-456. In this way, all identification records of each miner are aggregated together.

[0082] Step S149: Within each group corresponding to a miner's identity code, the complete identification result data records within the group are sorted in ascending order according to the order of the collection timestamps to generate the initial spatiotemporal sequence corresponding to that miner's identity code.

[0083] For each group after grouping, such as Group_M-123, each record within it is sorted in ascending order based on its attached Timestamp field (e.g., T1, T3, T5). After sorting, the order of records within the group represents the chronological order in which the miner was captured by iris scanning modules at different locations on the timeline. The output is the initial spatiotemporal sequence corresponding to the miner's identity code: Sequence_M-123=[Record_T1, Record_T3, Record_T5, ...].

[0084] Step S1410: Extract the acquisition timestamp and module identifier from each complete recognition result data record from the initial spatiotemporal sequence, and convert the module identifier into the fixed installation position coordinates of the iris acquisition module in the three-dimensional spatial coordinate system of the coal mine underground, generating an original set of individual spatiotemporal trajectories with the acquisition timestamp as the index and the fixed installation position coordinates as the elements.

[0085] Finally, from the initial spatiotemporal sequence Sequence_M-123, the information needed to construct the trajectory point is extracted for each record. For the first record in the sequence, Record_T1, its acquisition timestamp T1 and module identifier MZ-001 are extracted. By querying the iris acquisition module deployment location mapping table mentioned in step S122, MZ-001 is converted into its fixed installation location coordinates (X_001, Y_001, Z_001). Thus, the first original trajectory point Point_1 is generated, with the form (Timestamp: T1, Coordinate: (X_001, Y_001, Z_001)). This operation is repeated for subsequent records in the sequence, Record_T3, Record_T5, etc., to obtain Point_3, Point_5, respectively. All these raw trajectory points sorted by time constitute the individual spatiotemporal trajectory origin points of miner M-123: Set_raw_trajectory_M-123={Point_1, Point_3, Point_5, ...}.

[0086] Step S150: Project the original point set of the individual spatiotemporal trajectory onto the three-dimensional spatial coordinate system of the coal mine, and generate a continuous movement trajectory line of the miner underground by combining the fixed installation position coordinates of the iris acquisition module. Analyze the spatial positional relationship between the continuous movement trajectory line of the miner underground and the pre-defined boundary of the underground dynamic danger zone. When the analysis shows that there is a spatial intersection or entry trend between the continuous movement trajectory line of the miner underground and the boundary of the underground dynamic danger zone, trigger the generation operation of the safety control command bound to the underground dynamic danger zone.

[0087] Finally, using the original point set of individual spatiotemporal trajectories generated in step S1410, the continuous movement trajectory of the miner underground is constructed, and real-time spatial analysis is performed in conjunction with the pre-defined danger zone. Once a dangerous situation is detected, the corresponding safety control instructions are automatically triggered.

[0088] Step S151: Obtain the set of original points of the individual spatiotemporal trajectory corresponding to each miner's identity code. The set of original points of the individual spatiotemporal trajectory contains multiple original trajectory points sorted by the collection timestamp. Each original trajectory point consists of the collection timestamp and the coordinates of the fixed installation position of the corresponding iris collection module.

[0089] From the results of step S1410, extract the raw point set of the individual spatiotemporal trajectory of the miner identity code (e.g., M-123) to be analyzed: Set_raw_trajectory_M-123. This raw point set of the individual spatiotemporal trajectory contains a series of raw trajectory points arranged in ascending order of timestamp T, such as Point_A (T_A, X_A, Y_A, Z_A), Point_B (T_B, X_B, Y_B, Z_B), Point_C (T_C, X_C, Y_C, Z_C), etc.

[0090] Step S152: Calculate the time interval between adjacent original trajectory points in the set of individual spatiotemporal trajectory original points to obtain the acquisition time interval value between every two adjacent original trajectory points, and compare the acquisition time interval value with the preset maximum allowable time interval threshold.

[0091] For adjacent points in the point set, calculate their time difference. For example, calculate the time interval Δt_AB between Point_A and Point_B = T_B - T_A, and the time interval Δt_BC between Point_B and Point_C = T_C - T_B. A maximum allowed time interval threshold is preset, for example, T_gap_max = 3000ms. Each calculated time interval Δt is compared with this threshold.

[0092] Step S153: When the acquisition time interval value is less than or equal to the preset maximum allowable time interval threshold, the adjacent original trajectory points are directly connected to generate a preliminary trajectory line segment.

[0093] If Δt_AB≤3000ms, then the two points are considered to be continuous in time, and points A and B can be directly connected by a line segment to form a preliminary trajectory segment Segment_AB.

[0094] Step S154: When the acquisition time interval value is greater than the preset maximum allowable time interval threshold, it is determined that there is a missing acquisition time interval between adjacent original trajectory points. Based on the acquisition timestamps of the original trajectory points before and after the missing time interval and the coordinates of the fixed installation position, one or more virtual trajectory points are inserted in the missing time interval using a linear interpolation method, so that the time interval between adjacent original trajectory points after the insertion of virtual trajectory points is less than or equal to the preset maximum allowable time interval threshold.

[0095] If Δt_BC > 3000ms, for example, Δt_BC = 10000ms, then it is determined that there is a data gap between point B and point C. To generate a continuous trajectory, interpolation is required. First, calculate the number of virtual points to be inserted: N_interpolate = floor(Δt_BC / T_gap_max) = floor(10000 / 3000) = 3. Then, on the time axis, starting from T_B, insert a virtual time point every T_gap_max (3000ms), i.e., T_B+3000ms, T_B+6000ms, T_B+9000ms. For each virtual time point T_v, its corresponding coordinates (X_v, Y_v, Z_v) are obtained by linear interpolation between point B (X_B, Y_B, Z_B) and point C (X_C, Y_C, Z_C). For example, for T_v1 = T_B + 3000ms, its interpolation coefficient α = (T_v1 - T_B) / (T_C - T_B) = 3000 / 10000 = 0.3, then X_v1 = X_B + α × (X_C - X_B), and the same applies to Y_v1 and Z_v1. After inserting these virtual points, the time interval between point B and T_v1 is 3000ms, and the time interval between T_v1 and T_v2 is also 3000ms, satisfying the continuity requirement.

[0096] Step S155: Connect all original trajectory points and inserted virtual trajectory points in chronological order of their acquisition timestamps to generate a polyline of the miner's initial underground movement trajectory, consisting of multiple continuous line segments.

[0097] Merge all the original trajectory points (e.g., A, B, C) and the virtual trajectory points (e.g., V1, V2, V3) inserted in step S154, and sort them according to their timestamps from smallest to largest to obtain a point sequence [A, B, V1, V2, V3, C]. Then connect adjacent points in this sequence with line segments to obtain a preliminary, temporally continuous trajectory polyline Polyline_initial.

[0098] Step S156: Perform spatial smoothing filtering on the initial underground movement trajectory polyline of the miner, use a sliding window averaging algorithm to correct the fixed installation position coordinates of each trajectory point, and replace the coordinates of the current trajectory point with the average value of the coordinates of all trajectory points in the window to generate a smoothed sequence of continuous underground movement trajectory points of the miner.

[0099] To eliminate positioning noise and jagged fluctuations caused by interpolation, the initial trajectory polyline is smoothed. A sliding window averaging algorithm is used. A sliding window size is set, for example, window width W_size=3. For each point P_i (coordinates X_i, Y_i, Z_i) in the trajectory point sequence, a window is formed by taking itself and (W_size-1) / 2 points before and after it (less than this number if it's on a boundary). The average coordinate of all points within the window is calculated as the new smoothed coordinate X_i_smooth of point P_i. For example, for the middle point V2, five points (B, V1, V2, V3, C) are taken (if W_size=5), and the average of their X coordinates is calculated as the new X coordinate of V2. After traversing all points, a new smoothed trajectory point sequence [X_A_smooth, X_B_smooth, X_V1_smooth, ...] is obtained.

[0100] Step S157: Based on the smoothed sequence of continuous movement trajectory points of the miner underground, calculate the direction vector between adjacent trajectory points. The direction vector is obtained by subtracting the coordinates of the previous trajectory point from the coordinates of the next trajectory point, and record the magnitude of each direction vector as the length parameter of the trajectory segment.

[0101] For adjacent points in the smoothed trajectory point sequence, such as points P_i_smooth and P_j_smooth (P_j_smooth is the next point after P_i_smooth), calculate the direction vector Vector_ij = (X_j_smooth - X_i_smooth, Y_j_smooth - Y_i_smooth, Z_j_smooth - Z_i_smooth). Simultaneously, calculate the magnitude of this vector Length_ij = sqrt((X_j_smooth - X_i_smooth)^2 + (Y_j_smooth - Y_i_smooth)^2 + (Z_j_smooth - Z_i_smooth)^2), which is taken as the length of this trajectory segment.

[0102] Step S158: Combine the smoothed sequence of continuous underground movement trajectory points of the miner and the direction vectors between adjacent trajectory points to form the continuous underground movement trajectory line of the miner.

[0103] The smoothed sequence of trajectory points, along with the direction vector and length information between each pair of adjacent points, are encapsulated to form the final continuous underground movement trajectory line for the miner, Trajectory_M-123. This continuous underground movement trajectory line data structure contains an ordered list of coordinate points and information about the directed line segments connecting these points.

[0104] Step S159: Add a corresponding acquisition timestamp or virtual timestamp to each trajectory point in the continuous underground movement trajectory line of the miner to generate a complete trajectory data structure containing time and space dimensions.

[0105] For each point in the trajectory line Trajectory_M-123, whether it is a raw point or a virtual point, a time label is assigned. Raw points use their actual acquisition timestamp, while virtual points use the virtual timestamp assigned during interpolation. In this way, each point contains information about "when" (time) and "where" (spatial coordinates), forming a complete spatiotemporal trajectory data structure.

[0106] Step S1510: The generated continuous underground movement trajectory of the miner is classified and stored according to the miner's identity code, and an index mapping relationship is established between the miner's identity code and the continuous underground movement trajectory of the miner.

[0107] Finally, the generated trajectory line Trajectory_M-123 is stored in the trajectory database, and an index is created using the miner's identity code M-123 as the primary key. This allows for quick querying of historical or current continuous movement trajectories based on the miner's ID. For other miner identity codes (such as M-456), the corresponding trajectory line Trajectory_M-456 is also generated and stored.

[0108] Step S160: Analyze the spatial relationship between the miner's continuous underground movement trajectory and the pre-defined boundary of the underground dynamic danger zone.

[0109] Next, a real-time spatial relationship analysis is performed on the continuous movement trajectory line of the miner generated in step S1510 and the predefined underground dynamic danger zone to determine whether the miner has entered or is about to enter the danger zone.

[0110] Step S161: Obtain the pre-defined downhole dynamic hazard zone boundary data, which includes the closed polygon boundary coordinate sequence of each hazard zone, the hazard zone type label, and the effective time window of the hazard zone.

[0111] First, all current downhole dynamic hazard area information is loaded from a configuration database. Each hazard area is described by a set of data. For example, there is a hazard area Area_A, whose boundary is defined by a closed polygon consisting of a series of connected 3D coordinate points: [(X_A1, Y_A1, Z_A1), (X_A2, Y_A2, Z_A2), ..., (X_An, Y_An, Z_An)]. Its hazard area type label is "Mining Equipment Operation Area". Its effective time window is TimeWindow_A, for example, from 08:00:00 to 20:00:00 of the current day, indicating that this area is a dynamic hazard area during this period. Another hazard area, Area_B, has a boundary defined by another set of coordinates, its type label is "Gas Outburst Risk Area", and its effective time window is the entire day.

[0112] Step S162: Extract the coordinates of all trajectory points corresponding to the current miner's identity code from the continuous movement trajectory line of the miner underground, and extract the coordinates of each trajectory point in the order of the collection timestamps.

[0113] Taking miner M-123 as an example, all trajectory points, including original points and virtual points, are extracted from the trajectory line Trajectory_M-123 in chronological order. For example, the first extracted point is P_1 (T_1, X_1, Y_1, Z_1), the second point is P_2 (T_2, X_2, Y_2, Z_2), and so on.

[0114] Step S163: Use the ray method to determine whether the coordinates of the currently retrieved trajectory point are located inside the closed polygon of any underground dynamic danger zone boundary. If the coordinates of the trajectory point are located inside the closed polygon, it is determined that the current miner has entered the underground dynamic danger zone, and the collection timestamp of the entry time and the corresponding danger zone identifier are recorded.

[0115] For the extracted point P_1, check whether it is within any danger zone. Taking checking whether it is within the danger zone Area_A as an example, the ray casting method is used. A ray is cast from point P_1 in any direction (e.g., the positive X-axis direction), and the number of times this ray intersects the polygon boundary of Area_A is calculated. If the number of intersections is odd, the point is inside the polygon; if it is even, it is outside. If it is determined that P_1 is inside Area_A, then the entry event is recorded: Miner M-123 enters danger zone Area_A at time T_1.

[0116] Step S164: If the coordinates of the trajectory point are not located inside any closed polygon, then further calculate the shortest spatial distance between the current trajectory point coordinates and the boundary of each downhole dynamic hazard area, and generate a set of distance values.

[0117] If point P_1 is not inside any danger zone, calculate the shortest distance from it to the boundary of each danger zone. Taking the distance to Area_A as an example, calculate the shortest distance from point P_1 to each edge of the polygon Area_A, and take the minimum value, denoted as Dist_to_A. Similarly, calculate the shortest distance to Area_B (Dist_to_B) and the shortest distance to Area_C (Dist_to_C), forming a distance value set Dist_set={Dist_to_A, Dist_to_B, Dist_to_C}.

[0118] Step S165: Select distance values ​​that are less than the preset danger approach threshold from the set of distance values, and take the downhole dynamic danger area corresponding to the above distance value as the candidate set of possible approach danger areas.

[0119] A predefined danger proximity threshold is set, for example, D_alert = 5 meters. All distance values ​​less than 5 meters are filtered from Dist_set. Assuming Dist_to_A = 3 meters, Dist_to_B = 7 meters, and Dist_to_C = 10 meters, only Dist_to_A < 5 meters. Therefore, Area_A is added to the candidate danger zone set Candidate_set = {Area_A}.

[0120] Step S166: For each underground dynamic hazardous area in the candidate set of hazardous areas, calculate the direction vector of the current trajectory point coordinates pointing to the nearest point on the boundary of the hazardous area, and at the same time calculate the instantaneous motion direction vector of the miner at the current trajectory point. The instantaneous motion direction vector is obtained by the difference between the coordinates of the current trajectory point and the coordinates of the previous trajectory point.

[0121] For Area_A in Candidate_set, first find the point on the boundary of Area_A that is closest to point P_1, denoted as Nearest_Point_on_A, with coordinates (X_A_near, Y_A_near, Z_A_near). Calculate the direction vector Vector_to_A = (X_A_near - X_1, Y_A_near - Y_1, Z_A_near - Z_1) from P_1 to Nearest_Point_on_A. Simultaneously, it is necessary to know the miner's current movement direction. Take the previous trajectory point of P_1, i.e., point P_0 (T_0, X_0, Y_0, Z_0), and calculate the instantaneous movement direction vector Vector_motion = (X_1 - X_0, Y_1 - Y_0, Z_1 - Z_0). If P_1 is the first point, the movement direction cannot be calculated, and trend analysis is not performed at this time.

[0122] Step S167: Calculate the cosine value of the angle between the direction vector and the instantaneous motion direction vector, and determine whether the current movement direction of the miner points to the dynamic danger zone underground based on the cosine value of the angle.

[0123] Calculate the cosine of the angle between Vector_to_A and Vector_motion, Cos_theta. The formula is the product of the dot product and the magnitude: Cos_theta = (Vector_to_A · Vector_motion) / (|Vector_to_A| × |Vector_motion|). The closer the cosine value is to 1, the more consistent the directions of the two vectors, meaning the miner is moving towards the nearest point in the danger zone Area_A. A preset motion direction consistency threshold is used, for example, Cos_consistency = 0.8. If Cos_theta > 0.8, the miner's current motion direction is determined to be clearly pointing towards the danger zone Area_A.

[0124] Step S168: When the cosine value of the included angle is greater than the preset motion direction consistency threshold, it is determined that the current miner has a tendency to enter the underground dynamic danger zone, and the collection timestamp of the current trajectory point, the danger zone identifier and the estimated value of the predicted remaining entry time are recorded.

[0125] If an entry trend is determined, a trend event is recorded. The current timestamp T_1 and the danger zone identifier Area_A are recorded. Simultaneously, a rough estimate of the remaining entry time can be made. The remaining distance Dist_remaining = |Vector_to_A| = 3 meters. The instantaneous speed Speed_instant = |Vector_motion| / (T_1 - T_0). Therefore, the predicted remaining entry time T_eta = Dist_remaining / Speed_instant. This T_eta is also recorded.

[0126] Step S169: Summarize the determination results of the trajectory point being located inside the danger zone and the determination results of the trend of entering the danger zone, and generate a spatial position relationship analysis report including miner identity code, danger zone identifier, entry status mark, entry time or predicted entry time, and trajectory point coordinates.

[0127] All the above analysis results are integrated into a single report. The report includes: miner identification code M-123; for Area_A, the status marker Status_A is "Entered" or "Trending to Enter"; if entered, the entry time T_entry=T_1 is recorded; if trending to enter, the predicted entry time T_eta is recorded; and the coordinates (X_1, Y_1, Z_1) of the trajectory point that triggered the analysis. Areas that did not trigger any events are not recorded.

[0128] Step S1610: The spatial location relationship analysis report is pushed to the underground safety monitoring center of the coal mine in real time for display and storage.

[0129] Finally, this real-time generated spatial location analysis report is immediately sent to the surface or underground safety monitoring center via the underground network. The abnormal status of miner M-123 will be highlighted on the monitoring center's display screen, and the report will also be stored in a historical database for subsequent accident analysis and safety audits.

[0130] Step S170: When it is analyzed that the continuous movement trajectory line of the miner underground has a spatial intersection or entry trend with the boundary of the underground dynamic danger zone, the safety control command generation operation bound to the underground dynamic danger zone is triggered.

[0131] After step S160 analyzes and finds that a dangerous situation of "already entered" or "tending to enter" exists, the corresponding safety control process is immediately triggered to prevent accidents from occurring or to reduce accident losses.

[0132] Step S171: Receive the spatial location relationship analysis report, and extract the miner identification code, danger zone identifier and entry status marker from the spatial location relationship analysis report. The entry status marker includes an already entered marker and a trending entry marker.

[0133] The safety management module monitors and receives the spatial location relationship analysis report pushed in step S1610 in real time. From the received report Report_M-123, the key fields are parsed out: miner identification code M-123, danger area identifier Area_A, and entry status flag, assumed to be "entered".

[0134] Step S172: Query the pre-configured hazard area-control instruction mapping table according to the hazard area identifier, and obtain one or more safety control instruction templates bound to the underground dynamic hazard area. The safety control instruction templates include voice alarm instruction templates, vibration lamp instruction templates, equipment shutdown instruction templates, and area blockade instruction templates.

[0135] This embodiment maintains a hazardous area-control instruction mapping table, with the hazardous area identifier as the primary key. Querying this table using Area_A as the key retrieves all safety control instruction templates bound to that area. For example, the query results show that for the "mining equipment operation area" mentioned above in Area_A, the bound instruction templates are: the voice alarm instruction template Templet_voice, which reads, "Warning, miner [MinID], you have entered a hazardous area, please evacuate immediately!"; the vibration lamp instruction template Templet_lamp, which triggers a vibration mode of a specific frequency and intensity; and the equipment shutdown instruction template Templet_machine, which sends an emergency shutdown signal to the mining equipment in this area.

[0136] Step S173: If the entry status marker is marked as "entered", then directly obtain the complete instruction content from the danger zone-control instruction mapping table, fill the miner's identity code into the corresponding fields of the voice alarm instruction template and the vibration lamp instruction template, and generate personalized voice alarm instructions and personalized vibration lamp instructions for the miner.

[0137] Since the status flag in the report is "Entered the danger zone," the emergency handling process is initiated directly. The voice alarm command template `Templet_voice` is retrieved from the mapping table, and the placeholder `[MinID]` is replaced with the actual miner identification code `M-123` to generate the personalized voice alarm command: "Warning, miner M-123, you have entered a danger zone. Please evacuate immediately!". Similarly, a personalized vibration command for M-123 is generated from the vibration lamp command template `Templet_lamp`. This personalized vibration command includes the vibration frequency parameter `Freq_vibrate` and the duration parameter `Duration_vibrate`, which are predefined in the template.

[0138] Step S174: Send the personalized voice alarm command to the communication terminal carried by the miner for real-time broadcast, and send the personalized vibration lamp command to the lamp control module worn by the miner to trigger the lamp to vibrate according to the preset frequency and intensity.

[0139] The generated personalized voice alarm command is sent via network to the communication terminal (such as an explosion-proof mobile phone or smartwatch) carried by the miner M-123. Upon receiving the command, the terminal immediately plays the voice through its speaker. Simultaneously, a personalized vibration lamp command is sent via short-range wireless communication (such as Bluetooth) to the lamp control module on the miner's helmet. Upon receiving the command, the lamp control module controls the vibration motor inside the lamp to vibrate at a preset frequency (e.g., 5Hz) and intensity (e.g., high amplitude) to provide strong tactile feedback to alert the miner.

[0140] Step S175: If the underground dynamic danger zone is the operating area of ​​the mining equipment and the entry status is marked as entered, then extract the equipment shutdown instruction template from the danger zone-control instruction mapping table, and send the equipment shutdown instruction template to the mining equipment control system in the area to trigger the mining equipment to perform an emergency shutdown operation.

[0141] Since the danger zone Area_A is labeled "Mining Equipment Operation Area" and its status is "Entered," a shutdown operation is also required. The shutdown instruction template (Templet_machine) is extracted from the mapping table. This template may be a standardized shutdown command, such as containing the emergency shutdown instruction code CMD_EMERGENCY_STOP and the target equipment identifier DEV_ID_LIST. This instruction template is then sent directly to the mining equipment control system (e.g., the controller of a roadheader) within Area_A. Upon receiving the instruction, the controller immediately cuts off the equipment's power supply, performs emergency braking, and stops the equipment.

[0142] Step S176: If the entry status marker is a trend entry marker, then extract the estimated value of the predicted remaining entry time from the spatial location relationship analysis report, and compare the estimated value of the predicted remaining entry time with the preset early warning time threshold.

[0143] If the status flag parsed in step S171 is a "trend entry flag", then a different branch is executed. The predicted remaining entry time T_eta for that trend is extracted from the report. A preset warning advance time threshold is set, for example, T_advance = 10 seconds.

[0144] Step S177: When the estimated remaining entry time is less than or equal to the preset advance warning time threshold, it is determined that an advance warning command needs to be triggered in advance. The voice alarm command template and the vibration lamp command template are obtained from the danger zone-control command mapping table, and an advance warning command containing entry warning information is generated and sent to the communication terminal and the miner's lamp control module carried by the miner.

[0145] Compare T_eta with 10 seconds. If T_eta ≤ 10 seconds, it means the miner may enter the danger zone within 10 seconds, requiring immediate warning. At this point, retrieve the voice alarm command template and the vibration lamp command template from the mapping table, and generate a warning message different from the "Entered" warning. For example, the voice message could be filled with "Warning, miner M-123, you are heading towards danger zone Area_A, please stop immediately!" The vibration command can be set to a different warning vibration mode than the emergency entry warning, such as short, intermittent vibrations. Then, send the above warning commands to the miner's terminal and lamp.

[0146] Step S178: When the estimated remaining entry time is greater than the preset warning advance time threshold, no instruction is triggered temporarily, and the spatial relationship between subsequent trajectory points and the danger zone is monitored and the estimated remaining entry time is updated.

[0147] If T_eta > 10 seconds, it means there is still some time before the potential danger, and the warning is not triggered for now. The process returns to step S160 to continue analyzing the next trajectory point P_2, and recalculates its distance and trend from Area_A, updating the T_eta value. This continues until T_eta shrinks to within 10 seconds, or the miner changes direction and moves away from the danger zone.

[0148] Step S179: Track the execution status of all triggered safety control commands, receive the voice broadcast success confirmation signal returned by the communication terminal and the vibration execution success confirmation signal returned by the mine lamp control module, and add the execution status record to the spatial position relationship analysis report to generate a complete control event log.

[0149] For any instruction triggered in steps S174, S175, and S177, the executing end is required to return an acknowledgment signal. The communication terminal returns a "Voice broadcast successful" acknowledgment signal after successfully broadcasting the voice message. The mine lamp control module returns a "Vibration execution successful" acknowledgment signal after successfully initiating vibration. The equipment control system also returns a "Equipment stopped" acknowledgment signal after completing an emergency shutdown. Upon receiving these acknowledgment signals, the safety management module associates them with the initial spatial location analysis report and appends them to the report content. Finally, a comprehensive management event log is generated, recording the event triggering cause, triggering time, triggered instruction content, and the execution result of each instruction.

[0150] Step S180: After analyzing the spatial relationship between the miner's continuous underground movement trajectory and the pre-defined boundary of the underground dynamic danger zone, a series of global safety analysis and control steps are also included.

[0151] After completing the trajectory and danger zone analysis for individual miners, it is also necessary to conduct a comprehensive analysis of the trajectories of all miners from a global perspective, and combine it with environmental monitoring data and equipment status data to achieve more comprehensive safety management.

[0152] Step S181: Obtain the continuous underground movement trajectory lines of all miners corresponding to their identity codes, and overlay all the continuous underground movement trajectory lines of all miners in the same three-dimensional spatial coordinate system of the coal mine to generate a heat map of personnel distribution in the coal mine. The heat map of personnel distribution in the coal mine uses different color depths to represent the density of miners in different areas.

[0153] The latest position points of the continuous movement trajectories of all online miners are obtained from the trajectory database at the current moment. For example, the latest position point P_M123_latest for miner M-123, the latest position point P_M456_latest for miner M-456, and so on. All these position points are projected onto a three-dimensional spatial coordinate system in the coal mine. Then, a kernel density estimation algorithm is used to calculate the density of the point distribution in space. A fine grid is divided in space. For each grid cell, the number of points falling into that cell and its neighboring cells is calculated and smoothed to obtain a density value. Based on the density value, each grid cell is assigned a different color; the higher the density, the darker the color (e.g., red), and the lower the density, the lighter the color (e.g., blue), thus generating a real-time updated heatmap of personnel distribution in the coal mine.

[0154] Step S182: Spatially register the underground personnel distribution heat map with the underground ventilation network map, analyze the spatial correlation between high-density personnel gathering areas and key nodes of ventilation paths, and identify densely populated areas that may have poor ventilation risks.

[0155] The generated personnel distribution heatmap is overlaid and spatially registered with a pre-constructed underground ventilation network map. The ventilation network map includes the locations of key nodes such as ventilation roadways, air doors, and regulating windows. Through spatial overlay analysis, areas with high personnel density (darker color in the heatmap) that are also located at the end of the ventilation network, in return airways, or near ventilation bottlenecks are identified. For example, in the return airway area of ​​a certain tunneling face, the heatmap shows a large number of personnel gathered, while the design ventilation velocity in this area may be low. This area is then marked as a "densely populated area with potential ventilation failure risk."

[0156] Step S183: Send the boundary coordinates of the identified densely populated areas with poor ventilation risk to the underground gas concentration monitoring sensor network, triggering the gas concentration monitoring sensor network to perform encrypted sampling monitoring of the above-mentioned areas.

[0157] The list of boundary coordinates of the risk areas identified in step S182 is sent to the control unit of the underground gas concentration monitoring sensor network. Upon receiving the instruction, the control unit adjusts the sampling strategy of its subordinate sensors. For sensors that originally had a low sampling frequency, if their deployment location is in or near these risk areas, their sampling cycle is increased from the usual 30 seconds to once every 5 seconds to improve the monitoring sensitivity for potential gas accumulation risks.

[0158] Step S184: Receive encrypted sampling monitoring data returned by the gas concentration monitoring sensor network, compare the encrypted sampling monitoring data with a preset gas concentration safety threshold, and when the gas concentration monitoring data of any area exceeds the gas concentration safety threshold, generate a loudspeaker broadcast evacuation command for that area and send it to all miners' communication terminals in that area.

[0159] The system receives data in real time from the gas concentration monitoring sensor network, especially data from encrypted sampling areas. For example, it receives data `Gas_conc_1` from sensors in the risk area `Area_risk_1`, as well as regular data from other areas. `Gas_conc_1` is compared to a preset gas concentration safety threshold `Thresh_gas` (e.g., 1.0%). If `Gas_conc_1` ≥ 1.0%, the gas level is considered exceeded. At this point, a public address evacuation command is immediately generated for the `Area_risk_1` area. This command includes the message, "Gas level exceeded, all personnel in the area must evacuate immediately!" This command is sent to all public address terminals deployed within `Area_risk_1` and also sent via the underground communication network to the personal communication terminals of all miners currently located in the area.

[0160] Step S185: Overlay the underground personnel distribution heat map with the underground mining equipment operation status map, analyze the spatial intersection between the personnel trajectory line and the activity range of the operating mining equipment, and when it is found that the continuous underground movement trajectory line of a miner intersects with the boundary of the activity range of the operating mining equipment, generate a personnel avoidance instruction containing an equipment pause request and send it to the mining equipment control system.

[0161] A real-time personnel distribution heatmap (or more precisely, individual trajectory lines) is overlaid with an operational status map of the underground mining equipment. The equipment operational status map includes the current location of each piece of mining equipment, its dynamic range of movement (e.g., the swing range of the cutter head of a tunneling machine), and its current operational status (operating / stopped). Spatial analysis is used to check if any miner's continuous movement trajectory line enters the boundary of the operational range of an operating piece of equipment. For example, it is detected that the trajectory line of miner M-789 is about to intersect the operational range of the operating tunneling machine S-01. Once this impending spatial intersection is detected, a personnel avoidance command is immediately generated. This personnel avoidance command includes a request to stop the equipment and the miner's identification code M-789 who has entered the equipment area. The personnel avoidance command is sent to the control system of tunneling machine S-01.

[0162] Step S186: Receive the equipment pause execution confirmation signal returned by the mining equipment control system, associate the equipment pause execution confirmation signal with the corresponding miner identity code and record it to generate an equipment-personnel interaction avoidance event log.

[0163] After receiving the avoidance command, the control system of tunneling machine S-01 executes a pause procedure, stops the tunneling operation, and sends a confirmation signal of equipment suspension to the safety control center. Upon receiving the confirmation signal, the center associates it with the miner's identification code M-789 and records information such as the time of the event, equipment ID, and miner ID, generating a complete equipment-personnel interaction avoidance event log and storing it in the database.

[0164] Step S187: Merge all generated safety control instructions, gas concentration alarm events, and equipment-personnel avoidance events according to timestamps to generate a comprehensive coal mine safety control situation report, and push the comprehensive coal mine safety control situation report to the ground dispatch center for archiving and display.

[0165] Finally, all safety incidents occurring on the same day or during the current shift, including personal warnings / instructions triggered in step S170, gas over-limit alarms triggered in step S184, and equipment-personnel avoidance events recorded in step S186, are merged and sorted according to their timestamps to generate a comprehensive coal mine underground safety management and control situation report. This report, presented in the form of structured data and timeline charts, is pushed in real time to the large screen and archive server of the ground dispatch center, allowing dispatch and management personnel to fully grasp the underground safety situation.

[0166] Step S190: Before performing multimodal iris feature collaborative extraction on the miner's eye images in the set of iris images to be processed, a series of steps for optimizing and personalizing the identity recognition model are also included.

[0167] To ensure the accuracy and robustness of the iris recognition system in long-term operation, continuous model optimization and updates are necessary before feature extraction and during system operation. This includes initial registration, feature drift analysis, and personalized adjustments to model parameters.

[0168] Step S191: Obtain multiple sets of high-quality registration iris images collected during the miner onboarding registration phase, perform the multimodal iris feature collaborative extraction operation on each set of registration iris images, generate a registration multidimensional iris feature vector corresponding to each miner, and associate and store each miner's miner identity code with the registration multidimensional iris feature vector to construct an initial miner face and iris association database.

[0169] When a new miner joins the company, such as miner M-000, during the registration process, specialized equipment is used to collect multiple sets (e.g., 5 sets) of high-quality eye images under ideal lighting conditions. For these 5 sets of images, the multimodal iris feature collaborative extraction operation described in steps S131 to S139 is performed respectively, resulting in 5 multi-dimensional iris feature vectors: Feature_reg_1, Feature_reg_2, ..., Feature_reg_5. These vectors can be fused (e.g., averaged) to obtain a final representative registration feature vector, Feature_reg_M-000. Then, the miner's identity code M-000 is associated with this registration feature vector, Feature_reg_M-000, and stored in the initial miner face-iris association database.

[0170] Step S192: Obtain historical iris recognition success records accumulated during daily operations in the coal mine, and extract historical miner eye images and their corresponding historical multi-dimensional iris feature vectors and confirmed miner identity codes from each historical iris recognition success record.

[0171] After the system has been running for a period of time, a large number of daily successful identification records have been accumulated. For example, in the past three months, miner M-000 has been successfully identified hundreds of times. From these historical records, the miner's eye image (optional) corresponding to each successful identification, as well as, most importantly, the multi-dimensional iris feature vector Feature_hist_k generated at that time, and the miner's identity code M-000 confirmed in that identification, are extracted.

[0172] Step S193: Compare the historical multi-dimensional iris feature vector with the corresponding registered multi-dimensional iris feature vector of the miner identity code element by element, calculate the difference value on each feature dimension, and generate the drift curve of each miner's identity feature over time.

[0173] For miner M-000, all its historical feature vectors, Feature_hist_k (k=1, 2, ..., K), are compared element-wise with its registered feature vector, Feature_reg_M-000. Assume the feature vector dimension is D. For each dimension d (d=1 to D), calculate the absolute difference between the historical feature value and the registered feature value in that dimension, Delta_d_k=|Feature_hist_k[d]-Feature_reg_M-000[d]|. Then, plot a drift curve Curve_d for each dimension d, with time (or number of recognitions) on the horizontal axis and Delta_d_k on the vertical axis. This curve reflects the change of miner M-000's iris features in that dimension over time.

[0174] Step S194: Perform trend analysis on the drift curve, identify the feature dimensions whose drift exceeds the preset stability threshold, and mark the above feature dimensions as drift-sensitive feature dimensions.

[0175] Perform linear regression on each drift curve Curve_d or calculate its long-term trend. Calculate the average drift Avg_Delta_d of the curve over the most recent period. Preset a stability threshold Thresh_stable. If Avg_Delta_d > Thresh_stable, then the feature of dimension d is determined to be unstable and prone to drift over time. Mark the above dimensions as "drift-sensitive feature dimensions" and record their index list Sensitive_dims = [d1, d2, d3, ...].

[0176] Step S195: Based on the distribution of the easily drift-sensitive feature dimensions, dynamically adjust the input weight parameters of the corresponding feature dimensions in the miner identity deep association network, reduce the contribution of easily drift-sensitive feature dimensions in the identity matching process, and increase the contribution of stable feature dimensions.

[0177] The input layer weights of the miner identity deep association network are actually the weight matrix W1 of the first layer linear transformation. Each column of W1 corresponds to an input feature dimension and is connected to subsequent hidden layer nodes. W1 is adjusted according to the sensitive dimension list Sensitive_dims. For example, for a sensitive dimension d1, the weights of all its corresponding columns in W1 are multiplied by a decay factor Decay_factor less than 1 (e.g., 0.5) to reduce the influence of that dimension on subsequent network computations. Conversely, for very stable feature dimensions, their corresponding weights can be multiplied by an enhancement factor Enhance_factor greater than 1 (e.g., 1.2) to enhance their contribution. These adjustments are specific to a particular miner M-000, or they may be general adjustments for all miners, depending on the universality of the drift analysis.

[0178] Step S196: Solidify the adjusted input weight parameters into the miner identity deep association network to generate a personalized identity matching weight configuration for each miner.

[0179] The adjusted weight matrix W1' is saved to form a personalized network configuration Config_M-000 for miner M-000. When identity matching is needed for miner M-000, this specifically configured network is loaded, or the corresponding input layer weights are dynamically selected based on the miner ID in a shared network. In this way, the recognition process becomes adaptive to individual feature drift.

[0180] Step S197: Periodically acquire newly collected miner eye images and their successfully identified miner identity codes, and add the multi-dimensional iris feature vectors corresponding to the newly collected miner eye images as incremental data to the miner face-iris association database for subsequent drift curve updates and weight parameter fine-tuning.

[0181] The feature vector generated from each subsequent successful identification, such as Feature_new_k for miner M-000, along with its identification timestamp, is added as a new data point to the miner's historical record entry in the miner's face-iris association database. This continuously accumulating incremental data provides richer samples for the next drift curve analysis and weight adjustment.

[0182] Step S198: When a new miner is detected, repeat the operation of the miner registration stage to generate a registration multi-dimensional iris feature vector for the new miner and add it to the miner face-iris association database.

[0183] When a new miner joins the company, such as miner M-501, the system detects the new personnel event and automatically triggers the process of step S191 to collect a registration image, extract registration features, and add the new miner identity code M-501 and its registration feature vector Feature_reg_M-501 to the miner face and iris association database. At the same time, the output layer of the miner identity deep association network also needs to be dynamically expanded by a node to correspond to the new miner identity code.

[0184] Step S200: After generating the original set of individual spatiotemporal trajectories corresponding to each miner's identity code, a series of steps are also included to analyze and control personnel's work behavior.

[0185] After generating the original trajectory point set for each miner, in addition to real-time hazardous area analysis, the trajectory data can be combined with information such as production plans and equipment permissions to supervise and manage the compliance of mining operations.

[0186] For example, step S201: Perform time correlation analysis between the original set of individual spatiotemporal trajectories corresponding to each miner's identity code and the coal mine underground production operation schedule, and extract the planned operation area and planned operation time window corresponding to each miner's identity code.

[0187] Import the daily production operation schedule from the coal mine's production management system. For miner M-123, query his planned tasks for the current shift. For example, the schedule shows that miner M-123's planned work area from 08:00 to 12:00 today is Region_A (e.g., a coal mining face), and the planned work time window is TimeWindow_planned=[08:00:00, 12:00:00].

[0188] Step S202: Compare the coordinates of the original trajectory points in the individual spatiotemporal trajectory original point set with the spatial boundary of the planned operation area, and count the number of original trajectory points actually located within the planned operation area and the number of original trajectory points located outside the planned operation area for each miner's identity code within the planned operation time window.

[0189] Extract the raw point set `Set_raw_trajectory_M-123` representing the individual spatiotemporal trajectory of miner M-123. Filter out all raw trajectory points whose timestamps fall within the planned operation time window [08:00, 12:00] to obtain the point set `Planned_window_points`. For each point in this set, determine whether its coordinates are inside the boundary polygon of the planned operation region `Region_A` (using the ray casting method as well). Count the number of points inside (`Count_inside`) and the number of points outside (`Count_outside`).

[0190] Step S203: Calculate the work area deviation index for each miner based on the proportion of the number of original trajectory points located outside the planned work area to the total number of original trajectory points for that miner's identity code within the planned work time window.

[0191] Calculate the Deviation_ratio, the deviation from the work area index. Deviation_ratio = Count_outside / (Count_inside + Count_outside). This ratio reflects the proportion of track points that deviate from the work area that a miner should be in during the time they are supposed to be working.

[0192] Step S204: Compare the deviation index of the work area with the preset deviation tolerance threshold. When the deviation index of the work area exceeds the deviation tolerance threshold, it is determined that the miner has left his post or changed his post, and a warning record of leaving post or changing post containing the miner's identity code, deviation time window and deviation area coordinates is generated.

[0193] A preset deviation tolerance threshold is set, for example, Tolerance_deviation = 0.2 (i.e., 20%). The Deviation_ratio is compared to 0.2. If Deviation_ratio > 0.2, it is determined that miner M-123 has been outside their assigned area for more than 20% of their shift, indicating suspected absenteeism or dereliction of duty. An alert record is generated: {Miner_ID: M-123, TimeWindow: [08:00, 12:00], Deviation_Ratio: Deviation_ratio, Offending_Points: [List of coordinates of deviation points]}.

[0194] Step S205: Obtain real-time equipment operation data uploaded by equipment status monitoring sensors installed on key equipment in the coal mine. The equipment operation data includes equipment identifier, equipment operation timestamp, and equipment operation permission requirements.

[0195] In addition to personnel location, equipment also needs to be monitored. Real-time operational data should be obtained from sensors on critical equipment (e.g., coal mining machines, tunneling machines, and central substation switchgear). For example, operational data obtained from the coal mining machine Shearer_01 might be: {Device_ID: Shearer_01, Timestamp: T_op, Operation_Type: "Start", Permission_Required: "Coal Mining Machine Driver"}. This indicates that the coal mining machine Shearer_01 was started at time T_op, and operating this equipment requires the "Coal Mining Machine Driver" permission.

[0196] Step S206: Spatiotemporally correlate the original point set of the individual spatiotemporal trajectory with the equipment operation data to identify the miner identity code that appears near the installation location coordinates corresponding to the equipment identifier within a preset time range before and after the equipment operation timestamp.

[0197] For the Shearer_01's operation event at time T_op, a time window is set, for example, 30 seconds before and after, i.e., the time range [T_op-30s, T_op+30s]. Simultaneously, a spatial distance threshold is set, for example, 3 meters. Then, the original point set of the individual spatiotemporal trajectories of all miners is queried to identify those miners whose timestamps fall within this time window, and whose trajectory point coordinates are less than 3 meters away from the fixed installation position coordinates (X_s01, Y_s01, Z_s01) of the Shearer_01. For example, if miner M-456 has a trajectory point P_M456(T_op, X, Y, Z) at time T_op, and this point is 2 meters away from (X_s01, Y_s01, Z_s01), then miner M-456 is identified as being near the device during the device's operation time.

[0198] Step S207: Query whether the identified miner has permission to operate the device based on the miner's identity code. If a miner with no operating permission appears near the device, generate an unauthorized proximity alarm record containing the device identifier, the miner's identity code, and the timestamp of the occurrence.

[0199] Query the personnel permissions database to confirm whether miner M-456 has the "coal mining machine operator" permissions required to operate the Shearer_01. Assume the query result shows that M-456 is a support worker and does not have this permission. Therefore, this is determined to be an incident of unauthorized personnel approaching critical equipment. Generate an alarm record: {Alert_Type: "Unauthorized Approach", Device_ID: Shearer_01, Miner_ID: M-456, Timestamp: T_op}.

[0200] Step S208: Merge the off-duty and unauthorized access alarm records to generate a list of abnormal personnel work behavior events.

[0201] The off-duty and unauthorized access warning records generated in step S204 and the unauthorized access alarm records generated in step S207 are merged together in chronological order to form a list of abnormal personnel operation events, which comprehensively displays various personnel behavior violations that occurred on that day.

[0202] Step S209: Push the list of abnormal events in personnel work behavior to the handheld terminal of the underground management personnel in the coal mine, and receive the processing confirmation information of the abnormal events returned by the handheld terminal of the management personnel, and store the processing confirmation information and the corresponding abnormal events in the personnel work behavior history database.

[0203] The generated list of abnormal personnel operation events is pushed via network to the explosion-proof handheld terminal of the on-duty safety manager. The manager reviews these events on the terminal and, based on the actual situation (e.g., confirming that miner M-456 did indeed violate regulations, or explaining that he was simply passing by and conducting equipment inspection without operating it), confirms and processes each event, marking it as "Confirmed violation, verbal warning given" or "False alarm, ignored." The manager's terminal returns confirmation information including the processing opinion. Upon receiving the confirmation information, the system associates it with the original abnormal event record and stores both in the personnel operation behavior history database for long-term performance evaluation and safety trend analysis.

[0204] In one exemplary embodiment, an iris recognition-based underground personnel positioning and safety management system for coal mines is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2 As shown, this iris recognition-based underground personnel positioning and safety management system for coal mines includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements an iris recognition-based method for underground personnel positioning and safety management in coal mines. The display unit generates visually visible images and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of a coal mine personnel positioning and safety management system based on iris recognition, or an external keyboard, touchpad, or mouse, etc.

[0205] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for locating and managing personnel underground in coal mines based on iris recognition, characterized in that, The method includes: The system acquires raw iris image streams uploaded in real time by multiple iris acquisition modules deployed in the intersection area of ​​underground roadways, mining face area, and near key equipment in coal mines. The raw iris image streams contain images of the miner's eyes captured by each iris acquisition module according to a fixed acquisition cycle, each image having a module identifier and acquisition timestamp. The original iris image stream is subjected to dynamic scheduling processing of iris images. The acquisition frequency and acquisition resolution of the multiple iris acquisition modules in the next acquisition cycle are dynamically adjusted according to the personnel density distribution parameters of different areas in the coal mine, and a set of iris images to be processed after dynamic scheduling is generated. Multimodal iris feature collaborative extraction is performed on the miner's eye images in the set of iris images to be processed. At the same time, the micro-features of iris texture, the macro-features of iris gland distribution, and the topological features of iris vascular network are extracted to generate a multi-dimensional iris feature vector. The multi-dimensional iris feature vector is input into a pre-constructed deep association network for miner identity to perform identity matching and recognition, outputting a miner identity code that is uniquely corresponding to each miner's eye image, and sorting the miner's eye images carrying the miner identity code according to their module identifier and acquisition timestamp to generate an individual spatiotemporal trajectory original point set corresponding to each miner identity code; The original point set of the individual spatiotemporal trajectory is projected onto the three-dimensional spatial coordinate system of the coal mine. Combined with the fixed installation position coordinates of the iris acquisition module, a continuous movement trajectory line of the miner is generated underground. The spatial position relationship between the continuous movement trajectory line of the miner and the boundary of the pre-defined underground dynamic danger zone is analyzed. When the analysis shows that there is a spatial intersection or entry trend between the continuous movement trajectory line of the miner and the boundary of the underground dynamic danger zone, a safety control command generation operation bound to the underground dynamic danger zone is triggered.

2. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, The process involves dynamically scheduling the original iris image stream, adjusting the acquisition frequency and resolution of the multiple iris acquisition modules in the next acquisition cycle based on personnel density distribution parameters in different areas of the coal mine, to generate a dynamically scheduled set of iris images to be processed, including: The system receives the original iris image stream uploaded by the multiple iris acquisition modules in real time during the current acquisition cycle. It extracts the module identifier corresponding to each iris acquisition module and the number of miners' eye images actually acquired by each iris acquisition module in the current acquisition cycle from the original iris image stream as a preliminary estimate of the population density in the area where the iris acquisition module is located. Based on the module identifier, query the pre-stored iris acquisition module deployment location mapping table to obtain the specific installation location coordinates of each iris acquisition module in the underground roadway of the coal mine and its area type label. The area type label includes the main pedestrian passage area, mining equipment operation area, auxiliary operation area and restricted access area. The preliminary estimates of personnel density from multiple iris acquisition modules under the same area type label within the current acquisition period are summed to generate personnel density distribution parameters corresponding to each area type label. The personnel density distribution parameters corresponding to each area type label are compared with the preset high density threshold, medium density threshold and low density threshold for that area type label to determine the required personnel monitoring precision level for each area type label in the next collection cycle. Based on the personnel monitoring precision level, the iris acquisition module within each area type label is assigned a acquisition frequency adjustment coefficient and an acquisition resolution adjustment coefficient for the next acquisition cycle. The acquisition frequency adjustment coefficient is positively correlated with the personnel monitoring precision level, and the acquisition resolution adjustment coefficient is also positively correlated with the personnel monitoring precision level. Multiply the acquisition frequency adjustment coefficient by the reference acquisition frequency of each iris acquisition module to obtain the actual acquisition frequency of each iris acquisition module in the next acquisition cycle; Multiply the acquisition resolution adjustment coefficient by the reference acquisition resolution of each iris acquisition module to obtain the actual acquisition resolution of each iris acquisition module in the next acquisition cycle. The system sends an instruction to each iris acquisition module to adjust the acquisition parameters, which includes the actual acquisition frequency and the actual acquisition resolution. It also receives the miner's eye images newly acquired and uploaded by each iris acquisition module in the next acquisition cycle according to the adjusted acquisition parameters. The newly acquired miner's eye images, together with the miner's eye images that have been uploaded but not yet processed in the current acquisition cycle, are used to form the set of iris images to be processed. In the set of iris images to be processed, each frame of a miner's eye image is appended with the module identifier of the iris acquisition module from which it originates, the acquisition timestamp of the acquisition time, and the parameter record of the actual acquisition resolution used for that miner's eye image. All miner eye images in the set of iris images to be processed are globally sorted according to their acquisition timestamps, and the sorted miner eye images are divided into multiple batches to be processed. The number of miner eye images in each batch does not exceed the preset batch processing capacity limit.

3. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, The step involves performing a multimodal iris feature extraction operation on the miner's eye images in the set of iris images to be processed. This extraction simultaneously extracts microscopic features of iris texture, macroscopic features of iris gland distribution, and topological features of the iris vascular network, generating a multidimensional iris feature vector, including: Each frame of the miner's eye image is sequentially extracted from the set of iris images to be processed. The iris region of the miner's eye image is precisely located to determine the pixel coordinate range of the pupil region surrounded by the inner edge of the iris, the main annular region of the iris between the inner edge and the outer edge of the iris, and the sclera region outside the outer edge of the iris in the miner's eye image. The main annular region of the iris is subjected to polar coordinate transformation, and the main annular region of the iris is expanded from the original image coordinate system into a normalized rectangular iris image with fixed angular resolution and fixed radius resolution. The rows of the normalized rectangular iris image correspond to the angular dimension in the polar coordinate system, and the columns correspond to the radius dimension in the polar coordinate system. The normalized rectangular iris image is subjected to multi-scale two-dimensional Gabor filtering. A set of two-dimensional Gabor filters with different scales and directions are used to perform convolution operations on the normalized rectangular iris image to obtain the complex filter response values ​​of each pixel position at multiple scales and multiple directions. The complex filter response value is phase quantized and encoded, and the phase information of the complex filter response value at each pixel position is quantized into binary code. The binary codes of all pixel positions are concatenated in a preset order to generate an iris texture micro-feature encoding string. Local binary pattern texture analysis is performed on the main annular region of the iris, dividing the main annular region of the iris into multiple non-overlapping annular sub-bands. A local binary pattern histogram is calculated in each annular sub-band, and the local binary pattern histograms of all annular sub-bands are concatenated to generate a macroscopic feature vector of iris gland distribution. The vascular network enhancement process is performed on the main annular region of the iris. The Hessian matrix eigenvalue analysis method is used to detect the linear structures in the main annular region of the iris. The morphological skeleton of the detected linear structures is extracted to generate a binary skeleton image of the iris vascular network. Graph theory analysis is performed on the binarized skeleton image of the iris vascular network to extract the coordinates of branch nodes, intersection nodes, and the length and angle of the connecting edges between adjacent nodes, and to construct the topology diagram of the iris vascular network. The degree of all branch nodes and the degree of all intersection nodes in the iris vascular network topology diagram are statistically analyzed to generate a node degree distribution histogram that reflects the complexity of the vascular network. The node degree distribution histogram is then used as the topological feature vector of the iris vascular network. The iris texture micro-feature encoding string, the iris gland distribution macro-feature vector and the iris vascular network topology feature vector are spliced ​​and fused together in the feature dimension to generate the multi-dimensional iris feature vector. The module identifier and acquisition timestamp of the source miner's eye image are appended to the multidimensional iris feature vector to generate a complete multidimensional iris feature data unit carrying spatiotemporal information.

4. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, The process involves inputting the multi-dimensional iris feature vector into a pre-constructed deep association network for miner identity matching and recognition, outputting a miner identity code uniquely corresponding to each miner's eye image, and sorting the miner's eye images carrying the miner identity code according to their module identifier and acquisition timestamp to generate an original set of individual spatiotemporal trajectory points corresponding to each miner identity code, including: Obtain a pre-constructed deep association network for miner identities, which includes an input layer, multiple hidden layers, and an output layer. The dimension of the input layer is consistent with the dimension of the multi-dimensional iris feature vector, and the dimension of the output layer is consistent with the total number of miner identity codes. The multi-dimensional iris feature vector is input into the input layer of the miner identity deep association network. After forward propagation calculation through multiple hidden layers, the input data is subjected to linear transformation and nonlinear activation processing in each hidden layer to extract higher-order iris feature representations layer by layer. In the output layer, the softmax activation function is used to perform a normalized exponential transformation on the output of the last hidden layer to generate a probability distribution vector. Each element value in the probability distribution vector represents the probability value that the current miner's eye image belongs to the corresponding miner identity code. The miner's identity code corresponding to the maximum probability value in the probability distribution vector is selected as the preliminary identity matching result of the current miner's eye image, and the maximum probability value is recorded as the matching confidence score. The matching confidence score is compared with a preset acceptance threshold. If the matching confidence score is greater than or equal to the preset acceptance threshold, the preliminary identity matching result is determined as the miner identity code that uniquely corresponds to the current miner's eye image. If the matching confidence score is less than the preset acceptance threshold, a manual review process is triggered, and the miner's eye image and its multi-dimensional iris feature vector are pushed to the remote monitoring terminal. The manual confirmation miner identity code returned by the remote monitoring terminal is received, and the manual confirmation miner identity code is determined as the miner identity code that uniquely corresponds to the current miner's eye image. Associate each frame of the miner's eye image in the set of iris images to be processed with the finally determined miner identity code, and generate a complete recognition result data record containing the miner identity code, the miner's eye image itself, the module identifier and the acquisition timestamp; All complete identification result data records are grouped according to miner identity codes, and complete identification result data records with the same miner identity code are grouped together. Within each group corresponding to a miner's identity code, the complete identification result data records within the group are sorted in ascending order according to the order of collection timestamps to generate the initial spatiotemporal sequence corresponding to that miner's identity code. Extract the acquisition timestamp and module identifier from each complete recognition result data record from the initial spatiotemporal sequence, and convert the module identifier into the fixed installation position coordinates of the iris acquisition module in the three-dimensional spatial coordinate system of the coal mine underground, thereby generating an original set of individual spatiotemporal trajectories with the acquisition timestamp as the index and the fixed installation position coordinates as the elements.

5. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, The step of projecting the original set of individual spatiotemporal trajectories onto the three-dimensional spatial coordinate system of the coal mine, and generating a continuous movement trajectory line of the miner underground by combining the fixed installation position coordinates of the iris acquisition module, includes: Obtain the set of original points of the individual spatiotemporal trajectory corresponding to each miner's identity code. The set of original points of the individual spatiotemporal trajectory contains multiple original trajectory points sorted by the collection timestamp. Each original trajectory point consists of the collection timestamp and the coordinates of the fixed installation position of the corresponding iris collection module. The time interval is calculated for adjacent original trajectory points in the set of individual spatiotemporal trajectory original points to obtain the acquisition time interval value between every two adjacent original trajectory points, and the acquisition time interval value is compared with the preset maximum allowable time interval threshold. When the acquisition time interval is less than or equal to the preset maximum allowable time interval threshold, adjacent original trajectory points are directly connected to generate a preliminary trajectory line segment. When the acquisition time interval is greater than the preset maximum allowable time interval threshold, it is determined that there is a missing acquisition time interval between adjacent original trajectory points. Based on the acquisition timestamps of the original trajectory points before and after the missing time interval and the coordinates of the fixed installation position, one or more virtual trajectory points are inserted in the missing time interval using a linear interpolation method, so that the time interval between adjacent original trajectory points after the insertion of virtual trajectory points is less than or equal to the preset maximum allowable time interval threshold. Connect all the original trajectory points and the inserted virtual trajectory points in chronological order of their acquisition timestamps to generate a polyline of the miner's initial underground movement trajectory, which consists of multiple continuous line segments. Spatial smoothing filtering is applied to the initial underground movement trajectory polyline of the miner. The fixed installation position coordinates of each trajectory point are corrected using a sliding window averaging algorithm. The coordinates of the current trajectory point are replaced with the average of the coordinates of all trajectory points in the window to generate a smoothed sequence of continuous underground movement trajectory points of the miner. Based on the smoothed sequence of continuous movement trajectory points of the miner underground, the direction vector between adjacent trajectory points is calculated. The direction vector is obtained by subtracting the coordinates of the previous trajectory point from the coordinates of the next trajectory point, and the magnitude of each direction vector is recorded as the length parameter of the trajectory segment. The smoothed sequence of continuous underground movement trajectory points of the miner and the direction vectors between adjacent trajectory points are combined to form the continuous underground movement trajectory line of the miner. Add a corresponding acquisition timestamp or virtual timestamp to each trajectory point in the continuous underground movement trajectory of the miner to generate a complete trajectory data structure containing time and space dimensions. The generated continuous underground movement trajectory lines of the miners are classified and stored according to the miners' identity codes, and an index mapping relationship between the miners' identity codes and the continuous underground movement trajectory lines of the miners is established.

6. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, The step of analyzing the spatial relationship between the miner's continuous underground movement trajectory and the pre-defined boundary of the underground dynamic hazard zone includes: Acquire pre-defined downhole dynamic hazard zone boundary data, which includes the coordinate sequence of the closed polygon boundary of each hazard zone, the hazard zone type label, and the effective time window of the hazard zone; Extract the coordinates of all trajectory points corresponding to the current miner's identity code from the continuous movement trajectory line of the miner underground, and extract the coordinates of each trajectory point in the order of the collection timestamps. The ray method is used to determine whether the coordinates of the currently retrieved trajectory point are located inside the closed polygon of any underground dynamic danger zone boundary. If the coordinates of the trajectory point are located inside the closed polygon, it is determined that the miner has entered the underground dynamic danger zone, and the collection timestamp of the entry time and the corresponding danger zone identifier are recorded. If the coordinates of the trajectory point are not located inside any closed polygon, then the shortest spatial distance between the current trajectory point coordinates and the boundary of each downhole dynamic hazard area is further calculated to generate a set of distance values; From the set of distance values, select distance values ​​that are less than a preset danger approach threshold, and use the downhole dynamic danger areas corresponding to the above distance values ​​as a candidate set of potentially approachable danger areas; For each dynamic hazardous area in the candidate set of hazardous areas, calculate the direction vector of the current trajectory point coordinates pointing to the nearest point on the boundary of the hazardous area, and at the same time calculate the instantaneous motion direction vector of the miner at the current trajectory point. The instantaneous motion direction vector is obtained by the difference between the coordinates of the current trajectory point and the coordinates of the previous trajectory point. Calculate the cosine of the angle between the direction vector and the instantaneous motion direction vector, and determine whether the current movement direction of the miner points to the dynamic danger zone underground based on the cosine of the angle; When the cosine value of the included angle is greater than the preset motion direction consistency threshold, it is determined that the current miner has a tendency to enter the underground dynamic danger zone, and the collection timestamp of the current trajectory point, the danger zone identifier, and the estimated value of the predicted remaining entry time are recorded. The results of determining that the trajectory points are located inside the danger zone and the results of determining that there is a trend of entering the danger zone are summarized to generate a spatial positional relationship analysis report that includes miner identification code, danger zone identifier, entry status mark, entry time or predicted entry time, and trajectory point coordinates. The spatial location relationship analysis report is pushed to the underground safety monitoring center of the coal mine in real time for display and storage.

7. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, The step of generating a safety control instruction bound to the underground dynamic hazard zone when the analysis shows that the miner's continuous underground movement trajectory line intersects with or enters the boundary of the underground dynamic hazard zone includes: Receive the spatial location relationship analysis report, and extract the miner identification code, danger zone identifier, and entry status marker from the spatial location relationship analysis report. The entry status marker includes an already entered marker and a trending entry marker. Based on the hazardous area identifier, query the pre-configured hazardous area-control instruction mapping table to obtain one or more safety control instruction templates bound to the underground dynamic hazardous area. The safety control instruction templates include voice alarm instruction templates, vibration lamp instruction templates, equipment shutdown instruction templates, and area blockade instruction templates. If the entry status is marked as "entered", then the complete instruction content is directly obtained from the danger zone-control instruction mapping table, and the miner's identity code is filled into the corresponding fields of the voice alarm instruction template and the vibration lamp instruction template to generate personalized voice alarm instructions and personalized vibration lamp instructions for the miner. The personalized voice alarm command is sent to the communication terminal carried by the miner for real-time broadcast, and the personalized vibration lamp command is sent to the lamp control module worn by the miner to trigger the lamp to vibrate according to the preset frequency and intensity. If the underground dynamic danger zone is the operating area of ​​mining equipment and the entry status is marked as "entered", then the equipment shutdown instruction template is extracted from the danger zone-control instruction mapping table, and the equipment shutdown instruction template is sent to the mining equipment control system in the area to trigger the mining equipment to perform an emergency shutdown operation. If the entry status marker is a trend entry marker, then the estimated value of the predicted remaining entry time is extracted from the spatial location relationship analysis report, and the estimated value of the predicted remaining entry time is compared with the preset early warning time threshold. When the estimated remaining entry time is less than or equal to the preset advance warning time threshold, it is determined that an advance warning instruction needs to be triggered in advance. The voice alarm instruction template and the vibration lamp instruction template are obtained from the danger zone-control instruction mapping table, and an advance warning instruction containing entry warning information is generated and sent to the communication terminal and lamp control module carried by the miner. When the estimated remaining entry time is greater than the preset warning advance time threshold, no instruction is triggered temporarily, and the spatial relationship between subsequent trajectory points and the danger zone is monitored and the estimated remaining entry time is updated. The execution status of all triggered safety control commands is tracked, and the voice broadcast success confirmation signal returned by the communication terminal and the vibration execution success confirmation signal returned by the mine lamp control module are received. The execution status record is added to the spatial position relationship analysis report to generate a complete control event log.

8. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, After analyzing the spatial relationship between the miner's continuous underground movement trajectory and the pre-defined boundary of the underground dynamic hazard zone, the method further includes: Obtain the continuous underground movement trajectory lines of all miners corresponding to their identity codes, and overlay all the continuous underground movement trajectory lines of all miners in the same three-dimensional spatial coordinate system of the coal mine to generate a heat map of personnel distribution in the coal mine. The heat map of personnel distribution in the coal mine uses different color depths to represent the density of miners in different areas. Spatially register the underground personnel distribution heat map with the underground ventilation network map, analyze the spatial correlation between high-density personnel gathering areas and key nodes of ventilation paths, and identify densely populated areas that may have poor ventilation risks. The boundary coordinates of densely populated areas identified as having poor ventilation risks are sent to the underground gas concentration monitoring sensor network, triggering the gas concentration monitoring sensor network to perform encrypted sampling and monitoring of the aforementioned areas. The system receives encrypted sampling monitoring data returned by the gas concentration monitoring sensor network, compares the encrypted sampling monitoring data with a preset gas concentration safety threshold, and generates a loudspeaker broadcast evacuation command for the area when the gas concentration monitoring data in any area exceeds the gas concentration safety threshold, and sends it to all miners' communication terminals in the area. The underground personnel distribution heat map of the coal mine is overlaid with the underground mining equipment operation status map. The spatial intersection between the personnel trajectory line and the activity range of the mining equipment is analyzed. When it is found that the continuous movement trajectory line of the miner underground intersects with the boundary of the activity range of the mining equipment, a personnel avoidance instruction containing the equipment pause request is generated and sent to the mining equipment control system. Receive the equipment pause execution confirmation signal returned by the mining equipment control system, associate the equipment pause execution confirmation signal with the corresponding miner identity code and record it to generate an equipment-personnel interaction avoidance event log; All generated safety control instructions, gas concentration alarm events, and equipment-personnel avoidance events are merged according to timestamps to generate a comprehensive coal mine safety control situation report. This comprehensive coal mine safety control situation report is then pushed to the ground dispatch center for archiving and display.

9. The method for iris recognition-based personnel positioning and safety management in coal mines according to claim 1, characterized in that, Before performing multimodal iris feature collaborative extraction on the miner's eye images in the set of iris images to be processed, the method further includes: Multiple sets of high-quality registration iris images collected during the miners' onboarding registration phase are obtained. The multimodal iris feature collaborative extraction operation is performed on each set of registration iris images to generate a registration multidimensional iris feature vector corresponding to each miner. The miner's identity code is associated with the registration multidimensional iris feature vector and stored to construct an initial miner face and iris association database. Acquire historical iris recognition success records accumulated during daily operations in coal mines, and extract historical miner eye images and their corresponding historical multi-dimensional iris feature vectors and confirmed miner identity codes from each historical iris recognition success record; The historical multidimensional iris feature vector is compared element by element with the corresponding registered multidimensional iris feature vector of the miner's identity code, and the difference value on each feature dimension is calculated to generate the drift curve of each miner's identity feature over time. Trend analysis is performed on the drift curve to identify the feature dimensions whose drift exceeds a preset stability threshold, and these feature dimensions are marked as drift-sensitive feature dimensions. Based on the distribution of the easily drift-sensitive feature dimensions, the input weight parameters of the corresponding feature dimensions in the miner identity deep association network are dynamically adjusted to reduce the contribution of easily drift-sensitive feature dimensions in the identity matching process and increase the contribution of stable feature dimensions. The adjusted input weight parameters are solidified into the miner identity deep association network to generate a personalized identity matching weight configuration for each miner. Newly acquired images of miners' eyes and their successfully identified miner identification codes are periodically obtained. The multi-dimensional iris feature vectors corresponding to the newly acquired images of miners' eyes are added as incremental data to the miner face-iris association database for subsequent drift curve updates and weight parameter fine-tuning. When a new miner is detected, the operation of the miner registration stage is repeated to generate a registration multi-dimensional iris feature vector for the new miner and add it to the miner face-iris association database.

10. A coal mine underground personnel positioning and safety control system based on iris recognition, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the method for iris recognition-based personnel location and safety management in coal mines according to any one of claims 1 to 9 by executing the machine-executable instructions.

Images