Power transmission line construction personnel identity verification method based on face recognition

Abstract

The invention discloses a power transmission line constructor identity verification method based on face recognition, and relates to the technical field of electric power engineering safety management, and the method comprises the steps: collecting multispectral face data, and generating a spectral stereo feature matrix; micro blood flow pulsation characteristics, skin texture characteristics and thermal imaging temperature distribution characteristics are extracted from the spectrum stereo characteristic matrix, a three-layer cascade anti-counterfeiting verification mechanism is constructed, and comprehensive living body judgment is carried out; establishing a distributed feature matching network, and matching the spectrum stereo feature matrix with a pre-stored feature library; collecting a geographic position track and an operation behavior mode of a constructor, and carrying out multi-dimensional cross validation on the geographic position track and the operation behavior mode and an identity matching result to generate a multi-level safety evaluation index; and writing the identity verification process and the multi-level security evaluation index into a distributed account book, and generating a verification voucher. According to the invention, cross analysis is carried out on the identity matching result of the constructor and the behavior characteristics, so that the problem that a traditional face recognition system is easily falsely used by the identity is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power engineering safety management, and in particular to a method for verifying the identity of transmission line construction personnel based on face recognition. Background Art

[0002] Currently, transmission line construction sites are often located outdoors, at high altitudes, or in complex terrain, subject to interference from variable weather conditions such as strong light, shadows, rain, and snow. In these harsh environments, the accuracy of traditional facial recognition technology decreases significantly, making it difficult to meet the needs of construction safety management. Furthermore, transmission line construction areas are often scattered and vast, with construction workers in highly mobile locations. Achieving high-precision facial recognition at long distances, especially in areas with limited network coverage, is a recognized technical challenge in the industry.

[0003] Existing facial recognition systems are easily deceived by methods such as photos, videos, or 3D masks. This security vulnerability can have serious consequences in the high-risk field of power transmission line construction, necessitating the urgent need for more robust liveness detection and anti-spoofing technologies. Furthermore, relying solely on facial recognition is insufficient to address all scenarios. Integrating multimodal authentication technologies such as biometrics, RFID, and behavioral analysis to achieve seamless integration while maintaining system simplicity and reliability presents technical challenges. Summary of the Invention

[0004] The present invention provides a method for verifying the identity of transmission line construction personnel based on facial recognition, which solves the technical problems of traditional facial recognition technology, such as poor adaptability in complex environments, difficulty in long-distance recognition, insufficient protection against identity deception, and difficulty in integrating multimodal identity authentication.

[0005] In view of this, the first aspect of the present invention provides a method for verifying the identity of power transmission line construction personnel based on face recognition, including: collecting multi-spectral face data and processing it using an environmental compensation neural network to generate a spectral three-dimensional feature matrix; extracting micro-blood flow pulsation features, skin texture features and thermal imaging temperature distribution features from the spectral three-dimensional feature matrix, constructing a three-layer cascade anti-counterfeiting verification mechanism, and performing comprehensive liveness judgment; establishing a distributed feature matching network, matching the spectral three-dimensional feature matrix with a pre-stored feature library, and dynamically adjusting the feature compression level according to network conditions; collecting the construction personnel's geographic location trajectory and operation behavior pattern, performing multi-dimensional cross-validation with the identity matching results, and generating multi-level security assessment indicators; writing the identity authentication process and the multi-level security assessment indicators into a distributed ledger to generate a verification certificate for allocating construction personnel's work permissions.

[0006] Optionally, collecting multispectral facial data and processing it using an environmental compensation neural network includes: collecting multispectral facial data through a multispectral imaging module; collecting multidimensional environmental parameters and converting them into environmental parameter vectors, inputting the environmental parameter vectors into an environmental compensation neural network, and generating an environmental compensation factor matrix; using the environmental compensation factor matrix to process the multispectral facial data to obtain multi-band spectral features; and reconstructing the multi-band spectral features into a spectral three-dimensional feature matrix.

[0007] Optionally, comprehensive liveness judgment includes: extracting a near-infrared band feature submatrix from the spectral stereo feature matrix, detecting the light reflection change characteristics of the facial blood vessel area through a time series signal processing module, and constructing a micro-blood flow liveness feature vector; extracting a visible light band feature submatrix from the spectral stereo feature matrix, using a multi-scale feature enhancement network to extract skin microtexture information, and constructing a skin texture feature map; extracting a thermal infrared band feature submatrix from the spectral stereo feature matrix, and constructing a facial temperature topology map through hotspot detection and thermal gradient analysis algorithms; using a three-layer cascade judgment architecture to analyze and judge the micro-blood flow liveness feature vector, skin texture feature map and facial temperature topology map; and using a dynamic weighting strategy to fuse the three-layer judgment results to form a final liveness judgment result.

[0008] Optionally, establishing a distributed feature matching network includes: constructing a distributed feature matching network with a multi-level topology, including edge nodes, base station nodes and central nodes, and the three-level nodes are interconnected through heterogeneous links and collaboratively process feature matching tasks; clustering and partitioning the pre-stored feature library according to feature similarity, and distributing it among different computing nodes; inputting the spectral stereo feature matrix into the distributed feature matching network, and performing identity matching using a multi-mode feature matching engine; establishing a network status perception mechanism, monitoring the link quality between nodes in real time and calculating the network status index, and adaptively adjusting the compression level of the spectral stereo feature matrix based on the network status index; adopting a hierarchical federal matching strategy to dynamically switch the matching mode according to the network status.

[0009] Optionally, multi-dimensional cross-validation with the identity matching results includes: collecting position coordinate sequences and three-dimensional posture data to generate a position behavior characteristic spectrum including the work movement path and work activity distribution; collecting tool operation sequences and biomechanical characteristics to construct an operation behavior characteristic spectrum; constructing a behavior benchmark library, calculating the degree of deviation between the position behavior characteristic spectrum and the operation behavior characteristic spectrum and the behavior benchmark library, and generating a behavior matching score; through an inference analysis model, fusing the identity matching results and the behavior matching score to generate a multi-level security assessment indicator; if an abnormality in the security assessment indicator is detected, a progressive identity re-verification mechanism is triggered.

[0010] Optionally, writing the identity authentication process and multi-level security assessment indicators into the distributed ledger includes: executing distributed ledger consensus using a consensus mechanism based on trust weights; digitally encapsulating key data points in the identity authentication process and implementing privacy protection for biometric data; designing a hierarchical smart contract system to automatically execute identity authentication rules through composable smart contracts; generating verification credentials using a layered encryption architecture; allocating hierarchical work permissions to construction personnel in real time based on the verification credentials, and establishing a permission transfer mechanism to record permission changes in the distributed ledger.

[0011] Optionally, the multispectral facial data includes near-infrared spectrum data, visible light spectrum data, and thermal infrared spectrum data; the multi-level security assessment indicators include identity credibility, behavior coordination, risk warning level, and authority recommendation value.

[0012] The second aspect of the present invention provides a transmission line construction personnel identity verification system based on face recognition, including: a spectral data processing module for collecting multi-spectral face data and using an environmental compensation neural network to process it to generate a spectral three-dimensional feature matrix; a liveness detection module for extracting micro-blood flow pulsation features, skin texture features and thermal imaging temperature distribution features from the spectral three-dimensional feature matrix, constructing a three-layer cascade anti-counterfeiting verification mechanism, and performing comprehensive liveness judgment; an identity matching module for establishing a distributed feature matching network, matching the spectral three-dimensional feature matrix with a pre-stored feature library, and dynamically adjusting the feature compression level according to the network status; a behavior verification module for collecting the construction personnel's geographic location trajectory and operation behavior pattern, performing multi-dimensional cross-verification with the identity matching results, and generating multi-level security assessment indicators; a permission management module for writing the identity authentication process and multi-level security assessment indicators into a distributed ledger, generating verification credentials for allocating construction personnel's work permissions.

[0013] The beneficial effects of the present invention are as follows: the comprehensive liveness judgment technology of the present invention effectively solves the accuracy problem of traditional liveness detection in the transmission line construction environment through multimodal biometric analysis and cascade judgment mechanism; the multidimensional cross-validation technology effectively solves the problem that traditional face recognition systems are easily misused by identity by cross-analyzing the identity matching results of construction personnel with behavioral characteristics, realizes the transformation from "one-time verification" to "full process monitoring", and improves the security and reliability of identity authentication; the design of non-tamperable verification certificates based on distributed ledgers solves the problem that traditional identity authentication records are easily tampered with and difficult to trace, especially in high-risk operations such as transmission line construction that require clear responsibility. In summary, the present invention effectively improves the reliability, security and adaptability of identity verification of transmission line construction personnel through the innovative combination of multispectral data acquisition, multi-layer cascade anti-counterfeiting, network adaptive feature matching, behavioral pattern analysis and distributed ledger technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0015] Figure 1 The following is a framework flow chart of a method for identity verification of power transmission line construction personnel based on face recognition.

[0016] Figure 2 This is a comprehensive liveness determination flowchart for a method for verifying the identity of transmission line construction workers based on face recognition.

[0017] Figure 3 This is a flowchart of a multi-dimensional cross-verification method for verifying the identity of transmission line construction personnel based on face recognition. DETAILED DESCRIPTION

[0018] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] Example 1, with reference to Figures 1 to 3 , which is the first embodiment of the present invention, provides a method for verifying the identity of transmission line construction personnel based on face recognition. The framework flow chart is as follows Figure 1 Shown, including: S1: Collect multispectral face data and process it using an environmental compensation neural network to generate a spectral stereo feature matrix.

[0020] In a preferred embodiment of the present invention, step S1 includes the following sub-steps: S1.1: Collect multispectral facial data through a multispectral imaging module.

[0021] In this embodiment, a multispectral imaging module is integrated on the inner forehead of the construction helmet, including a near-infrared sensor, a visible light sensor, and a thermal infrared sensor. The synchronous sampling of multiple sensors is achieved through a hardware synchronization trigger circuit and a clock synchronization mechanism. A spatial registration algorithm based on feature point matching and affine transformation is used to align the three spectral images to the same coordinate system to form a multispectral data cube.

[0022] Among them, multispectral facial data includes near-infrared spectrum data, visible light spectrum data and thermal infrared spectrum data.

[0023] S1.2: Collect multi-dimensional environmental parameters and convert them into environmental parameter vectors, input the environmental parameter vectors into the environmental compensation neural network, and generate an environmental compensation factor matrix.

[0024] In order to cope with the complex and changeable environmental conditions at the transmission line construction site, the present invention sets an environmental perception module on the outside of the safety helmet to collect environmental parameters including light intensity, light angle, temperature, humidity and atmospheric particle concentration, and constructs an environmental feature vector after normalization and dimensionality reduction processing.

[0025] Furthermore, the environmental compensation neural network is designed based on the Transformer architecture and consists of an encoder module and three decoder modules (one for each of the three spectral channels). This network uses a multi-head self-attention mechanism to calculate the correlation between environmental parameters and a cross-attention mechanism to analyze the degree of impact of environmental parameters on each spectral channel. After mapping the environmental feature vector to the feature space, the decoder module generates compensation factors for the three spectral channels, forming an environmental compensation factor matrix.

[0026] S1.3: Use the environmental compensation factor matrix to process the multispectral face data to obtain multi-band spectral features.

[0027] Specifically, the multispectral facial data is compensated at the pixel level using the environmental compensation factor matrix, including adaptive adjustment of brightness, contrast, and color balance. The compensated image is then adaptively denoised, where bilateral filtering is used in the near-infrared channel to retain edges while suppressing noise, a combination of guided filtering and non-local mean filtering is used in the visible light channel, and wavelet domain threshold denoising is used in the thermal infrared channel. The denoising parameters are dynamically adjusted within a preset range according to the noise estimation value in the environmental feature vector.

[0028] Furthermore, a multi-scale residual convolutional network is used to extract deep features from the processed multispectral images. The network consists of a shared feature extraction backbone with multi-layer residual connections and batch normalization, multiple channel-specific feature extraction branches, and a feature pyramid fusion module with convolution and upsampling operations, and finally outputs feature maps of three channels.

[0029] S1.4: Reconstruct the multi-band spectral features into a spectral stereo feature matrix.

[0030] Furthermore, a channel attention mechanism is performed on the multispectral feature map to generate channel weights; a cross-spectral feature fusion module based on a bidirectional gated recurrent unit is adopted to integrate the complementary information of the three spectral domains through the forward and backward GRU layers to generate a feature tensor; through the tensor reconstruction operation, the fused features are reorganized into a multi-dimensional spectral stereo feature matrix, where the first dimension represents the feature type, and the last three dimensions constitute a three-dimensional spatial structure, which is used to represent the distribution characteristics of facial information.

[0031] Advantageously, the present invention significantly improves recognition accuracy under complex and changing environmental conditions by integrating a multispectral imaging system with an environmental compensation neural network. This effectively addresses the issue of reduced accuracy in traditional facial recognition under complex conditions (such as those encountered during power line construction, such as strong light, low light, and inclement weather). Compared to existing facial recognition technologies that rely solely on visible light, the present invention utilizes multispectral acquisition combined with an environmental compensation neural network, enabling it to adapt to the diverse and extreme environmental conditions found at power line construction sites and improving recognition robustness.

[0032] S2: Extract micro-blood flow pulsation features, skin texture features, and thermal imaging temperature distribution features from the spectral stereo feature matrix, construct a three-layer cascade anti-counterfeiting verification mechanism, and perform comprehensive liveness judgment.

[0033] In a preferred embodiment of the present invention, the comprehensive living body judgment flow chart is as follows Figure 2 As shown, it includes the following sub-steps: S2.1: Extract the near-infrared band feature submatrix from the spectral stereo feature matrix, detect the light reflection change characteristics of the facial blood vessel area through the time series signal processing module, and construct the microblood flow living feature vector.

[0034] It should be noted that near-infrared light has strong tissue penetration ability and can capture information on blood vessel changes below the skin surface.

[0035] Specifically, the system locates key facial regions, such as the forehead, cheeks, and nose. Multiple frames of near-infrared feature data are continuously captured for each key region. A bandpass filter with a frequency range of 0.5Hz-4Hz is used to filter out ambient noise, retaining signal components within the normal human heart rate range. Independent component analysis based on the FastICA algorithm is used to separate the weak light intensity variation signal caused by cardiac pulsation, extract waveform features, and construct a micro-blood flow liveness feature vector. This feature vector effectively characterizes the unique physiological response patterns of living faces, providing a basis for the first level of liveness verification.

[0036] Preferably, through micro-blood flow pulsation feature analysis technology, the system can improve the recognition accuracy of physiologically active faces in the complex lighting environment of the transmission line construction site, overcome the limitations of traditional blinking and nodding verification that is easily disturbed by the construction environment, and improve the anti-interference ability and reliability of liveness detection.

[0037] S2.2: Extract the visible light band feature submatrix from the spectral stereo feature matrix, use a multi-scale feature enhancement network to extract skin microtexture information, and construct a skin texture feature map.

[0038] Furthermore, a multi-scale feature enhancement network is used to process the feature sub-matrix of the visible light band to improve the recognizability of texture details. The network contains multiple convolutional layers, which use 3×3, 5×5 and 7×7 convolution kernels to extract texture features of different scales respectively; on the enhanced texture image, the local texture features extracted by the LBP descriptor are combined with the directional features extracted by the multi-directional Gabor filter (using 4 scales and 8 directions) to construct a skin texture feature map. This feature map contains multiple randomly distributed microstructural feature points. The distribution pattern and local eigenvalue combination of these feature points form an individual biometric feature that is difficult to replicate, which is used for the second layer of anti-counterfeiting verification.

[0039] Preferably, the multi-scale skin texture feature analysis technology helps the system adapt to different lighting conditions in high-altitude operation scenarios on transmission towers. By capturing the unique microscopic optical properties of real skin, it enhances the system's defense against deceptive methods such as high-definition photos and video playback of construction sites.

[0040] S2.3: Extract the thermal infrared band feature submatrix from the spectral stereo feature matrix and construct the facial temperature topology map through hotspot detection and thermal gradient analysis algorithms.

[0041] It should be noted that the human face exhibits a unique temperature distribution pattern due to blood vessel distribution and metabolic activity, which is difficult to be accurately imitated by non-biological materials.

[0042] Specifically, a hotspot detection algorithm is used to identify areas of higher facial temperature (such as the inner corners of the eyes, the wings of the nose, and the lips), and a thermal gradient analysis algorithm is used to construct a facial temperature topology map. This algorithm uses the OTSU adaptive threshold method to segment the facial area into multiple temperature zones and calculate the temperature gradient between adjacent temperature zones. By analyzing continuous temperature change sequences, the algorithm captures subtle temperature fluctuations caused by changes in blood flow. It also monitors the synchronization and fluctuation patterns of temperature changes in different facial regions, enhancing the recognition of natural physiological thermal distribution patterns and effectively distinguishing 3D masks that lack physiological regulation.

[0043] Preferably, the thermal infrared temperature distribution analysis technology improves the recognition ability of advanced deception methods such as 3D mask attacks in response to severe weather conditions during transmission line construction, helps the system maintain reliable identity authentication performance in extreme construction environments such as high cold and high temperature, and enhances the environmental adaptability of the system.

[0044] S2.4: A three-layer cascade decision architecture is used to analyze and judge the micro-blood flow living feature vector, skin texture feature map and facial temperature topology map.

[0045] Specifically, the first layer uses micro-blood flow pulsation features for verification, primarily targeting static photo attacks; the second layer uses skin texture features for verification, defending against attacks involving high-quality photos and screen playback; and the third layer uses temperature distribution features for verification, identifying advanced attacks such as 3D masks and silicone models. All three layers of verification utilize a deep capsule network for feature learning and judgment. The capsule network consists of a main routing layer and two capsule layers, with the routing iteration count set to 3. The unique structure of the capsule network better preserves the spatial relationships between features, improving the recognition of complex patterns. Each layer outputs a confidence score between 0 and 1, indicating the likelihood of passing the corresponding verification.

[0046] Optionally, the inference process of the three-layer network is optimized by using low-precision quantization (reducing floating-point operations to 8-bit integers) and channel pruning (pruning channels whose contributions are lower than a set threshold) to reduce computational complexity and achieve efficient operation in resource-constrained environments.

[0047] S2.5: A dynamic weighting strategy is used to fuse the three-layer judgment results to form the final liveness judgment result.

[0048] Specifically, an environmental condition scoring matrix and a deception risk scoring matrix are constructed. The environmental condition scoring takes into account factors such as light intensity, ambient temperature, and humidity. In particular, parameter compensation is performed for special conditions such as strong light, shadow changes, extreme temperatures, and electromagnetic interference that may occur in the transmission line construction environment. Each factor is assigned a score of 0-1 based on the degree of impact on the performance of the corresponding verification layer; the deception risk score is determined based on historical deception attempt statistics and security level settings. The final weight is calculated using the following formula: in, is the weight vector of the three-layer verification, and are the environment and risk score vectors respectively, is the base weight vector (usually set to [0.3, 0.3, 0.4]), and is the weight factor, satisfying .

[0049] Furthermore, the final liveness judgment score is calculated as the weighted sum of the verification scores at each layer and the corresponding weights. When the weighted sum exceeds a preset threshold, it is determined to be a true live body. The present invention adopts an adaptive threshold mechanism to dynamically adjust the judgment threshold through a linear combination function based on the environmental condition score, the system security level, and the verification accuracy rate over a period of time. The value range is usually 0.75-0.85. The dynamic weighted fusion strategy is designed specifically for the transmission line construction environment, enabling the system to adaptively adjust the verification strategy according to different construction scenarios (such as high altitude, tunnels, and the wild). This helps to maintain good liveness detection performance under typical environmental conditions of transmission line construction, such as strong light, shadows, low temperatures, and high humidity, and keeps the overall false positive rate within a small range.

[0050] The integrated liveness assessment of the three-tiered cascaded anti-counterfeiting verification mechanism addresses the vulnerability of traditional liveness detection to photo, video, and 3D masks, a particular safety hazard in the high-security environment of power transmission line construction. Existing technologies typically rely on single or dual liveness detection methods. This new method combines three distinct dimensions: physiological characteristics (micro-blood flow pulsation), physical characteristics (skin texture), and thermal characteristics (temperature distribution), enhancing anti-counterfeiting efforts. This multi-dimensional liveness verification mechanism can be applied to any scenario requiring high-security identity verification, effectively preventing identity theft and fraud.

[0051] S3: Establish a distributed feature matching network to match the spectral stereo feature matrix with the pre-stored feature library, and dynamically adjust the feature compression level according to the network status.

[0052] In a preferred embodiment of the present invention, step S3 includes the following sub-steps: S3.1: Construct a distributed feature matching network with a multi-level topology, including edge nodes in construction helmets, on-site mobile base station nodes, and cloud center nodes. The three-level nodes are interconnected through heterogeneous links and collaboratively process feature matching tasks.

[0053] Edge nodes are responsible for preliminary feature extraction and local matching, mobile base station nodes are responsible for data aggregation and relay, and cloud nodes store the complete feature library and execute complex matching algorithms. Dual-mode communication between edge nodes and mobile base stations utilizes Bluetooth Low Energy (BLE) and ZigBee, supporting dynamic switching. Between mobile base stations and cloud nodes, 4G / 5G and dedicated wireless backhaul networks support bandwidth optimization and data compression. Secure encryption protocols are introduced to ensure seamless collaboration and secure data transmission between the three levels of nodes.

[0054] S3.2: Cluster and partition the pre-stored feature library according to feature similarity, and distribute it among different computing nodes.

[0055] Specifically, the K-means clustering algorithm is used to cluster the feature vectors of all construction workers. The number of clusters is dynamically determined based on the total number of construction workers and system resources. During the clustering process, cosine similarity is used as the distance metric between feature vectors. After clustering is completed, the feature data are distributed to different nodes based on the number of members in each cluster center and the storage capacity of the node.

[0056] Furthermore, cloud nodes store a complete feature library, including complete feature information for all construction workers. Mobile base station nodes store a medium-sized feature subset of personnel in the current construction area and adjacent areas. Edge nodes store only a small feature subset of current construction team members and high-frequency contact personnel. Furthermore, the feature subsets on edge nodes are dynamically updated through incremental learning.

[0057] S3.3: Input the spectral stereo feature matrix into the distributed feature matching network and perform identity matching using the multimodal feature matching engine.

[0058] Specifically, the multimodal feature matching engine performs matching according to a three-level cascade architecture. The first level is a fast retrieval module based on cosine similarity. By calculating the cosine similarity between the input features and the candidate features, it selects the top N candidate sets whose similarity exceeds the initial screening threshold (usually 0.75 to 0.85) and passes the candidate features and their similarities to the second level. The second level is an accurate matching module based on deep metric learning. This module uses a deep neural network trained with triplet loss to project features into a more discriminative high-dimensional space, calculate the Mahalanobis distance between features, and select the top-K identities with the highest confidence and their feature vectors to pass to the third level. The network structure is optimized to balance computational complexity and recognition accuracy. This module is suitable for execution in mobile base station nodes or cloud nodes. The third level is a relationship verification module based on graph neural networks. It models the collaborative relationship between construction workers as a dynamic graph structure, uses a lightweight graph convolutional network to extract relationship features, performs relationship reasoning and verification on the top-K identities, and outputs a relationship matching score. This module is particularly suitable for handling identity spoofing attacks and is primarily executed in cloud nodes.

[0059] The matching results of the three cascaded modules are fused according to an initial weighting of {Quick Search Weight: 0.2, Exact Match Weight: 0.5, Relationship Verification Weight: 0.3}. The weighting coefficients are dynamically adjusted based on the confidence of the module output and the current network conditions. If any module fails, the weights of the other modules are automatically adjusted to compensate. Based on the final fusion result, the system outputs a comprehensive match confidence score. By comparing this score with the verification decision threshold, the verification status (passed / further verification required / rejected) is determined, forming a complete identity match result.

[0060] Finally, the construction worker identity matching results are output, including the identity verification result (whether the person is an authorized person), specific identity identification information (the corresponding authorized person identity), matching confidence score, and verification status (passed / needs further verification / rejected).

[0061] S3.4: Establish a network status perception mechanism to monitor the link quality between nodes in real time and calculate the network status index, and adaptively adjust the compression level of the spectral stereo feature matrix based on the network status index.

[0062] The link quality between nodes includes signal strength, bandwidth, delay, and packet loss rate, and the Network Status Index (NSI) is calculated as follows: in, is the normalized signal intensity, is the normalized bandwidth, is the normalized delay, is the normalized packet loss rate, is the weight coefficient and satisfies .final The value range is 0 to 1. A larger value indicates a better network condition.

[0063] Furthermore, according to the network status index The system divides the feature compression level into multiple levels and selects the compression strategy adaptively according to the network conditions to maximize the compression efficiency while ensuring the recognition accuracy. When , lossless compression is used to retain all the information of the original feature matrix; when When , low-loss compression is used to retain 95% of the main components, and the compression rate is about 40%; when When using lossy compression, 85% of the main components are retained, and the compression rate is about 65%; when When high loss compression is used, 75% of the main components are retained, and the compression rate is about 80%.

[0064] It should be noted that the compression process utilizes a principal component analysis algorithm to reduce the dimensionality of the spectral stereo feature matrix. The system pre-trains on large-scale data to obtain the feature transformation matrix. Compression requires only simple matrix multiplication, which reduces the computational burden and is suitable for edge nodes. Decompression is performed at the receiving end, where the feature data is restored through an inverse transform.

[0065] S3.5: Adopt a hierarchical federated matching strategy and dynamically switch matching modes based on network conditions.

[0066] Specifically, when the network condition is good ( ) Using the cloud-edge collaborative mode, the edge node first searches for matching results in the local feature subset. If the confidence reaches the matching confidence threshold (usually 0.9), the result is returned directly. Otherwise, the compressed feature query request is sent to the mobile base station or cloud node, and the upper node completes a more accurate match and returns the result to the edge node. When the network condition deteriorates ( Activate local-first mode, where edge nodes rely primarily on their local feature subsets for matching, and adjust the match confidence requirement appropriately (typically lowering it to 0.75). At the same time, enhance the local processing capabilities of edge nodes, such as activating more complex feature transformation algorithms to compensate for the limitations of the feature subset. When the network recovers, the edge nodes upload the matching results to the cloud for verification and correction.

[0067] To handle complete network disconnections, this invention incorporates an offline caching mechanism, maintaining a historical record of the last 50 identity verification attempts on edge nodes, including timestamps, location information, verification results, and credibility scores. Once the network is restored, this cached data is uploaded to the cloud for consistency checks. Any anomalies detected trigger a security alert. Furthermore, the system regularly synchronizes signature libraries across all nodes to ensure effective identification of the latest individuals even in offline conditions.

[0068] This invention solves the identity authentication problem under unstable network conditions through edge computing and a multi-level caching mechanism; utilizes a multi-modal feature matching engine and relational graph reasoning to improve long-distance recognition capabilities; employs a graph neural network to analyze collaborative relationship patterns to enhance system security; and optimizes system resource utilization through an adaptive feature compression mechanism, reducing network transmission overhead and energy consumption. The distributed feature matching network of the present invention effectively improves identity authentication performance under limited network conditions at transmission line construction sites, shortens response time while ensuring recognition accuracy, improves system availability and verification accuracy, meets the needs of safety management for transmission line construction personnel, and has application value in remote areas with poor network conditions.

[0069] S4: Collect the geographic location trajectories and operational behavior patterns of construction personnel, perform multi-dimensional cross-validation with identity matching results, and generate multi-level safety assessment indicators.

[0070] In a preferred embodiment of the present invention, the identity multi-dimensional cross-verification flow chart is as follows: Figure 3 As shown, it includes the following sub-steps: S4.1: Collect position coordinate sequences and three-dimensional posture data to generate a position behavior feature spectrum including the work movement path and work activity distribution.

[0071] Specifically, an inertial measurement unit and a geographic location chip are integrated into the construction workers' safety helmets to collect position coordinate sequences and three-dimensional posture data. The collected data is preprocessed by the edge computing module, and Kalman filtering and complementary filtering are applied to eliminate noise. Features are extracted through a sliding time window. The processed data are fused to generate a position behavior characteristic spectrum, which includes the work movement path and work activity distribution (including height change curve and stop hotspot distribution).

[0072] S4.2: Collect tool manipulation sequences and biomechanical characteristics to construct manipulation behavior profiles.

[0073] Specifically, a lightweight sensor network is deployed on construction tools and equipment, including pressure sensors, acceleration sensors, bioelectric sensors and infrared thermal imaging sensors. Based on these sensors, the tool operation sequence and biomechanical characteristics are collected, including grip strength, operation speed, movement frequency and energy consumption. The collected original features are processed through feature engineering, including wavelet transform, statistical feature calculation, sequence pattern extraction, and dimensionality reduction, to form an operation behavior characteristic spectrum.

[0074] S4.3: Construct a behavioral benchmark library, calculate the degree of deviation between the position behavior characteristic spectrum and the operation behavior characteristic spectrum and the behavioral benchmark library, and generate a behavioral matching score.

[0075] Specifically, the behavioral benchmark library stores the standard behavioral patterns for each authorized construction worker in different work scenarios. The library utilizes a hierarchical storage structure: the first layer indexes the worker's ID; the second layer categorizes the work type; and the third layer stores standard behavioral feature templates. For each work type, the system uses supervised learning to create multiple behavioral templates, each representing an effective operating mode.

[0076] Furthermore, the calculation process of behavior matching includes: retrieving the corresponding behavior template set from the behavior benchmark library based on the current construction worker ID and job type; calculating the similarity between the currently collected location behavior feature spectrum and operation behavior feature spectrum and each template; and combining the two similarities to obtain the initial behavior matching score.

[0077] To improve matching accuracy, the present invention also employs an anomaly detection algorithm to identify behavioral deviations, combining statistical methods with machine learning techniques. For detected anomalies, the degree of abnormality is additionally calculated to form a behavioral anomaly index. The final behavioral matching score is determined by combining the initial behavioral matching score and the behavioral anomaly index. The system also implements an adaptive learning mechanism that dynamically updates behavioral templates based on changes in construction workers' long-term behavioral data, adapting to improvements in skills and changes in work habits.

[0078] S4.4: Through the reasoning analysis model, the identity matching results and behavior matching scores are integrated to generate multi-level security assessment indicators.

[0079] Specifically, a neural symbolic reasoning engine was designed, employing a three-stage reasoning architecture. In the first stage, feature preprocessing converts the authentication result in the identity matching result into a binary feature (1 for yes / 0 for no), maps the identity information into a one-hot encoded vector, and normalizes the matching confidence score and verification status to the [0, 1] range, where the verification status is mapped to {1 for passed, 0.5 for further verification required, 0 for rejected}. Furthermore, the location and operation behavior matching components of the behavior matching score are each normalized to the [0, 1] range.

[0080] In the second stage, multi-layer reasoning first calculates identity trustworthiness based on a weighted combination of the authentication result (weight 0.4), match confidence (weight 0.4), and verification status (weight 0.2). The behavioral match score is then used as the behavioral coordination score. The risk warning level is then determined by monitoring the temporal changes in the two scores (low risk when both are within a reasonable range and have consistent trends, medium risk when one indicator is abnormal or has inconsistent trends, high risk when both indicators are abnormal, and severe risk when there are significant fluctuations or persistent anomalies). Finally, based on the standard permission level corresponding to the identity information and the current risk warning level, permissions are downgraded (low risk maintains standard permissions, medium risk downgrades by one level, high risk downgrades by two levels, and severe risk downgrades to the lowest permissions), generating recommended permissions. The third stage generates results by integrating identity trustworthiness, behavioral coordination, risk warning levels, and recommended permissions to form a multi-level security assessment metric and generate interpretable explanations. The calculation parameters and thresholds for each evaluation metric can be adaptively adjusted based on actual application scenarios and historical data distribution.

[0081] S4.5: If an abnormality in the security assessment indicator is detected, a progressive identity re-verification mechanism will be triggered.

[0082] Specifically, identity re-verification strategies of varying intensities are adopted based on the current risk warning level. When the risk warning level is medium, construction workers are required to undergo quick verification (such as facial recognition or fingerprint recognition); when the risk warning level is high, dual verification (such as facial recognition plus fingerprint recognition) is required; when the risk warning level is severe, full multimodal identity verification is required, and on-site management personnel are notified for manual verification. The re-verification results will update the security assessment indicators and adjust the recommended permission values accordingly. If re-verification fails or the verification timeout is not completed, the system automatically reduces the operating permissions until the relevant permissions are locked. The verification frequency and verification mode can be dynamically adjusted according to the construction scenario and task type.

[0083] Preferably, through multi-dimensional cross-validation, the traditional single identity recognition is expanded to a multi-modal verification of "identity + behavior", which solves the problem in the existing technology that relying solely on biometrics is prone to deception or occasional misjudgment. By cross-validating the position behavior feature spectrum and operation behavior feature spectrum collected in real time with the identity matching results, combined with the risk warning level determination based on time series change characteristics, it can effectively identify the security risk scenario of "correct identity but abnormal behavior", thereby improving the reliability and security of identity verification. At the same time, the progressive re-verification mechanism based on the risk warning level realizes the dynamic adjustment of verification strength, balances the needs of security and operation continuity, avoids unnecessary operation interruptions due to verification failures, and improves the availability and adaptability of the system in practical applications.

[0084] S5: Write the identity authentication process and multi-level security assessment indicators into the distributed ledger to generate tamper-proof verification credentials for allocating work permissions to construction personnel.

[0085] Specifically, the method includes the following steps: S5.1: Enforce distributed ledger consensus using a trust-weighted consensus mechanism.

[0086] Specifically, a practical Byzantine fault-tolerant algorithm is used for efficient consensus, and the amount of consensus verification computation is reduced by introducing a trust weight mechanism.

[0087] S5.2: Digitally encapsulate key data points in the identity verification process and implement privacy protection for biometric data.

[0088] Among them, key data points include multispectral facial data collection timestamp, liveness detection results, identity matching results and multi-level security assessment indicators.

[0089] S5.3: Design a hierarchical smart contract system to automatically execute identity authentication rules through composable smart contracts.

[0090] Specifically, the hierarchical smart contract system includes identity authentication contracts, security assessment contracts, and permission allocation contracts. The combinable smart contracts can dynamically adjust verification thresholds and permission strategies according to different job types and risk warning levels, and the contracts are linked through event triggering mechanisms.

[0091] S5.4: Generate tamper-proof authentication credentials using a layered encryption architecture.

[0092] Specifically, authentication results and multi-level security assessment indicators are linked together via cryptographic hashing, combined with the timestamp and geographic location of the construction operation to generate a spatiotemporal proof. Multiple digital signatures from management and verification nodes are used to form a verification credential. The credential is encrypted and encapsulated based on the basic information layer, authentication layer, security assessment layer, and permission policy layer. Hierarchical key management allows for the display of information at different levels based on the querying party's permission level. Validity periods and verification rules are also embedded in the verification credential to support offline verification.

[0093] S5.5: Allocate hierarchical work permissions to construction personnel in real time based on verification credentials, establish a permissions transfer mechanism, and record permissions changes in the distributed ledger.

[0094] Specifically, based on the authentication results and security assessment indicators in the verification credentials, and in combination with the construction task type, work permissions are divided into four levels: general work permissions, restricted work permissions, high-risk work permissions, and emergency operation permissions. When abnormal security assessment indicators or emergency tasks occur, the corresponding permission adjustment mechanism is triggered, and the permission change record is written to the distributed ledger.

[0095] Furthermore, this embodiment also provides a transmission line construction personnel identity verification system based on face recognition, including: a spectral data processing module, which is used to collect multi-spectral face data and use an environmental compensation neural network to process it to generate a spectral three-dimensional feature matrix; a liveness detection module, which is used to extract micro-blood flow pulsation features, skin texture features and thermal imaging temperature distribution features from the spectral three-dimensional feature matrix, construct a three-layer cascade anti-counterfeiting verification mechanism, and perform comprehensive liveness judgment; an identity matching module, which is used to establish a distributed feature matching network, match the spectral three-dimensional feature matrix with a pre-stored feature library, and dynamically adjust the feature compression level according to the network status; a behavior verification module, which is used to collect the construction personnel's geographic location trajectory and operation behavior pattern, perform multi-dimensional cross-verification with the identity matching results, and generate multi-level security assessment indicators; a permission management module, which is used to write the identity authentication process and multi-level security assessment indicators into a distributed ledger, generate an unalterable verification certificate, and use it to allocate construction personnel's work permissions.

[0096] In summary, the comprehensive liveness judgment technology of the present invention effectively solves the accuracy problem of traditional liveness detection in the transmission line construction environment through multimodal biometric analysis and cascade judgment mechanism; the multidimensional cross-validation technology effectively solves the problem that traditional face recognition systems are easily misused by identity by cross-analyzing the identity matching results of construction personnel with behavioral characteristics, realizes the transformation from "one-time verification" to "full process monitoring", and improves the security and reliability of identity verification; the design of non-tamperable verification certificates based on distributed ledgers solves the problem that traditional identity verification records are easily tampered with and difficult to trace, especially in high-risk operations such as transmission line construction that require clear responsibility. In summary, the present invention effectively improves the reliability, security and adaptability of identity verification of transmission line construction personnel through the innovative combination of multispectral data acquisition, multi-layer cascade anti-counterfeiting, network adaptive feature matching, behavioral pattern analysis and distributed ledger technology.

[0097] Example 2 is the second embodiment of the present invention. This embodiment provides a method for verifying the identity of transmission line construction personnel based on face recognition. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0098] To validate the effectiveness of the proposed facial recognition-based identity verification method for power transmission line construction workers, a research team conducted a three-month simulation experiment at a power transmission line construction site of a provincial power grid company. The experimental scenarios encompassed three typical power transmission line construction environments: mountainous, plain, and suburban. These scenarios covered varying climate conditions (sunny, cloudy, and rainy) and work hours (daytime and dusk). A total of 60 construction workers participated, ranging in age from 25 to 55, with a balanced mix of skin tones and gender. The experimental hardware included a modified smart helmet (integrated with a multispectral imaging module, an environmental perception module, an inertial measurement unit, and a communication module); an on-site mobile base station (supporting 4G / 5G dual-mode communication and edge computing); and a cloud server (equipped with an Intel Xeon E5-2680 v4 processor, 128GB of RAM, and an NVIDIA Tesla V100 GPU). The software system utilizes a microservices architecture, deployed using Docker containers, and employs Kubernetes for service orchestration and resource management.

[0099] In experiments involving multispectral data acquisition and environmental compensation, researchers collected multispectral facial data under various environmental conditions. The environmental compensation neural network parameters were set as follows: the Transformer encoder consisted of six multi-head self-attention layers, each with eight attention heads and a hidden layer dimension of 512; the three decoder modules each contained four cross-attention layers. When tested in bright light environments (light intensity >60,000 lux), the accuracy of face recognition after environmental compensation increased from the original 76.2% to 92.8%. In low light environments (light intensity <100 lux), the accuracy increased from 68.5% to 90.3%. In high temperature environments (ambient temperature >35°C), the accuracy increased from 79.1% to 93.5%. In low temperature environments (ambient temperature <-10°C), the accuracy increased from 72.3% to 91.7%. The dimension of the spectral stereo feature matrix is 256×128×128×3. After environmental compensation processing, the feature distribution is more concentrated, the intra-class variance is reduced by 42.8%, and the inter-class distance is increased by 31.5%, indicating that the environmental compensation neural network effectively improves the discriminative ability and robustness of the features.

[0100] Liveness detection experiments were evaluated using a three-layer cascaded anti-counterfeiting verification mechanism. Five typical attack methods were employed: a high-definition photo attack, a high-quality video playback attack, a 3D-printed mask attack, a silicone mask attack, and an electronic screen playback attack. The dimension of the microblood flow liveness feature vector was set to 128, and the heart rate signal extracted using the FastICA algorithm achieved an average signal-to-noise ratio (SNR) of 18.6 dB. Skin texture feature maps were extracted using a 12-layer multi-scale feature enhancement network with convolution kernel sizes of 3×3, 5×5, and 7×7. The facial temperature topology map was segmented into eight temperature zones using the OTSU adaptive thresholding method. Experimental results showed that, using a single liveness detection method, the recognition accuracy for the photo attack was 95.3%, the video playback attack was 93.1%, the 3D-printed mask attack was 82.7%, the silicone mask attack was 78.5%, and the electronic screen playback attack was 94.8%. After adopting a three-tiered cascaded anti-counterfeiting verification mechanism, the overall recognition accuracy increased to 99.2%, 98.7%, 97.3%, 96.8%, and 99.1%, respectively. The dynamic weighted fusion strategy adaptively adjusts weights within a range of ±0.15 under different environmental conditions, keeping the system's false positive rate for liveness detection below 2.3% in typical environmental conditions such as strong light, shadows, low temperatures, and high humidity. This significantly outperforms traditional single or dual liveness detection methods.

[0101] The distributed feature matching network experiment evaluated the system's performance under different network conditions. The edge node is equipped with an ARM Cortex-A72 quad-core processor and 4GB of RAM, running a lightweight feature matching algorithm; the mobile base station node uses an Intel Core i7-8565U processor and 16GB of RAM; and the cloud node uses the aforementioned high-performance configuration. The initial screening threshold of the fast retrieval module of the multi-modal feature matching engine is set to 0.78, the precise matching module uses a 128-dimensional feature vector, and the graph neural network constructed by the relationship verification module contains three layers of graph convolution layers. When the network is good ( ) the average system response time is 186ms and the recognition accuracy is 99.3%; under normal network conditions ( ) case, using medium feature compression (retaining 85% of the principal components), the average response time is 235ms and the recognition accuracy is 97.8%; in poor network conditions ( ) case, using high feature compression (retaining 75% of the principal components), the average response time is 312ms and the recognition accuracy is 94.2%; when the network is almost disconnected ( ) in the absence of a local priority mode, the system switches to local priority mode, reducing average response time to 156ms and achieving 86.5% recognition accuracy, including 92.3% for cached features. Compared to traditional centralized recognition systems, the distributed feature matching network of this invention maintains high availability even in extreme network environments, reduces network bandwidth requirements by 65.3%, and reduces energy consumption by 47.8%.

[0102] Multidimensional cross-validation and permission management experiments evaluated the system's security and reliability. The sampling frequency of the position behavior profile was set at 10 Hz, with a sliding window size of 5 seconds. Features such as movement trajectory, height change curve, and hotspot distribution were extracted. The sampling frequency of the operation behavior profile was set at 100 Hz, and wavelet transform was used to extract features such as grip strength, operation speed, movement frequency, and energy consumption. The behavioral benchmark library established behavioral templates for five typical operation types for each construction worker, each of which includes 3-5 standard operation modes. Experimental results show that under normal operation, the average identity trustworthiness was 0.92, the average behavioral coordination was 0.88, the risk warning level was low, and permissions remained at standard levels. However, in a simulated identity impersonation scenario (authorized person A using the credentials of authorized person B), despite the identity trustworthiness being 0.90, the behavioral coordination was only 0.32, triggering a medium-risk warning and initiating a rapid reverification mechanism. In a simulated abnormal behavior scenario (authorized person performing unauthorized operations), the identity trustworthiness was 0.91, but the behavioral coordination was 0.41, also triggering a medium-risk warning. In a simulated composite attack scenario (unauthorized person using fake credentials to perform high-risk operations), the system detected identity trustworthiness of 0.45 and behavioral coordination of 0.28, triggering a severe risk warning. Permissions were reduced to the lowest level, and on-site management was notified. The distributed ledger verification credentials were generated using a weighted practical Byzantine fault-tolerant algorithm, which improved consensus efficiency by 38.2%. The average generation time for verification credentials was 126ms, and the validity period for offline verification was set to 4 hours. The entire permission management system demonstrates high adaptability and reliability in responding to various security risks, with an accuracy rate of permission allocation reaching 98.7%.

[0103] Based on the above simulation experiment results, we compared the performance differences between the present invention and the existing mainstream technical solutions, as shown in Table 1.

[0104] Table 1 Performance comparison between the present invention and the prior art As can be seen in Table 1, the present invention's solution significantly outperforms traditional single-modal facial recognition and deep learning multimodal recognition solutions across all key performance indicators. This comparison table clearly demonstrates the significant advantages of the present invention in four key dimensions: adaptability to complex environments, liveness detection security, network adaptability, and identity verification reliability. In particular, in the unique application scenario of power transmission line construction, the present invention achieves highly accurate, low-latency, and robust identity verification, providing a strong guarantee for the safety management of construction personnel.

[0105] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for verifying the identity of power transmission line construction personnel based on face recognition, characterized in that: include: Collect multispectral face data and process it using an environmental compensation neural network to generate a spectral stereo feature matrix; Extracting micro-blood flow pulsation features, skin texture features, and thermal imaging temperature distribution features from the spectral stereo feature matrix, a three-layer cascade anti-counterfeiting verification mechanism is constructed for comprehensive liveness determination. Establish a distributed feature matching network to match the spectral stereo feature matrix with the pre-stored feature library, and dynamically adjust the feature compression level according to the network status; Collect construction workers' geographic location trajectories and operational behavior patterns, conduct multi-dimensional cross-validation with identity matching results, and generate multi-level safety assessment indicators; The identity authentication process and multi-level security assessment indicators are written into the distributed ledger to generate verification credentials for allocating work permissions to construction personnel.

2. The method for verifying the identity of power transmission line construction personnel based on face recognition according to claim 1, characterized in that: The collecting of multispectral face data and processing using an environment compensation neural network comprises: Collect multispectral facial data through a multispectral imaging module; Collecting multidimensional environmental parameters and converting them into environmental parameter vectors, inputting the environmental parameter vectors into an environmental compensation neural network to generate an environmental compensation factor matrix; Processing multispectral face data using the environmental compensation factor matrix to obtain multi-band spectral features; The multi-band spectral features are reconstructed into a spectral stereo feature matrix.

3. The method for verifying the identity of power line construction personnel based on face recognition according to claim 1, characterized in that: The comprehensive living body judgment comprises: Extract the near-infrared band feature sub-matrix from the spectral stereo feature matrix, detect the light reflection change characteristics of the facial blood vessel area through the time series signal processing module, and construct the micro-blood flow living feature vector; Extract the visible light band feature submatrix from the spectral stereo feature matrix, use a multi-scale feature enhancement network to extract skin microtexture information, and construct a skin texture feature map; Extract the thermal infrared band feature submatrix from the spectral stereo feature matrix, and construct the facial temperature topology map through hotspot detection and thermal gradient analysis algorithms. A three-layer cascade decision architecture is used to analyze and decide the micro-blood flow living feature vector, skin texture feature map and facial temperature topology map; A dynamic weighting strategy is used to fuse the three-layer judgment results to form the final liveness judgment result.

4. The method for verifying the identity of power transmission line construction personnel based on face recognition according to claim 1, characterized in that: The establishing of a distributed feature matching network includes: Build a distributed feature matching network with a multi-level topology, including edge nodes, base station nodes, and central nodes. The three-level nodes are interconnected through heterogeneous links and collaboratively process feature matching tasks. The pre-stored feature database is clustered and partitioned according to feature similarity, and distributed across different computing nodes; The spectral stereo feature matrix is input into a distributed feature matching network and identity matching is performed using a multimodal feature matching engine; Establishing a network status perception mechanism to monitor the link quality between nodes in real time and calculate a network status index, and adaptively adjusting the compression level of the spectral stereo feature matrix based on the network status index; A hierarchical federated matching strategy is adopted to dynamically switch matching modes according to network conditions.

5. The method for verifying the identity of power transmission line construction personnel based on face recognition according to claim 1, characterized in that: The multi-dimensional cross-validation with the identity matching result includes: Collect position coordinate sequences and three-dimensional posture data to generate position behavior feature spectra including work movement paths and work activity distribution; Collect tool operation sequences and biomechanical characteristics to construct an operation behavior profile; Build a behavioral benchmark library, calculate the degree of deviation between the position behavior characteristic spectrum and the operation behavior characteristic spectrum and the behavioral benchmark library, and generate a behavioral matching score; Through the reasoning analysis model, the identity matching results and behavior matching scores are integrated to generate multi-level security assessment indicators; If an abnormal security assessment indicator is detected, a progressive identity re-verification mechanism will be triggered.

6. The method for verifying the identity of power transmission line construction personnel based on face recognition according to claim 1, characterized in that: Writing the identity verification process and multi-level security assessment indicators into the distributed ledger includes: Enforce distributed ledger consensus using a trust-weighted consensus mechanism; Digitally encapsulate key data points in the identity verification process and implement privacy protection for biometric data; Design a hierarchical smart contract system to automatically execute identity authentication rules through composable smart contracts; Generate tamper-proof authentication credentials using a layered encryption architecture; Based on the verification credentials, construction personnel are assigned hierarchical work permissions in real time, and a permission transfer mechanism is established to record permission changes in the distributed ledger.

7. The method for verifying the identity of power transmission line construction personnel based on face recognition according to claim 1, characterized in that: The multispectral facial data includes near-infrared spectrum data, visible light spectrum data, and thermal infrared spectrum data; the multi-level security assessment indicators include identity credibility, behavior coordination, risk warning level, and authority recommendation value.

8. A system for verifying the identity of power transmission line construction personnel based on face recognition, characterized in that: include: Spectral data processing module, used to collect multi-spectral facial data and process it using an environmental compensation neural network to generate a spectral three-dimensional feature matrix; The liveness detection module is used to extract micro-blood flow pulsation characteristics, skin texture characteristics, and thermal imaging temperature distribution characteristics from the spectral stereo feature matrix, build a three-layer cascade anti-counterfeiting verification mechanism, and perform comprehensive liveness judgment; The identity matching module is used to establish a distributed feature matching network, match the spectral stereo feature matrix with the pre-stored feature library, and dynamically adjust the feature compression level according to the network status; The behavior verification module is used to collect the construction workers' geographic location trajectories and operational behavior patterns, perform multi-dimensional cross-validation with the identity matching results, and generate multi-level safety assessment indicators; The authority management module is used to write the identity authentication process and multi-level security assessment indicators into the distributed ledger, generate verification credentials, and allocate construction personnel's work permissions.

Images