Unmanned education training method and system for traffic industry workers

Abstract

The invention discloses an unmanned education training method and system for traffic industry workers, and relates to the technical field of education management, and the method comprises the steps: obtaining the eye image data of a worker; verifying the identity of the worker according to the eye image data, and determining a target worker passing the identity verification; according to the learning record of the target worker, obtaining a knowledge point mastering probability vector of the worker through a knowledge tracking model; screening out a course candidate set by taking maximization of course knowledge coverage and minimization of course difficulty deviation as targets; according to the knowledge point mastering probability vector and the course candidate set, through a collaborative filtering recommendation model, recommending a course for the target worker; monitoring the learning state of the target worker; and performing integral excitation on the target worker according to the learning state. According to the invention, personalized curriculum recommendation of individualized teaching can be realized, and the learning enthusiasm of workers and the overall training effect are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of education management technology, and in particular to an unmanned education and training method and system for transportation industry workers. Background Technology

[0002] The unmanned education and training method for transportation industry workers refers to a training approach that utilizes technologies such as artificial intelligence, online learning platforms, and virtual reality for workers engaged in front-line positions such as transportation, road management, and vehicle maintenance. This method eliminates the need for teachers to provide on-site guidance throughout the training process, breaking through the limitations of traditional training in terms of time, space, and teachers. It enables workers to acquire professional skills and industry knowledge flexibly and efficiently, thereby meeting the rapid needs of the transportation industry's digital and intelligent development.

[0003] As the transportation industry rapidly develops towards digitalization and intelligence, frontline transportation workers need to continuously acquire new skills to adapt to job changes. Unmanned education and training methods utilize technologies such as artificial intelligence, online learning platforms, and virtual reality to provide workers with personalized, accessible vocational training anytime and anywhere without the need for teachers to provide full-time on-site guidance.

[0004] However, existing unmanned education and training methods rely heavily on account passwords or facial recognition for identity verification, which are easily misused and affected by factors such as lighting and occlusion, resulting in insufficient security. The course recommendation mechanism is also relatively simple, failing to balance the matching of knowledge coverage and course difficulty, which can easily lead to recommendation imbalances. Furthermore, there is a lack of effective incentive mechanisms based on multiple dimensions of behavior such as learning time, focus, and performance, resulting in insufficient worker motivation and poor overall training effectiveness. Summary of the Invention

[0005] To address the technical problems of existing unmanned education and training methods that rely heavily on account passwords or facial recognition for identity verification, which are easily impersonated and affected by factors such as lighting and occlusion, resulting in insufficient security, and the relatively simple course recommendation mechanism that fails to balance knowledge coverage with course difficulty, leading to recommendation imbalances, and the lack of an effective incentive mechanism based on multi-dimensional behaviors such as learning time, focus, and performance, resulting in insufficient learning motivation among workers, this invention provides an unmanned education and training method and system for transportation industry workers.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: First aspect: This invention provides an unmanned education and training method for transportation industry workers, including: S1: Acquire worker's eye image data; S2: Based on the eye image data, verify the identity of the target worker and determine the target worker who has passed the identity verification; S3: Based on the learning records of the target worker, the knowledge point mastery probability vector of the target worker is obtained through the knowledge tracking model. The knowledge tracking model specifically includes a hypergraph module, a self-attention module, a time dynamic module, a gating fusion module, and a prediction module. S4: Select a set of candidate courses with the goal of maximizing course knowledge coverage and minimizing course difficulty deviation; S5: Based on the knowledge point mastery probability vector and the course candidate set, recommend courses to the target worker through a collaborative filtering recommendation model; S6: Monitor the learning status of the target worker; S7: Based on the learning status, the target worker is given an integral incentive.

[0007] The second aspect: This invention provides an unmanned education and training system for transportation industry workers, comprising: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the unmanned education and training method for transportation industry workers as described in the first aspect.

[0008] Third aspect: An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the unmanned education and training method for transportation industry workers as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, identity verification based on iris recognition ensures the uniqueness of worker identities and the security of the training process, preventing impersonation. A knowledge tracking model dynamically and precisely characterizes workers' mastery of each knowledge point, improving the accuracy of knowledge tracking. With the goal of maximizing course knowledge coverage and minimizing course difficulty deviation, course selection is conducted. The selected course set ensures comprehensive knowledge coverage while achieving a balanced level of difficulty. A collaborative filtering recommendation model embeds workers' knowledge mastery vectors and the course candidate set into a unified space for interactive matching, thereby achieving personalized course recommendations tailored to individual needs. Real-time monitoring of learning status, combined with an incentive mechanism, incorporates learning time, focus, and assessment results into a comprehensive evaluation, effectively enhancing workers' learning enthusiasm and the overall training effect. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an unmanned education and training method for transportation industry workers, provided as an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of the structure of an unmanned education and training system for transportation industry workers, provided as an embodiment of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0014] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0015] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0016] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] Reference manual attached Figure 1 The diagram illustrates a flowchart of an unmanned education and training method for transportation industry workers provided by an embodiment of the present invention.

[0019] This invention provides an unmanned education and training method for transportation industry workers. This method can be implemented by an unmanned education and training system for transportation industry workers, which can be a terminal or a server. The processing flow of the unmanned education and training method for transportation industry workers may include the following steps:

[0020] S1: Obtain image data of the worker's eyes.

[0021] Among them, eye image data refers to digital image information containing features such as pupil, iris, and eyelids, acquired by near-infrared or visible light cameras.

[0022] S2: Verify the worker's identity based on eye image data and identify the target worker who has passed the identity verification.

[0023] It should be noted that identity verification through eye images can achieve high-precision verification of workers' identities with a low false positive rate, effectively preventing situations such as substitute learning or impersonation, ensuring the safety and fairness of the training process, and enhancing the credibility of unmanned education and training.

[0024] In one possible implementation, S2 specifically includes: S201: Verify the worker's identity based on eye image data.

[0025] In one possible implementation, S201 specifically includes: S2011: Based on eye image data, the boundaries of the pupil and iris are located by circular Hough transform to obtain the pupil boundary and iris boundary.

[0026] The circular Hough transform is an image processing algorithm based on parameter space voting, used to detect the boundary positions of circular objects in an image. The pupil boundary is the dividing curve between the pupil region and the iris region, and is usually approximately circular. The iris boundary is the dividing curve between the iris and the sclera (white part of the eye), and is usually circular.

[0027] Specifically, by using circular Hough transform to locate the boundary between the pupil and iris, key eye features can be accurately extracted under complex backgrounds and lighting conditions, avoiding the problem of traditional threshold segmentation being easily affected by noise. This provides a highly robust and accurate foundation for subsequent iris region extraction and identity verification.

[0028] S2012: Generate a ring-shaped ROI region based on the pupil boundary and iris boundary.

[0029] Among them, the annular ROI region refers to the annular area between concentric or nearly concentric circles jointly defined by the pupil boundary and the iris boundary. This region contains the main texture information of the iris and is a key area for subsequent feature extraction and identity verification.

[0030] Specifically, by generating a ring-shaped ROI region based on the pupil boundary and iris boundary, the effective texture area of ​​the iris can be effectively focused, irrelevant information such as the white of the eye and the pupil can be eliminated, noise interference can be reduced, thereby ensuring the stability and accuracy of subsequent feature extraction and improving the reliability of identity verification.

[0031] S2013: Based on the parameters of the pupil boundary and iris boundary, the eyelid boundary is located using parabolic Hough transform to obtain the eyelid boundary:

[0032] in,( x , y ) represents the Cartesian coordinates of pixels in the eye image data. h g , k g ) indicates the first g The coordinates of the vertex of the eyelid parabola Indicates the parabola of the eyelid g The rotation angle relative to the horizontal axis, cos represents the cosine function. a g Indicates the parabola of the eyelid g The curvature control parameter is denoted by sin, which represents the sine function.

[0033] Specifically, by using parabolic Hough transform to locate the eyelid boundary, the eyelid occlusion range can be accurately identified even in the presence of complex interferences such as eyelashes and light reflection. This effectively separates the usable area of ​​the iris, avoids invalid pixels from interfering with feature extraction and identity verification results, and thus significantly improves the robustness and accuracy of iris recognition.

[0034] S2014: Generate the eyelid mask area based on the eyelid boundary.

[0035] In this embodiment of the invention, a fixed height (e.g., 5-10 pixels) is extended downward (upper eyelid) or upward (lower eyelid) along the fitted eyelid parabola to form a rectangular area, namely the eyelid mask area.

[0036] S2015: Combining the annular ROI region and the eyelid mask region, the iris annular region is obtained.

[0037] Specifically, by combining the annular ROI region with the eyelid mask region, invalid pixels caused by interference from eyelids, eyelashes, or highlights are precisely eliminated, leaving only the effective texture area of ​​the iris. This significantly improves the stability of iris feature extraction and the accuracy of identity verification while ensuring feature integrity.

[0038] In this embodiment of the invention, the eye image data is first Gaussian smoothed and thresholded to remove specular highlights. Then, a circular Hough transform is used to locate the pupil boundary (outer circle) and the iris boundary (inner circle), respectively, thus solving for the centers and radii of the two circles. Next, a parabolic Hough transform is used to fit the arcs of the upper and lower eyelids, generating the eyelid occlusion region. Finally, the segmentation output is: ① The effective iris region, i.e., the annular ROI region between two concentric or nearly concentric circles. ② The eyelid noise mask region, i.e., the pixels covered by highlights, eyelashes, and eyelids. The annular iris region is the annular ROI region minus the eyelid noise mask region.

[0039] S2016: Using a rubber sheet model, the iris ring region is mapped from Cartesian coordinates to polar coordinates.

[0040] Specifically, by mapping the iris ring region from Cartesian coordinates to polar coordinates using a rubber sheet model, the normalized unfolding of the iris texture is achieved, effectively eliminating geometric distortions caused by differences in eye size, shooting distance, and pupil contraction. This ensures that subsequent feature extraction is performed at a uniform scale and orientation, improving the accuracy and robustness of the comparison.

[0041] S2017: Perform feature encoding on the mapped iris ring region to obtain a binary feature template.

[0042] Specifically, by encoding the features of the mapped iris ring region and generating a binary feature template, complex iris texture information can be transformed into a stable and compact coded representation, which can enhance the ability to resist noise and illumination changes, thereby significantly improving the efficiency and accuracy of identity verification.

[0043] In one possible implementation, S2017 specifically includes: S2017A: The mapped iris ring region is split into multiple one-dimensional signals.

[0044] S2017B: Complex response matrix is ​​obtained by performing complex convolution on each one-dimensional signal using a 1D Log-Gabor filter.

[0045] Among them, the 1D Log-Gabor filter is a bandpass filter based on a log-Gaussian distribution, possessing zero DC component and good frequency selectivity, and is often used to extract texture and edge features. A one-dimensional signal refers to the strip data obtained by splitting the iris annular region along the angular direction; each strip can be considered a one-dimensional gray-level intensity sequence. Complex convolution refers to using complex-form filtering operations to convolve the input signal with the filter, thereby simultaneously obtaining amplitude and phase information. The complex response matrix is ​​a matrix composed of the convolution results, where each element is a complex number containing the amplitude and phase characteristics of the iris texture at different frequencies.

[0046] Specifically, by performing complex convolution on the iris strip signal using a 1D Log-Gabor filter, subtle changes in iris texture in the frequency domain can be fully extracted. In particular, the stability of phase features is stronger than that of amplitude features, thus maintaining high discriminability and robustness under conditions of illumination, contrast changes, or partial occlusion.

[0047] S2017C: Calculate the phase angle of each complex response in the complex response matrix to obtain the phase angle matrix.

[0048] S2017D: The phase angle matrix is ​​quantized using Gray code to obtain a binary feature template.

[0049] The phase angle matrix is ​​a matrix composed of phase angle information extracted from the response obtained by complex convolution, reflecting the subtle structural features of the iris texture. Gray code quantization is a coding method where adjacent codes differ by only one bit, used to reduce abrupt changes and error accumulation during the quantization process.

[0050] Specifically, by quantizing the phase angle matrix using Gray code to obtain binary features, it is possible to reduce quantization errors and bit flipping risks while maintaining the distinguishability of phase information, making the generated feature template more stable and robust, thereby improving the accuracy and consistency of identity comparison.

[0051] S2018: Retrieve reference binary feature templates for registered workers in the database.

[0052] S2019: Determine whether the Hamming distance between the binary feature template and the reference binary feature template is less than a preset distance; if yes, determine that the worker has passed the identity verification; otherwise, determine that the worker has not passed the identity verification.

[0053] It should be noted that those skilled in the art can set the size of the preset distance according to actual needs, and the present invention does not limit this.

[0054] The Hamming distance is specifically:

[0055] Where HD represents the Hamming distance. X c The first character representing the binary feature template c Bit binary value, Y c Indicates the first reference binary feature template c Bit binary value, C This indicates the total number of bits in the binary feature template. Represents symbolic XOR, The symbols represent AND, Represents the symbol or, Xn c The first noise mask, aligned with the binary feature template. c Mask position, Yn c The reference noise mask representing the alignment of the reference binary feature template. c Mask position.

[0056] S202: Workers who pass identity verification are identified as target workers, and workers who fail identity verification are identified as non-target workers.

[0057] In this embodiment of the invention, eye image data is acquired using a (near-infrared preferred) camera. The iris is then segmented—the circular Hough localization of the pupil circle and the outer iris circle are used, and parabolic Hough fitting is applied to the upper and lower eyelids. Eyelashes / highlights are also mapped as noise masks. The resulting annular iris is normalized into fixed-size polar coordinate strips using a Daugman rubber membrane model. Phase is extracted along the angular direction on the strips using 1D Log-Gabor filtering, and binary templates and their aligned masks are generated using four-quadrant Gray code quantization (invalid bits = 1). During matching, the on-site template and the registered template in the library are cyclically shifted within a small range angular direction, and the Hamming distance with the mask is calculated only on the valid bits (the minimum value is taken as the matching score). Finally, a threshold decision is made: if the score meets the threshold, it is "passed"; otherwise, it is "failed".

[0058] S3: Based on the learning records of the target worker, the knowledge mastery probability vector of the target worker is obtained through the knowledge tracking model. The knowledge tracking model specifically includes a hypergraph module, a self-attention module, a time dynamic module, a gating fusion module, and a prediction module.

[0059] It should be noted that by introducing a knowledge tracking model that includes modules such as hypergraph, self-attention, temporal dynamics, gating fusion, and prediction, it is possible to fully capture the high-order relationships and temporal dependencies between knowledge points in the workers' historical answer sequences, effectively improving the dynamic accuracy of workers' knowledge mastery, thereby providing more scientific and accurate data support for subsequent course recommendations and personalized learning path planning.

[0060] In one possible implementation, the learning record is specifically the target worker's historical question-answering sequence.

[0061] Among them, the historical answer sequence refers to the record of the questions completed by workers during the learning process and their answers (correct / incorrect, chronological order, etc.).

[0062] S3 specifically includes: S301: In the hypergraph module, the historical answer sequence is mapped to a hypergraph to obtain the hypergraph association matrix.

[0063] A hypergraph is a generalization of a graph, where a single "hyperedge" can connect multiple nodes simultaneously, used to characterize complex multivariate relationships. The hypergraph incidence matrix represents the matrix of relationships between nodes and hyperedges in the hypergraph, used for subsequent computation and feature propagation.

[0064] Specifically, by mapping historical answer sequences to hypergraphs and generating hypergraph association matrices, we can overcome the limitations of traditional graph structures that can only describe pairwise relationships, fully express the high-order associations between knowledge points in workers' answering behavior, and thus more comprehensively and accurately characterize workers' learning features, laying a solid foundation for knowledge state modeling and mastery probability prediction.

[0065] In this embodiment of the invention, nodes in the hypergraph represent interaction instances between workers and questions, such as "a worker's answer to a question." Hyperedges represent knowledge points and can simultaneously connect multiple interaction nodes related to that knowledge point. The hypergraph can naturally express complex relationships where "a knowledge point may correspond to multiple questions, and multiple answer records are simultaneously associated with a knowledge point." Rows in the hypergraph association matrix correspond to nodes, and columns correspond to hyperedges. If an answer record involves a knowledge point, a value of 1 is assigned to the corresponding position in the matrix; otherwise, it is 0. Combined with weight information, different weights can be set for different knowledge points to reflect their importance in the learning process.

[0066] S302: Based on the hypergraph correlation matrix, perform convolutional updates on the interaction feature matrix corresponding to the historical answer sequence to obtain the interaction embedding.

[0067] Among them, the interaction feature matrix ( The feature representation matrix is ​​generated from the worker's historical answer sequence, containing information such as correct / incorrect answers, question numbers, and knowledge point associations. The hypergraph association matrix (H) describes the relationships between answer nodes and knowledge point hyperedges, used to characterize higher-order associations. Convolutional update is an operation that propagates and aggregates information in the hypergraph structure, equivalent to weighting and updating node features according to hyperedge connections. Interactive embedding (...) () is a low-dimensional dense representation obtained through hypergraph convolution, used to characterize the potential features of worker-knowledge point interaction.

[0068] Specifically, by updating the interaction feature matrix through hypergraph convolution, multiple interaction node features can be aggregated under the relational constraints of high-order hyperedges, thereby obtaining an embedded representation that better reflects the global dependence and multi-dimensional relationships of knowledge points, significantly improving the accuracy of characterizing the worker's knowledge state.

[0069] Specifically, the interactive embedding is as follows:

[0070]

[0071] in, Indicates interactive embedding, D v Represents a diagonal matrix of node degrees. D e Represents a hyperdiagonal matrix. H Represents the hypergraph incidence matrix. W Represents the hyperedge weight matrix. T Indicates transpose. Representing the interaction feature moments, This indicates a learnable linear transformation. This represents a hyperedge in a hypergraph. Denotes the set of superedges. () indicates the weight of the superedge. Represents a node v Does it belong to the superedge? e , =1, then node v Does it belong to the superedge? e , =0, then node v Does it belong to the superedge? e , v This represents a node in the hypergraph. V Represents a set of nodes.

[0072] In this embodiment of the invention, the target worker is the first m The correct answer to the exercise is recorded as ν m+ Incorrect answers are recorded as ν m- These two interactive nodes will be mapped to embedding vectors. and All these interactive embeddings are stacked together to form an interaction feature matrix. .

[0073] S303: In the self-attention module, global dependency modeling is performed on the interaction embedding to obtain the temporal state.

[0074] The self-attention module is a neural network mechanism that models global dependencies by calculating the correlation weights between any two positions in a sequence. Global dependency modeling refers to capturing the potential connections between any positions in a sequence during sequence modeling. The temporal state is a sequence representation obtained through the self-attention mechanism, containing dynamic information about the worker's knowledge acquisition as their interaction changes over time.

[0075] It should be noted that by using the self-attention mechanism to model the global dependency of interactive embeddings, we can break through the limitation of traditional time series models that only rely on neighboring information, fully capture the long-distance dependency relationship between different worker answering behaviors, and thus more comprehensively reflect the dynamic changes in knowledge mastery, significantly improving the modeling accuracy and expressive ability of learning states.

[0076] In this embodiment of the invention, the self-attention module adopts a Transformer-based encoder-decoder structure, and residual connections and layer normalization are used in each layer of the self-attention module. First, the interaction embedding updated by convolution is used as the query (Q), key (K), and value (V) of the encoder. The attention weights between interactions are calculated and output. That is, in each layer, the layers are stacked in the order of "multi-head self-attention (including causal mask) → residual connection + layer normalization → feedforward network (FFN) → residual connection + layer normalization", so that each time step establishes a global dependency with its entire history and produces a context representation Oenc. Subsequently, the query sequence embedding (i.e., the question the worker is about to answer) is right-shifted and positionally encoded to obtain the decoder input (as query Q). At each layer, masked multi-head self-attention is first performed to model the temporal dependency within the query. Then, encoder-decoder cross-attention is performed with Oenc as the key / value pair. The information most relevant to the current step is selected from the historical global representation and fused into the query representation. Next, it is passed through a feedforward network (two linear + nonlinear layers), and residual connections and layer normalization are applied outside each sub-layer to obtain the output of that layer. Several layers are stacked to form the final decoder representation Dec. Finally, temporal readout is performed on Dec (e.g., taking the last time Dec, or performing attention pooling on the entire segment) to obtain the temporal state, which is used for subsequent gating fusion or prediction.

[0077] S304: In the time dynamics module, perform time-series modeling on the interaction feature matrix to obtain the knowledge state.

[0078] The temporal dynamics module refers to a computational unit in the model specifically designed to capture the evolution of the learning process over time, implemented as a gated recurrent unit (GRU). Temporal modeling involves explicitly considering the time dimension during the modeling process, analyzing the temporal dependencies and evolutionary patterns of learning behavior. The knowledge state is a representation vector obtained through temporal modeling, reflecting the worker's mastery level and trends of various knowledge points at different time stages.

[0079] Specifically, by using a time dynamic module to perform time-series modeling of the interaction feature matrix, it is possible to effectively characterize the pattern of changes in workers' learning behavior over time, capture the accumulation and forgetting characteristics of learning effects, and thus make the knowledge state representation closer to the real learning process, significantly improving the prediction accuracy and reliability of workers' knowledge mastery.

[0080] In this embodiment of the invention, the interaction feature matrix is ​​first expanded over time to obtain the time-series input. x t Then, GRU is used for temporal modeling, the hidden state is updated step by step, and the latest hidden state is mapped to the knowledge point space to obtain the knowledge state of the "GRU branch".

[0081] S305: In the gating fusion module, the temporal state and the knowledge state are fused to obtain the fused knowledge state:

[0082]

[0083] in, Indicates the first i The fusion knowledge state of each knowledge point, where gate represents the fusion weight vector. This indicates element-wise multiplication. Indicates the first i The knowledge status of each knowledge point Indicates the first i The temporal state of each knowledge point This represents the Sigmoid activation function. W g1 and W g2 Both represent weight matrices. b g1 and b g2 Both represent bias vectors.

[0084] Specifically, by dynamically weighting and integrating temporal states and knowledge states through a gating fusion module, the one-sidedness of single feature representation can be avoided. While taking into account global dependencies and temporal evolution patterns, the integrity and robustness of knowledge representation can be improved, thereby more accurately reflecting the true mastery level of workers.

[0085] S306: In the prediction module, the fused knowledge state is mapped to probability to obtain the knowledge point mastery probability vector of the target worker.

[0086]

[0087]

[0088] in, Indicates the target worker in the i The probability of mastering each knowledge point w 0 represents the weight vector. b 0 indicates the bias term. This represents the probability vector of the target worker's mastery of certain knowledge points. i =1,2,…, K , K This indicates the total number of knowledge points.

[0089] Among them, the knowledge point mastery probability vector is a vector form representing the degree of mastery of each knowledge point by the worker. The value ranges from 0 to 1, and the larger the value, the higher the degree of mastery.

[0090] Specifically, by mapping the fused knowledge state to probabilities through the prediction module, an intuitive knowledge point mastery probability vector is obtained. This can accurately depict the worker's mastery level on different knowledge points in a quantitative form, which not only facilitates personalized learning path planning and course recommendation, but also improves the interpretability and operability of the system for evaluating learning effectiveness.

[0091] S4: Select a set of candidate courses with the goal of maximizing course knowledge coverage and minimizing course difficulty deviation.

[0092] Among them, course knowledge coverage refers to the scope and proportion of knowledge points that the course set can cover. The higher the coverage, the broader the knowledge that workers can learn. Course difficulty deviation refers to the difference between the overall difficulty of the course and the set target difficulty. The smaller the deviation, the better the course combination matches the actual level of the workers.

[0093] It should be noted that by aiming to maximize the coverage of course knowledge and minimize the deviation in course difficulty, we can ensure the comprehensiveness of the learning content for workers while avoiding situations where the course difficulty is too high or too low. This ensures that the selected course set is both broad and adaptable, thereby significantly improving the scientific nature and personalized effect of the training plan.

[0094] In one possible implementation, S4 specifically includes: S401: Set the first objective function with the goal of maximizing the coverage of course knowledge; set the second objective function with the goal of minimizing the deviation in course difficulty. F1

[0095]

[0096] Where F1 represents the first objective function and F2 represents the second objective function. max To maximize, min Indicates minimization. A iIndicates the first i Each knowledge point corresponds to a chapter. w i Indicates the first i The weight of each knowledge point corresponding to the chapter. M Indicates the difficulty of the objective. express i The difficulty of each knowledge point.

[0097] S402: Based on the first objective function and the second objective function, the leech optimization algorithm is used to select a set of candidate courses.

[0098] Among them, the leech optimization algorithm is a biomimetic intelligent optimization algorithm that simulates the foraging and movement behavior of leeches. It quickly finds the optimal or near-optimal solution by combining global search and local search.

[0099] Specifically, by employing the leech optimization algorithm, global search and local exploration are performed under the dual constraints of the first and second objective functions. This enables efficient finding of course combinations that balance knowledge coverage and difficulty suitability in large-scale course resources, avoiding the pitfalls of traditional greedy or random methods in local optima, thereby ensuring the scientific nature and optimization effect of the course candidate set.

[0100] S5: Based on the knowledge point mastery probability vector and the course candidate set, recommend courses to the target workers through a collaborative filtering recommendation model.

[0101] It should be noted that by combining the probability of workers mastering knowledge points with the candidate course set through collaborative filtering recommendation models, a balance can be achieved between individual differences and group preferences, realizing personalized and dynamic course recommendations. This avoids a "one-size-fits-all" approach to course allocation and significantly improves the relevance and efficiency of learning.

[0102] In one possible implementation, the collaborative filtering recommendation model specifically includes: an embedding layer, an embedding propagation layer, a cascading layer, a matching layer, and a recommendation layer.

[0103] In one possible implementation, S5 specifically includes: S501: In the embedding layer, the knowledge point mastery probability vector and the course candidate set are linearly mapped to the dense embedding space to obtain the initial embedding matrix. The initial embedding matrix includes the target worker embedding sequence and the course embedding sequence. The target worker embedding sequence includes multiple worker embedding vectors, and the course embedding sequence includes multiple course embedding vectors.

[0104] in, E Represents the initial embedding matrix. Indicates target workers nThe embedding vector is obtained by mapping the probability vector of knowledge points mastered. Indicates course m The embedding vector is obtained by mapping from the course candidate set. n =1,2,…, U , U This represents the total number of target workers. m =1,2,…, M , M This represents the total number of courses in the candidate course set.

[0105] Among them, the dense embedding space is a space composed of continuous vectors of fixed dimension d, which is used to uniformly represent workers and courses, facilitating subsequent graph propagation and matching calculations.

[0106] Specifically, by mapping worker mastery vectors and course attributes to the same dense space, the sparsity and dimensional differences of features are significantly reduced, and the alignability and separability of workers and courses in graph propagation and matching are enhanced, thus providing a stable and learnable unified representation foundation for personalized recommendations.

[0107] S502: In the embedding propagation layer, based on the initial embedding matrix, construct the self-connection information graph of the target worker, the self-connection information graph of the course, and the transitive information graph between the target worker and the course:

[0108]

[0109] ,

[0110] in, l Indicates the index number of the embedded propagation layer. Indicates the first l Layer from course v Passed to target workers u The message vector, N u Indicates target workers u Neighborhood course collection, N v Indicates course v The neighboring workers gathered. W 1. W 2. W 3 and W 4 indicates the first l The message linear transformation matrix of the layer, Indicates course v Attention score Indicates the first l Layered course embedding vectors, Indicates the firstl The target worker embedding vector of the layer, w 1 represents the weight vector. This indicates element-wise nonlinearity. b 3 and b 4 indicates bias. Indicates the first l A self-connected information graph of the target workers in the layer. Indicates the first l A self-connected infographic of courses in a layer.

[0111] Among them, the self-connected infographic is a self-looping graph within each worker or course, used to maintain the continuity of its own characteristics during the update process. The transitive infographic represents the graph structure of the interaction relationship between the target worker and the course, reflecting the worker's interest in the course or the course's suitability for the worker.

[0112] Specifically, by simultaneously constructing worker self-connection graphs, course self-connection graphs, and worker-course transfer graphs, it is possible to retain the characteristics of nodes themselves while integrating cross-domain interactive information during information dissemination, thereby achieving multi-level and multi-directional feature updates for workers and courses, and thus improving the expressiveness of the recommendation model and the accuracy of personalized recommendations.

[0113] S503: Perform message synthesis on the transit information graph and the self-connected information graph of the target worker to update the target worker embedding sequence.

[0114] S504: Perform message synthesis on the delivery information graph and the self-connected information graph of courses to update the course embedding sequence.

[0115] Specifically, by synthesizing and updating the self-connection information with the messages passed from neighboring nodes, it is possible to ensure that the target worker and the course retain their own characteristics while fully absorbing the interaction information of each other during the embedding and propagation process, thereby achieving dynamic fusion of cross-domain information and improving the expressive power of worker-course embedding and the accuracy of recommendation results.

[0116] In this embodiment of the invention, message synthesis is performed on the transit information graph and the self-connected information graph of the target worker using the nonlinear function LeakyReLU, and the target worker embedding sequence is updated. Similarly, message synthesis is performed on the transit information graph and the self-connected information graph of the course using the nonlinear function LeakyReLU, and the course embedding sequence is updated.

[0117] S505: In the cascaded layer, the updated target worker embedding sequence and the course embedding sequence are concatenated to obtain the final worker embedding and the final course embedding:

[0118] in, This indicates that the final worker is embedded. This represents the initial embedding of the target worker. This indicates the final course embedding. This indicates the initial embedding of the course. L This indicates the total number of embedded propagation layers.

[0119] Specifically, by cascading worker and course embeddings at different levels, not only is the original semantic information of the initial features preserved, but also the higher-order interaction features after multi-layer propagation are integrated, thus forming a multi-granularity, multi-perspective final representation, which significantly improves the model's expressive power and discrimination accuracy in course matching and recommendation.

[0120] S506: In the matching layer, the final worker embedding and the final course embedding are interactively matched to obtain the matching degree between the target worker and the course:

[0121] in, G Indicates the degree of matching. a L Indicates the first L The activation function of the layer, h L Indicates the first L The fully connected weight matrix of the layer, b L Indicates the first L The bias corresponding to the fully connected weight matrix of the layer.

[0122] Specifically, by performing interactive calculations on the final worker embedding and course embedding through the matching layer, not only can multi-level feature information be fully utilized, but also the deep relationship between workers and courses can be captured through non-linear mapping, thereby obtaining an accurate matching degree. This effectively improves the targeting and accuracy of course recommendations and avoids the recommendation results from being based on superficial similarity or single-dimensional judgments.

[0123] S507: In the recommendation layer, the matching degree is sorted in descending order, and courses are recommended to the target workers according to the sorting order.

[0124] S6: Monitor the learning status of target workers.

[0125] S7: Incentivize target workers with points based on their learning progress.

[0126] It should be noted that by quantifying status factors such as learning time, focus, and assessment results into points and incentivizing target workers, it is possible to effectively enhance workers' learning motivation in an unmanned training environment, establish a positive feedback mechanism between learning and rewards, thereby increasing workers' participation and willingness to continue learning, and ensuring the effectiveness and completion rate of the training process.

[0127] In one possible implementation, S7 specifically includes: S701: Based on the learning status, a weighted algorithm is used to calculate the learning score, where the learning status specifically includes: learning duration, concentration level, and assessment score.

[0128] In this embodiment of the invention, a learning score is obtained by weighted summation of learning time, concentration, and assessment results.

[0129] S702: Reward target workers with points based on their learning achievements.

[0130] In this embodiment of the invention, the points and goods redemption system are linked, allowing workers to independently redeem rewards and incentivize continuous learning. After completing the learning process, the system automatically pushes assessment questions corresponding to the learned courses, including multiple-choice questions, scenario simulations, and other question types. The system automatically evaluates the assessment results and archives them in individual electronic files, which can be accessed by managers at any time. The aggregated data is periodically used to generate personalized learning and assessment reports, providing a basis for managers to make precise training and staffing decisions.

[0131] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, identity verification based on iris recognition ensures the uniqueness of worker identities and the security of the training process, preventing impersonation. A knowledge tracking model dynamically and precisely characterizes workers' mastery of each knowledge point, improving the accuracy of knowledge tracking. With the goal of maximizing course knowledge coverage and minimizing course difficulty deviation, course selection is conducted. The selected course set ensures comprehensive knowledge coverage while achieving a balanced level of difficulty. A collaborative filtering recommendation model embeds workers' knowledge mastery vectors and the course candidate set into a unified space for interactive matching, thereby achieving personalized course recommendations tailored to individual needs. Real-time monitoring of learning status, combined with an incentive mechanism, incorporates learning time, focus, and assessment results into a comprehensive evaluation, effectively enhancing workers' learning enthusiasm and the overall training effect.

[0132] Reference manual attached Figure 2 The diagram shows a structural schematic of an unmanned education and training system for transportation industry workers provided by the present invention.

[0133] This invention also provides an unmanned education and training system 20 for transportation industry workers, applied to the aforementioned unmanned education and training method for transportation industry workers, comprising: Processor 201.

[0134] The memory 202 stores computer-readable instructions, which, when executed by the processor 201, implement the unmanned education and training method for transportation industry workers as described in the method embodiment.

[0135] The unmanned education and training system 20 for transportation industry workers provided by this invention can execute the above-mentioned unmanned education and training method for transportation industry workers and achieve the same or similar technical effects. To avoid repetition, this invention will not elaborate further.

[0136] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0137] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0138] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0139] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0140] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0141] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0142] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0143] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0144] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0145] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0146] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0147] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the unmanned education and training method for transportation industry workers as described in the method embodiment.

[0149] The computer-readable storage medium provided by this invention can realize the steps and effects of the unmanned education and training method for transportation industry workers in the above-described method embodiments. To avoid repetition, this invention will not repeat them.

[0150] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0151] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0152] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0153] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0154] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for unmanned education and training of transportation industry workers, characterized in that, include: S1: Acquire worker's eye image data; S2: Based on the eye image data, verify the worker's identity and identify the target worker who has passed the identity verification; S3: Based on the learning records of the target worker, the knowledge point mastery probability vector of the target worker is obtained through the knowledge tracking model. The knowledge tracking model specifically includes a hypergraph module, a self-attention module, a time dynamic module, a gating fusion module, and a prediction module. S4: Select a set of candidate courses with the goal of maximizing course knowledge coverage and minimizing course difficulty deviation; S5: Based on the knowledge point mastery probability vector and the course candidate set, recommend courses to the target worker through a collaborative filtering recommendation model; S6: Monitor the learning status of the target worker; S7: Based on the learning status, the target worker is given an integral incentive.

2. The unmanned education and training method for transportation industry workers according to claim 1, characterized in that, S2 specifically includes: S201: Verify the worker's identity based on the eye image data; S202: Workers who pass the identity verification are identified as target workers, and workers who fail the identity verification are identified as non-target workers.

3. The unmanned education and training method for transportation industry workers according to claim 2, characterized in that, S201 specifically includes: S2011: Based on the eye image data, the pupil and iris are located by circular Hough transform to obtain the pupil boundary and iris boundary. S2012: Generate an annular ROI region based on the pupil boundary and the iris boundary; S2013: Based on the parameters of the pupil boundary and the iris boundary, the eyelid boundary is located by parabolic Hough transform to obtain the eyelid boundary; S2014: Generate an eyelid mask area based on the eyelid boundary; S2015: Combining the annular ROI region and the eyelid mask region, the iris annular region is obtained; S2016: Using a rubber sheet model, the iris annular region is mapped from Cartesian coordinates to polar coordinates; S2017: Perform feature encoding on the mapped iris ring region to obtain a binary feature template; S2018: Retrieve reference binary feature templates for registered workers in the database; S2019: Determine whether the Hamming distance between the binary feature template and the reference binary feature template is less than a preset distance; if yes, determine that the worker has passed the identity verification; otherwise, determine that the worker has not passed the identity verification.

4. The unmanned education and training method for transportation industry workers according to claim 3, characterized in that, S2017 specifically includes: S2017A: The mapped iris ring region is split into multiple one-dimensional signals; S2017B: By using a 1D Log-Gabor filter, complex convolution is performed on each of the aforementioned one-dimensional signals to obtain a complex response matrix; S2017C: Calculate the phase angle of each complex response in the complex response matrix to obtain the phase angle matrix; S2017D: The phase angle matrix is ​​quantized using Gray code to obtain the binary feature template.

5. The unmanned education and training method for transportation industry workers according to claim 1, characterized in that, The learning record is specifically the target worker's historical question-answering sequence; S3 specifically includes: S301: In the hypergraph module, the historical answer sequence is mapped to a hypergraph to obtain the hypergraph association matrix; S302: Based on the hypergraph association matrix, perform convolutional updates on the interaction feature matrix corresponding to the historical answer sequence to obtain the interaction embedding; S303: In the self-attention module, global dependency modeling is performed on the interaction embedding to obtain the temporal state; S304: In the time dynamic module, the interaction feature matrix is ​​modeled temporally to obtain the knowledge state; S305: In the gated fusion module, the temporal state and the knowledge state are fused to obtain a fused knowledge state; S306: In the prediction module, the fused knowledge state is mapped to a probability to obtain the knowledge point mastery probability vector of the target worker.

6. The unmanned education and training method for transportation industry workers according to claim 1, characterized in that, S4 specifically includes: S401: Set a first objective function with the goal of maximizing the knowledge coverage of the course; set a second objective function with the goal of minimizing the difficulty deviation of the course; S402: Based on the first objective function and the second objective function, the leech optimization algorithm is used to select the course candidate set.

7. The unmanned education and training method for transportation industry workers according to claim 1, characterized in that, The collaborative filtering recommendation model specifically includes: an embedding layer, an embedding propagation layer, a cascading layer, a matching layer, and a recommendation layer.

8. The unmanned education and training method for transportation industry workers according to claim 7, characterized in that, S5 specifically includes: S501: In the embedding layer, the knowledge point mastery probability vector and the course candidate set are linearly mapped to a dense embedding space to obtain an initial embedding matrix, wherein the initial embedding matrix includes a target worker embedding sequence and a course embedding sequence, the target worker embedding sequence includes multiple worker embedding vectors, and the course embedding sequence includes multiple course embedding vectors; S502: In the embedding propagation layer, based on the initial embedding matrix, a self-connection information graph of the target worker, a self-connection information graph of the course, and a transmission information graph between the target worker and the course are constructed; S503: Perform message synthesis on the transmission information graph and the self-connected information graph of the target worker to update the target worker embedding sequence; S504: Perform message synthesis on the transmission information graph and the self-connection information graph of the course to update the course embedding sequence; S505: In the cascaded layer, the updated target worker embedding sequence and course embedding sequence are cascaded to obtain the final worker embedding and the final course embedding. S506: In the matching layer, the final worker embedding and the final course embedding are interactively matched to obtain the matching degree between the target worker and the course; S507: In the recommendation layer, the matching degree is sorted in descending order, and the courses are recommended to the target worker according to the sorting order.

9. The unmanned education and training method for transportation industry workers according to claim 1, characterized in that, Specifically, S7 includes: S701: Based on the learning state, a weighted algorithm is used to calculate the learning integral, wherein the learning state specifically includes: learning duration, concentration level, and assessment score; S702: Based on the learning points, reward the target worker with points.

10. An unmanned education and training system for transportation industry workers, characterized in that: include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the unmanned education and training method for transportation industry workers as described in any one of claims 1 to 9.

Images