458 results about "Proactive learning" patented technology

The invention provides a knowledge base and business system collaboration method under an AI platform, and belongs to the technical field of AI platform digital data processing.The method includes the steps that triple semantic analysis is conducted on query by constructing a multi-level semantic vector representation module, a parallel data retrieval engine is started, vector retrieval, graph reasoning and real-time data pulling are executed at the same time, and the query efficiency is improved; establishing a dynamic confidence evaluation mechanism to evaluate the quality of the data source, executing a weight distribution algorithm based on reinforcement learning, dynamically calculating the weight of the data source according to a query type by adopting a graph convolutional network intelligent routing decision model, and implementing multi-source data fusion and consistency verification to solve data conflicts through a weighted voting mechanism. An intelligent result sorting and filtering system is established, a multi-dimensional evaluation strategy is adopted to output high-quality answers, a continuous learning and feedback optimization loop is constructed, system performance is continuously optimized through active learning and a graph shortest path algorithm, and the technical problem that knowledge base data and a service system cannot effectively and uniformly make decisions is solved.

A knowledge graph construction method based on active learning and incremental learning comprises the following steps: S1, preprocessing data from a plurality of heterogeneous data sources, and extracting entities, relationships and attributes to form an initial knowledge network; s2, vectorizing elements in the initial knowledge network by using a knowledge graph embedding model, and performing entity alignment based on vector similarity to obtain an initial knowledge graph; s3, screening out candidate knowledge triples with high uncertainty and / or high representativeness from the initial knowledge graph by adopting an active learning strategy, and obtaining user labeling information corresponding to the candidate knowledge triples; s4, performing iterative optimization on a knowledge extraction model and / or a knowledge graph embedding model according to the user labeling information; s5, new data are fused into the optimized knowledge graph in an incremental learning mode, and knowledge conflict detection and resolution are carried out in the fusion process; and S6, circularly executing the steps S3 to S5 until the knowledge graph meets a preset quality condition.

The invention discloses an active learning and physical loss-based fatigue crack growth rate prediction method, which comprises the following steps of: splicing and fusing a preprocessed stress intensity factor, a stress ratio, pre-strain and stress amplitude, and taking the fused characteristics as the input of a model; embedding the Jones model into a loss function of the neural network, constructing a physical information time sequence model based on the Jones model, setting a parameter set, selecting a most valuable training sample from a training set by using active learning, retraining the model by using the selected most valuable training sample, selecting an optimal model, and evaluating the optimal model by using a test set; fusing the actually measured stress intensity factor, stress ratio, pre-strain and stress amplitude of the to-be-predicted material, and inputting into the obtained prediction model to obtain a predicted value of the fatigue crack growth rate. By adopting the technical scheme of the invention, the fatigue crack growth rate of the material in the whole life period can be efficiently and accurately predicted at low cost.

The invention discloses an Internet of Vehicles CAN bus intrusion detection method based on noise perception active learning, and belongs to the technical field of Internet of Vehicles safety and machine learning. The invention aims to solve the technical problems of false label noise interference, high manual labeling cost, high attack missing report rate caused by class imbalance and the like. The core of the method is to execute a noise sensing mixed query strategy in an iterative loop: firstly, generating a pseudo tag through clustering and correcting by using an integrated noise detector; secondly, calculating uncertainty scores and noise probabilities of the samples, fusing the uncertainty scores and the noise probabilities to obtain a comprehensive score, and preferentially selecting the samples with high uncertainty and low noise probabilities; and then adaptively selecting a sampling strategy according to the model performance and applying category balance constraint. The query batch is used to iteratively update the model while dynamically adjusting the classification threshold to reduce the missing report rate. According to the method, the influence of pseudo label noise can be effectively suppressed, and the attack detection precision and generalization capability are remarkably improved with extremely low labeling cost.

The invention provides a collaborative classification method and system fusing advantages of large and small models, and the method comprises the steps: inputting to-be-classified data into a trained zero-sample classification small model, and outputting candidate label domain screening information and an initial classification result; inputting the to-be-classified data, the candidate label domain screening information and the initial classification result into the trained large model, and outputting a final classification result; the training process of the small model comprises the following steps: inputting training data into the zero sample classification small model to obtain a preliminary prediction result; screening and obtaining pseudo label data based on an active learning strategy; checking and re-marking the pseudo-label data by using the large model to obtain a modified pseudo-label data set; and carrying out iterative training on the zero sample classification small model by utilizing the modified pseudo label data set. The method has the advantages that a small model has classification performance close to that of a large model while keeping lightweight calculation characteristics; the calculation burden of a large model is reduced, and the accuracy of classification decision is improved.

The invention relates to the technical field of internet big data and information security, and designs a gearbox fault diagnosis method based on multi-source agent federal field generalization by combining the interaction capability of wide-area data and a local diagnosis agent. According to the method, a dual regularization mechanism is introduced in the training process of each local diagnosis agent, and the problem of negative migration is solved by actively learning domain-invariant fault features, so that the cross-working-condition and cross-equipment generalization diagnosis performance of a final model is remarkably enhanced on the premise of protecting data privacy; meanwhile, a central server is replaced by a decentralized coordination network realized by a block chain technology, and a committee consensus protocol is executed to ensure that the generation process of the global model is open, transparent and non-tampering, so that the single-point fault risk is fundamentally eliminated, the security and robustness guarantee is provided for the whole coordination process, and the method has the advantages of being high in practicability and the like. And the long-term applicability and the fault diagnosis accuracy and reliability of the method in a dynamically changing industrial environment are ensured.

The invention belongs to the technical field of artificial intelligence, and particularly relates to an artificial intelligence data annotation and cue word automatic construction engine system, which comprises a data annotation module for firstly carrying out preliminary annotation on data based on a pre-training model, then automatically annotating a data sample through an active learning algorithm, and meanwhile, monitoring the quality of annotated data in real time; and the cue word automatic construction module generates cue words based on task analysis, optimizes the cue words by using a reinforcement learning technology, and performs classified storage and management on the generated cue words. By adopting the semi-automatic labeling and active learning labeling functions, not only can the workload be greatly reduced, but also the unnecessary labeling work can be reduced, so that the labeling efficiency is remarkably improved, the labeling time is shortened, the large-scale data labeling requirement is met, and the problem of efficiency bottleneck caused by slow manual labeling is solved.

The invention relates to the technical field of new energy power system instability risk assessment, in particular to a new energy power system frequency instability risk assessment method based on a heterogeneous graph attention network. The method comprises the following steps: constructing a wind power-photovoltaic heterogeneous graph model according to a power grid topology; performing combined sampling on different operation modes and anticipated disturbances, and determining corresponding input feature vectors; calculating a frequency stability index value label of the sample set; expanding the training set through an active learning iteration process, and carrying out model training; and inputting the collected operation data into the trained heterogeneous graph attention model, outputting a frequency stability index value, and evaluating the system frequency instability risk in combination with the risk matrix. By adopting the frequency instability risk assessment method for the new energy power system based on the heterogeneous graph attention network, the problem of low efficiency of risk assessment in a high-dimensional uncertain scene is solved, and the heterogeneous graph attention network can reflect the influence of different types of devices at different positions and disturbance types on the dynamic frequency of the system; and the accuracy of frequency instability risk assessment is improved.

The invention discloses a concrete strength detection method based on image recognition. The method comprises the following steps: generating a corrected concrete image data set; outputting a shared feature map set; obtaining a semantic mask set by adopting a rasterized pyramid decoder and a channel attention mechanism; performing microstructure feature statistics based on the semantic mask set to obtain a microstructure index set; outputting a concrete strength predicted value set through a transfer learning and fine tuning regression mechanism; performing incremental updating on the strength prediction network by using a pseudo label-active learning loop mechanism to obtain an updated strength prediction network; and reasoning the corrected concrete image data set by using the updated strength prediction network to generate a concrete strength visualization result map set. According to the method, the interference of a non-intensity correlation noise region on a segmentation result is effectively suppressed, and the accuracy of a microstructure index is remarkably improved.

The invention provides a multi-objective optimization method fusing a classifier and an active learning strategy, and belongs to the technical field of artificial intelligence and optimization design, and the method specifically comprises the steps: obtaining a sample data set containing a plurality of target performance function values; performing non-dominated sorting on the sample data set to extract a Pareto frontier point set, and extracting geometric features from the point set as a geometric feature vector sequence; using the geometric features and the corresponding convex marks to train a classifier model of a Transform architecture, wherein the classifier model supports a multi-head attention mechanism and a position coding structure; judging the convex confidence coefficient of the current Pareto front by using a classifier model, and obtaining a candidate scheme; and inputting the candidate scheme into a double-attention mechanism prediction model based on context reasoning, and predicting the multi-target performance of the candidate scheme. According to the method, the experiment frequency can be reduced, the target performance can be accelerated to be achieved, and the intelligence and efficiency of the multi-target optimization process are improved.

The invention discloses a dynamic active learning and knowledge graph agent-based airworthiness document review method, which comprises the following steps of: collecting and processing a multi-modal heterogeneous data source to obtain a standardized knowledge base; based on the standardized knowledge base, constructing an aviation domain knowledge system by using dynamic active learning and a man-machine cooperation mechanism; constructing a knowledge graph and performing knowledge enhancement; and dynamically constructing a review context based on a retrieval enhancement generation method in combination with the knowledge graph, and injecting an intelligent agent to perform intelligent data review and report generation. According to the method, automatic review of airworthiness documents is realized through a dynamic active learning-agent review method, and the manual review efficiency is remarkably improved; domain knowledge is injected into an intelligent agent by utilizing an RAG technology, so that the accuracy of a general large language model in an airworthiness review task is improved; through knowledge semantic driving and human-in-the-loop knowledge enhancement, the problem of scarcity of high-quality data in the field is relieved; models of different parameter scales are cooperatively utilized, and computing power resources are saved.

The invention discloses a semi-automatic labeling method and system for rail transit engineering construction video images, and the method comprises the steps: removing redundant frames from a video stream, extracting key frames, and forming a block index set; and driving the multi-modal large model to pre-annotate the image by using the constructed cue word, and carrying out binarization processing on an annotation result. Based on active learning, migrating unmarked samples to a marked set for multiple times and training a key sample classifier CD until the scale of the marked set reaches the standard; meanwhile, a confidence classifier CC is trained by comparing manual and pre-labeling results. And for residual samples in the unmarked set, after the residual samples reach the standard through a CC precision test, marking tasks are divided according to a threshold value theta: high-confidence samples are pre-marked, and low-confidence samples are manually marked. And finally, updating the set of the manually labeled samples, and determining whether to retrain the model or not according to category distribution. On the premise of ensuring the labeling quality, the blindness of manual labeling is effectively reduced, and the efficiency of the labeling process is improved.

The invention provides a river flow online measurement and calculation method based on multi-source data fusion, and belongs to the technical field of river flow measurement. A Bayesian neural network and active learning collaborative hydrological memory reconstruction model is adopted to carry out high-confidence data interpolation on a sensor failure period and quantify uncertainty, fractional calculus is introduced to model a river memory effect, and flow evolution is analyzed through a fractional order water balance equation. And selecting a steady-state or non-constant flow calculation mode according to the flow change rate, inverting an optimal flow field for the non-constant flow by adopting a four-dimensional variational data assimilation method in combination with regularization constraint and time smoothing constraint, and outputting a flow measurement result and an uncertainty quantitative index. The technical problems that flow measurement and calculation data are missing and accurate interpolation is difficult due to sensor failure under the extreme hydrological condition are solved.

The invention discloses a power system multi-scene adaptive transient stability evaluation method based on active transfer learning, and the method comprises the following steps: constructing a transient stability evaluation model based on a Transform encoder, building a mapping relation between data and a stable state, and obtaining the source domain data of a power system; constructing a pre-training model based on the source domain data, and when it is detected that the pre-training model enters a new electrical system scene, performing short-term simulation to obtain an unlabeled target domain sample; calculating the information amount of each target domain sample through an active learning strategy; and carrying out fine adjustment on parameters of the model by utilizing transfer learning, selecting a transfer scheme by considering a difference degree between data distribution, enabling the transient stability evaluation model to quickly finish final updating training, evaluating the transient stability of the power system by using the updated model, and outputting a transient stability evaluation result. According to the method, the transient stable state of the power system can be accurately evaluated, and the model updating efficiency is higher.

The invention relates to the technical field of visual identification, and discloses an annotation, evaluation and learning integrated visual identification platform, which comprises a data acquisition module, an annotation enhancement module, a detection model training module, a dynamic evaluation module, an active learning module and an optimization closed loop module. Through integration of data acquisition, annotation enhancement, detection model training, dynamic evaluation, active learning, closed loop optimization and other modules, full-process collaboration is realized based on an improved YOLOv8 algorithm, and the annotation efficiency and quality are improved through auxiliary annotation, quality verification and cross-modal annotation association functions of the annotation enhancement module. The ICR algorithm and the sample screening strategy of the active learning module solve the cold start problem and reduce the labeling workload, and the dynamic evaluation module realizes model interpretability quantification through a class activation thermodynamic diagram, a MobileSAM segmentation mask, N-IoU, N-Recall and other indexes, so that an efficient and accurate one-stop solution is provided for a visual identification task.

Provided in the present application are an active learning and augmentation method for a complex skill in a heterogeneous scenario and a sim-to-real transfer method. The active learning and augmentation method comprises: first, collecting state data by means of the interaction of multiple executors and an environment; storing same in a shared experience replay buffer; after sampling, alternately training the executors and discriminators; and finally selecting the optimal executor to be deployed, so as to implement active learning and augmentation of a complex skill. The sim-to-real transfer method comprises: in a 3C assembly digital twin environment, collecting multi-modal data to construct a skill knowledge base; generating a policy sequence; using technologies such as residual reinforcement learning to perform optimization; and, by means of a coding and decoding model, transferring a skill policy from simulation to reality. Without a teacher model or teaching data, the solution provided in the present application improves skill learning efficiency by means of combining reinforcement learning and knowledge distillation techniques, and implements the sim-to-real transfer of operating skills from a simulation environment to a real assembly environment.

The invention belongs to the technical field of steel defect detection in a real industrial scene, and particularly relates to a steel defect detection method combined with an active learning strategy, which comprises the following steps: constructing a steel surface defect data set; performing initial training based on the lightweight YOLO model; screening out a high-value sample from an unlabeled steel image through an uncertainty evaluation strategy; preferentially performing expert labeling on the selected samples to form an updated training set; performing model training on the updated training set and dynamically optimizing the network weight; a new sample is introduced into each round of iteration, an active learning closed-loop system is continuously improved, and the detection precision under the small sample condition is improved. According to the method, the manual marking cost is effectively reduced, the cross-scene adaptive capacity is improved, and a feasible path and theoretical support are provided for construction of a large-scale defect detection system in the iron and steel industry.

The invention provides a double-criterion active learning carbonate rock lithology classification method and system. The method comprises the steps of obtaining conventional logging information of multiple wells; in the data preprocessing step, a logging curve is intercepted into sample segments, and data are made to conform to the input shape of the model; building a hybrid model Bi-LSTM-DANN combining a bidirectional long-short term memory neural network and a domain adversarial neural network, and inputting the intercepted samples into the network; domain adversarial pre-training: defining a loss function to determine hyper-parameters and training the network to obtain a pre-training model; the active learning fine tuning adopts a dual-criterion active learning method combining KMeans and a minimum confidence method, and the network is further fine-tuned to obtain a final model; and testing and applying the model. According to the method, two sample query strategies are fused, so that the prediction performance of the model is greatly improved; the performance of the model is further improved through domain confrontation, accurate identification of the lithology of the carbonate rock is achieved, and the system has good mobility and robustness.

The invention provides a multi-objective optimization method for material performance, and belongs to the technical field of material science and artificial intelligence, and the process of the method provided by the invention takes a small sample learning model as a core, combines an active learning optimization strategy, and realizes quantitative optimization of multi-objective performance indexes in material research and development in an extreme service environment. According to the method, limited test data are fully utilized for learning, and the number of physical tests is remarkably reduced; through intelligent optimization search, the efficiency and the success rate of searching for the material design scheme meeting the multi-target performance requirement are improved, and the method has important significance in accelerating development of new materials in a severe environment.

The invention discloses a drainage basin monitoring network layout optimization method based on mutual information active learning. The method comprises the following steps: data processing; constructing an integrated learning prediction model composed of a plurality of LSTMs, initializing each LSTM model parameter by setting different random number seeds, and iteratively updating the model parameters; constructing a covariance matrix of prediction results of the potential monitoring points by using the trained integrated learning prediction model; on the basis of an active learning algorithm of mutual information maximization, mutual information gains of all candidate sites added into the current monitoring network are calculated, candidate points enabling the mutual information gains to be maximum are selected, and information value sorting is carried out on potential monitoring points according to the sizes of the mutual information gains; and outputting a monitoring station optimization suggestion list. According to the method, uncertainty is scientifically quantified by adopting an ensemble learning method, and global information value is evaluated by adopting a mutual information maximization strategy, so that point selection decision does not depend on subjective experience any more, but strict calculation based on data and information theory, and the scientificity of decision is remarkably improved.

The invention relates to the technical field of aeronautical text data processing, and discloses an aeronautical text content cleaning and labeling method, system, device and medium wherein the method comprises: noise filtering: identifying and removing noise in an aeronautical text in combination with static cleaning and a general large model; format standardization: converting the aviation text after noise removal into a standardized format text; duplicate removal and error correction: detecting duplicate contents based on a hash algorithm, and correcting spelling errors and grammar errors based on a general large model to obtain an aviation text subjected to duplicate removal and error correction; entity identification: key entities are extracted based on the general large model, and the extracted key entities are labeled; active learning: screening high-value samples in the marked key entities based on an uncertainty query strategy; and dynamic optimization: performing verification and iterative optimization on the marking result in combination with the aviation knowledge base. According to the method, the automation level, the labeling accuracy and the system self-adaptive capability of aviation text processing can be remarkably improved.

The invention belongs to the technical field of artificial intelligence and power system crossing, and discloses a marketing and distribution fusion scene-oriented multi-modal large model efficient fine tuning method and system, and the method comprises the steps: firstly constructing a small amount of high-quality multi-modal seed data with a thinking chain by field experts; performing domain knowledge injection and multi-modal alignment on the base model in stages by adopting a parameter efficient fine tuning technology; the method comprises the following steps of: selecting an unlabeled sample, further introducing an active learning iterative loop based on hybrid uncertainty perception, automatically screening the unlabeled sample with the most rich information amount by quantifying cognitive uncertainty and accidental uncertainty of a model, and labeling the unlabeled sample by an expert, so as to expand a data continuous optimization model and form a'fine tuning-evaluation-labeling 'closed loop. According to the method, rapid and accurate adaptation of the multi-modal large model in a marketing and distribution fusion complex scene is realized with extremely low expert labeling cost, the service reliability and safety of model output are ensured, and a long-acting mechanism of sustainable evolution of the model is established.

The invention discloses an OCR (Optical Character Recognition) optimization method based on multi-label classification and active learning, and relates to the technical field of image recognizing.The method comprises the steps that an original image-text image is obtained and preprocessed, problem type classification is performed on problems existing in the preprocessed image based on a multi-label classification model, corresponding image problem types are obtained, and meanwhile, priorities are generated; performing targeted enhancement on the image according to the priority of the image problem classification result, repairing various problems existing in the image, and identifying the repaired image to obtain a structured character identification result list; and verifying an identification result, screening out difficult sample data, entering an active learning link according to the number of the sample data, and optimizing a multi-label classification model at the same time, thereby realizing dynamic cooperation of image quality improvement and OCR performance optimization. The problem that the accuracy of OCR recognition is reduced due to the multi-source quality degradation problem of the image in the prior art is effectively solved.

The invention provides an image semantic segmentation active domain adaptation method and system based on a segmentation cutting model. The method comprises the following steps: generating a full-image mask for an image of a target domain by using a pre-trained segmentation cutting model; the images of the target domain are sampled in proportion, initial labeling is carried out on the full-image mask of each sampled image, an initial target domain labeling data set is obtained, and labeling is that a semantic category is given to a full-image mask area generated by each image; training the semantic segmentation model through semi-supervised learning; wherein in the training iteration of the semi-supervised learning, active learning is triggered according to a preset period, and an updated target domain labeling data set is obtained and used for continuous training of the semi-supervised learning. According to the method, the technical purpose of obtaining initial annotation data with complete semantics and clear structure at relatively low labor cost is achieved, and the technical problems that existing semantic segmentation training depends on large-scale pixel-level manual annotation, the annotation efficiency is low and the cost is high are solved.

The invention discloses a solid electrolyte performance prediction optimization method and system and a medium, and belongs to the technical field of solid electrolytes. The method comprises a multi-source data acquisition and standardization stage, a feature extraction stage, a multi-feature collaborative prediction stage, a multi-target optimization design stage and a physical simulation verification and active learning stage. According to the method, a whole chain from data acquisition, prediction, optimization to verification is integrated through an automatic process, and the problems of blindness and fragmentation of a traditional trial and error method are solved. The machine learning model can quickly screen out massive candidate materials, and the range needing experimental verification is reduced by several orders of magnitude, so that the research and development cycle of the solid electrolyte is expected to be shortened to several months from several years in the prior art, and high-throughput and intelligent material design is realized. A multi-model fusion strategy is adopted, an uncertainty evaluation mechanism is introduced, a confidence interval is provided for a prediction result, and decision scientificity and risk control ability are enhanced.

The invention discloses an intelligent sand table identification system based on an artificial intelligence multi-modal technology, which belongs to the field of sand table identification and comprises a multi-modal data acquisition module, a multi-modal feature fusion module, an intelligent analysis and interaction module and a system integration optimization module. According to the method, the problems of insufficient recognition robustness, poor dynamic analysis real-time performance, weak cross-scene generalization ability, high data labeling cost and single interaction mode caused by dependence on a single sensor and lack of multi-dimensional perception ability in the prior art are solved, the limitation of traditional single-mode recognition is broken through by collecting vision, force sense, voice and environment data, and the recognition efficiency is improved. The real-time intelligent analysis and dynamic interaction technology is adopted, real-time intelligent analysis and dynamic interaction of sand table operation are achieved, the system further has strong generalization ability, a meta-learning framework and an active learning strategy are adopted, dependence on annotation data is reduced, the system is reasonable in architecture, distributed collection, heterogeneous calculation and micro-service deployment are adopted, and the system has good application prospects. And high-efficiency operation and expandability are ensured.

The invention discloses a casting defect diagnosis and closed-loop control system and method based on causal anti-fact and multi-agent arbitration. The system collects multi-source data such as temperature, flow, cooling, vision and the like, learns a causal structure, establishes an agent model and solves an adjustment scheme which enables the defect probability to be remarkably reduced in a security domain; performing evidence verification on the suggestions through verifiable generation and a knowledge graph, fusing vision, process, metallurgy and simulation agent arbitration, executing in an edge-cloud double ring, and triggering security rollback abnormally; and when the uncertainty is increased, active learning and incremental updating are implemented. According to the scheme, the defects of shrinkage cavities, air holes and the like can be remarkably reduced, the first-pass yield is improved, and the method is suitable for aluminum alloy sand molds, metal molds and low-pressure casting production lines.

The present disclosure relates to systems, non-transitory computer-readable media, and methods that a implement a framework for active learning to discover pairwise interactions via representation learning. Indeed, in one or more implementations, the disclosed systems generate a first individual perturbation embedding from a first representation of a first cell exposed to a first perturbation and a second individual perturbation embedding, from a second representation of a second cell exposed to a second perturbation. For instance, the disclosed systems combine the first individual perturbation embedding and the second individual perturbation embedding to determine a predicted pairwise embedding. Moreover, in some instances, the disclosed systems generate a pairwise embedding from a representation of a cell exposed to both the first and second perturbation. Additionally, from comparing the predicted pairwise embedding with the pairwise embedding, the disclosed systems generate a measure of biological interaction of the first and second perturbation.

The invention provides a goaf tunnel surrounding rock stability progressive evaluation method and system, and relates to the technical field of geotechnical engineering. The method comprises the following steps: constructing a virtual training sample set by utilizing an optimization process fused with active learning based on goaf and tunnel engineering key parameters and a Knoth time function; establishing a two-stage prediction architecture comprising a rapid screening model and an accurate rechecking model; carrying out risk primary judgment and shunting on the target section by utilizing a rapid screening model; for a section needing to be rechecked, outputting a surrounding rock safety coefficient and a probability prediction value of convergence displacement by using an accurate rechecking model in combination with Bayesian inference; outputting a differentiated support scheme in combination with the multi-level decision matrix; and executing parameter reverse reconstruction and model increment evolution based on the form deviation index of the field actual measurement data and the prediction data. The problem that the stability of the surrounding rock is difficult to accurately quantify under the small sample condition is solved, and self-adaptive evolution of an evaluation system is realized.

The invention discloses a crack detection model training method and device based on dynamic multi-strategy active learning, and the method comprises the steps: collecting and carrying out the preprocessing of a crack image, and constructing a data set containing a labeled subset and an unlabeled subset; performing preliminary training on the double-branch deep learning model by using the labeled subset; then, through an iterative active learning framework, comprehensively evaluating the value of an unlabeled sample from three dimensions of uncertainty, difficulty and representativeness, and dynamically adjusting the weight of each dimension according to a training stage to perform sample screening; and a mixed domain attention module fusing space and frequency domain information is introduced to realize more accurate difficulty perception, and the segmentation precision of the crack edge is improved by including boundary optimization loss. According to the method, the problems of high data labeling cost and low efficiency in deep learning crack detection can be solved, a detection model with better performance can be obtained with less labeling quantity, and the training efficiency and the detection accuracy are remarkably improved.