4323 results about "Labeled data" patented technology

The invention discloses a grassroots power supply enterprise compliance risk intelligent system and method based on big data analysis. The data acquisition unit is used for acquiring business operation data, historical violation records and policy and regulation update data of basic power supply enterprises to form a unified compliance data set. And the natural language processing unit performs text word segmentation and correlation analysis on the policy and regulation and violation record data, extracts key risk factors and labels compliance risk labels. And the risk feature construction unit performs multi-dimensional feature fusion on the business operation data and the compliance risk label data to generate a feature matrix for risk identification. And the intelligent risk assessment unit performs real-time analysis on the feature matrix by using a pre-trained machine learning model, identifies compliance risk categories and levels, and generates early warning information. According to the method, the accuracy of compliance risk identification is improved by using big data analysis and an intelligent algorithm, the compliance management cost of basic-level power supply enterprises is reduced, and the operation safety and compliance of the enterprises are improved.

The invention relates to the technical field of data security, in particular to a data security sharing method and system. The method comprises the following steps: acquiring original shared data, carrying out reverse mapping and block marking, generating block marking data, carrying out inter-block random arrangement on the block marking data, carrying out differential disturbance simulation based on a random sequence, forming disturbance block data, and encrypting the disturbance block data by applying a progressive encryption technology. The method comprises the following steps: carrying out fragmentation storage, generating storage shared data, carrying out access monitoring on the storage shared data, collecting shared access data, and constructing an evidence chain and generating an access traceability data chain according to the data, thereby realizing the security and traceability of data sharing. According to the invention, a safer and more credible data security sharing method is realized.

The invention belongs to the technical field of target detection and recognition, and particularly relates to a thinking chain-based large and small model collaborative target detection and recognition method. According to the method, the small model is responsible for most of easy-to-detect targets, the calculation pressure of the large model is reduced, the large model is responsible for suspected samples, vision and language multi-mode reasoning is combined, the overall false detection rate and the omission ratio are both reduced, confidence evaluation is conducted through the joint probability, automatic screening and manual rechecking of uncertain results are achieved, the reliability of key results is guaranteed, and the method is suitable for large-scale popularization and application. According to the'pseudo thinking chain + pseudo label 'method, by means of reasoning and labels generated by the model, data dependence on manual labeling is reduced, only low-confidence samples are manually confirmed, the human intervention range is narrowed, the human cost is remarkably saved, and semantic information with finer granularity is provided for the model by introducing phrase-level feature descriptors. And the identification capability of complex target attributes and states is improved.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a time series data processing method, device and equipment and a medium. Performing a data enhancement operation on the original time series data to generate enhanced time series data; training a feature encoder based on the enhanced time sequence data in a comparative learning mode to obtain a pre-trained feature encoder; connecting the pre-training feature encoder with a sparse attention mechanism module to construct a time sequence modeling network; target time series data is processed using the time series modeling network to generate a processing result. According to the method, the enhanced view is constructed on the unlabeled data and the contrast learning training feature encoder is introduced, so that the time sequence representation with generalization ability is obtained, effective modeling of the long dependency relationship is realized in combination with a sparse attention mechanism, and the accuracy of time sequence modeling is improved under the condition of not depending on a large amount of labeled data.

The invention discloses a field specialized large model fine tuning and deployment method and system based on adaptive optimization, belongs to the technical field of artificial intelligence and machine learning, and aims to solve the technical problem of how to improve the rapid adaptation capability of a large model on small-scale field data. In order to meet the requirements of special fields of law, medical treatment, finance and the like on professional terms and complex contexts, the technical scheme adopted by the invention comprises the following steps of: data processing and sample generation: performing cleaning, feature extraction and small sample expansion on field data, and generating a high-quality training sample through a field feature guide mechanism; small-sample fine tuning and migration: realizing rapid field adaptation of a large model through a small-sample learning technology and dynamic Prompt optimization, and reducing dependence on large-scale annotation data in combination with cross-field migration learning; compressing and optimizing the model; performing automatic deployment and collaborative reasoning; and monitoring and adaptive optimization are carried out.

The invention discloses an intelligent substation communication link fault accurate positioning method and system, and the method comprises the steps: obtaining a configuration file and equipment state data, carrying out the processing of the configuration file and the equipment state data, and generating a standardized link feature vector and a marking data set; constructing a hybrid deep learning model, and optimizing parameter configuration of the hybrid deep learning model by adopting an optimization algorithm to obtain a parameter-optimized hybrid deep learning model; training by using a real fault sample in combination with a virtual fault sample generated by a generative adversarial network, optimizing a time sequence prediction capability through an echo state network, and generating a fault positioning model; in combination with the link state data, outputting a fault link positioning result and confidence evaluation through multi-stage confidence evaluation and topological correlation analysis; and carrying out virtual-real corresponding verification in combination with the configuration file, carrying out parameter optimization on the fault positioning model, and outputting a fault positioning system. The problems that the fault positioning precision is low, the response speed is low, and complex fault scenes cannot be processed are solved.

The invention relates to the technical field of computer vision, and discloses a real-time video analysis method based on deep learning. The method comprises the following steps: acquiring a real-time video stream through image acquisition equipment, and performing frame segmentation processing to generate a continuous video frame sequence; and extracting features of the video frame sequence by using a pre-trained convolutional neural network to obtain a multi-dimensional feature vector, inputting the multi-dimensional feature vector into the time sequence analysis model to calculate dynamic relevance, and outputting an inter-frame movement track and object behavior features. And constructing a scene understanding map containing a spatial position and a time evolution relationship according to the above-mentioned data, and carrying out abnormal event detection and generating event marking data based on the map. And performing semantic analysis on the event marking data, determining an abnormal event type and a confidence score, triggering a real-time alarm signal according to a result, and updating a historical event database. In the analysis process, the resource occupancy rate of the system is continuously monitored, the calculation precision is dynamically adjusted, a degradation processing mechanism is started when a preset threshold value is exceeded, and key area analysis is preferentially guaranteed.

The invention discloses a semi-supervised medical image segmentation method based on uncertainty-driven dynamic correction and multi-scale consistency learning, and the method comprises the steps: carrying out the preprocessing of a medical image data set, and dividing the medical image data set into a training set and a test set; constructing a semi-supervised segmentation model of dynamic correction and multi-scale consistency learning based on uncertainty driving; inputting the training set into a semi-supervised segmentation model, and performing iterative training and parameter optimization to obtain a trained semi-supervised segmentation model; inputting the test set into the trained semi-supervised segmentation model to obtain a medical image segmentation result; wherein the semi-supervised segmentation model adopts a mean teacher model of a V-Net network, a prediction block is added behind each up-sampling block of a V-Net decoder, and a dropout layer is added; according to the method, the problems that the existing semi-supervised learning method is difficult to adapt to the complexity of annotated data and unannotated data distribution, so that effective information is lost; meanwhile, a traditional uncertainty estimation method needs multiple times of forward transmission, and the calculation cost is high.

The invention discloses a semi-supervised multi-temporal satellite image time-varying information extraction method, and belongs to the technical field of remote sensing image processing. The semantic change detection performance of the model and the detection precision of complex shape change ground objects are improved. The method comprises the following steps: constructing a semantic change detection model, carrying out full-supervised training on the semantic change detection model by using a binary change detection supervised loss function and a semantic segmentation supervised loss function by using a small amount of labeled dual-temporal remote sensing images to obtain an initial model, and obtaining a semantic change detection prediction result of each pair of images; using a pseudo label optimization strategy to optimize the semantic change detection prediction result of each pair of images; combining the obtained pseudo label data with high confidence and a small amount of labeled dual-temporal remote sensing images into a new training set, using the new training set to perform semi-supervised training on the initial model in a semantic change detection model, and using a consistency regularization combination loss function to perform supervised training to obtain a new model; and until a preset number of iterations is reached.

Various embodiments of the present disclosure provide automated encryption and data de-identification techniques for improving computer security. The techniques apply machine learning and encryption techniques to transform input data objects to tagged data objects that may be locally decrypted using encrypted element representation stored within the tagged data objects. The techniques may include determining a protected data element from an input data object based on privacy criteria and generating the tagged data object from the input data object by replacing the protected data element with an anonymized privacy tag that identifies a privacy type of the protected data element. The techniques may further include generating an encrypted element representation of the protected data element and inserting the encrypted element representation to a portion of the tagged data object to enable decryption of the tagged data object by authorized entities.

The system and methods for determining the representative samples from a large imperfectly labeled dataset to support data processing and inferences for machine-learning applications. The method includes accessing data samples that may be processed to generate embedded vectors along with a set of reference labels. For each label, clustering is performed to group at least some of the embedded vectors together into clusters based on the associated inherent patterns, followed by a refinement process to select relevant clusters from the clustered patterns. One or more embedded vectors from the selected clusters are passed to a statistical technique to generate representative embedded vectors for each label. The statistical technique is configured such that the weights of selected embedded vectors within each of the cluster are the same. These representative embedded vectors may be further fed into a machine-learning model to predict a label from the set of reference labels for a given prompt.

The invention provides a risk abnormal behavior event identification method based on multi-source risk abnormal behavior data fusion analysis. The method comprises the following steps: S1, obtaining abnormal behavior label data; s2, constructing a knowledge graph ontology structure; s3, extracting entities, attributes and relationships involved in the structured data of the abnormal behavior label data into the constructed knowledge graph ontology structure; s4, for the constructed knowledge graph ontology structure, encoding graph data to obtain corresponding modal features; aiming at the structured data of the knowledge graph ontology structure, coding each source by adopting a corresponding feature coding method to obtain a corresponding modal feature; s5, the obtained modal features are input and mapped to the same vector space for alignment fusion; s6, performing fine tuning training to obtain an abnormal risk behavior recognition model LLM; and S7, superposing the fused multi-modal features, inputting the superposed multi-modal features to the LLM, and guiding the LLM to generate a corresponding output or decision according to the prompt of the specified input.

The invention relates to the technical field of project cost management, and discloses a project cost dynamic management method. The method comprises the following steps: marking a timestamp of a data updating node in a historical cost database, extracting a consumption index of each subentry project in a specified interval before and after the timestamp, and constructing a consumption feature set dynamically associated with the cost; inputting the consumption feature set into an optimized neural network, calculating the deviation degree of the same sub-project in the construction stage and the design stage through time sequence comparative analysis, and generating the difference representation of the actual cost and the plan cost; a material price fluctuation coefficient is used as an adjusting variable, a fuzzy comprehensive evaluation model is adopted to carry out multi-dimensional reasoning on the difference representation, and the influence weight value of each cost component on the cost deviation is obtained; key parameters related to the market supply and demand index and the construction progress deviation rate are mined, the key parameters are input into a pre-trained evaluation model, a screening threshold value is dynamically adjusted according to a model output numerical value, and core parameters with influence weight values higher than the adjusted threshold value are screened out.

The invention relates to the technical field of electromagnetic spectrum monitoring, and particularly discloses a distance-based electromagnetic spectrum monitoring abnormal data detection method, which comprises the following steps of: performing short-time Fourier transform and normalization processing on an acquired original signal to generate an energy density distribution characteristic graph; constructing a multivariate Gaussian distribution model based on non-abnormal historical data; during real-time monitoring, the mahalanobis distance between the collected data and the mean vector of the historical model is calculated after the collected data is preprocessed. And comparing the distance metric value with a preset threshold value to preliminarily judge abnormity, and calculating a distance fluctuation variance through a sliding window mechanism to perform secondary verification. And finally, processing and positioning anomalies by using image morphology, and dividing the degree of anomalies according to the relative deviation between the energy density and the mean value of the historical model. According to the method, the mahalanobis distance and the multivariate Gaussian distribution are introduced, secondary verification and abnormal positioning are combined, the limitation of a traditional method is overcome, the detection accuracy and reliability are effectively improved, the misjudgment and missing judgment rate is reduced, and the method does not depend on a large amount of labeled data and is high in practicability.

The invention discloses a radar signal modulation identification method and device for self-supervised contrast mask reconstruction, and belongs to the technical field of radar signal modulation identification, and the method comprises the following steps: obtaining a radar signal, constructing a radar modulation signal data set containing label data and label-free data, and converting the radar modulation signal data set into a time-frequency image; building a self-supervised contrast mask image reconstruction model comprising an online branch and a target branch; the method comprises the following steps: pre-training a self-supervised contrast mask image reconstruction model by using label-free data, performing data enhancement and random mask operation on a time-frequency image to generate double views, respectively inputting an online branch and a target branch, updating model parameters by jointly optimizing reconstruction loss and contrast loss, and obtaining a pre-training weight; and migrating the pre-training weight to a downstream identification network, freezing part of encoder parameters, and performing fine tuning by using a small amount of labeled data to obtain a radar signal modulation identification model. According to the invention, the radar signal modulation identification precision in a complex electromagnetic environment is improved.

The invention discloses an uncertainty and representativeness combined active learning method and system for medical image data annotation, and relates to the technical field of medical image data. The method comprises the following steps of: training a variational auto-encoder infoVAE on an image pool; m samples are randomly extracted from the image pool and labeled, an initial labeled image set is constructed, and a segmentation model is trained on the labeled image set; t rounds of active learning circulation are carried out, and in each round of t active learning circulation, the following steps are specifically executed: screening candidate samples based on a representative method; screening a final sample based on an uncertainty method; updating the labeled data set and the unlabeled data set; retraining the segmentation model on the updated labeled image set, and optimizing model parameters; and obtaining final model parameters. According to the method, by designing a sample selection strategy integrating uncertainty and representativeness, the global performance of the model and the difficult sample segmentation capability are improved.

The present disclosure provides an example method for obtaining labeled trajectories. The example method can include obtaining log data describing a trajectory of a vehicle traveling through an environment. The example method can include determining a suboptimal condition associated with the trajectory. The example method can include generating label data that characterizes the suboptimal condition along one or more constraint dimensions of a motion planner of the autonomous vehicle control system. The example method can include generating a training example for training the one or more machine-learned models of the autonomous vehicle control system to decrease a probability of the autonomous vehicle control system inducing the suboptimal condition.

The invention discloses a complex exponential signal joint spectrum reconstruction and parameter estimation method and device, and relates to the field of signal processing, and the method comprises the steps: S1, constructing noise-containing complex exponential signal training data and label data; s2, constructing a dual-module neural network model comprising a super-resolution denoising reconstruction module and a parameter prediction module; s3, training the dual-module neural network model by using the training data and the annotation data to obtain a trained dual-module neural network model; and S4, performing frequency spectrum reconstruction and parameter prediction by using the trained dual-module neural network model, and performing signal post-processing on the output to obtain estimation parameters of the angular frequency, the attenuation factor, the real part amplitude and the imaginary part amplitude. According to the invention, a cascade neural network architecture of a super-resolution denoising reconstruction module and a parameter prediction module is designed, and angular frequency detection is converted into a Gaussian distribution heat map regression task; meanwhile, a sparse activation labeling mechanism is adopted, the parameter truth value is only reserved at the spectrum peak position, and the model learning complexity is remarkably reduced.

The invention discloses a multi-source heterogeneous medical data fusion and intelligent diagnosis method, and relates to the technical field of medical data processing and intelligent diagnosis, and the method comprises the specific steps: firstly, synchronously collecting medical images and clinical text data of a patient, and carrying out the correlation and integration to form a heterogeneous diagnosis data set; performing standardized feature extraction to obtain a feature set in a unified format; then constructing a parallel model, fusing features by using a cross-modal attention alignment technology, and guiding correction by means of a knowledge graph; and finally, the cross-modal diagnosis features are input into the reference model, automatic focus positioning is realized through processing, and a visual marker graph is output. Heterogeneous data of medical images and clinical texts are synchronously integrated, and the diagnosis feature reliability is improved through standardization processing, feature fusion and the like; a focus sensing mask is generated through comparison with a normal model, a multi-scale feature fusion technology is combined to realize automatic and accurate positioning of the focus, a large amount of labeled data is not needed, the process is simplified, and the diagnosis efficiency and accuracy are improved.

The invention relates to a visual safety risk early warning system for pseudo-classic architecture mixed structure construction, in particular to the field of safety risk early warning of pseudo-classic architecture construction, through multi-source data fusion and intelligent algorithm cooperation, the construction safety risk management and control capability is remarkably improved, the system collects mechanical and environmental data of key nodes of a high-support formwork body in real time, and the construction safety risk early warning capability is improved. Load distribution is dynamically updated in combination with the building information model, and a high-precision space-time label data stream is generated; predicting a short-term instability risk based on a two-channel neural network, constructing a complex network model to quantify a node failure linkage effect, and generating a visual risk propagation thermodynamic diagram; finally, a grading alarm strategy is triggered through a dynamic threshold value, augmented reality visual warning and equipment linkage control is achieved, the system breaks through the limitation of traditional manual monitoring, full-process closed-loop management of risk perception, prediction, positioning and response is achieved, the occurrence rate of safety accidents is effectively reduced, and the safety and reliability of construction of the complex structure of the pseudo-classic architecture are guaranteed.

The invention relates to the technical field of product traceability, in particular to a traceability system and method based on a block chain technology, and the system comprises a right confirmation registration module, a parameter verification module, a quality inspection label module, a logistics locking module and a traceability auditing module. According to the method, compliance evaluation is carried out by comparing environmental parameters bound with the operation time period and the land parcel information with an agricultural standard boundary, systematic judgment of land parcel growth conditions can be realized, confidence modeling is carried out on a quality inspection data layer based on a difference value amplitude and a batch mean value, credible label data of batch quality is formed, and the reliability of the quality inspection data layer is improved. Through sequence comparison of storage time periods and carrying timestamps and environmental parameter change frequency deviation judgment, abnormal carrying behaviors and storage fluctuation conditions can be identified, a traceability identification value and quality traceability code bidirectional verification mechanism is introduced in a query response stage, and credible labels, quality inspection states and position information are displayed in an aggregated manner. And the completeness, the accuracy and the result credibility of the traceability query result are obviously improved.

The invention provides an image noise mark feature selection method and system, a storage medium and a computer. The method comprises the steps of obtaining a to-be-processed image noise mark data set; embedding a sample set in the image noise mark data set into a multi-granularity fuzzy cluster to construct a dynamic fuzzy membership evaluation matrix; dynamically evolving a multi-level high-precision granular ball cluster; obtaining mark distribution with high identification degree; constructing a rough perception feature evaluation framework based on granular ball topology driving, extracting decision equivalence classes by combining rough set upper and lower approximation and extended positive domain models, and determining and measuring the contribution degree of each feature to a decision system by fusing multi-granular-ball decision boundary information based on a dependency degree quantitative model; a particle and ball structure consistency verification mechanism is introduced, and multi-level evaluation is carried out on the importance of the features through dependency and consistency. According to the method, the optimal feature subset with strong anti-noise performance and high discrimination capability is obtained, and stable and efficient input support is provided for a subsequent image noise mark learning model.

The invention relates to the technical field of data processing, and discloses a knowledge graph incremental updating and consistency checking method and system. The method and the device are used for solving the problem of low incremental updating and consistency checking efficiency of a large-scale knowledge graph. The method comprises the following steps: collecting a to-be-updated data source, sorting and separately storing data, and marking priority labels; performing change detection on the marked data source, identifying change items by comparing entities and relationships, and generating a log; incremental updating is executed based on the log, and nodes and edges are processed in a hierarchical fusion mode; preliminary consistency verification is carried out, and attribute uniqueness and relation directivity are checked; expanding a verification range, traversing an association path through cascade check, and recording problems; and optimizing storage according to the record, merging the update area, updating the index and cleaning the log. The method solves the problem of low efficiency of incremental updating and consistency verification of the large-scale knowledge graph, improves the response speed and the data accuracy of the system, and is suitable for a high-frequency dynamic data environment.

The invention belongs to the technical field of intelligent ocean and remote sensing image processing, and discloses an ocean remote sensing coastline segmentation method based on a text-guided semi-supervised pseudo tag, which comprises the following steps: collecting ocean remote sensing coastline image data, dividing into tag data and non-tag data, and constructing a text prompt; constructing a segmentation model comprising a shared image encoder, a text encoder, two decoders and a pseudo label calibration module, cooperatively extracting image features and text features by all the parts, generating a pseudo label and a prediction mask, and optimizing the pseudo label through uncertainty calibration; training the model in a supervised stage and an unsupervised stage, and adding loss function values of the two stages to a back propagation optimization model; and finally, based on the trained model, precise segmentation of the ocean remote sensing coastline image is realized. According to the method, the efficiency and the accuracy of cross-regional ocean remote sensing coastline segmentation are effectively improved by utilizing text guidance and semi-supervised pseudo labels, and the dependence on a large amount of labeled data is reduced.

Provided is a method for perceiving and eliminating an abnormal state of active distribution network based on data enhancement, including: acquiring, by synchrophasor measurement device, data of each node of active distribution network in target domain in real-time and transmitting to processor; inputting the acquired data into a classification model, and outputting abnormal detection and classification results in real time; and analyzing the abnormal detection and classification results, and transmitting an abnormal state eliminating instruction to a distribution terminal to eliminate the abnormal state. Wherein, hidden distribution features in node data of active distribution network are mined through dynamic clustering, a large amount of unlabeled data are clustered, a data label is generated through self-coding and label correction rule, training samples with balanced category distribution is generated through data enhancement and is used to train the classification model based on dynamic graph attention network by domain adaption method.

The invention discloses a medical image segmentation method based on uncertainty estimation and multistage distillation, which relates to medical image segmentation, and comprises the following steps: quantifying the prediction uncertainty of a teacher model on an input medical image through an uncertainty estimation module, and generating an uncertainty weight; high-confidence-coefficient pseudo labels are screened based on uncertainty weights, and knowledge of the teacher model is transmitted to the student model in stages through a multi-stage knowledge distillation module; embedding a mixed attention module in the teacher model and the student model, and enhancing feature capture of a focus structure and a boundary in combination with a channel attention mechanism and a space attention mechanism; optimizing the training process of the student model by adopting a composite loss function fusing cross entropy loss, Dice loss and distillation loss based on uncertainty; and applying the trained network to a medical image segmentation scene to complete medical image segmentation based on uncertainty estimation and multistage distillation. According to the invention, the problem of scarcity of high-quality annotation data in medical image segmentation is solved.

The invention relates to a wind power equipment blade fault detection technology, discloses an unsupervised wind power equipment blade fault detection method based on a phase perception parallel attention mechanism, and solves the problems that an existing wind power equipment blade fault detection method is high in dependence on labeled data, insufficient in generalization ability under strong noise and variable working conditions and high in fault detection efficiency. And a weak transient fault signal and a dynamic change characteristic are difficult to capture robustly. According to the scheme of the invention, the method comprises the steps: collecting a blade operation audio signal, and extracting a dual-channel time-frequency feature containing an amplitude spectrum and a phase spectrum through improved short-time Fourier transform; a deep adversarial auto-encoder is constructed by using an encoder containing a phase perception parallel attention module, a decoder and an auxiliary encoder, and normal working condition feature distribution is learned by reconstructing an error loss, potential representation consistency loss, adversarial loss and phase consistency loss optimization model during off-line training; in the reasoning stage, the fault is judged based on the feature distance score and the reconstruction error score.

The invention relates to the technical field of computer vision, and discloses a high-speed human motion trail motion mode recognition system. Through multi-modal sensor fusion and an adaptive space-time attention network (ASTAN), the problems that a traditional optical system depends on mark points and IMU accumulative errors exist are effectively solved, and the high-speed movement track reconstruction precision and the environmental adaptability are remarkably improved; self-supervised data enhancement and dynamic model compression technologies are combined, so that the dependence on labeled data is greatly reduced, lightweight edge deployment is realized, and the real-time requirement is met; the system can synchronously output multi-mode feedback instructions, supports multi-scene application such as sports action correction, medical rehabilitation evaluation, security and protection anomaly detection and man-machine interaction control, and solves the core pain points of insufficient precision, real-time performance, robustness and generalization ability in the prior art. And an efficient, reliable and low-cost comprehensive solution is provided for high-speed motion analysis.

The invention discloses a two-stage infrared and visible light image registration method, system and device. The method comprises the following steps of image preprocessing, contour extraction, angular point detection, feature description and matching, affine transformation estimation, multi-scale optical flow estimation, optical flow constraint and loss function, reverse resampling and fusion and error evaluation. According to the invention, rough registration is carried out by using contour angular point features, so that a preliminary alignment result can be quickly obtained; refined alignment is carried out in combination with an unsupervised optical flow network, and sub-pixel-level registration precision is achieved. The contour angular points are based on shape information of an image target, are natural and are not influenced by spectral differences, and the matching stability is enhanced through main direction and angle features. The unsupervised depth optical flow model estimates a pixel displacement field by learning consistency characteristics of an input image, and does not need to depend on annotation data. The combination can effectively eliminate the difference between infrared light and visible light, and improves the robustness and adaptability of registration.

The invention discloses a defect classification and segmentation method and system based on unsupervised and weak supervised combination, and the method comprises the steps: employing a training data set only containing a defect-free sample, extracting the feature representation of the defect-free sample through a feature extractor, and storing the feature representation in a feature memory library; comparing the mixed data set without labels including defect and defect-free samples with the features in the feature memory bank to determine false classification labels; designing a class activation graph-based weak supervision network for extracting features in the image, a semantic segmentation network and a loss function thereof, and training the semantic segmentation network by using a training data set of a defect-free sample and pseudo-classification label data; according to the method, the advantages of weak supervision and unsupervised learning algorithms are combined, and efficient defect detection and positioning can be realized.