6726 results about "Classification methods" patented technology

The invention provides a battery assembly process defect real-time detection and classification method and system, and relates to the technical field of intelligent manufacturing, and the method comprises the steps: carrying out the standardization preprocessing of collected battery surface images and three-dimensional point cloud data; a distributed edge computing architecture is adopted for feature extraction, the distributed edge computing architecture comprises a defect propagation prediction module based on a graph neural network, inter-process defect features are constructed into a dynamic graph structure, a defect evolution rule is learned through the graph network, and feature extraction parameters are dynamically adjusted; inputting the extracted features into a deep learning detection model, carrying out feature adaptive mapping and alignment through a multi-modal spatial-temporal feature coding layer, constructing a process association graph through a dynamic reasoning layer based on causal reasoning, and calculating a propagation influence degree; carrying out hierarchical precision detection on different defect evolution probability regions by a prediction perception detection layer with a self-adaptive threshold value, and outputting a defect detection result; and classifying defect detection results based on defect classification standards.

The invention provides an urban functional area gridding identification and classification method based on multi-source spatio-temporal data, and relates to the technical field of area identification, and the method comprises the steps: carrying out the dynamic grid division of an urban area through employing a self-adaptive grid division algorithm, obtaining the multi-source spatio-temporal data of a gridding area, carrying out the space-time alignment, and generating a multi-dimensional time sequence feature set; extracting multi-dimensional feature vectors of the grid units through a multi-modal feature fusion algorithm, acquiring long-range time sequence features by using a bidirectional gating recursion unit, calculating spatial correlation features based on a multi-head attention mechanism, and inputting the two types of features into a feature fusion network with jump connection to generate enhanced spatial-temporal features; and an optimization strategy is adopted, a functional area clustering center point is determined through density peak value clustering in sequence to obtain an initial division result, the initial result is optimized by combining spectral clustering of spatial adjacency constraints, and the boundary of the functional area is optimized by using a simulated annealing algorithm, so that accurate identification and classification of the urban functional area are realized.

The invention discloses an intention classification method and device based on vector retrieval and context awareness and a medium, and relates to the technical field of artificial intelligence. The method comprises the steps of extracting business metadata and associating the business metadata with typical problem examples to generate a standardized service description document; encoding the standardized service description document into a high-dimensional semantic vector through a pre-training language model, and constructing a neighbor search index to store the high-dimensional semantic vector; splicing the user identity information and the current question text into an enhanced query statement, and encoding the enhanced query statement into a context-aware dynamic query vector through a semantic model; performing similarity retrieval based on the dynamic query vector to obtain candidate intelligent services, performing business domain filtering, context weighted sorting and dynamic priority rearrangement, and outputting target recommendation services; by collecting interactive behavior data of a target recommendation service, quality scoring is performed on service descriptions and problem examples based on a preset evaluation rule, and the service descriptions and the problem examples of which the quality scores are lower than a quality threshold value are updated.

The invention provides an artificial intelligence-based ultrasonic image data classification method and system, and the method comprises the steps: firstly obtaining a real-time ultrasonic scanning signal sequence containing the time sequence change characteristics of a tissue elastic parameter and a hemodynamic parameter, carrying out the noise suppression and motion artifact compensation processing, generating a standardized ultrasonic image sequence, and marking the coordinates of an anatomical boundary; then performing multi-scale anatomical structure decomposition on the ultrasonic image to obtain a local feature map set of different organization levels, inputting the local feature map set into a cascade deep classification network, and realizing cross-frame feature fusion and dynamic weight adjustment through spatial-temporal feature alignment and a multi-granularity attention distribution module to obtain a spatial-temporal feature fusion model; and the abnormal region classification probability distribution and the spatial topological relation graph are output, finally, a multi-modal diagnosis report is generated according to the abnormal region classification probability distribution and the spatial topological relation graph, an interactive three-dimensional visual interface containing risk level labels and treatment suggestions is generated after the multi-modal diagnosis report is compared with historical cases, and ultrasonic image classification accuracy and diagnosis efficiency are improved.

The invention provides a hyperspectral image open set spectral spatial feature extraction and classification method, and aims to improve the classification precision of a known category and an unknown category in a hyperspectral image and the robustness of a classification model. According to the method, input data are preprocessed through fractional Fourier transform, and rotation analysis of a signal time-frequency plane is achieved. Multi-scale features of multi-branch cavity convolution are fused through an enhanced spectrum space residual module, deep separable convolution enhancement nonlinear representation of a lightweight convolution enhancement block is combined, and self-adaptive separation and weighted fusion of high / low frequency features are realized by innovating a dual-frequency enhancement module. Proposing a class perception comparison loss function, integrating anchoring loss, triple loss and regularization terms, and collaboratively optimizing intra-class compactness, inter-class separability and an open set decision boundary. The model framework provided by the invention is combined with the improved loss function, so that the classification precision of the known class and the unknown class of the hyperspectral image is remarkably improved.

The invention provides an engineering construction defect automatic detection and classification method based on deep learning, and the method comprises the steps: obtaining a welding seam surface image through the shooting of an unmanned plane, and carrying out the denoising and illumination normalization processing of the welding seam surface image, and obtaining a standardized image; welding seam surface texture features are extracted from the standardized image, a convolutional neural network is adopted to analyze the spatial distribution characteristics of textures, and vectorization processing is carried out to obtain texture feature vectors; segmenting a weld surface corresponding to abnormal region distribution by adopting a region growing algorithm, and analyzing pore and weld discontinuity in combination with the texture feature vector to obtain a defect candidate region; performing threshold division on the sizes and the numbers of the defects according to the defect types and the feature vectors of the candidate regions to obtain a severity grading result of each type of defects; and severity features are extracted from a grading result, and a Bayesian network is adopted to fuse texture feature vectors and defect type labels to obtain a welding quality evaluation score.

The invention discloses a multi-label electrocardiogram classification method based on self-supervised pre-training and multi-modal semantic alignment, which belongs to the technical field of artificial intelligence, and comprises the following steps: realizing self-supervised pre-training of unlabeled data through a single-modal contrast enhancement network, generating global and local contrast views by adopting a multi-scale random cutting strategy, and classifying the global and local contrast views in a multi-scale random cutting mode; in combination with a teacher-student network architecture, the potential invariance features of the ECG signals are learned while negative sample dependence is avoided, the problem of annotation data scarcity is effectively relieved, and the feature robustness is improved. A multi-modal fusion mechanism based on label semantic guidance is provided, a time domain signal and a frequency domain time-frequency graph are mapped to a unified semantic space through fine-grained semantic alignment, local feature enhancement and cross-modal complementary information fusion are realized by using a cross attention mechanism, and the problem of semantic difference caused by modal heterogeneity in a traditional method is overcome. A multi-label comparison loss function based on a disease co-occurrence relation is proposed, a category discrimination boundary is dynamically optimized by modeling a label co-occurrence probability, the feature separability of a tail category is improved while the head category discrimination ability is enhanced, and the problem of sample category imbalance in a multi-label scene is remarkably relieved.

The invention provides an image-based three-dimensional point cloud traffic marker classification method and system, and relates to the technical field of computer vision, and the method comprises the steps: collecting a multi-frame multi-angle image of a traffic marker, converting the multi-frame multi-angle image into point cloud data, constructing a histogram based on reflection intensity and geometric features, determining an optimal segmentation threshold value, and carrying out the segmentation of the point cloud data; the method comprises the following steps: carrying out region merging by combining spatial distribution characteristics of scattering coefficients and absorption coefficients to obtain point cloud sub-regions with uniform materials, determining a core region by utilizing a curvature entropy value and a normal vector entropy value, establishing a local coordinate system, carrying out region growth and dynamic merging through spatial position and geometric structure constraints to obtain candidate point cloud sub-regions, and carrying out point cloud distribution on the candidate point cloud sub-regions. And extracting the position offset and the attitude variation to construct a time sequence motion feature, thereby realizing accurate classification of the traffic markers.

The invention relates to the technical field of text classification, and provides a text classification method based on a large model and a rule engine, and the method comprises the steps: S1, storing multi-level rule classification labels, and constructing a classification rule template library; s2, receiving text data from various data sources, and preprocessing the text data; s3, performing rule matching on the text data based on the classification rule through a rule engine, and outputting a rule classification result; and S4, when any one of the following conditions is met, large language model classification is triggered: a, a classification rule is not matched; b, matching a classification rule, wherein the rule confidence is smaller than a rule confidence threshold; c, the text data length exceeds the preset text data length; d, matching a specific business scene label; outputting a model classification result; and S5, when the rule engine classification in the S3 and the large language model classification in the S4 are parallel, executing the strategy. The output reliability and the service continuity are guaranteed, and the method is suitable for scenes with high accuracy requirements such as financial compliance examination and the like.

The invention provides a machine vision defect real-time detection and classification method and system based on deep learning, and relates to the field of machine vision detection.The method comprises the steps that regional enhancement weights are determined by calculating local entropy and gradient direction consistency, and regional self-adaptive enhancement is carried out; establishing a feature transfer sequence and progressively fusing features; generating and correcting a defect area probability distribution diagram; and constructing a dynamic decision matrix to calculate a comprehensive score for defect grading. According to the method, the defect detection accuracy under a complex background can be improved, false detection and missing detection are reduced, and real-time defect positioning and accurate classification are realized.

The invention discloses an automatic label labeling and classifying method and system oriented to unstructured system documents, and relates to the technical field of artificial intelligence. The method comprises the steps that semantic structure pre-analysis is conducted on an original system text, and a system semantic structure tree is constructed; establishing a system semantic enhancement vector space based on the semantic units and the logic relationship thereof; performing semantic deconstruction on the preset tag and extracting a feature vector; realizing cross-space semantic matching of the document and the tag through a system semantic attention mechanism; a confidence evaluation module is introduced to screen high-confidence labels from the three dimensions of structural integrity, coverage and logic consistency; and outputting a final label and a score through semantic conflict detection and resolution. According to the method, the problems that in the prior art, unstructured system text labeling accuracy is low and large-scale labeling samples are dependent on polysemy ambiguity, high context dependency, complex semantic structure and the like are solved, and labeling accuracy and robustness are remarkably improved.

The invention discloses a convergent media content intelligent label generation and target classification method, and belongs to the technical field of artificial intelligence and convergent media. In order to solve the problems of low labor efficiency, poor rule label flexibility and the like of traditional fusion media content label generation, the deep learning and multi-modal fusion technology is innovatively adopted. In a data processing layer, through deep integration of NLP and CV technologies, feature extraction of multi-source data such as texts, pictures, videos, audios and the like is realized. In a label generation layer, an intelligent label engine is constructed based on a CNN and RNN hybrid model, and a dynamic label library is established in cooperation with a rule matching strategy. And in the classification optimization level, the K-means clustering and hierarchical classification methods are combined to realize adaptive classification and optimization of the labels. Through continuous data training and a user feedback mechanism, the label generation accuracy, the classification efficiency and the system adaptability are remarkably improved, and an innovative solution is provided for intelligent management of the convergence media content.

The invention discloses an image classification method and system based on CNN-Transform staged fusion, and the method specifically comprises the steps: inputting an input image into a lightweight CNN backbone network after preprocessing, and extracting the local features of a shallow layer, a middle layer and a deep layer; non-overlapping image blocks are divided on the middle-layer feature map, learnable position codes are embedded, and global semantic features are generated through linear attention Transform; a dynamic gating fusion module is designed, channel attention and space attention branches are combined to generate adaptive weights, and local and global features are subjected to weighted fusion; the CNN shallow detail features and Transform deep semantic features are fused through cross-level jump connection, and the classification robustness is improved. The system supports model quantization, operator fusion and edge device deployment, and the classification precision is remarkably improved while the calculation complexity is reduced.

The invention relates to the technical field of trajectory tracking, in particular to an unmanned aerial vehicle trajectory tracking target classification method and system based on a multi-frequency-band radar, and the method comprises the following steps: synchronously collecting original radar echo data of a millimeter wave frequency band, a Ku frequency band and an L frequency band, and generating a multi-frequency-band data set comprising a timestamp, an azimuth angle and a pulse sequence; extracting multi-dimensional features of the target from the synchronized multi-band data set, and generating a multi-band feature vector; constructing a dynamic trajectory association matrix, and determining the space-time continuity of a target trajectory through a trajectory association algorithm based on an attention mechanism; constructing a lightweight classification model; performing online classification on the real-time multi-band feature vectors by using a lightweight classification model, and outputting a target type and confidence; according to the method, capture of the motion state of the target is enhanced, rapid response can be made to tiny motion and change of the target in a complex scene, and therefore more comprehensive and accurate time sequence context information is provided for subsequent classification tasks.

A colorectal lesion multi-modal classification method based on pathological attention and multi-instance learning constructs an efficient automatic diagnosis model by fusing visual features and a textual prototype defined by pathology experts. The method comprises the following steps: collecting histopathological image data, segmenting the histopathological image data into standardized image blocks, and extracting visual features by using a pre-training feature extraction network after color standardization and noise processing; a multi-instance learning framework and a pathological attention mechanism are combined, feature space distribution of a text prototype is adjusted in a self-adaptive mode through a dynamic prototype optimization module, and optimization targets of visual clustering and cross-modal semantic alignment are balanced by adopting a gradient perception double-loss dynamic weighting strategy; and after the model is trained in stages, the generalization performance is verified in an external data set. According to the method, the classification precision is remarkably improved, the method can adapt to dyeing difference and tissue heterogeneity without pixel-level labeling, the accuracy rate in cross-center verification is superior to that of an existing reference model, and the efficiency of pathological diagnosis is greatly improved.

The invention relates to the technical field of image processing and mode recognition, in particular to a classification method and device based on user video data and a medium, and the method comprises the steps: collecting video data uploaded by a user, and extracting a plurality of key frames from the video data; using a video feature extraction network and an image-level spatial feature extraction network with time sequence modeling capability to respectively extract time sequence features between the key frames and spatial features of each key frame, and generating video visual features based on the time sequence features and the spatial features; collecting behavior data of the user in the process of watching the video data, and constructing a final user behavior embedding vector based on the behavior data; fusing the final user behavior embedding vector with the video visual features to generate joint feature representation; and inputting the joint feature representation into a pre-trained graph neural network classification model for classification, and outputting one or more semantic category labels to which the video data belong.

The invention relates to the technical field of data processing, and discloses a motor vehicle safety performance classification method and system based on multi-factor coupling influence. The method comprises the steps of collecting multidimensional safety data of a motor vehicle, calculating a mutual information value and transfer entropy to generate a coupling relation network diagram, constructing a three-layer safety index system to form a comprehensive score, obtaining low-dimensional features through phase-space mapping, determining a dynamic boundary value by applying density clustering, and classifying real-time states to output risk early warning information. According to the technical method for realizing accurate classification and early warning, the defects of neglect of factor coupling relation, lack of scene adaptability, insufficient dynamic prediction capability and the like in the prior art are overcome, and the accuracy and practicability of safety performance evaluation and early warning of the motor vehicle are improved.

The invention provides a natural language text data intelligent classification method and system based on deep learning, and relates to the technical field of natural language processing, and the method comprises the steps: 1, employing a context awareness mechanism to analyze the real semantics of a target vocabulary according to an antagonistic variant existing in a text, and obtaining a target vocabulary; in combination with a word meaning library and a pre-training process of a dynamic learning rate adjustment strategy, generating a candidate replacement vocabulary set with consistent semantics; and step 2, based on the candidate replacement vocabulary set, performing multi-dimensional semantic similarity calculation and emotional tendency discrimination, determining applicable vocabularies conforming to an original culture background through a context adaptation strategy, and generating a standardized text sequence. According to the method, through multi-dimensional semantic analysis, cultural context fusion, cross-granularity feature construction and dynamic parameter correction, the accuracy and adaptability of natural language text classification are realized.

The invention discloses a multi-modal electroencephalogram classification method under multi-source interference based on a multi-head attention mechanism, and relates to the technical field of electroencephalogram signal processing. The method comprises the following steps: synchronously acquiring three types of electroencephalogram modal signals of motor imagery, steady-state visual evoked and event-related potentials, and synchronously acquiring multi-source interference data; performing data preprocessing on the multi-mode signal data; time domain, frequency domain and spatial domain three-dimensional features of each electroencephalogram mode are extracted, and meanwhile, physical features of interference data are quantified; the multi-domain features of the three types of electroencephalogram modes serve as parallel query vectors, the multi-source interference features serve as key value vectors, interference suppression weights are dynamically distributed through a layered attention mechanism, and anti-interference enhanced multi-mode fusion features are generated; and the multi-modal fusion features are processed through a gating circulation unit and a time domain attention module, a multi-modal classification result is output, and online learning and weight updating are carried out. According to the method, a robust decoding scheme can be provided for a brain control interaction system in a complex environment.

The invention relates to a Han-Cross industry text classification method and system based on prompt learning and adaptive loss weighting, and belongs to the technical field of natural language processing. The method comprises the following steps: designing and constructing a universal prompt template; the method comprises the following steps: recombining a Han-Cross cross-border industry text classification data set, namely converting an original single sample into paired samples; in a few-sample and multi-language scene, related vocabularies are adopted as external knowledge resources, and vocabularies most related to the mapping labels are retrieved from the related vocabularies; expanding the vocabulary mapper by introducing synonyms and associated vocabularies; adopting a dynamic mixed loss function and applying the dynamic mixed loss function to a pre-training language model to optimize a few-sample classification task; and classifying Chinese and Vietnamese cross-border industry texts by using the optimized pre-training language model. The method shows a remarkable effect in Chinese and Vietnamese industry text classification tasks, and is particularly suitable for a few-sample learning scene with data scarcity and language imbalance.

The invention relates to the technical field of databases, in particular to an intelligent mapping and classifying method based on heterogeneous data sources, which comprises the following steps: collecting heterogeneous data streams through an API (Application Program Interface) gateway and converting the heterogeneous data streams into structured data packets; using a semantic topology engine to fuse BERT semantic extraction, a graph convolutional network and a dynamic time warping technology to generate a cross-source association graph; constructing a field type clustering center by adopting a meta-learning framework based on the atlas, generating an initial classification rule through mode compatibility measurement, and dynamically updating the rule by means of adversarial training; outputting a DSL configuration script in combination with a template engine and an AST compiling technology; and dynamically adjusting a graph convolution weight and classifier parameters by using a strategy gradient algorithm through a reinforcement learning agent, and establishing a mapping-classification-verification collaborative optimization mechanism. According to the method, cross-source data semantic association accuracy is improved, small sample adaptive classification is realized, and system robustness and efficiency are improved.

The invention relates to a zero sample text classification method and device based on cross-modal information completion. According to the method, cross-modal information complemented image construction is realized by designing a new image label mapping mechanism and based on context text generation, finally, a multi-modal collaborative reasoning framework is designed in a reasoning stage, text semantic perception capability is enhanced by prompting project optimization text input, and zero sample reasoning capability of a CLIP model is effectively improved. Aiming at the semantic deviation problem of automatic label generation, a cross-modal information complemented image is constructed based on context text generation so as to optimize the text semantic representation integrity, and meanwhile, a prompt design is introduced in a reasoning stage to enhance the text semantic perception capability; according to the two-stage keyword automatic selection mechanism, firstly, a keyword candidate set is generated through a large model, secondly, through multi-modal matching and selection, the optimal keyword phrases are selected to serve as text information complemented by cross-modal information, and semantic accuracy and field adaptability are effectively improved.

The invention discloses a data asset classification method and system based on big data, and the method comprises the steps: carrying out the standardization processing of multi-source data assets in an enterprise business system, constructing a knowledge graph through correlation analysis, dynamically updating the graph according to the flow condition of the data assets, determining key data assets according to the use frequency of the data assets, and carrying out the classification of the key data assets. The method comprises the steps of obtaining a knowledge graph, obtaining an association path and association strength from the knowledge graph, carrying out foreground prediction, clustering prediction results according to data complexity, calculating a data asset evaluation average value of each cluster as a clustering center index, classifying all data assets according to the clustering center index, and forming a data asset portrait. And mutually isolated distributed storage is executed, and a minimum storage principle is followed to optimize life cycle management of data assets. According to the method, the utilization rate of data assets is increased, data security management is enhanced, an enterprise is supported to make a more intelligent data-driven decision, and meanwhile, the method has relatively good interpretability.

The embodiment of the invention provides a wafer defect classification method, a model training method, a system, equipment and a medium. According to the wafer defect classification scheme provided by the invention, the multi-modal test information of the wafer can be acquired, so that a plurality of test maps generated based on the multi-modal test information of the wafer are used as the basis of wafer defect classification, and the test information of different modals (namely, different dimensions) is considered during wafer defect classification; therefore, the accuracy of the classification result can be improved, and an actual manual wafer defect analysis mode can be met. Wherein the plurality of test maps comprise at least two types of maps, and one type of map is generated based on one type of modal test information.

The invention discloses an intelligent classification method and system for defects of a strain clamp of a power transmission line based on a two-dimensional image, and belongs to the field of defect classification of the strain clamp, and the method comprises the steps: carrying out the preprocessing of the two-dimensional image of the strain clamp, extracting related features, and constructing a defect region candidate set; the method comprises the following steps: firstly, acquiring edge information by using an edge detection algorithm, and calculating curvature and roughness to generate a morphological feature vector; then, training a support vector machine model based on the defect sample set, and classifying the first candidate set to obtain a first defect classification result; thirdly, density feature correction is carried out on wire clamp images of different materials, gray gradient features are extracted, a second candidate set is obtained, and the second candidate set is input into the trained SVM model for second defect classification; and finally, integrating the first classification result and the second classification result to generate a final defect classification result. According to the method, morphological characteristics and material characteristics are effectively combined, and the accuracy and robustness of defect detection are improved.

The invention discloses a node classification method and system based on a hybrid expert model.The node classification method comprises the steps that node data are input into a pre-trained hybrid expert model to obtain a classification result of the node data, and the hybrid expert model comprises a node mode extractor, a gating model and a plurality of classification models; the node mode extractor is used for extracting node mode information from given node data, and the gating model is used for receiving the node mode information and generating a model combination weight corresponding to each node; and dynamically selecting and combining the classification results of the node data by the plurality of different types of classification models according to the model combination weights corresponding to the nodes so as to obtain the final classification result of the nodes. The invention aims to solve the problem of node classification under a complex graph structure, accurately define a node mode under the condition that the node mode is complex, and properly combine different node predictors for different nodes.

The invention discloses a carbon emission source intelligent identification and classification method based on deep learning, and particularly relates to the technical field of carbon emission identification, and the method comprises the steps: collecting a carbon emission spatio-temporal data set of a target monitoring region, and forming a standardized spatio-temporal data set through a spatio-temporal alignment algorithm; based on a clustering algorithm, dividing the acquired carbon emission spatio-temporal data set into a plurality of time regions and a plurality of spatial positions; the data source corresponding to each time-space position is divided into multiple levels according to the resolution gradient, the time-space features of each time-space position are extracted by using a multi-scale convolutional neural network, and the time-space features comprise global features and local features; calculating to obtain an alignment priority of each time-space position; adjusting the alignment granularity of the data source according to the alignment priority; and matching fine-grained alignment for the space-time position with a large alignment priority value, or performing fine-grained alignment on the basis of the coarse alignment.

The invention provides an open set cross-domain hyperspectral classification method and system based on multi-mode and optimal transmission, and relates to the technical field of hyperspectral image classification. The method specifically comprises the following steps: acquiring and utilizing an original hyperspectral image to construct a labeled source domain data set and an unlabeled target domain data set; utilizing a spatial-spectral feature extractor to respectively obtain spatial-spectral features of the source domain data and spatial-spectral features of the target domain data; and determining a source domain prototype, calculating a similarity matrix between the target domain data and the source domain prototype, solving an optimal transmission matching matrix, distributing a pseudo tag for each target domain data, and performing text correction on all the source domain data and all the target domain data by using a contrast learning loss function to obtain the target domain data. And the source domain prototype and the spatial spectrum feature extractor are optimized by constructing the total loss of minimizing hyperspectral image classification, so that classification prediction can be carried out on the hyperspectral image to be classified, and target domain data can be classified more accurately.

The invention discloses a radar multi-angle multi-feature fusion target classification method, and relates to the technical field of radar signal processing, and the method comprises the steps: obtaining a multi-angle multi-feature fusion network target classification model; the multi-angle multi-feature fusion network target classification model comprises a time sequence feature extraction module, an attention fusion module and a feature fusion module; acquiring radar echo data of a target in real time; processing the radar echo data of the target to obtain processed radar echo data; the processed radar echo data comprises a time domain signal component graph of the target and a time frequency graph of the target; and inputting the processed radar echo data into a multi-angle multi-feature fusion network target classification model to obtain a target classification result. According to the invention, the accuracy of target classification can be improved.

The invention discloses an intelligent manufacturing defect automatic detection and classification method based on machine vision, and particularly relates to the technical field of defect automatic detection and classification, by constructing a high-resolution multi-source sample data set and introducing image preprocessing operation, defect expressions under different manufacturing batches, surface states and illumination conditions are covered, and the defect detection and classification accuracy is improved. Generating a defect probability heat map through an image segmentation network, extracting a primary defect candidate region, calculating a pseudo defect high-frequency interference coefficient by combining a high-frequency pseudo defect feature tensor, and calculating a multi-class defect overlapping coupling coefficient based on multi-classification confidence distribution and semantic adjacency; pseudo defect interference intensity and multi-class defect boundary fuzzy degree in the defect candidate area are accurately described, a sample label pollution risk assessment model is constructed to realize automatic identification and screening of high pollution risk samples in training data, and interference of mistakenly labeled samples on deep model training is significantly reduced; and erosion of error feature-label mapping on the generalization ability of the model is effectively prevented.