CN121033062B - Cross-domain gear surface defect detection method and system based on open set unknown separation - Google Patents
Cross-domain gear surface defect detection method and system based on open set unknown separationInfo
- Publication number
- CN121033062B CN121033062B CN202511575523.3A CN202511575523A CN121033062B CN 121033062 B CN121033062 B CN 121033062B CN 202511575523 A CN202511575523 A CN 202511575523A CN 121033062 B CN121033062 B CN 121033062B
- Authority
- CN
- China
- Prior art keywords
- class
- domain
- target domain
- unknown
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
- G06V10/765—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects using rules for classification or partitioning the feature space
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V10/7753—Incorporation of unlabelled data, e.g. multiple instance learning [MIL]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/778—Active pattern-learning, e.g. online learning of image or video features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of gear surface defect detection and discloses a cross-domain gear surface defect detection method and system based on open set unknown separation, wherein the method comprises the steps of obtaining surface images of gears of different types, dividing the surface images into a training set and a testing set, dividing the training set into a source domain sample and a target domain sample, and taking the images in the testing set as the target domain sample; the method comprises the steps of constructing a gear surface defect detection model comprising a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-field discriminator and a classifier, extracting source domain and target domain feature data by the feature extractor, dividing clear classes and fuzzy classes by the fuzzy class separation module, training the model by using feature data corresponding to the source domain, the clear classes and the fuzzy classes, applying the difference of a target domain sample on the classification confidence of the known classes as weight in the training process, and inputting a test set into the trained model to obtain a gear surface defect detection result in the target domain. The invention can improve the unknown class identification capability and the detection precision of the known class.
Description
Technical Field
The invention relates to the technical field of gear surface defect detection, in particular to a cross-domain gear surface defect detection method and system based on open set unknown separation.
Background
Gears are used as important mechanical transmission elements and widely applied to various mechanical equipment in industrial production, and the performance and reliability of the gears directly influence the operation efficiency and service life of a mechanical system. Due to the complex manufacturing process and high strength working environment, various defects such as fatigue cracks, wear, scratches, etc., which may develop gradually under adverse conditions, eventually leading to failure of the gear, are often present on the gear surface. Therefore, defect detection of the gear surface is particularly important.
The defect detection of the gear surface is usually carried out manually or mechanically, and the method has high cost and low efficiency. In order to replace the traditional manual and mechanical detection, the prior art has adopted the practice of using computer vision technology to detect the defects on the surface of the gear. Conventional machine vision surface defect detection techniques typically employ a combination of image processing and shallow machine learning, and the core challenge is how to extract good feature representations in order to be able to accurately distinguish between defect and non-defect areas, which often requires engineers in the professional field to manually select, design feature extraction methods and suitable classifiers according to different gear types and defect situations in practice, which results in such methods not being applicable to different gears or different defect situations. Moreover, conventional machine vision surface defect detection techniques often perform poorly when dealing with complex and varying defect types, as the artificial design features may not cover all cases.
Deep learning based models can automatically learn feature representations from data without relying on artificial design features, and thus the prior art has developed using deep learning models for gear surface defect detection. However, in an actual industrial detection scene, defects on the surface of the gear are complex and various and are easy to generate new defects because of interference of various external factors such as illumination, resolution, equipment and the like, and acquired images on the surface of the gear lack knowledge of unknown defects, so that the accuracy of the deep learning model in detecting the defects on the surface of the gear is affected.
Disclosure of Invention
Therefore, the technical problem to be solved by the invention is to overcome the defects in the prior art, and provide the cross-domain gear surface defect detection method and system based on open set unknown separation, which can improve the unknown class identification capability and the detection precision of the known class.
In order to solve the technical problems, the invention provides a cross-domain gear surface defect detection method based on open set unknown separation, which comprises the following steps:
the method comprises the steps of obtaining surface images of gears of different types, dividing the images in the training set into a source domain sample and a target domain sample according to the types of the gears, wherein the source domain sample is labeled, the target domain sample is unlabeled, and the images in the testing set are all target domain samples;
Constructing a gear surface defect detection model, wherein the gear surface defect detection model comprises a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-domain discriminator and a classifier, the feature extractor is used for extracting features of a source domain sample and a target domain sample to obtain source domain feature data and target domain feature data, and the fuzzy class separation module estimates a known class feature space range according to the source domain feature data and the target domain feature data and divides the samples in a training set into clear classes and fuzzy classes based on a class definition scoring mechanism;
Training a gear surface defect detection model by using a training set, namely training the unknown class discriminator by using source domain feature data, clear class and feature data corresponding to fuzzy class to realize class decision boundaries of the known class and the unknown class, training the classifier by using the source domain feature data and the feature data corresponding to the fuzzy class to distinguish all classes by using the model, training the sub-domain discriminator by using the source domain feature data and the feature data corresponding to the clear class to realize fine granularity distribution alignment of class layers, and applying the difference of a target domain sample on the classification confidence coefficient of the known class as a weight in the training process of the unknown class discriminator, the classifier and the sub-domain discriminator, performing risk calibration on a separation and adaptation process to adapt to alignment difficulty and dynamically adjusting the class decision boundaries;
inputting the test set into a trained gear surface defect detection model to obtain a gear surface defect detection result in the target domain.
Further, the fuzzy class separation module estimates a known class feature space range according to the source domain feature data and the target domain feature data and classifies samples in the training set into a clear class and a fuzzy class based on a class definition scoring mechanism, specifically:
Inputting the source domain feature data and the target domain feature data into the fuzzy class separation module, and calculating the ratio of the distance from the target domain feature data of each target domain sample to the corresponding class prototype to the space robust boundary of all class features;
And for each target domain sample, selecting the ratio with the smallest ratio in all the categories as a category definition score, and classifying the target domain sample into a definition category and a fuzzy category according to the size of the category definition score.
Further, the calculating the ratio of the distance from the target domain feature data of each target domain sample to the corresponding class prototype to the space robust boundary of all class features specifically includes:
The distance from the target domain characteristic data of each target domain sample to the corresponding class prototype is calculated as follows:
,
Wherein, the For the Euclidean distance between the target domain feature data of the ith target domain sample and the corresponding class prototype, c k is the kth class prototype of the target domain sample,Target domain feature data of the ith target domain sample,Is Euclidean distance;
the class feature space robust boundary of the k class is calculated as follows:
,
Wherein, the For the class-k feature space robust boundary, r is the to-be-solved variable,Accumulating a distribution function for a kernel density function corresponding to the kth class, wherein q is a control coefficient;
The ratio of the distance from the target domain feature data of the ith target domain sample to the corresponding class prototype to the class feature space robust boundary of the kth class is calculated as follows: 。
Further, the said The calculation method of (1) is as follows:
,
wherein N k is the total number of samples of the kth type of target domain, Is the bandwidth parameter of the gaussian kernel function.
Further, the loss function of the unknown class identifier during training is as follows:
,
Wherein, the For the loss function when the unknown class arbiter is trained,The total number of samples is the source domain; a label of the jth source domain sample, if the label is a fuzzy class Marked 1, if clear classMarked 0; for the jth source domain sample, () For the output of the feature extractor,() Is the output of the unknown class discriminator; for the total number of samples of the target domain, For the label of the i-th target domain sample,For the i-th sample of the target field,For a known class probability weight for the ith target domain sample,The unknown class probability weight for the ith target domain sample.
Further, the saidThe calculation method of (1) is as follows:
,
where mu represents the scaling factor, Is a preset constant value, and the preset constant value is set,In order for the domain of interest to be a target,Confidence for the ith target domain sample;
The said The calculation method of (1) is as follows:
。
Further, the said The calculation method of (1) is as follows:
,
where K is the number of known categories, Output for classifierBelongs to the category ofIs a probability of (2).
Further, the loss function when the classifier is trained is as follows:
,
Wherein, the For the loss function in the training of the classifier,For the total number of samples in the source domain,For the i-th source domain sample,In order to be a source domain,() In order to cross-entropy loss function,() For the output of the feature extractor, C () is the output of the classifier,For the label of the ith source domain sample,In order to control the parameters of the device,In order to be able to determine the number of samples of an unknown class,In the case of an unknown class of samples,In order for the domain of interest to be a target,Representing unknown class samples in the target domain; an open set pseudo tag assigned to the feature data corresponding to the unknown class, Unknown class probability weights for all target samples;; for the total number of samples of the target domain, The unknown class probability weight for the ith target domain sample.
Further, the loss function of the sub-field arbiter during training is:
,
Wherein, the A loss function aligned for the distribution when the sub-domain arbiter is trained,K is the total number of the source domain discriminators and is the total number of the sub-domain discriminators,For the i-th source domain sample,Is a source domain; A label is predicted for the class of the ith source domain sample, Whether the class prediction label representing the ith source domain sample is class k, if so=1, If not category k=0;() As an output of the sub-field arbiter,() Is the output of the feature extractor; In order for the domain of interest to be a target, The number of samples after filtering out ambiguities in the target domain,In order to filter out the samples after the blurring class,Representing the samples in the target domain after filtering out the ambiguous classes,Is thatCategory predictive labels of (c); For all known class probability weights, ;For the total number of samples of the target domain,Is a known class probability weight for the ith target domain sample.
The invention also provides a cross-domain gear surface defect detection system based on open set unknown separation, which comprises:
The data acquisition module acquires surface images of gears of different types and divides the surface images into a training set and a testing set, the images in the training set are divided into a source domain sample and a target domain sample according to the types of the gears, the source domain sample is labeled, the target domain sample is unlabeled, and the images in the testing set are all target domain samples;
The detection model construction module is used for constructing a gear surface defect detection model, the gear surface defect detection model comprises a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-domain discriminator and a classifier, the feature extractor is used for extracting features of a source domain sample and a target domain sample to obtain source domain feature data and target domain feature data, and the fuzzy class separation module is used for estimating a known class feature space range according to the source domain feature data and the target domain feature data and dividing the samples in a training set into clear classes and fuzzy classes based on a class definition scoring mechanism;
The training module trains the gear surface defect detection model by using a training set, specifically, trains the unknown class discriminator by using the source domain feature data, the feature data corresponding to the clear class and the fuzzy class, and realizes class decision boundaries for dividing the known class and the unknown class; training the classifier by using the source domain feature data and the feature data corresponding to the fuzzy class so as to distinguish all classes of the model, training the sub-domain discriminators by using the source domain feature data and the feature data corresponding to the clear class to realize fine granularity distribution alignment of class layers;
And the detection module inputs the test set into a trained gear surface defect detection model to obtain a gear surface defect detection result in the target domain.
Compared with the prior art, the technical scheme of the invention has the following beneficial effects:
According to the invention, the fuzzy class separation module is used for separating the fuzzy class and the clear class, so that the false alignment of the unknown class in the target domain can be avoided on the premise of ensuring the alignment precision of the known class, the fine granularity alignment is realized, the unknown class discriminator, the sub-domain discriminator and the classifier are trained on the basis, the risk calibration is carried out on the separation and adaptation process, the classifier confidence is introduced into the separation and adaptation process to adapt to the alignment difficulty and dynamically adjust the class decision boundary, thereby improving the detection precision of the known class defects on the gear surface and improving the identification capability of the unknown class defects.
Drawings
In order that the invention may be more readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings, in which:
FIG. 1 is a flow chart of a method in a preferred embodiment of the invention.
Fig. 2 is a block diagram of a gear surface defect detection model in a preferred embodiment of the present invention.
FIG. 3 is a schematic diagram of a fuzzy class separation module in accordance with a preferred embodiment of the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings and specific examples, which are not intended to be limiting, so that those skilled in the art will better understand the invention and practice it.
The unsupervised domain adaptation method expands the model from the source domain to the target domain by training the model on the labeled source domain data and the unlabeled target domain data, so that the knowledge of the source domain data is utilized to make up the shortage of the unlabeled data in the target domain, and the generalization capability and the adaptability of the model are improved. The existing unsupervised domain adaptation method is mostly limited to the closed set assumption, however, in the actual cross-domain industrial detection scene, the defects on the surface of the gear are complex and various and are easy to generate novel defects, so that the detection precision and the generalization capability of the traditional domain adaptation method are seriously affected. Therefore, the invention introduces an Open set domain adaptation (Open-Set Domain Adaptation, OSDA) method to solve the new problem caused by the occurrence of unknown categories in the target domain.
Referring to fig. 1, the invention discloses a cross-domain gear surface defect detection method based on open set unknown separation, which comprises the following steps:
S1, acquiring surface images of gears of different types, dividing the images in the training set into a source domain sample and a target domain sample according to the types of the gears, wherein the source domain sample is labeled, the target domain sample is unlabeled, and the images in the testing set are all target domain samples.
In the embodiment, the surface images of the gears with four defects of cracks, bumps, missing teeth and powder removal of the I-type gear and the II-type gear are obtained, and the I-type gear and the II-type gear are respectively gears with different processing technologies. The divided training set comprises 500 crack I-shaped gear images, 500 bumped I-shaped gear images, 300 missing tooth I-shaped gear images, 400 powder-removed I-shaped gear images, 500 crack II-shaped gear images, 500 bumped II-shaped gear images, 300 missing tooth II-shaped gear images and 400 powder-removed II-shaped gear images, wherein the test set comprises 100 crack II-shaped gear images, 100 bumped II-shaped gear images, 70 missing tooth II-shaped gear images and 80 powder-removed II-shaped gear images. In the test set, the image sample corresponding to the I-type gear is used as a source domain, and the image sample corresponding to the II-type gear is used as a target domain to divide the source domain sample and the target domain sample.
S2, constructing a gear surface defect detection model shown in FIG. 2, wherein the gear surface defect detection model comprises a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-domain discriminator and a classifier, the feature extractor is used for extracting features of a source domain sample and a target domain sample to obtain source domain feature data and target domain feature data, and the fuzzy class separation module estimates a known class feature space range according to the source domain feature data and the target domain feature data and classifies the samples in a training set into clear classes and fuzzy classes based on a class definition scoring mechanism.
The feature extractor maps the source domain samples and the target domain samples to a feature space, and the classifier is used for outputting a prediction result. In order to solve the problem that unknown class knowledge in a target domain does not exist in a source domain and therefore unknown classes are difficult to identify, a fuzzy class separation module and an unknown class discriminator are designed, and the purpose of autonomously exploring the class boundaries of the target domain is achieved instead of relying on a preset experience threshold. The sub-domain discriminant solves the domain offset problem between domains and performs fine granularity alignment on public classes.
S2-1 in this embodiment, the feature extractor and classifier are initialized by pre-training. And training the feature extractor and the classifier by utilizing the source domain sample in the pre-training stage, learning partial unchanged features in the migration task, improving the initial performance of the model for formal training, and accelerating the convergence rate. Meanwhile, reliable class feature space can be established in the early stage of formal training, and a foundation is laid for subsequent unknown class separation and known class alignment tasks. And then in the constructed gear surface defect detection model, the model realizes collaborative optimization by gradually learning unknown class knowledge from the fuzzy classes obtained by the fuzzy class separation module and combining the separation of the unknown classes with an alignment mechanism of the known classes. The model is continuously optimized on decision boundaries between unknown classes and known classes and inside the known classes, so that the migration performance of the model is remarkably improved.
S2-2, inputting the labeled source domain sample and the unlabeled target domain sample into a feature extractor of a gear surface defect detection model to respectively obtain source domain feature data and target domain feature data.
S2-3, the fuzzy class separation module estimates the space range of the known class features according to the source domain feature data and the target domain feature data and divides the samples in the training set into clear classes and fuzzy classes based on a class definition scoring mechanism, wherein the fuzzy class separation module comprises the following specific steps:
s2-3-1, inputting the source domain feature data and the target domain feature data into the fuzzy class separation module, wherein a schematic diagram of the fuzzy class separation module is shown in FIG. 3, and calculating the ratio of the distance from the target domain feature data of each target domain sample to the corresponding class prototype to the space robust boundary of all class features;
s2-3-1-1, calculating the distance from the target domain characteristic data of each target domain sample to the corresponding class prototype as follows:
,
Wherein, the For the Euclidean distance between the target domain feature data of the ith target domain sample and the corresponding class prototype, c k is the kth class prototype of the target domain sample,Target domain feature data of the ith target domain sample,The class prototypes are representative vectors obtained by averaging the characteristic data of all samples in a certain class in a specific characteristic space, and are marked as prototypes, and each class is provided with a corresponding prototype. Since each target domain sample belongs to a specific class, the corresponding class prototype corresponding to the sample refers to a representative vector of the class to which the sample belongs.
S2-3-1-2, calculating a class feature space robust boundary of the k class as follows:
,
Wherein, the For the class-k feature space robust boundary, r is the to-be-solved variable,Accumulating a distribution function for a kernel density function corresponding to a kth class, q being a control coefficient for controlling a level of truncation of a distribution tail resulting from the density estimation to exclude outlier samples within the class and enhance compactness of the feature distribution,Is defined as a cumulative distribution functionA minimum r value equal to 1-q;
In this embodiment, a gaussian kernel function is used to model the distance distribution of the kth class, The calculation method of (1) is as follows:
,
wherein N k is the total number of samples of the kth type of target domain, Bandwidth parameters that are gaussian kernel functions;
S2-3-1-3, calculating the ratio of the distance from the target domain feature data of the ith target domain sample to the corresponding class prototype to the class feature space robust boundary of the kth class as follows: 。
S2-3-2, selecting the smallest ratio in all the categories as the category definition score for each target domain sample, and classifying the target domain samples into clear categories and fuzzy categories according to the size of the category definition score.
S2-3-2-1, selecting the smallest specific value in all the categories as the category definition score, namely the category definition score of the ith target domain sample is:
,
wherein CCS i is the class definition score for i target domain samples;
S2-3-2-2, if CCS i is larger than 1, the corresponding target domain sample class definition is lower and difficult to judge as a known class, and is further defined as a fuzzy class, otherwise, if CCS i is smaller than or equal to 1, the class is defined as a clear class. During training, the clarity class is taken as a known class (denoted as ) Processing is performed to treat the fuzzy class as an unknown class (denoted as) And (5) processing.
The method comprises the steps of S3, training a gear surface defect detection model by using a training set, namely training the unknown class discriminator by using source domain feature data, clear classes and feature data corresponding to fuzzy classes to achieve class decision boundaries of the known classes and the unknown classes, training the classifier by using the source domain feature data and the feature data corresponding to the fuzzy classes to distinguish all classes, training the sub-domain discriminator by using the source domain feature data and the feature data corresponding to the clear classes to achieve fine grain distribution alignment of class layers, taking the difference of a target domain sample on the confidence coefficient of classification of the known classes as a main measurement index for measuring the contribution performance of the target domain sample on a network, applying the contribution performance as a weight in the training process of the unknown class discriminator, the classifier and the sub-domain discriminator, performing risk calibration on the separation and adaptation process to adapt to alignment difficulty and dynamically adjusting class decision boundaries, continuously optimizing loss and updating model parameters by using random gradient descent in the reverse propagation process, and storing the model parameters when the number of times reaches the optimal, and finishing training.
Dynamically adjusting classification decision boundaries of known classes and unknown classes by using the confidence of the classifier for carrying out known class recognition on the sample, and weighting the probability of the known classesAnd unknown class probability weightsIntroducing an unknown class identifier, wherein a loss function of the unknown class identifier during training is as follows:
,
Wherein, the For the loss function when the unknown class arbiter is trained,The total number of samples is the source domain; a label of the jth source domain sample, if the label is a fuzzy class Marked 1, if clear classMarked 0; a fuzzy class probability predictor output for the unknown class arbiter, For the jth source domain sample,() For the output of the feature extractor,() Is the output of the unknown class discriminator; for the total number of samples of the target domain, For the label of the i-th target domain sample,For the i-th sample of the target field,For a known class (i.e. defect class) probability weight of the ith target domain sample,The unknown class probability weight for the ith target domain sample.
The saidThe calculation method of (1) is as follows:
,
where mu represents the scaling factor, Is a preset constant, i.e. a small positive value preventing the denominator from being zero,In order for the domain of interest to be a target,Confidence for the ith target domain sample;
The said The calculation method of (1) is as follows:
。
And (3) introducing confidence coefficient of the classifier to perform risk calibration on the separation and adaptation process, and performing sample-level weighting by utilizing information entropy to adapt to the alignment difficulty of the known class samples and dynamically adjusting the classification decision boundaries of the known class and the unknown class. The output dimension of the classifier is K+1 (K is consistent with the number of the sub-field discriminators), wherein the first K classes belong to known classes, the unknown classes are classified into the last class, and only the output result of the classifier on the known classes is considered, so that the confidence of the classifier on the known classes is reflected more accurately. The said The calculation method of (1) is as follows:
,
where K is the number of known categories, Output for classifierBelongs to the category ofC () is the output of the classifier.
In the target domain classification learning, the classification process of the fuzzy class is easily affected by the domain distribution difference. In particular, target domain known class samples that fail to align well with the source domain distribution may be misclassified into fuzzy classes. If an open set pseudo tag were directly assigned to all fuzzy samples, this would negatively impact the learning of the unknown class features, possibly resulting in the model erroneously fusing some known class features into the unknown class distribution. To alleviate this problem, an unknown class weight is introduced to the classification lossAnd dynamically adjusting the contribution degree of the fuzzy class sample to the unknown class feature learning. The loss function during classifier training is as follows:
,
Wherein, the For the loss function in the training of the classifier,For the total number of samples in the source domain,For the i-th source domain sample,In order to be a source domain,() In order to cross-entropy loss function,() For the output of the feature extractor, C () is the output of the classifier,For the label of the ith source domain sample,To control parameters for controlling the learning ability of the model to unknown class features in the target domain,In order to be able to determine the number of samples of an unknown class,In the case of an unknown class of samples,In order for the domain of interest to be a target,Representing unknown class samples in the target domain.The method comprises the steps that an open set pseudo tag is endowed for feature data corresponding to an unknown class, the pseudo tag refers to a tag (a real tag which is not manually marked) generated by a model on unlabeled data, the tag is used for assisting training or reasoning, and the open set emphasizes that the tag aims at the unknown class in an open set scene;=[0,...0,1], including K0 s, K being the total number of known classes, 0 s indicating that it does not belong to a known class, and the last 1 s indicating that it belongs to an unknown class. For the unknown class probability weights of all target samples,,The total number of samples is the target domain; The smaller the value, the more confident the sample is for classifying the known class, the model can be prompted to ignore the influence of the sample which is wrongly classified into the unknown class, otherwise, the fuzzy class sample is difficult to be identified into the known class by the classifier, and the model preferentially classifies the fuzzy class sample into the unknown class. Classifier minimizes source domain label supervision loss and fuzzy class prediction and pseudo-labels by combining And the unknown class identifier to distinguish all classes together.
And constructing K sub-field discriminators for condition distribution alignment, filtering fuzzy samples in the target field through probability prediction information of the classifier, dividing the fuzzy samples into corresponding sub-field discriminators, and distributing source field samples to the corresponding sub-field discriminators according to the real labels. Each sub-domain arbiter focuses on the characteristic distribution of a specific class and reduces the alignment interference among classes by independently optimizing the weights, thereby realizing the fine granularity distribution alignment of class layers.
In the sub-domain distribution alignment process of the known class, there is a risk that the unknown class sample is misclassified into the known class. Sub-domain alignment loss introductionThe sub-domain discriminant is focused on the alignment of the high-confidence known class features, and the influence of the difficult-to-migrate sample is effectively ignored, so that the incorrect alignment of the unknown class features and the source domain features is avoided. By dynamically adjusting the alignment strategy of the known samples, the self-adaptive optimization training according to the alignment difficulty of the samples can be realized. The loss function of the sub-field arbiter during training is as follows:
,
Wherein, the A loss function aligned for the distribution when the sub-domain arbiter is trained,K is the total number of the source domain discriminators and is the total number of the sub-domain discriminators,For the i-th source domain sample,A label is predicted for the class of the ith source domain sample,Whether the class prediction label representing the ith source domain sample is class k, if so=1, If not category k=0;() Is the output of the sub-domain arbiter; In order for the domain of interest to be a target, The number of samples after filtering out ambiguities in the target domain,In order to filter out the samples after the blurring class,Representing the samples in the target domain after filtering out the ambiguous classes,Is thatIs a category predictive tag of (a),For all known class probability weights,;For the total number of samples of the target domain,Is a known class probability weight for the ith target domain sample.
Each sub-domain arbiter focuses on the characteristic distribution of a specific class and reduces the alignment interference among classes by independently optimizing the weights, thereby realizing the fine granularity distribution alignment of class layers. By optimizing the loss of each sub-domain discriminator, the features in the target domain can be better aligned with the known class features of the source domain, and the interference of the unknown class is reduced, so that the classification precision and the adaptability of the model are improved.
And S4, inputting the test set into a trained gear surface defect detection model to obtain a gear surface defect detection result in the target domain.
The invention also discloses a cross-domain gear surface defect detection system based on open set unknown separation, which comprises the following steps:
The data acquisition module acquires surface images of gears of different types and divides the surface images into a training set and a testing set, the images in the training set are divided into a source domain sample and a target domain sample according to the types of the gears, the source domain sample is labeled, the target domain sample is unlabeled, and the images in the testing set are all target domain samples;
The detection model construction module is used for constructing a gear surface defect detection model, the gear surface defect detection model comprises a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-domain discriminator and a classifier, the feature extractor is used for extracting features of a source domain sample and a target domain sample to obtain source domain feature data and target domain feature data, and the fuzzy class separation module is used for estimating a known class feature space range according to the source domain feature data and the target domain feature data and dividing the samples in a training set into clear classes and fuzzy classes based on a class definition scoring mechanism;
The training module trains the gear surface defect detection model by using a training set, specifically, trains the unknown class discriminator by using the source domain feature data, the feature data corresponding to the clear class and the fuzzy class, and realizes class decision boundaries for dividing the known class and the unknown class; training the classifier by using the source domain feature data and the feature data corresponding to the fuzzy class so as to distinguish all classes of the model, training the sub-domain discriminators by using the source domain feature data and the feature data corresponding to the clear class to realize fine granularity distribution alignment of class layers;
And the detection module inputs the test set into a trained gear surface defect detection model to obtain a gear surface defect detection result in the target domain.
The invention also discloses a computer readable storage medium, on which a computer program is stored, which when executed by a processor implements a cross-domain gear surface defect detection method based on open set unknown separation.
The invention also discloses a device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the cross-domain gear surface defect detection method based on open set unknown separation when executing the computer program.
According to the method, firstly, a fuzzy class separation module and an unknown class discriminator are designed aiming at the problem that a model cannot construct a decision boundary of known class defects and unknown class defects due to the fact that a source domain lacks of unknown class knowledge in a target domain, a known class feature space range is estimated by using center distance distribution, and fuzzy classes are separated based on a class definition scoring mechanism. Secondly, in order to avoid the misalignment of the unknown class in the target domain on the premise of ensuring the alignment precision of the known class, a sub-domain discriminator is constructed to realize fine granularity alignment. And finally, performing risk calibration on the separation and adaptation process, introducing the classifier confidence into the separation and adaptation process to adapt to the alignment difficulty and dynamically adjusting the class decision boundary. When the method is used for detecting the surface defects of the gears in the cross-domain open set scene, the identification capability of unknown classes can be improved, and the detection precision of the known classes can be remarkably improved.
In order to further prove the beneficial effects of the invention, the method of the invention and the existing OSBP are respectively used in the embodiment (see paper "Saito K, Yamamoto S, Ushiku Y, et al. Open set domain adaptation by backpropagation[C] Proceedings of the European conference on computer vision (ECCV). 2018: 153-168."）、STA（ for details "Liu H, Cao Z, Long M, et al. Separate to adapt: Open set domain adaptation via progressive separation[C] Proce-edings of the IEEE/CVF conference on computer vision and pattern recognition. 2019: 2927-2936."）、UADAL（ for details for paper "Jang J H, Na B, Shin D H, et al. Unknown-aware domain adversarial learning for open-set domain adaptation[J]. Advances in Neural Information Processing Systems, 2022, 35: 16755-16767."） for experiments, defect detection is carried out on the obtained gear surface image, universal evaluation indexes in an open-set domain adaptation task are used, the universal evaluation indexes comprise OS, UNK and HOS, which are used for measuring the performance of the model from different angles respectively: . The test results are shown in Table 1.
Table 1 results table of gear surface image defect detection by different methods
In table 1, "type I gear→type II gear" indicates that the image sample corresponding to the type I gear is used as the source domain and the image sample corresponding to the type II gear is used as the target domain, and "type II gear→type I gear" indicates that the image sample corresponding to the type II gear is used as the source domain and the image sample corresponding to the type I gear is used as the target domain. From table 1, it can be seen that the method of the present invention can be used to achieve both the detection accuracy of the known class and the detection accuracy of the unknown class, and the harmonic mean index is optimal, thereby proving the advantages of the present invention.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It is apparent that the above examples are given by way of illustration only and are not limiting of the embodiments. Other variations and modifications of the present invention will be apparent to those of ordinary skill in the art in light of the foregoing description. It is not necessary here nor is it exhaustive of all embodiments. And obvious variations or modifications thereof are contemplated as falling within the scope of the present invention.
Claims (6)
1. The method for detecting the surface defects of the cross-domain gear based on the unknown separation of the open set is characterized by comprising the following steps:
the method comprises the steps of obtaining surface images of gears of different types, dividing the images in the training set into a source domain sample and a target domain sample according to the types of the gears, wherein the source domain sample is labeled, the target domain sample is unlabeled, and the images in the testing set are all target domain samples;
Constructing a gear surface defect detection model, wherein the gear surface defect detection model comprises a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-domain discriminator and a classifier, the feature extractor is used for extracting features of a source domain sample and a target domain sample to obtain source domain feature data and target domain feature data, and the fuzzy class separation module estimates a known class feature space range according to the source domain feature data and the target domain feature data and divides the samples in a training set into clear classes and fuzzy classes based on a class definition scoring mechanism;
Training a gear surface defect detection model by using a training set, namely training the unknown class discriminator by using source domain feature data, clear class and feature data corresponding to fuzzy class to realize class decision boundaries of the known class and the unknown class, training the classifier by using the source domain feature data and the feature data corresponding to the fuzzy class to distinguish all classes by using the model, training the sub-domain discriminator by using the source domain feature data and the feature data corresponding to the clear class to realize fine granularity distribution alignment of class layers, and applying the difference of a target domain sample on the classification confidence coefficient of the known class as a weight in the training process of the unknown class discriminator, the classifier and the sub-domain discriminator, performing risk calibration on a separation and adaptation process to adapt to alignment difficulty and dynamically adjusting the class decision boundaries;
inputting the test set into a trained gear surface defect detection model to obtain a gear surface defect detection result in a target domain;
The method comprises the steps of applying the difference of the target domain sample on the confidence coefficient of the known class classification as a weight in the training process of the unknown class discriminator, the classifier and the sub-domain discriminator, and performing risk calibration on the separation and adaptation process to adapt to the alignment difficulty and dynamically adjust the class decision boundary, wherein the method comprises the following steps:
The loss function of the unknown class arbiter training is as follows:
,
Wherein, the For the loss function when the unknown class arbiter is trained,The total number of samples is the source domain; a label of the jth source domain sample, if the label is a fuzzy class Marked 1, if clear classMarked 0; for the jth source domain sample, () For the output of the feature extractor,() Is the output of the unknown class discriminator; for the total number of samples of the target domain, For the label of the i-th target domain sample,For the i-th sample of the target field,For a known class probability weight for the ith target domain sample,Unknown class probability weights for the ith target domain sample;
The said The calculation method of (1) is as follows:
,
where mu represents the scaling factor, Is a preset constant value, and the preset constant value is set,In order for the domain of interest to be a target,Confidence for the ith target domain sample;
The said The calculation method of (1) is as follows:;
The loss function during classifier training is as follows:
,
Wherein, the For the loss function in the training of the classifier,For the i-th source domain sample,In order to be a source domain,() For the cross entropy loss function, C () is the output of the classifier,For the label of the ith source domain sample,In order to control the parameters of the device,In order to be able to determine the number of samples of an unknown class,In the case of an unknown class of samples,Representing unknown class samples in the target domain; an open set pseudo tag assigned to the feature data corresponding to the unknown class, Unknown class probability weights for all target samples;;
The loss function of the sub-field arbiter during training is as follows:
,
Wherein, the The loss function is aligned for distribution during training of the sub-field discriminators, and K is the total number of the sub-field discriminators; A label is predicted for the class of the ith source domain sample, Whether the class prediction label representing the ith source domain sample is class k, if so=1, If not category k=0;() As an output of the sub-field arbiter,The number of samples after filtering out ambiguities in the target domain,In order to filter out the samples after the blurring class,Representing the samples in the target domain after filtering out the ambiguous classes,Is thatCategory predictive labels of (c); For all known class probability weights, 。
2. The method for detecting the surface defects of the cross-domain gear based on the unknown separation of the open set according to claim 1, wherein the fuzzy class separation module estimates the spatial range of the known class features according to the source domain feature data and the target domain feature data and classifies the samples in the training set into the clear class and the fuzzy class based on a class definition scoring mechanism, specifically comprising the following steps:
Inputting the source domain feature data and the target domain feature data into the fuzzy class separation module, and calculating the ratio of the distance from the target domain feature data of each target domain sample to the corresponding class prototype to the space robust boundary of all class features;
And for each target domain sample, selecting the ratio with the smallest ratio in all the categories as a category definition score, and classifying the target domain sample into a definition category and a fuzzy category according to the size of the category definition score.
3. The method for detecting surface defects of cross-domain gears based on open set unknown separation according to claim 2, wherein the calculating the ratio of the distance from the target domain feature data of each target domain sample to the corresponding class prototype to the spatial robust boundary of all class features is specifically as follows:
The distance from the target domain characteristic data of each target domain sample to the corresponding class prototype is calculated as follows:
,
Wherein, the For the Euclidean distance between the target domain feature data of the ith target domain sample and the corresponding class prototype, c k is the kth class prototype of the target domain sample,Target domain feature data of the ith target domain sample,Is Euclidean distance;
the class feature space robust boundary of the k class is calculated as follows:
,
Wherein, the For the class-k feature space robust boundary, r is the to-be-solved variable,Accumulating a distribution function for a kernel density function corresponding to the kth class, wherein q is a control coefficient;
The ratio of the distance from the target domain feature data of the ith target domain sample to the corresponding class prototype to the class feature space robust boundary of the kth class is calculated as follows: 。
4. the method for detecting surface defects of cross-domain gears based on unknown separation of open set according to claim 3, wherein the method is characterized in that The calculation method of (1) is as follows:
,
wherein N k is the total number of samples of the kth type of target domain, Is the bandwidth parameter of the gaussian kernel function.
5. The method for detecting surface defects of cross-domain gears based on unknown separation of open sets according to claim 1, wherein the method is characterized in thatThe calculation method of (1) is as follows:
,
where K is the number of known categories, Output for classifierBelongs to the category ofIs a probability of (2).
6. A cross-domain gear surface defect detection system based on open set unknown separation, comprising:
The data acquisition module acquires surface images of gears of different types and divides the surface images into a training set and a testing set, the images in the training set are divided into a source domain sample and a target domain sample according to the types of the gears, the source domain sample is labeled, the target domain sample is unlabeled, and the images in the testing set are all target domain samples;
The detection model construction module is used for constructing a gear surface defect detection model, the gear surface defect detection model comprises a feature extractor, a fuzzy class separation module, an unknown class discriminator, a sub-domain discriminator and a classifier, the feature extractor is used for extracting features of a source domain sample and a target domain sample to obtain source domain feature data and target domain feature data, and the fuzzy class separation module is used for estimating a known class feature space range according to the source domain feature data and the target domain feature data and dividing the samples in a training set into clear classes and fuzzy classes based on a class definition scoring mechanism;
The training module trains the gear surface defect detection model by using a training set, specifically, trains the unknown class discriminator by using the source domain feature data, the feature data corresponding to the clear class and the fuzzy class, and realizes class decision boundaries for dividing the known class and the unknown class; training the classifier by using the source domain feature data and the feature data corresponding to the fuzzy class so as to distinguish all classes of the model, training the sub-domain discriminators by using the source domain feature data and the feature data corresponding to the clear class to realize fine granularity distribution alignment of class layers;
The detection module inputs the test set into a trained gear surface defect detection model to obtain a gear surface defect detection result in the target domain;
The method comprises the steps of applying the difference of the target domain sample on the confidence coefficient of the known class classification as a weight in the training process of the unknown class discriminator, the classifier and the sub-domain discriminator, and performing risk calibration on the separation and adaptation process to adapt to the alignment difficulty and dynamically adjust the class decision boundary, wherein the method comprises the following steps:
The loss function of the unknown class arbiter training is as follows:
,
Wherein, the For the loss function when the unknown class arbiter is trained,The total number of samples is the source domain; a label of the jth source domain sample, if the label is a fuzzy class Marked 1, if clear classMarked 0; for the jth source domain sample, () For the output of the feature extractor,() Is the output of the unknown class discriminator; for the total number of samples of the target domain, For the label of the i-th target domain sample,For the i-th sample of the target field,For a known class probability weight for the ith target domain sample,Unknown class probability weights for the ith target domain sample;
The said The calculation method of (1) is as follows:
,
where mu represents the scaling factor, Is a preset constant value, and the preset constant value is set,In order for the domain of interest to be a target,Confidence for the ith target domain sample;
The said The calculation method of (1) is as follows:;
The loss function during classifier training is as follows:
,
Wherein, the For the loss function in the training of the classifier,For the i-th source domain sample,In order to be a source domain,() For the cross entropy loss function, C () is the output of the classifier,For the label of the ith source domain sample,In order to control the parameters of the device,In order to be able to determine the number of samples of an unknown class,In the case of an unknown class of samples,Representing unknown class samples in the target domain; an open set pseudo tag assigned to the feature data corresponding to the unknown class, Unknown class probability weights for all target samples;;
The loss function of the sub-field arbiter during training is as follows:
,
Wherein, the The loss function is aligned for distribution during training of the sub-field discriminators, and K is the total number of the sub-field discriminators; A label is predicted for the class of the ith source domain sample, Whether the class prediction label representing the ith source domain sample is class k, if so=1, If not category k=0;() As an output of the sub-field arbiter,The number of samples after filtering out ambiguities in the target domain,In order to filter out the samples after the blurring class,Representing the samples in the target domain after filtering out the ambiguous classes,Is thatCategory predictive labels of (c); For all known class probability weights, 。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202511575523.3A CN121033062B (en) | 2025-10-31 | 2025-10-31 | Cross-domain gear surface defect detection method and system based on open set unknown separation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202511575523.3A CN121033062B (en) | 2025-10-31 | 2025-10-31 | Cross-domain gear surface defect detection method and system based on open set unknown separation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN121033062A CN121033062A (en) | 2025-11-28 |
| CN121033062B true CN121033062B (en) | 2026-04-07 |
Family
ID=97776646
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202511575523.3A Active CN121033062B (en) | 2025-10-31 | 2025-10-31 | Cross-domain gear surface defect detection method and system based on open set unknown separation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN121033062B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN121724996A (en) * | 2026-02-26 | 2026-03-24 | 中国科学院西安光学精密机械研究所 | Open set industrial defect detection method based on dynamic fuzzy prototype network |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116894187A (en) * | 2023-07-12 | 2023-10-17 | 山东省计算中心（国家超级计算济南中心） | A gearbox fault diagnosis method based on deep transfer learning |
| CN117892203A (en) * | 2024-03-14 | 2024-04-16 | 江南大学 | Defective gear classification method, device and computer readable storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11379967B2 (en) * | 2019-01-18 | 2022-07-05 | Kla Corporation | Methods and systems for inspection of semiconductor structures with automatically generated defect features |
| WO2023155069A1 (en) * | 2022-02-16 | 2023-08-24 | 佛山市南海区广工大数控装备协同创新研究院 | Deep-learning-based surface defect detection method for mobile phone battery |
| CN118212240B (en) * | 2024-05-22 | 2024-07-30 | 山东华德重工机械有限公司 | Automobile gear production defect detection method |
-
2025
- 2025-10-31 CN CN202511575523.3A patent/CN121033062B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116894187A (en) * | 2023-07-12 | 2023-10-17 | 山东省计算中心（国家超级计算济南中心） | A gearbox fault diagnosis method based on deep transfer learning |
| CN117892203A (en) * | 2024-03-14 | 2024-04-16 | 江南大学 | Defective gear classification method, device and computer readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN121033062A (en) | 2025-11-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110211097B (en) | Crack image detection method based on fast R-CNN parameter migration | |
| CN111860106B (en) | Unsupervised bridge crack identification method | |
| CN119229372A (en) | Battery thin film production automation control method and system based on visual inspection | |
| CN121033062B (en) | Cross-domain gear surface defect detection method and system based on open set unknown separation | |
| CN114694028B (en) | Ship weld defect detection method based on convolution confidence generation countermeasure network model | |
| CN113724233B (en) | Transformer equipment appearance image defect detection method based on fusion data generation and transfer learning technology | |
| CN117349595A (en) | An extreme few-sample fault diagnosis method for industrial scenarios | |
| CN113988215A (en) | Power distribution network metering cabinet state detection method and system | |
| CN114219763A (en) | Infrared picture detection method for abnormal heating point of power distribution equipment based on fast RCNN algorithm | |
| CN119089297B (en) | A cross-domain fault diagnosis method and system based on pseudo-label learning under time-varying working conditions | |
| CN119848622B (en) | A robust cross-condition migration fault diagnosis method under time-varying operating conditions with multi-noise characteristics | |
| CN113870254A (en) | Target object detection method, device, electronic device and storage medium | |
| CN119831992A (en) | Self-iterative membrane electrode assembly defect detection method, device and equipment | |
| CN117649387B (en) | Defect detection method suitable for object with complex surface texture | |
| CN118967474A (en) | A defect image enhancement method based on traditional data enhancement | |
| CN117808748A (en) | Battery defect detection method based on richer convolutional feature edge detection network | |
| CN121504862A (en) | A Deep Learning-Based Method and System for Detecting Appearance Defects in Carbon Fiber Prepreg | |
| CN121147219A (en) | A Weld Defect Identification Method Based on Densely Connected Convolutional Network Model | |
| CN119004044B (en) | A bearing fault diagnosis method, device and equipment based on weighted confrontation | |
| CN120541577A (en) | Self-supervised detection and classification method and system for concept drift in network data | |
| CN120316564A (en) | A Fault Diagnosis Method and System Based on Hybrid Attention Domain Adaptation Network | |
| CN117292119B (en) | Power transmission multi-scale target detection method and system | |
| CN118333988A (en) | Defect detection method and system based on difference enhancement and adaptive meta-learning | |
| Barberena et al. | Handling Drift in Industrial Defect Detection Through MMD-Based Domain Adaptation | |
| CN118797539B (en) | Intelligent detection early warning method based on multi-level dynamic analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |