23 results about "Invariant feature extraction" patented technology

The invention discloses a photovoltaic array fault diagnosis and positioning method and system based on digital twinning and deep learning. The method comprises the following steps: constructing a photovoltaic array digital twin, synchronously collecting multi-source monitoring data and converting the multi-source monitoring data into a time-frequency spectrum; constructing a twin network model, carrying out feature learning through a triple loss function, and extracting high-discrimination depth features; an adversarial transfer learning mechanism is introduced, cross-working-condition domain invariant feature extraction is realized through adversarial training of a feature generation unit and a domain discriminator, and the diagnosis robustness is improved; a multi-source domain generalization strategy is adopted, domain invariant features and domain private features are extracted through a double-branch network, difference constraints are applied, adaptive fusion is carried out, and a generalization diagnosis model oriented to unknown working conditions is constructed; and integrating the model to a digital twinborn body to realize accurate positioning and visualization of a fault component. According to the method, the problems of low fault diagnosis precision and poor generalization ability of the photovoltaic array in data scarcity, variable working conditions and unknown environments are solved.

The invention discloses a non-contact palm vein multi-mode identification system based on deep learning. The system comprises a multi-mode image acquisition and processing terminal, a mode matching and identification decision terminal, a safety protection and anti-counterfeiting terminal and an iterative optimization terminal. The multi-modal image acquisition and processing terminal is used for acquiring palm print and palm vein images, performing dynamic calibration, image quality enhancement and preprocessing, and outputting multi-modal images; the mode matching and identification decision terminal is used for carrying out feature extraction, multi-modal feature fusion, mode matching and identification decision and outputting a final matching result; the safety protection and anti-counterfeiting terminal is used for carrying out living body detection, anti-counterfeiting monitoring and data encryption; and the iterative optimization terminal is used for realizing model lightweight, knowledge distillation, end-to-end optimization and continuous learning. According to the invention, through deformation invariant feature extraction and cross-modal attention fusion, the recognition accuracy and environmental adaptability in a complex state are improved.

The invention relates to the technical field of monitoring, in particular to an intelligent camera monitoring data storage and data enhancement processing method, which effectively solves the image quality problem under a complex illumination condition through the combination of illumination invariant feature extraction and a weighted Gaussian probability model, especially the influence of dynamic shadow, uneven illumination and the like. According to the dynamic shielding synthesis mechanism, the dynamic shielding in a real scene is simulated through a Poisson fusion method based on a semantic segmentation result and a physically reasonable shielding object generation algorithm, the limitation of a conventional fixed template shielding method is avoided, the diversity and spatial rationality of shielding objects can be better reflected, and the real-time performance of the real scene is improved. The robustness of the model to shielding is improved; according to the invention, through combination of the cross-scale residual aggregation module and the gating channel-space attention mechanism, multi-scale feature reservation and deep fusion are realized, and detail information under small targets and complex illumination can be effectively captured in a complex monitoring scene.

The invention relates to the technical field of medical image processing, in particular to a pulmonary nodule follow-up visit intelligent management system which comprises a pulmonary nodule intelligent matching and change analysis engine, a pulmonary nodule intelligent follow-up visit scheme generation engine, a cloud storage and collaborative service module, an early warning module and a user management and operation module. The surface of the pulmonary nodule is modeled into a differential manifold, accurate matching of the pulmonary nodule and quantitative analysis of small changes are achieved through a multi-scale differential invariant feature extraction framework and a geodesic distance-based change analysis technology, the system generates a personalized follow-up visit scheme based on a risk assessment model, and through a multi-stage early warning mechanism and multi-channel notification distribution, the risk assessment accuracy of the pulmonary nodule is improved. The system also adopts a GPU acceleration computing architecture, so that the processing efficiency is remarkably improved, multi-terminal cooperative operation is supported, the accuracy, intelligence and individuation of pulmonary nodule follow-up visit management are realized, and the follow-up visit efficiency and the early detection rate of lung cancer are improved.

The invention provides a fault diagnosis method for open set domain generalization under a continuous variable working condition. The method comprises the following steps: firstly, constructing a training sample with time domain data, a working condition index and a fault category index; converting the time domain data into a time-frequency domain and extracting semantic features; a class-specific semantic reconstruction module is adopted to classify the semantic features; a cross-domain alignment module is adopted to estimate mutual information of the semantic features of the training samples and the working condition indexes, and the smaller the mutual information is, the lower the working condition dependence degree is; constructing joint loss including cross-domain alignment loss; the cross-domain alignment loss constrains the feature distribution consistency between continuous domains based on mutual information minimization; back propagation training is carried out by adopting joint loss, so that the class-specific semantic reconstruction module learns and extracts the working condition invariant feature extraction capability; and performing fault classification by using the trained feature extraction module and the class-specific semantic reconstruction module. According to the method, fault diagnosis can be realized under a cross-continuous change working condition, and meanwhile, the method has the capability of identifying unknown fault types.

This invention discloses a rotation- and scale-invariant feature extraction method and system based on dual-coordinate system collaboration, belonging to the field of image processing technology. Addressing the poor matching performance of existing image feature extraction methods under rotation and scale changes, this invention employs the following scheme: The input image undergoes a logarithmic polar coordinate transformation to generate a polar coordinate image; a dual-branch network is constructed, where the feature extraction branch extracts keypoint location information in a Cartesian coordinate system, and the feature description branch extracts a polar coordinate feature map in a logarithmic polar coordinate system; the keypoint locations in the Cartesian coordinate system are mapped to their corresponding positions in the polar coordinate feature map using a coordinate mapping function, and interpolation sampling is performed to generate feature descriptors that integrate precise location information and rotation / scale invariance information; based on the feature descriptors, matching point pairs are calculated, and the spatial transformation relationship between images is determined. This invention achieves highly robust feature extraction under rotation and scale changes while maintaining high real-time processing efficiency.

The invention discloses a cross-domain steganographic text recognition and analysis method and system based on multi-adversarial domain adaptation, and belongs to the technical field of network security. The method comprises the following steps: constructing a feature extractor of double heterogeneous branches, and performing fine-grained semantic feature extraction on a real text and a text in a public data set; constructing a steganography text analyzer based on the combination of a multi-adversarial domain self-adaptive domain invariant feature extractor and a discriminator based on a full connection layer, allocating a unique domain discriminator for each class, realizing cross-domain feature alignment, and updating parameters of the feature extractor and the discriminator; for the trained steganography text analyzer, a mutual learning mechanism and a feature alignment means are further used to optimize the model; and deploying a steganography text analyzer and carrying out real-time detection. Compared with a traditional single-branch method or a method without a mutual learning mechanism, the method has the advantages that the cross-domain knowledge migration efficiency is improved, and stable recognition performance can still be maintained even in a scene with remarkable inter-domain distribution difference.

The application discloses a kind of non-contact palm vein multi-modal identification system based on deep learning, including multi-modal image acquisition and processing terminal, mode matching and identification decision terminal, security protection and anti-fake terminal, iterative optimization terminal;Multi-modal image acquisition and processing terminal are used to collect palm print and palm vein image, carry out dynamic calibration, image quality enhancement and pre-processing, output multi-modal image;Mode matching and identification decision terminal are used to carry out feature extraction, multi-modal feature fusion, mode matching and identification decision, and output final matching result;Security protection and anti-fake terminal are used to carry out living body detection, anti-fake monitoring, data encryption;Iterative optimization terminal is used to realize model light weight, knowledge distillation, end-to-end optimization and continuous learning.The application improves the recognition accuracy and environmental adaptability under complex state by deformation invariant feature extraction and cross-modal attention fusion.

The application discloses a hyperspectral image anti-attack method based on invariant features, which comprises the following steps: step one, constructing a sample set; step two, pre-training a deep convolutional neural network classification model; step three, building an anti-attack model; step four, constructing a loss function of the anti-attack model; step five, iteratively training the anti-attack model; and step six, testing the trained anti-attack model. The anti-attack method can enhance the robustness of a convolutional neural network and improve the classification accuracy of a hyperspectral classification model against an anti-attack.

This disclosure provides a method, device, electronic equipment, and storage medium for HVAC fault diagnosis, relating to the field of fault diagnosis technology. It acquires multi-source time-series operational data of an HVAC system and performs standardized preprocessing. A multi-scale feature extraction network is constructed, containing convolutional branches with different receptive fields set in parallel and achieving adaptive fusion of features from each branch based on an attention mechanism. This network is then optimized using domain adversarial training to align feature distributions on the source and target domain datasets, thereby obtaining a domain-invariant feature extractor. Finally, this domain-invariant feature extractor is combined with a few-sample learning paradigm of metric learning to calculate feature prototypes for each category in the dataset. The fault category is determined based on the distance metric between the fault query data and the feature prototypes. Therefore, this method can solve the problems of poor model generalization and difficulty in accurately diagnosing faults in small-sample scenarios in existing technologies.

The application discloses a single-source domain target recognition generalization method, product, medium and equipment, relates to the field of domain adaptive target recognition, and comprises the following steps: generating a stylized image corresponding to an original image through a style feature space; encoding original image and stylized image features; jointly decoupling the original image and the stylized image features into domain-invariant features and domain-unique features; training a region candidate network using the domain-invariant features, and optimizing the network using orthogonal loss and target recognition loss functions; and inputting a complex unknown weather sea target image into the optimized network to obtain the category and position of the sea target in the complex unknown weather sea target image. The application can overcome the problems of difficulty in extracting domain-invariant features in complex weather data sets, single style generation and data generated being biased towards source domain distribution, and difficulty in method model generalization, improve the generalization ability of the model, and effectively improve the recognition ability of sea targets under different complex unknown weather conditions.

This invention provides a cross-device domain adaptation method based on domain decoupling and class confusion minimization feature alignment. The main steps include: collecting vibration signals from different devices to construct labeled source domain and unlabeled target domain datasets, and dividing them into training and testing sets; constructing a domain decoupling module based on a feature extractor with convolutional channel separation and a decoder reconstructing the input data, and constructing a classifier and domain discriminator on this basis; constraining the model's learning behavior through sample reconstruction loss, conditional adversarial domain adaptation loss, classification loss, and class confusion minimization loss, the model decouples features into domain-specific features and domain-invariant features to enhance the extraction effect of domain-invariant features, aligns domain-invariant features, and simultaneously minimizes inter-class confusion in the target domain; completing model training on the training set, and finally establishing a high-precision fault diagnosis model to achieve fault diagnosis of the target device. This invention enhances the representation performance of domain-invariant features through the decoupling module, improves feature alignment by combining conditional domain adaptation loss and class confusion minimization loss, and optimizes the classifier's classification behavior in the target domain, effectively addressing the problem of completely missing labels in the target device dataset.

The invention discloses a low-duration calibration electroencephalogram decoding method and system based on invariant feature extraction, and relates to the technical field of electroencephalogram signal decoding, and the method comprises the steps: obtaining electroencephalogram data with labels, and carrying out Fourier transform to obtain amplitude information and phase information; inputting the phase information and the tag into a teacher network for training to obtain feature output of the teacher network; performing random pairing and reconstruction on the labeled electroencephalogram data to obtain reconstructed electroencephalogram data, inputting the reconstructed electroencephalogram data into a student network, and learning phase information extracted from a teacher network through the student network to obtain feature output of the student network; and acquiring real-time data, and inputting the real-time data into a student network to obtain an electroencephalogram decoding recognition result. Similarity features among different individuals are extracted through an invariant feature extraction method, when electroencephalogram signals of a new user are analyzed, the similarity features can be directly used for electroencephalogram analysis of a new subject, and the use efficiency of a non-intrusive brain-computer interface and the electroencephalogram recognition accuracy are remarkably improved.

The invention discloses an image invariant feature extraction method based on feature decoupling and a network thereof, which are used for detecting a copy image, and the method mainly comprises two parts of neural network construction and network training. The method comprises the following steps: firstly, constructing a decoupling neural network based on five network modules including an invariant feature encoder, a distortion feature encoder, a decoder, a mutual information estimation network and a feature compression module; when the network is trained, a mutual information loss function and a reconstruction loss function are used to decouple the invariant features of the image, a feature loss function is used to constrain the invariant features of the copy image to be close to the invariant features of the original image, and the invariant features of the irrelevant images are far away from each other. Further separating the features of the copy image and the irrelevant image by using a nearest neighbor sample separation loss function; according to the method, the robust low-dimensional invariant features can be extracted only by retaining the invariant feature encoder and the feature compression module in the reasoning stage.

This invention relates to the field of medical equipment management technology and discloses an intelligent surgical instrument identification and real-time inventory image analysis system. The system includes an image acquisition module, an image preprocessing module, a topological feature extraction module, an instrument identification module, a spatiotemporal trajectory analysis module, a state prediction module, and a decision support module. This system achieves high-precision identification of surgical instruments through topological invariant feature extraction technology; accurately identifies instrument usage anomalies using spatiotemporal topological manifold trajectory analysis; and achieves forward-looking prediction of instrument states based on topological manifold learning prediction technology. This overcomes the bottleneck of low accuracy in complex environments inherent in traditional image recognition methods, providing a new technical guarantee for surgical safety.

The application discloses a photovoltaic array fault diagnosis and positioning method and system based on digital twinning and deep learning. The method comprises the following steps: constructing a photovoltaic array digital twinning body, synchronously collecting multi-source monitoring data and converting the multi-source monitoring data into time-frequency spectrum; constructing a twinning network model, performing feature learning through a triple loss function, and extracting high-discriminative deep features; introducing an adversarial transfer learning mechanism, performing adversarial training through a feature generation unit and a domain discriminator, realizing cross-condition-domain invariant feature extraction, and improving the diagnosis robustness; adopting a multi-source domain generalization strategy, extracting domain-invariant features and domain-private features through a double-branch network, applying difference constraints and adaptively fusing, and constructing a generalization diagnosis model for unknown conditions; and integrating the model into the digital twinning body to realize accurate positioning and visualization of fault components. The application solves the problems of low fault diagnosis precision and poor generalization ability of photovoltaic arrays in the conditions of data scarcity, variable working conditions and unknown environments.

The invention provides a multivariate time sequence classification domain adaptive method and device, a storage medium and a program product, and the method comprises the steps: collecting multivariate time sequence data, and carrying out the multivariate multi-scale feature extraction, multi-scale maximum pooling, special convolution fusion and L1 regularization constraint of the multivariate time sequence data, constructing a convolutional neural network structure for extracting domain invariant features to obtain a time sequence classification model; and based on the time sequence classification model, through source domain pre-training, target domain pseudo label generation and iterative fine tuning, optimizing classification prediction of the time sequence classification model in a target domain. According to the method, domain invariant feature extraction is realized based on multivariable multi-scale feature extraction fusion and an L1 regularized deep neural network, cross-domain migration is better realized by further cooperating with a self-training technology, and the classification prediction effect of a time sequence classification model in an actual application scene is improved.

The application discloses a kind of mask image modeling guide domain adaptation cross-domain remote sensing scene classification method, comprising: constructing domain adaptation network, the self-encoding of unlabelled self-supervised pre-training is carried out, and the model parameter of pre-training self-encoding is obtained;Self-encoding model parameters are loaded into self-encoder, data are input into domain adaptation network, and the mask image modeling of source domain image and target domain image is carried out using self-encoder;High-level semantic feature distribution of source domain and target domain is aligned using feature adaptation module;And based on data, overall target loss function is constructed, and the iterative training of domain adaptation network is carried out by optimizing overall target loss function, the decoder part of self-encoder is removed, the encoder of self-encoder and feature adaptation module are used to test target domain image, and good scene classification result is obtained.The application preserves domain-specific features in the process of extracting domain-invariant features, and further improves the classification generalization ability for unlabelled data target domain.

The invention provides a low-illumination scene target detection method based on hierarchical feature enhancement. The method comprises the following steps: acquiring a to-be-detected image; an illumination invariant feature extraction module IRFE is adopted to extract features of the to-be-detected image, fusion convolution is carried out on the features and the to-be-detected image, and an illumination robust image is generated; inputting the image containing the illumination invariant feature into a hierarchical feature enhancement network HFENet to reconstruct a high-resolution image; and inputting the generated high-resolution image into a YOLO detection network, and extracting a scene target. According to the low-illumination scene target detection method based on hierarchical feature enhancement, parameters are updated through back propagation, the detection loss guides the network to focus on target area features more effectively, and image processing and feature extraction are optimized.

The invention discloses an HRRP target identification method and system based on angle guidance, and the method comprises the steps: constructing a target identification network model, obtaining HRRP data of a target vehicle in a static state, carrying out the dynamic convolution operation of the HRRP data through introducing azimuth angle information, and extracting the local scattering features of angle self-adaption. Further acquiring global structure features by using a global attention module; and finally, carrying out cross fusion on the local dynamic convolution features and the global attention features, and outputting a target category through a classification module. The azimuth angle information is introduced into the convolution kernel generation process through the angle condition kernel hybrid convolution module, so that the convolution kernel can be adaptively adjusted according to the azimuth angle, the change rule of the HRRP along with the azimuth angle is effectively described, the traditional azimuth invariant feature extraction or angular domain division is not needed, and the influence of the azimuth sensitivity on the recognition performance is relieved; and the identification precision and the generalization ability are further improved.

The invention discloses a target monitoring system based on optical remote sensing. The system comprises a monitoring data acquisition module, a data preliminary processing module, a preliminary screening model construction module, a monitoring model construction module and a target monitoring module. The invention relates to the technical field of optical remote sensing intelligent image processing, in particular to a target monitoring system based on optical remote sensing. A data primary processing method of multispectral data correction, data enhancement, data standardization and data set segmentation is adopted; an improved joint decision forest model is adopted as a preliminary screening model, and a target and an interference region can be distinguished under a complex background by positioning a potential abnormal region, reducing the redundant calculation amount of subsequent processing and fusing multi-dimensional spectrum and spatial features; an improved convolutional neural network model is adopted as a monitoring model, and the sensitivity to target rotation and scale change is overcome by introducing a rotation invariant feature extraction and dynamic instance decoding mechanism.

The invention discloses a domain generalization remote sensing image change detection method based on domain invariant feature extraction, and the method comprises the following steps: carrying out the geometric enhancement of obtained synthetic dual-phase remote sensing image data, and dividing an obtained enhanced synthetic data set into a training set and a test set; constructing a domain generalization remote sensing image change detection network based on semantics; selecting a change detection algorithm as a feature extractor to obtain an output feature of an original image pair and an output feature of an enhanced image pair which are consistent with the output in spatial size, and respectively inputting the two groups of features into a prediction head of 3 * 3 convolution sharing weight to obtain two detection result images; constructing an auto-covariance matching loss function and a cross-covariance diagonal loss function; and setting an overall loss function, iteratively training and optimizing network parameters, and inputting a detection image into the trained neural network after the loss is stable to obtain a final detection result graph. According to the method, the domain offset between the training domain data set and the unseen domain data set can be effectively reduced, so that the generalization ability of an existing convolutional neural network model is effectively improved.

The invention discloses a few-sample multi-modal image unsupervised registration method based on modal adaptive optimization. The method is implemented based on a multi-modal registration network, and the construction method of the network comprises the steps that firstly, a multi-modal pre-training data set with rich modals is constructed, pre-training is conducted on the multi-modal registration network through the data set, and the multi-modal registration network comprises a modal invariant feature extraction module and a homography registration module; when few-sample multi-modal image data is processed, the initial pre-training weight of the network is frozen, a modal self-adaptive optimizer is introduced into a modal invariant feature extraction and homography registration module, and optimization training is carried out for specific few-sample data; unsupervised learning is realized through Gram matrix feature loss and self-supervised homography loss, and enhanced training is performed on a modal adaptive optimizer in combination with simulation deformation so as to improve the network adaptability and robustness. The method can be used for solving the problem of multi-modal image registration under the scene of sample missing or data acquisition difficulty.