88 results about "Label free" patented technology

The invention discloses a multi-agent dynamic task allocation and collaborative path-finding system based on label-free distributed deep reinforcement learning. The system comprises the following steps: step 1, receiving state information and environment perception data of each agent in a multi-agent system based on distributed deep reinforcement learning; step 2, extracting feature representations of the environmental perception data and the intelligent agent state information, and performing multi-source heterogeneous information fusion through an attention mechanism to obtain state-task matching features; 3, transmitting the state-task matching characteristics to a multi-agent network in real time, realizing task allocation negotiation among agents by adopting a hierarchical scheduling and state exchange mechanism based on task priorities, dynamically detecting newly added task types, and performing incremental learning; the online updating iteration of the model is realized to assist the multi-agent optimization task allocation strategy and the path planning action; compared with the prior art, the method has the advantages that by applying the distributed deep reinforcement learning technology and a state exchange mechanism between intelligent agents, the system can quickly adapt to environment changes and task dynamics, and the resource utilization rate and the task completion efficiency are improved.

The invention provides a graphical interface agent training method and device and a storage medium, and the method comprises the steps: deducing an initial state and a transition state corresponding to an action in a first unlabeled graphical interface interaction track based on a dynamic model, and enabling the initial state and the transition state to serve as a supervision signal of the first unlabeled graphical interface interaction track; on the basis of an inverse dynamic model, inferring an action corresponding to state transition in the second label-free graphical interface interaction track, and taking the action as a supervision signal of the second label-free graphical interface interaction track; and training the graphical interface agent based on the first untagged graphical interface interaction track with the supervision signal and the second untagged graphical interface interaction track with the supervision signal. Therefore, the supervision signal is automatically extracted from the unlabeled data, rich information contained in the unlabeled data is fully utilized, the economic and time cost of data labeling is reduced, and the performance of the GUI intelligent agent on key technical indexes such as multi-step task planning and interface understanding is improved.

The invention discloses an unsupervised industrial defect detection method, system and device based on a multi-mode expert system. The unsupervised industrial defect detection method comprises the following steps: acquiring a label-free industrial defect image data set and a text cue word; inputting into a multi-modal expert system to obtain a corresponding prediction result; screening out a preliminary labeling result; filtering the preliminary labeling result based on a prior constraint to obtain a fine label; inputting the industrial defect image data set and the fine label into a self-supervised model for training, and outputting a pseudo label after training is completed; performing double filtering on the pseudo labels by adopting a dynamic threshold value and a prior constraint; inputting the industrial defect image data set and the pseudo label after the current training double filtering into the self-supervised model after the last training for re-training, outputting the pseudo label, and returning to execute the double filtering until the trained self-supervised model is obtained; and predicting an industrial defect image to be detected by adopting the trained self-supervised model to obtain a corresponding industrial defect detection result. And the detection accuracy and efficiency are improved.

The invention provides a single base station error suppression and positioning method based on unsupervised learning, which automatically models to reach error distribution of a phase difference without label data through a deep learning network, and does not need manual labeling. The system extracts signal features, and non-line-of-sight propagation errors are effectively identified and corrected by using unsupervised learning. Compared with a traditional multi-base-station method, high-precision positioning can be achieved in a complex indoor environment only through a single base station. The method combines the feature extraction capability of deep learning and the distribution modeling of unsupervised learning, significantly reduces the deployment cost, improves the positioning precision and robustness, is especially suitable for scenes with significant dynamic and multipath effects, provides an innovative solution for ultra-wideband positioning, and has a wide application prospect.

The invention discloses an underwater in-situ fish label-free sample detection method and equipment. The method comprises the following steps: acquiring an in-situ image shot by a monocular camera of an underwater observation platform, and constructing an image-text structured knowledge base; extracting a training and verification data set of the detection model from the knowledge base, and dividing the data set into a closed set category and an open world category; receiving a natural language text instruction which is input by a user and is used for describing the to-be-detected data set, and generating a text feature vector of a target category; a targeting auxiliary model is obtained through vision-text semantic space training, dynamic interactive fusion is carried out on the visual features of the in-situ image and the text features of the category, and end-to-end joint optimization is carried out on the target model; and performing label-free detection on fishes appearing in the underwater in-situ image by using the optimized target model. According to the method, new species which are not seen during training can be effectively identified, cross-domain reasoning can be implemented on a label-free observation sample, and the retraining and maintenance cost is reduced.

The invention provides a semi-supervised target detection model training method and system based on a hybrid learning framework. The method comprises the following steps: acquiring a teacher model and a student model; obtaining a training set; performing strong enhancement operation on any labeled data, and inputting the labeled data into the student model to obtain a labeled detection result; calculating supervised loss; performing weak enhancement operation and strong enhancement operation on the label-free data, inputting the label-free data into a teacher model for detection to generate a pseudo label, and inputting the label-free data into a student model for detection to obtain a label-free detection result; extracting a pseudo negative label; calculating pseudo label loss and pseudo negative label loss to form unsupervised loss; and iteratively updating parameters of the student model and the teacher model. In the hybrid learning strategy provided by the invention, the positioning accuracy of the model on the target is enhanced by positive learning; the negative learning effectively improves the utilization rate of label-free data, reduces the propagation of noise in pseudo labels, and finally remarkably improves the training efficiency and the accuracy of model detection.

The invention discloses a graph backdoor encoder defense method and device, and a readable storage medium. The method comprises the steps that at least two label-free enhanced data sets are generated based on a preset training data set, a to-be-processed encoder is trained based on the enhanced data sets to obtain a teacher encoder, and the training data set is a downstream data set or a subset of the downstream data set; defining identical attention operators in corresponding layers of the teacher encoder and the to-be-processed encoder, wherein the attention operators are used for outputting layer attention maps of corresponding encoder layers; calculating the distribution offset of the layer attention map of the teacher encoder and the layer attention map of the encoder to be processed in the same encoder layer; and fixing the parameters of the teacher encoder, and updating the parameters of the to-be-processed encoder based on the distribution offset. Compared with the prior art, through lightweight deployment, on the premise that the model precision is not reduced, the attack success rate of backdoor attack on graph self-supervised learning is effectively reduced, and the backdoor defense capability of the system is improved.

Some embodiments can include methods and related systems to project coarse representations of natural language interactions into fine-grained representations using contrastive learning projections. Some embodiments can maximize a first set of distances between dissimilar points and anchor points in the fine-grained representations, and minimize a second set of distances between similar points and the anchor points in the fine-grained representations. Some embodiments can then propagate labels from labeled projections to unlabeled projections based on a similarity metric.

The invention discloses a semi-supervised image annotation method based on a limited label data set, and the method comprises the steps: taking a FixMatch frame as a basis, and integrating a learnable batch normalization module, a dual-scale parallel convolution module, a content and style separation dual-branch module and a dynamic residual gating module in a ResNet backbone network, the stability of feature extraction and the adaptive capacity to enhanced disturbance are improved. For a label-free sample, a multi-level pseudo-label fusion mechanism is provided, prediction distribution of weak, medium and strong enhanced views is synthesized, and high-confidence pseudo-labels are generated through confidence weighted fusion of multi-level enhanced views and comparison and screening with a category threshold. On the basis, a joint loss function composed of label supervision loss and pseudo label consistency loss is constructed, and a plurality of key control parameters in FixMatch + + are adjusted and optimized in a pre-experiment and grid search combined mode to obtain a group of optimal parameters of the model. Finally, a user inputs a label-free image into the trained FixMatch + + model, and the model can automatically generate a high-confidence pseudo label, so that the number of labeled samples in a limited labeled image set is increased, and the classification precision is improved. By implementing the method, the manual annotation cost can be reduced, and efficient and reliable support is provided for image analysis and recognition tasks.

The invention relates to the field of three-dimensional vision, in particular to a label-free three-dimensional point cloud segmentation method based on a visual large model, which comprises the following steps: acquiring three-dimensional point cloud data, and extracting a boundary point set according to the three-dimensional point cloud data in combination with curvature and normal vector to construct a target function; solving the objective function by using an optimization algorithm, selecting an optimal projection visual angle combination, and extracting two-dimensional features of all two-dimensional projection images and three-dimensional features of three-dimensional point cloud data by using a visual large model; performing consistency matching on the two-dimensional features and the three-dimensional features through a matching algorithm to obtain a plurality of matching results; acquiring a definition weight factor and a coverage rate weight factor to calculate an importance weight; fusing the two-dimensional features according to the weight factors to obtain overall fused two-dimensional features; and inversely mapping the integrally fused two-dimensional features to a three-dimensional space to realize three-dimensional point cloud unmarked segmentation. The three-dimensional point cloud segmentation method has the effect of realizing three-dimensional point cloud automatic segmentation without labels.

The invention discloses a gesture posture estimation method based on multiple visual angles and semi-supervised learning, and belongs to the technical field of man-machine interaction, the gesture posture estimation method comprises the following steps: collecting gesture images through multiple visual angles, constructing a label-free data set, labeling a small amount of gesture posture data to form a label data set with hand key points; then randomly selecting one from a plurality of visual angles as input, forming a hidden code through an encoder constructed by a neural network, mapping the hidden code into a gesture image of another visual angle through a decoder to realize consistency constraint of different visual angles, mapping the hidden code into 2D key point coordinates through a projection converter, and finally obtaining a gesture image of the other visual angle; the method comprises the following steps of: randomly inputting a multi-view label-free sample or a small number of label samples for training under the combined action of two constraint mechanisms, and generating 3D key point coordinate estimation of a gesture posture only by inputting a gesture image into an encoder in practical application.

The invention discloses a label-free data deep learning method based on an auto-encoder structure, and the method comprises the steps: obtaining a label-free data set, carrying out the unified normalization preprocessing, and constructing a standardized input tensor; respectively extracting potential variable representation and semantic representation through a first branch network of a variational auto-encoder structure and a second branch network extracted by semantic representation; on the basis of the corresponding relation between the two branch networks, potential space alignment loss is constructed, and the potential space alignment loss, reconstruction loss and semantic comparison loss jointly form a joint loss function, so that joint optimization of the two branch networks is realized. And finally outputting the optimized label-free data feature representation for data input of various downstream tasks. According to the method, the representation quality of the label-free data is effectively improved, the feature extraction and representation capability of the deep learning model in an unsupervised scene is enhanced, and the method has a wide application prospect.

A particulate matter detection device takes holographic images of flowing particulate matter concentrated by a virtual impactor, which selectively slows down and guides larger particles to fly through an imaging window. The flowing particles are illuminated by a pulsed laser diode, casting their inline holograms on a CMOS image sensor in a lens-free mobile imaging device. The illumination contains three short pulses with a negligible shift of the flowing particle within one pulse and triplicate holograms of the same particle are recorded at a single frame revealing different perspectives of each particle. A deep neural network classifies the particles based on the acquired holographic images. The device was tested using different types of pollen and achieved a blind classification accuracy of 92.91%. This mobile and cost-effective device weighs ˜700 g and can be used for label-free sensing and quantification of various bio-aerosols over extended periods.

The invention discloses a data matching method, apparatus and device, and a storage medium. The method comprises the steps of constructing a to-be-discriminated data set and a training data set according to a plurality of sample pairs corresponding to a plurality of to-be-matched data sets; determining a plurality of pseudo label sample pairs and a plurality of to-be-discriminated sample pairs according to the initial discriminator and the to-be-discriminated data set; performing real label labeling on the plurality of to-be-discriminated sample pairs to obtain a plurality of labeled sample pairs, and updating the training data set in combination with a plurality of pseudo label sample pairs; and performing iterative training on the initial discriminator through the updated training data set, obtaining a sample label of each label-free sample pair according to the initial discriminator after the iterative training, and further performing data matching and storage on data samples existing in the data set. By means of the mode, the labeling cost is remarkably reduced, semi-supervised cooperative training is combined, the problem of data sample imbalance is effectively solved, the value of labeling data on model training is improved, and precise matching of complex data attributes is achieved.

The present invention discloses a mobile path planning method, device, robot and storage medium. The method comprises: when the robot triggers navigation resumption, determining the label type of the current label corresponding to the current position of the robot based on the collected image data; the label type includes at least one of: no label, legal label and illegal label; and planning the movement path of the robot based on the label type. By running the technical solution provided by the embodiment of the present invention, the problem of the robot passively leaving the originally planned movement path during autonomous movement, resulting in the interruption of the robot's navigation state and the inability to continue moving, which reduces the robot's movement efficiency and task execution efficiency, can be solved, thereby achieving the beneficial effect of improving the robot's movement success rate and movement efficiency.

In order to make full use of a large amount of label-free face data to cope with the actual problems of slow labeling progress, label shortage and the like in the initial stage of face data set construction, the invention provides a body constitution recognition method based on a label-free face image. Taking the facial constitution data as a target domain for feature extraction to obtain source domain features and target domain features; according to the category center feature vector of the source domain feature sample, respectively determining optimal transmission matching streams of the source domain feature sample and the category center feature vector as well as the target domain feature sample and the category center feature vector; and determining the classification probability of the source domain feature sample according to the optimal transmission matching flow, and identifying the physique according to the classification probability.

The invention discloses a model training method, an information release method and device, equipment and a medium, and belongs to the technical field of computers. The method comprises the steps that a positive sample, a negative sample and a label-free sample are obtained, the positive sample comprises object features of a put object and a positive sample label, the negative sample comprises object features of the put object and a negative sample label, and the label-free sample comprises object features of a non-put object; training a label prediction model based on the positive samples and the negative samples; predicting a pseudo label of the label-free sample through a label prediction model obtained through training; and based on the positive sample, the negative sample, the unlabeled sample and the pseudo label of the unlabeled sample, training a delivery prediction model, the delivery prediction model being used for predicting the delivery score of any object. The training process of the delivery prediction model can globally learn the features of the delivery object and the non-delivery object, so that the accuracy of the delivery prediction model is improved, and the delivery precision can be improved.

The present invention relates to a semi-supervised model generalization method that integrates label denoising, comprising the following steps: 1. Using a pseudo-label generation model to generate pseudo-labels for unlabeled source domain data; 2. Using a dual calibration generalization model to learn on the source domains of pseudo-labels and true labels, selecting clean samples for exchange based on a small-loss strategy, and updating both sides on the selected clean samples. Simultaneously, a style confusion module is inserted to improve the model's generalization capability; 3. In an intermediate domain comprising a labeled source domain and an unlabeled source domain with clean samples, a progressive intermediate domain generation module linearly mixes these two samples, which are then fed into the next cycle as new labeled source domain data for the pseudo-label generation model. This method is beneficial for improving the model's generalization capability.

The application belongs to the field of artificial intelligence model compression and edge deployment in intelligent operation and maintenance and fault diagnosis of power equipment, and discloses a transformer cooperative distillation incremental learning method and system based on time sequence consistency, which comprises the following steps: a teacher model is used to screen a no-label sample set to obtain a pseudo-label sample set, the pseudo-label sample set is combined with an original label sample set to obtain a distillation training sample set; a multi-mechanism cooperative distillation training scheme containing soft and hard label joint distillation, time sequence consistency distillation and multi-task distillation is constructed, and a student model is subjected to distillation training through the multi-mechanism cooperative distillation training scheme; and a cooperative distillation incremental learning scheme in which a cloud end continuously updates a teacher model to adapt to new multi-source monitoring data and a student model after edge distillation generates a pseudo-label to expand the distillation training sample set is constructed. The application can be used for realizing high-precision and low-cost training and continuous updating of a student model in a resource-limited terminal or an online scene.

The invention provides a few-label industrial process quality prediction method based on a DCMSViT model, and relates to the technical field of industrial process monitoring, and the method comprises the steps: carrying out the preprocessing of historical data, and dividing the historical data into a labeled sample set and an unlabeled sample set; then, a dynamic collaborative regression module is utilized to generate pseudo labels for unlabeled samples based on labeled samples, and high-confidence pseudo label samples are screened out through a dynamically calculated confidence threshold, so that a labeled training set is expanded; then, the expanded labeled sample set and the remaining unlabeled sample set are input into a multi-scale visual Transform module together for training, and the multi-scale visual Transform module is used for extracting global feature vectors of the data so as to carry out quality prediction; and finally, the trained DCMSViT model is utilized to realize real-time and online prediction of the key quality variables of the industrial process. According to the method, a small amount of label data and a large amount of label-free data are effectively utilized, and the accuracy of industrial process quality prediction and the model generalization ability in a small-label scene are remarkably improved.

The invention discloses a multivariable time series data anomaly detection method based on a coupling network, and relates to the technical field of adversarial learning. Comprising the following steps: acquiring labeled data samples and label-free data samples, performing label mixing on the labeled data samples according to a preset proportion, combining the label-free data samples according to a specific rule to form a sample data set, preprocessing the sample data set by adopting a data specification and data dimension reduction mode, and obtaining the label-free data set. Generating a training data set according to a certain proportion, constructing a deep learning model, inputting original data in the training data set into the deep learning model to learn data sample features, dynamically adjusting model parameters, obtaining the trained deep learning model, and generating feature data; and comparing the difference between the original data and the feature data as an abnormal score, and judging abnormal data and difference data based on the abnormal score. According to the invention, multivariable time series data anomaly detection based on the coupling network can be realized.

The invention provides a liquid gate graphene transistor antibiotic sensor for realizing real-time detection of bleomycin. According to the method, a probe ssDNA is fixed on the surface of a gate electrode of a graphene transistor through an Au-S bond. Wherein coexistence of Fe (II) and bleomycin can cause formation of a binary compound BLMFe (II), oxygen can be reduced into free radicals at recognition sites, and DNA degradation capacity is achieved. A BLMFe (II) complex is added into an electrolyte, ssDNA and BLMFe (II) react at a recognition site, so that part of ssDNA fragments fall off from the surface of a gate electrode, the double-electrode-layer interface characteristic between a transistor and a sample solution is changed, and Dirac point voltage and current in a graphene channel are changed. According to the graphene transistor antibiotic sensor provided by the invention, the drain voltage is fixed, the grid potential is changed, the lowest detection limit of bleomycin can reach 10 <-18 > M, bleomycin solutions with different concentrations are added, the current of the sensor is changed immediately, and the graphene transistor antibiotic sensor has very high sensitivity.

The invention discloses a camera shooting focusing optimization method based on a self-supervised learning model. The method comprises the following steps: firstly, shooting at least M frames of label-free image data according to preset continuous shooting parameters; extracting label-free image features in each frame of label-free image data; combining the label-free image features into a label-free image feature sequence according to the sequence of obtaining the label-free image data; then, inputting the unlabeled image feature sequence into a preset self-supervised learning model to obtain a focus prediction parameter of a target object in unlabeled image data output by the preset self-supervised learning model; and finally, adjusting a focusing parameter of the camera device according to the focus prediction parameter to obtain a continuous shooting focusing parameter. Therefore, the airborne camera shooting equipment can be enabled to quickly adapt to a new scene under the condition of unmarked continuous shooting samples, the focus prediction capability is optimized in real time, and the focal length of the camera shooting equipment is dynamically adjusted, so that the quick adaptation of camera shooting focusing is realized, and the quick focusing capability of the camera shooting equipment for quickly adapting to different scene types is further improved.

The present disclosure relates to a model training method, an instant pushing method, a device, a medium and an electronic device, and belongs to the technical field of computers, and can realize model training in an end intelligent scene. A model training method based on federated learning comprises: multiple first participants with labels, the same features and different samples, encrypting the label values possessed by the multiple first participants and transmitting the encrypted label values to a second participant without labels, the second participant possessing all the samples of the first participants but having different features from the first participants; the second participant determining a second gradient based on the encrypted label values and feature values of the features possessed by the second participant and transmitting the second gradient to the first participants; the first participants determining a first gradient and a first histogram based on the feature values of the features possessed by the first participants and the label values possessed by the first participants and transmitting the first gradient and the first histogram to the second participant; the first participants determining a first optimal split point based on the second gradient and transmitting the first optimal split point to the second participant; and the second participant determining a global optimal split point based on the first gradient, the first histogram and the first optimal split point and transmitting the global optimal split point to the first participants.

An optical system includes an optical resonant cavity, an optical source, an optical detector, and a signal processing circuit. The optical source is arranged to provide a source beam of light to be at least partially coupled into the optical resonant cavity. The optical detector is arranged to detect light from the source beam of light after the source beam of light has coupled into the optical resonant cavity to provide a detection signal. The signal processing circuit is configured to communicate with the optical detector to receive the detection signal. The optical resonant cavity has a bioconjugatable lipid membrane on a surface thereof. The bioconjugatable lipid membrane is functionalized to capture a specific biomolecule, and the signal processing circuit is further configured to determine a presence of the specific biomolecule when captured by the functionalized bioconjugatable lipid membrane based on processing the detection signal.

The invention relates to a vibration meter error monitoring and calibrating method and system based on data analysis, and the method comprises the steps: obtaining a plurality of paired original vibration signals, and screening and marking credible samples and suspicious samples; calculating a first error value of the credible sample, generating a first training set with a credible label according to the first error value, and training a first error model; predicting the suspicious sample according to the first error model, and generating a pseudo label; calculating the prediction uncertainty of a pseudo label, screening a pseudo label sample, combining the pseudo label sample with the first training set, generating a second training set, and training a second error model; inputting the vibration signal to be calibrated into the second error model to obtain a real-time error; and the real-time error is calibrated. According to the technical scheme, the true label is generated by using the high-credibility sample, a large amount of label-free field data is converted into available training data by means of pseudo label self-training, the model accuracy is improved, and therefore the error monitoring and calibration precision of the vibration meter is improved.

A method, device and system (10) for labeling data, relating to the field of computers, effectively improving the efficiency of labeling data. The method is applied to a computing device (20), comprising: determining a target control parameter based on first labeled data and first unlabeled data; sending the target control parameter to a quantum annealing device (30), the target control parameter being used to instruct the quantum annealing device (30) to perform quantum annealing to obtain a target annealing result, wherein the target annealing result includes the state of quanta in a quantum group (51, 52, 53, 54, 55, 56, 57, 58) corresponding to the first unlabeled data, and the state of quanta in the quantum group (51, 52, 53, 54, 55, 56, 57, 58) corresponding to the first unlabeled data is used to characterize the label of the first unlabeled data.

The invention relates to the technical field of machine learning and classification, in particular to a semi-supervised classification learning method and system based on double reconstruction similarity, and the method comprises the steps: representing feature information through a weighted undirected graph and label information through one-hot coding, and obtaining an initial similarity graph based on the weighted undirected graph; through pairwise typicality, in combination with the initial similarity graph, determining a first reconstruction similar graph; calculating the non-affinity of any node pair to obtain a propagation probability; determining a second reconstruction similarity graph, and converting the second reconstruction similarity graph into a standardized propagation matrix; partitioning the standardized propagation matrix, calculating label predicted values of all label-free samples in the label-free sample set by combining one-hot coding, and determining corresponding label categories; by introducing the pairwise typicality and the propagation probability, the classification precision and the operation speed of the semi-supervised classification task are improved, no extra physical law or statistical law is needed for supporting, and self-adaptive operation can be achieved.

The application discloses a pseudo-label model training method and device, a storage medium and electronic equipment. The method comprises the following steps: obtaining a plurality of sample communication relationship data with added sample labels; using a first classification model to analyze the sample communication relationship data, and generating a first classification label, wherein the first classification model is trained using clean label data; using a second classification model to analyze the sample communication relationship data, and generating a second classification label, wherein the second classification model is trained using mixed data, and the mixed data comprises clean label data, noise label data and unlabeled data; determining that the difference between the sample label and the second classification label is a first difference, and the difference between the first classification label and the second classification label is a second difference; and using a preset loss function model to analyze the first difference and the second difference, and determining a pseudo-label model. The application solves the technical problem that unlabeled data cannot be used for model training.

The present invention provides a face recognition method for missing label data of the user side, including selecting an initial model and determining a sample selection strategy. Supplementary samples are screened out from the unlabeled sample set based on the sample selection strategy to obtain a supplementary sample set. The supplementary sample set is added to the labeled sample set to obtain a first training set. The first training set includes labeled samples and unlabeled samples, and the initial model is trained hybridly using the first training set to obtain a face recognition model. In semi-supervised training, first, the labeled samples are used to learn the mapping relationship between the input features and the output labels, and then the sample types of the unlabeled samples are predicted. The unlabeled samples contain a large amount of unstructured or unannotated data. The present invention combines the labeled samples and the unlabeled samples, which can solve the problem of falling into a local optimal solution when using active learning alone, accelerate the convergence process of the initial model in the case of lack of label data, and thus improve the efficiency and accuracy of training the initial model.