26 results about "Insufficient Sample" patented technology

The invention discloses a composite material reflectivity spectral information classification method and system based on a PCA-SVM algorithm. The method comprises the following steps: collecting reflectivity spectral data of a composite material sample; preprocessing data to eliminate measurement deviation and unify numerical scale; carrying out dimensionality reduction on the preprocessed high-dimensional spectral data through principal component analysis, and extracting feature components retaining main variance information; based on dimension reduction features, a support vector machine is adopted to construct a multi-label classification model according to a'one-to-other 'strategy; predicting the test sample, and generating a multi-label classification result through probability output and threshold processing; and analyzing the classification performance by using the multi-label evaluation index. The data processing module of the system executes preprocessing, dimension reduction, modeling, prediction and evaluation operations. The method is suitable for lossless identification of multi-component composite samples, both interpretability and identification precision are considered, the performance bottleneck of a traditional method under the conditions of feature overlapping, insufficient samples and the like is effectively overcome, and efficient and accurate classification of the composite materials is achieved.

The invention discloses a hydroelectric generating set fault diagnosis data enhancement method based on a diffusion model and generative adversarial training, and belongs to the field of industrial equipment fault diagnosis. According to the method, the data enhancement model combining the diffusion model and the generative adversarial network is constructed, a multi-modal non-Gaussian distribution mechanism and a latent variable control generation process are introduced, the limitation of a single Gaussian hypothesis of a traditional diffusion model is broken through, and richer and more real sample generation is realized. And a condition discriminator and a time sequence modeling mechanism are introduced, so that the training of the model under different noise levels is more stable, and the authenticity judgment capability of the discriminator on the generated sample is enhanced. The problems that an existing hydroelectric generating set fault diagnosis system is insufficient in sample, low in generated sample quality and unreal in sample distribution are solved, and the diversity and quality of data are remarkably improved.

This application relates to the field of equipment performance testing and evaluation technology, and provides a method and system for equivalent reduction of equipment test sample size. By constructing a data fusion evaluation model based on actual and numerical test data, the fidelity of actual test response estimation under small sample conditions is improved. An equivalent measure is designed and calculated by fusing the mean square error of the evaluation model from a single data source with a preset test confidence level, and a box plot of the equivalent measure from a single data source is obtained. By connecting the lower quartile and upper quartile of each obtained box plot, equivalent composite confidence intervals for actual and numerical simulation tests are plotted. Based on these equivalent composite confidence intervals, the theoretically required actual test sample size is equivalently reduced to the numerical simulation test sample size, achieving an equivalent reduction of the actual test sample size by the numerical simulation test, thus solving the problem of equipment performance evaluation under insufficient actual test samples.

The invention belongs to the technical field of artificial intelligence and network predictive maintenance, relates to a fault prediction method based on a multi-source migration echo state network, and aims to solve the problem of low prediction precision caused by insufficient early samples of a fault, and the method comprises the following steps: firstly, extracting migration knowledge from a plurality of historical fault source domains by using the echo state network; secondly, measuring the similarity between the source domains and the target domain and the difference between the source domains by adopting a dynamic time warping distance, and selecting a plurality of similar source domains with complementary information; then, a plurality of prediction sub-models are established in the target domain in combination with migration knowledge of the selected source domain, and a final prediction model is obtained through integrated learning; and finally, performing future value prediction on the key variables acquired in real time by using the model. According to the method, multi-source historical information can be fully utilized under the small sample condition, the fault prediction precision and generalization ability are remarkably improved, and the method is suitable for fault prediction scenes with multi-source time series data in industrial processes, mechanical equipment, power systems and the like.

The application discloses a high-speed moving target recognition method based on data enhancement and belongs to the field of target recognition. The application introduces a high-speed moving target background data set, the confidence of a pseudo target obtained by a classification network after the pseudo target generated by a differentiable generative adversarial network, and the target instances with a confidence greater than a set threshold and the target instances in the initial data set are segmented objects and are enhanced together in the high-speed moving target background data set, solving the problems of insufficient samples and inconsistency between the high-speed moving target training background and the working environment. When the pseudo target generated by the differentiable generative adversarial network is trained, the confidence of the pseudo target and the CIoU value of the positive sample boundary box and the real box are weighted and summed to form a new confidence loss function in the YOLOv7 target detection algorithm, and the improved loss function can more accurately measure the authenticity of the target. The method can realize accurate recognition of a specific type of high-speed moving target under small samples.

The invention discloses an MEMS inertial sensor reliability analysis method, which comprises the following steps: step 1, failure mode and mechanism analysis: carding environmental factors influencing the work of an MEMS inertial sensor, identifying failure modes caused by the environmental factors, and determining internal action mechanisms corresponding to the failure modes; 2, selecting a representative sample, and formulating a sample selection rule according to an application scene, a production batch, a structure type and performance parameter distribution of the sensor; the method has the beneficial effects that the system carding the association relationship between various environmental factors and failure modes, defining the internal mechanism of failure, breaking through the limitation that the existing method only pays attention to a single factor, enabling the analysis result to be more fit with the actual working scene of the sensor, and remarkably improving the comprehensiveness of reliability analysis; representative samples are selected based on scientific sample selection rules, key variables such as production batches, structure types and performance parameters are covered, and the problem of insufficient sample representativeness of an existing method is solved.

The invention provides a degradation data enhancement method based on combination of a conditional encoder and a diffusion model, which is characterized in that a combined network of the conditional encoder and the diffusion model is designed on the basis of a physical priori guidance generation principle, and a physical rule of monotonically decreasing residual service life is embedded into a potential space through a two-stage conditional encoder network. The high-fidelity degradation data conforming to the physical law is efficiently generated by using the space-time condition diffusion model, the problems of data scarcity and unbalanced distribution faced by the existing data-driven prediction method are solved, the generalization ability and robustness of the prediction model in a complex scene are remarkably improved, and the method can be applied to prediction of the residual service life of the equipment and has a good application prospect. The problem that the prediction precision is reduced due to insufficient samples and unbalanced distribution when a traditional residual service life prediction method faces data scarcity and complex working conditions is solved.

This invention discloses a method for rapid training of neural network models in a small-sample environment, comprising the following steps: S1, acquiring a small-sample dataset containing several categories and preprocessing it to obtain a standardized training sample set, wherein the number of samples in each category of the small-sample dataset satisfies the small-sample distribution characteristics; S2, training an initial neural network classifier based on the training sample set, and constructing prototype point-boundary corridor topology maps for each category in the feature latent space. This invention relates to the field of neural network model training technology. This method for rapid training of neural network models in a small-sample environment, by constructing prototype point-boundary corridor topology maps, can clearly understand the relative positions of each category in the feature latent space and the distribution characteristics of inter-class classification boundaries, thereby accurately identifying boundary gap areas with insufficient sample coverage and prone to misclassification, providing a clear target area for subsequent supplementary sampling operations.

The invention relates to the technical field of hydrologic ecology and water resource protection, in particular to a basin physical attribute fused short sequence hydrologic data ecological flow estimation method, and aims to solve the problems of low ecological flow estimation precision and poor regional transportability caused by insufficient statistical samples and lack of physical mechanisms under short sequence hydrologic data. According to the method, a nonlinear exponential function model with monthly median flow and rainwater collecting area as independent variables and ecological flow as a dependent variable is constructed, the rainwater collecting area extracted by a DEM and the confluence speed inverted by a Manning formula are introduced to serve as physical scale parameters, model parameters are optimized through a Lyon method reference value, and high-precision estimation is achieved. The method has the advantages that the estimation precision equivalent to that of a long sequence can be achieved only through short-sequence data, meanwhile, the regional transportability of the model is remarkably improved by introducing drainage basin physical characteristics, and reliable technical support is provided for ecological flow estimation of regions deficient in data.

The invention provides a reservoir prediction large sample data set construction method and system, and relates to the technical field of geophysical exploration and reservoir evaluation, and the method comprises the following steps: obtaining multi-dimensional parameters of reservoir rock samples, and carrying out the preprocessing to obtain a sample parameter data set; screening the rock physical model by adopting a machine learning algorithm to obtain a first rock physical model; performing preliminary optimization on the first rock physical model to obtain a second rock physical model; performing Bayesian inversion based on the sample parameter data set and the second rock physical model to obtain an inversion result; calculating a model prediction deviation based on an inversion result, and iteratively optimizing the second rock physical model to obtain a third rock physical model; and geological constraint conditions are defined, and a reservoir prediction large sample data set is constructed in combination with the sample parameter data set and the third rock physical model. The generated samples are close to actual geological conditions, the availability is high, the problem of insufficient samples is effectively solved, and key support is provided for intelligent evaluation of the reservoir.

The application provides a negative sample enhancement APT attack detection method based on graph structure learning, mainly solving the problem that the existing data set compatibility, edge information utilization and insufficient negative samples of the data set lead to poor detection effect. The scheme comprises the following steps: 1) obtaining a heterogeneous data set and constructing a visual traceability graph; 2) preprocessing the traceability graph, dividing it into snapshots according to the time stamp, and constructing a training set and a test set; 3) designing an encoder and a decoder of the graph neural network, taking the time window, the node feature, the adjacency matrix and the edge feature as the input of the encoder, and generating the reconstructed adjacency matrix through the decoder; 4) designing a loss function, and training the graph neural network through RNN; and 5) inputting the to-be-detected data into the trained model, identifying abnormal attack traffic, and completing the detection. The application can effectively improve the accuracy and robustness of APT attack detection, and can be used for the development and deployment of the APT attack detection and defense system in the network security field.

This invention proposes a small-sample reservoir classification method based on meta-transfer learning, primarily involving artificial intelligence, machine learning, and the petroleum exploration field. The main steps include: designing two case studies based on actual oilfield conditions, utilizing samples from neighboring and more distant blocks respectively to address the problem of insufficient target block samples; designing a masked attention mechanism to avoid interference from non-reservoir elements while learning the relationship between reservoirs and non-reservoir elements, improving the model's feature extraction capability; designing a meta-transfer learning strategy to achieve rapid parameter optimization during transfer learning, improving the model's convergence speed; and designing a value-aware module within meta-transfer learning to enable the model to focus on challenging tasks, learning valuable transfer knowledge and avoiding negative transfer caused by geological differences between blocks. This invention addresses the issue of limited target block data, which hinders model training. By leveraging meta-transfer learning, the reservoir classification performance under two practical case studies is evaluated, and the meta-transfer learning method is optimized, effectively solving the small-sample problem in reservoir classification.

This invention provides a two-stage intelligent analysis method and system for small-sample casting defects. The method includes generating morphological text anchors based on image visual features and generating process cause text anchors based on historical process data; mapping class condition reference anchors to a language embedding space; extracting real visual features and unfreezing low-rank fitting parameters in a large language model to obtain an optimal detection model; performing similarity retrieval between the visual features to be detected and knowledge entries based on the optimal detection model to obtain retrieval results and calculating consistency; if consistent, outputting structured results including defect category, evidence description, and confidence level. This invention effectively solves the problems of insufficient samples of similar defects and easy confusion of similar defects under small-sample conditions.

This invention discloses a SAR target recognition method based on game-theoretic adversarial sample generation. The method involves determining simulated and measured samples, constructing a recurrent generative adversarial network (RGAN) based on an encoder, converter, and decoder, and training the RGAN with the simulated and measured samples to obtain a generator and a discriminator. The simulated samples are then input into the generator and processed to obtain multiple virtual measured samples, forming a virtual measured sample set. The last layer of the discriminator is adjusted, and the virtual measured samples are input into the discriminator to train a target recognition network E. The measured samples are used to fine-tune or correct the target recognition network E to obtain the final target recognition network Q. The target recognition network Q is then used to test validation samples, which are measured data acquired by the radar at a preset or specified time. By utilizing prior target information, this method compensates for the lack of sufficient SAR measured samples, improving SAR target recognition performance under small sample conditions.

The application discloses a kind of dynamic working condition under rotating machinery early fault diagnosis method and system, it is related to intelligent fault diagnosis technical field, it includes: S1, data preparation;2, training DA-WGAN model and carrying out data expansion, it includes: S21, generating adversarial network;S22, introduce the conditional version of GAN model: C-GAN model, and define the objective function of C-GAN, realize the input of class label;S23, use Wasserstein distance as the distance between the measurement of real data distribution and generated data distribution, and satisfy Lipschitz continuity condition using gradient penalty method (GP), define the objective function of GP S24, realize domain adversarial loss function by binary cross-entropy loss function, and define DA-WGAN loss based on domain adversarial loss function and improved objective function, complete the construction of domain recognizer in generation model and the setting of domain adversarial loss.The present application has the effect of improving the problem of insufficient rotating machinery early fault diagnosis label sample.

The invention relates to the technical field of equipment performance test evaluation, and provides an equipment test sample size equivalent conversion method and system. By constructing a digital-real test data fusion evaluation model, the fidelity of real-installation test response estimation under the small sample condition of the real-installation test is improved. The equivalent measure is designed and calculated by fusing the mean square error of the single data source evaluation model and the preset test credibility, and a single data source equivalent measure box diagram is obtained. And drawing a real installation test equivalent composite confidence interval and a digital simulation test equivalent composite confidence interval by respectively connecting a lower quartile and an upper quartile for each obtained box type graph. On the basis of the equivalent composite confidence interval of the real installation test and the equivalent composite confidence interval of the digital simulation test, equivalently converting a real installation test sample size required by a theory into a digital simulation test sample size, so that the equivalent conversion of the digital simulation test to the real installation test sample size is realized; the problem of equipment performance evaluation under the condition of insufficient real installation test samples is solved.

The invention discloses an ANPC inverter fault diagnosis method based on CGAN sample expansion and an adaptive wavelet neural network, and belongs to the technical field of intelligent fault diagnosis of power electronic equipment. Firstly, voltage signals of upper, middle and lower bridge arms of an inverter are collected, energy spectrum entropies are extracted through wavelet packet decomposition to construct 24-dimensional feature vectors, and the 24-dimensional feature vectors are reduced to 6-dimensional feature vectors through kernel principal component analysis (KPCA); aiming at the problem that the diagnosis precision is insufficient under the small sample condition, generating a high-quality fault sample by adopting a CGAN to expand a training set; in the classification stage, AWNN is introduced as a classifier, and a Mexican Hat wavelet basis function and an Adam optimization algorithm are used for training and testing. Experiments show that the method significantly improves the convergence speed and classification precision of the model, and the test accuracy reaches 96.92%. Through the method, the problem of difficult diagnosis caused by insufficient samples can be effectively solved under the condition of not increasing extra hardware cost, and the method has relatively high engineering application value and popularization prospect.

The application provides a target detection method and device, electronic equipment and a readable storage medium. The method comprises: acquiring a current frame point cloud and a previous frame point cloud; using a target detection algorithm based on deep learning to perform target detection on the current frame point cloud to obtain a target detection result; performing clustering operation on the current frame point cloud to obtain a clustering result; tracking a third target of the previous frame point cloud to obtain a tracking result of each third target; matching the tracking result with each first target to determine whether there is a first target belonging to the same object as the tracking result; if there is, outputting information of the tracking result; if there is not, matching the tracking result with each second target to determine whether there is a second target belonging to the same object as the tracking result; if there is, outputting information of the tracking result. The application solves the problem of missed detection of a detection method based on deep learning due to insufficient samples.

The application relates to the technical field of natural language processing, and provides a document question answering method, device and system, electronic equipment and a storage medium. The method uses a coarse ranking model to obtain a plurality of candidate documents in a target document library, uses a fine ranking model to obtain the similarity of each candidate document and a user question, and determines a target document, and further determines a target answer corresponding to the user question. The coarse ranking model and the fine ranking model are based on a question document pair set in a target field, a first difficult negative sample and a second difficult negative sample, and are obtained through multi-round iterative training of a basic coarse ranking model and a basic fine ranking model. The introduction of the difficult negative sample can effectively improve the sample quality, solve the problems of insufficient sample data and difficult labeling in the target field, reduce the difficulty of model training, and further improve the robustness of the coarse ranking model and the fine ranking model obtained through training, and can also improve the accuracy of the target answer.

The invention relates to the technical field of computer vision, in particular to a remote sensing image small sample target detection method based on prototype guide generation, and the method comprises the steps: carrying out the sample balance and enhancement processing of a training data set in a fine tuning stage; extracting multi-scale features of the input image by using a backbone network and a multi-scale feature fusion module, and generating a region-of-interest proposal and corresponding region-of-interest features; based on the region-of-interest proposals and the corresponding region-of-interest features, constructing a dynamic prototype queue; based on the dynamic prototype queue and the teacher-student network architecture, performing label credibility evaluation and correction on the proposals marked as negative samples in the training process; based on a generative adversarial network guided by a prototype condition, new region-of-interest features are synthesized through a trans-attention mechanism, and the new region-of-interest features are added into a training batch to expand a feature space of a new target category. The method aims to solve the target detection limitation of insufficient sample labeling so as to improve the detection precision.

The invention relates to the technical field of multi-modal training, in particular to a multi-modal model training method, an image-text matching method, a device, equipment and a medium. According to the method, the unlabeled samples are constructed on the basis of the unlabeled images and the label texts in the training samples, the number of negative samples needed by multi-modal model training is increased through the unlabeled samples, and under the condition that the number of the unlabeled images is sufficient, sufficient unlabeled samples can be constructed; the multi-modal model is trained on the basis of the unlabeled sample and the training sample, the trained multi-modal model is finally obtained, and compared with the prior art, the problem of insufficient negative samples is solved, and the technical effect of improving the alignment performance of the multi-modal model is achieved.

The invention relates to a welding spot quality detection-oriented data synthesis and labeling optimization method based on element level transformation. The method comprises the following steps of: obtaining a total sample set and a trained welding nugget target detection model; on the basis of the total sample set, double-source collaborative screening is carried out, then a weld nugget target detection model is utilized, labeling optimization is achieved through a diffusion map generation-image fusion collaborative driving method, and the weld nugget target detection model is added into the total sample set; defining full-coverage gradient design parameters of the welding spot defects, constructing a segmentation element set based on the total sample set, and forming a gradient expansion data set by adopting a full-coverage gradient design fusion method; and defining full-coverage composite design parameters of the welding spot defects, and forming a composite defect expansion data set by adopting a full-coverage composite design method based on the total sample set. Compared with the prior art, the method solves the problems that defect samples are natural and scarce, traditional enhancement is limited, boundary labeling subjectively pollutes, samples are insufficient, a model is difficult to learn element-level common rules, composite defects cover blank, and detection confidence is low.

The invention discloses a debris flow susceptibility evaluation method based on kernel transfer learning and safety coefficient constraint, and belongs to the technical field of debris flow prediction and evaluation, and the method comprises the steps: S1, obtaining the environment factor data of a source domain and a target domain, calculating the safety coefficient of a research area through a TRIGRS model, and constructing a negative sample set based on the safety coefficient; s2, constructing a kernel regression model of the debris flow occurrence probability based on a kernel function by using the source domain debris flow sample and the environmental factor; s3, taking a prediction result obtained by inputting a target domain sample into the source domain model as a migration feature, and jointly training a projection model with the existing label of the target domain; and S4, standardizing the safety coefficient calculated by the TRIIGRS, fusing the safety coefficient with the migration model to correct the susceptibility probability, and outputting the regional susceptibility level. According to the method, the geological stability physical index and the kernel transfer learning mechanism are fused, and the model generalization ability and the prediction reliability are remarkably improved under the conditions of insufficient samples and remarkable regional difference.

The application discloses a tool breakage state in-situ real-time monitoring method, and belongs to the technical field of tool breakage state monitoring.The method is based on multi-source information fusion features and multi-algorithm combination, aims to realize efficient and accurate tool breakage state real-time monitoring, and can comprehensively utilize multi-dimensional information in a cutting process by integrating multi-dimensional force signals and multi-dimensional vibration signals, so that the monitoring precision and stability are improved.The application adopts a generative adversarial network to enhance unbalanced samples, effectively solves the problem of insufficient minority class samples in a traditional monitoring method, and improves the generalization ability and judgment accuracy of the model.Deep feature extraction is combined with multi-layer threshold decision and a multi-tooth tool breakage intelligent identification system, so that the accuracy and efficiency of tool breakage type identification are further improved.The method is suitable for various mechanical processing scenes, especially in advanced manufacturing fields with high precision and high stability, and can effectively prevent production accidents caused by tool breakage.

The invention discloses a composite material performance prediction method based on small sample learning, and belongs to the field of composite material performance prediction.The method comprises the steps that original data containing multiple parameters such as calcium carbonate content, stearic acid content, modification temperature and particle size distribution are obtained through an orthogonal experiment; then, an extended data set is generated in an experimental data interval by adopting a cubic spline interpolation technology, so that the problem of insufficient sample size is effectively solved; and finally, training a machine learning model based on the extended data, and establishing a mapping relationship between the process parameters and the maximum bending load.