20results about How to "Solve overfitting" patented technology

The invention discloses a machine vision-based belt scratch and offset monitoring method, which comprises the following steps of: acquiring image data in a belt running process to form a training data set; performing sample generation based on a generative adversarial network algorithm of conjugate gradient feature fusion to obtain an expanded training data set; inputting the expanded training data set into a feature extraction model for training to obtain a training data set after feature extraction; inputting the training data set after feature extraction into a feature dimension reduction model for training to obtain a training data set after dimension reduction; inputting the dimensionality-reduced training data set into a classifier model for training to obtain a trained model; and processing a new sample by using the model trained in the above steps, identifying and monitoring belt scratches and deviation, and outputting identification and monitoring results. According to the method, the weight adjustment process is optimized, so that the model is more suitable for data changes, the over-fitting risk is remarkably reduced, and the generalization ability of the model is improved.

The invention discloses a power distribution network state estimation method fusing a neural network and an attention mechanism, and the method comprises the steps: firstly constructing a power distribution network state estimation model fusing a neural network algorithm and the attention mechanism, then constructing input data for training, and enabling the input data to be processed in two ways: one is directly connected with a conventional DNN, and the other is directly connected with a conventional DNN; in the other mode, a time sequence dimension needs to be added, so that the time sequence dimension becomes a three-dimensional data set to be connected with the LSTM network; and then combining the output results of the two parallel modules into the input of the attention module, and distributing the weight through the method to realize the accurate estimation of the state of the power distribution network.

The invention discloses a pilot fatigue classification method based on an EEG signal, and the method comprises the following steps: carrying out the frequency band decomposition of the EEG signal, and extracting differential entropy features corresponding to different frequency bands; a linear dynamic system (LDS) is used for filtering features to remove components irrelevant to fatigue, a deep ResTaylor Network (DRN) model is constructed, the problems of overfitting and gradient disappearance in high-degree polynomial processing are effectively reduced through a residual connection mechanism, and adaptive adjustment of model parameters is carried out in combination with an Adam optimization algorithm; and training the DRN model based on the training set, and verifying on the test set to obtain a pilot fatigue classification result. Through the pilot fatigue state classification method and device, accurate classification of the pilot fatigue state can be achieved, and compared with the prior art, the pilot fatigue state classification method and device have higher classification accuracy and robustness.

The invention provides a fitting parameter determination method and temperature determination method of electronic equipment and the electronic equipment, training sets are classified, a sub-training set corresponding to each use scene of the electronic equipment is acquired, the sub-training set comprises the actually measured temperature of the electronic equipment and the temperature of a corresponding thermistor in the electronic equipment, and the actually measured temperature of the electronic equipment and the temperature of the thermistor in the electronic equipment are acquired; and then fitting parameters corresponding to each use scene are formed based on the sub-training set corresponding to each use scene, so that the problems of incomplete coverage and over-fitting of the fitting scene can be solved, and the temperature of the electronic equipment can be described more accurately.

The invention discloses a sparse time sequence Bayesian network construction method and system based on power distribution network topological constraints, and the method specifically comprises the steps: constructing a power distribution network topological graph, and calculating an adjacent matrix between nodes and a k-hop neighborhood matrix Nk; based on the power distribution network topology distance information and the matrix Nk, generating a hard constraint rule, and constructing a topology dependence mask matrix M; calculating an electrical influence range of the fault point on surrounding nodes to obtain an electrical influence matrix E; if the electrical influence coefficient of one node on the other node is smaller than a set value, deleting the corresponding dependent edge; introducing a data source reliability matrix R, and performing hard deletion or soft weakening on edges with reliability lower than a threshold value; combining the matrixes M, E and R to synthesize a sparse structure matrix S; and taking the matrix S as a space skeleton, adding a time dimension autoregression edge, and constructing a complete sparse time sequence Bayesian network. According to the invention, by guiding the rarefaction of the network structure, the number of network edges and the number of parameters are effectively reduced, and the trainability and reasoning efficiency of the model are improved.

The application discloses a subject-driven personalized generation method based on decoupled mask prompt attention fine-tuning, and belongs to the field of deep learning, computer vision and artificial intelligence generated content. A mask prompt decoupling module is designed to decompose the unified text prompt into a text prompt containing only the subject and a text prompt containing only the context. A subject attention focusing module and a context attention adjusting module are introduced, and independent subject feature extraction paths and context semantic adaptation paths are established, respectively. The subject identity learning and the context scene modeling are explicitly separated, and the double constraint loss guided by the mask is used for joint optimization to further prevent feature coupling. Finally, the algorithm effectively preserves the fine appearance features of the subject, significantly enhances the adaptability of the model to novel context instructions, and effectively improves the quality of personalized generation and the flexibility of context editing.

The invention discloses a complex scene traffic perception method and system based on fuzzy logic and data enhancement, and relates to the technical field of automatic driving environment perception. The method comprises the steps of obtaining an RGB image and point cloud data, and performing preprocessing operation to obtain a sample set; expanding the sample set by using a forced enhancement mechanism to obtain an expanded sample set; performing exclusive feature enhancement operation on the expanded sample set according to the scene complexity to obtain enhanced features; performing dynamic weight calculation and feature fusion on the enhanced features based on factors influencing traffic perception by using a fuzzy logic algorithm to obtain fusion features; and performing perceptual prediction on the fused features by using a perceptual prediction model to obtain a perceptual prediction result. The method combines forced weather enhancement and fuzzy logic fusion technologies, adapts to complex scenes, and solves the problem that multi-modal sensing robustness and generalization are insufficient in severe weather.

The present invention provides a method and system for evaluating the quality of a light field image without reference based on deep meta-learning, respectively constructing a training set including a quality evaluation task without reference and a test set including a target evaluation task; constructing a deep meta-learning model including a Swin Transformer network structure; based on the training set, learning prior knowledge of distortion to obtain a quality prior model; fine-tuning the quality prior model on the test set to obtain a final light field image quality evaluation model; and using the light field image quality evaluation model to evaluate the quality of the light field image to be evaluated. The present invention overcomes the problem of insufficient combination of spatial and angular information of light field images, which leads to a large difference between experimental results and real results, as well as the fitting problem and inadaptability to various distortions that may be caused by using deep learning, and adapts to each type of distortion, with greatly enhanced performance.

The invention provides an abnormal biomarker prediction method and system based on electrocardiogram information. The abnormal biomarker prediction method comprises the steps that S1, a server receives to-be-analyzed electrocardiogram data sent by a client; s2, preprocessing the electrocardiogram data to be analyzed to obtain preprocessed electrocardiogram data to be analyzed; s3, the server performs biomarker prediction on the preprocessed to-be-analyzed electrocardiogram data by using the trained task self-adaptive convolutional transformer network to obtain a recognition result; and S4, the server transmits the identification result to the terminal device. According to the method, by adopting the convolution-transformer model of the task adaptive strategy, the problem of model overfitting when a small-scale training set is used is solved, and the performance of the predicted AUROC, the positive predicted value (PPV) and the negative predicted value (NPV) of the subject working characteristic curve is remarkably improved.

This invention discloses a spatial resilience assessment method and system for public health risk prevention and control. An automated data acquisition and fusion platform is created to obtain sets of building information, environmental information, and spatial operation and maintenance resilience indicators, generating a unified spatial resilience feature vector. This feature vector is then input into a constructed spatiotemporal attention-based multi-task network model, which outputs a comprehensive resilience index that is adaptively dual-state ("normal-emergency"), thus improving the accuracy of spatial resilience assessment.

This disclosure relates to the field of face recognition technology, and provides a training method and apparatus for a face recognition model that avoids the long tail problem of data. The method includes: constructing a face recognition model; obtaining a training dataset, and executing the following loop to train the face recognition model in multiple rounds: sampling the current round of training from the training dataset using a dynamic sampler to obtain a sample set used for the current round of training; inputting the sample set into a feature extraction network to obtain a feature vector set corresponding to the sample set; inputting the feature vector set into a normalization network to normalize the feature vectors in the feature vector set; calculating a classification loss using a classification network and a contrastive loss using a contrastive network based on the feature vector set processed by the normalization network; updating the model parameters of the face recognition model based on the classification loss and contrastive loss; incrementing the training round number corresponding to the current round of training by one; and ending the loop when the training round number equals a preset round number.

The invention belongs to the technical field of gas cylinder pressure detection, and provides an aviation gas cylinder high-precision pressure detection method based on computer assistance. According to the method, firstly, comprehensive data are collected through monitoring equipment, a high-precision sensor and pressure detection equipment, feature engineering processing is carried out on the comprehensive data, and high-quality feature data are extracted; afterwards, an AE-MSGRU model is adopted for pressure prediction, the model captures short-term, middle-term and long-term dependency relationships of gas cylinder pressure data through a multi-stage GRU layer, key points in the pressure data are accurately recognized in combination with an attention mechanism, and time sequence characteristic data are further optimized; through the above steps, the method effectively improves the precision and reliability of the pressure detection of the aviation weather cylinder, ensures the safety of the gas cylinder under different use conditions, and provides powerful technical support for the pressure detection of the aviation weather cylinder.

The invention discloses an analysis method and system for large-scale multi-type customer order data, and the method comprises the following steps: extracting a remark demand text of the order data, carrying out the data preprocessing, extracting a hidden layer feature vector, decomposing the hidden layer feature vector into a plurality of business semantic features, outputting a decoupling enhancement feature, and carrying out the data preprocessing; calculating the similarity between the intention semantic feature and a preset intention prototype, and outputting an intention activation distribution vector; calculating a matching confidence coefficient for each knowledge domain, and generating a knowledge feature vector; separating a corresponding semantic subspace feature from the decoupling enhancement feature, calculating a gating coefficient with an intention activation distribution vector, performing element-by-element product operation on the gating coefficient and a knowledge feature vector to obtain knowledge representation, and splicing the knowledge representation and the decoupling enhancement feature to obtain a fused semantic feature; and performing intention recognition and entity extraction based on the fused semantic features, and outputting intention probability distribution and a service entity sequence. According to the invention, the automation degree and reliability of customer special demand analysis can be improved.

The invention discloses a resonator inverse design method and a resonator inverse design device based on a residual error automatic encoder method, which adopt lumped electroacoustic analogy to generate training data to train a residual error automatic encoder model and automatically generate accurate training data. The model adopts an auto-encoder architecture, learns low-dimensional representation of data in an unsupervised learning mode, compresses input data into low-dimensional implicit representation and then reconstructs the low-dimensional implicit representation, effectively extracts main features of the data, and meanwhile, the model introduces residual connection on the basis of an auto-encoder to ensure that gradient depends on a chained rule in a back propagation process, so that the accuracy of back propagation is improved. And transmission can be carried out through jump connection, so that the problems of gradient disappearance and explosion are effectively relieved. The model is subjected to operator fusion, perceptual training is quantified, the size of the model is compressed, and then a decoder part is cut off, so that the problem that the existing model is easy to fall into local optimum and over-fitting is solved, meanwhile, the calculation efficiency is greatly improved, the size of the model is compressed, flexible deployment can be carried out at multiple terminals with different performances, and the hardware performance is fully exerted.

The embodiment of the present invention proposes a smart meter fault classification method based on global sample rebalancing, including: taking fault history data of smart meters under different categories as input data sets, and dividing them into multiple two-category data sets; for each two-category data set, constructing a fusion model of VAE and GAN, taking each sample as the input of the model, and dividing its latent code into important feature code and secondary feature code; obtaining its variant latent code through latent code reconstruction technology, and generating multiple reliable similar variant samples considering important features of input samples through decoder restoration, mutual information constraint and discriminator confrontation; designing a feature repulsion technology acting on the latent codes of two types of samples to perform supervised feature representation learning; superimposing the reconstruction errors of each dimension of the sample through the hybrid coding technology as a supplement to the important feature code, and judging the classification result of the sample to be tested in each two-category data set accordingly, and obtaining its fault category through hard voting.

The application discloses a medical document image classification method based on multi-modal fusion, relates to the technical field of medical information processing and artificial intelligence, and realizes automatic recognition and classification of nine types of documents; single-label annotation and consistency review of medical document images are performed based on annotation software; standardization preprocessing is performed on the images, an OCR engine is called to extract text to form an image mode and a text mode; a document image Transformer classification model and a Chinese pre-training language model classifier are trained respectively; under the same hierarchical cross-validation division, two types of features are extracted and late fusion training is performed to obtain a fusion classifier; single, batch and directory level inputs are supported in the reasoning stage, and the output category and confidence are output, and difficult samples can be located through cross-modal divergence analysis to guide data iteration; the application has higher robustness and explainability under long-tail category distribution, and is suitable for medical document automatic archiving, quality control auditing and data governance and the like.

The invention provides an electrocardiogram diagnosis model personalization method and system based on differentiated knowledge distillation, and the method comprises the steps: S1, constructing a differentiated data set according to a student model and a teacher model, and carrying out the random sampling of the data set and the replacement of individual electrocardiogram data; s2, performing individualized knowledge extraction according to the differentiated data set, wherein the knowledge extraction comprises a knowledge distillation process, a temperature coefficient and a soft label; and S3, carrying out classification loss calculation and total loss function design. According to the method, through a differentiation knowledge distillation method, the problems of model overfitting and performance reduction caused by sample scarcity in the individualization process of an existing ECG diagnosis model are solved.

The application provides a GSA-based optimization neural network representation method for evaluating biological cell mechanical characteristics, and belongs to the fields of micro-nano manipulation technology and cell biomechanics. The method obtains the surface mechanical characteristic parameters of biological cells under physiological conditions by using a nanoindentation technology, establishes a single-factor multi-level cell biomechanics sample set by using a factorial design method, and carries out neural network optimization on the sample set to inversely calculate a cell mechanics model. The neural network clustering analysis of the mechanical data of different types of cells in different regions solves the difficulty that the surface mechanical parameters in the cell mechanical characteristics cannot be subjected to clustering analysis. The network weight values of the neural network are optimized by combining a simulated annealing algorithm with a genetic algorithm, so that the problems of being easily trapped in a local minimum and a convergence speed in the weight value optimization process are solved. The optimization neural network is used for modeling, so that the cell mechanics evaluation result has good generalization ability and conforms to the biological cell mechanics law.

The application discloses a smart contract vulnerability detection method based on multi-view learning, and the method obtains three representation modes of a smart contract source code, an abstract syntax tree, a control flow graph and a data flow graph through static analysis of the smart contract; noise codes outside called external functions and variable positions are pruned for different representation modes, and features of the noise codes are obtained; abstract syntax tree features are learned through an extended recurrent neural network, and control flow graph and data flow graph features are learned through a graph attention network; and finally, features obtained through fusion of the three kinds of features are used to detect smart contract vulnerabilities. The smart contract vulnerability detection method based on multi-view learning can more comprehensively capture indicative features of vulnerabilities from codes, simplify redundant noise, and improve the performance and effect of smart contract vulnerability detection.

The present invention relates to the field of machine learning and computer vision technology, and in particular to an intelligent detection system for the sand spreading state of a locomotive in a maintenance yard, comprising a camera, a license plate recognition device, an LED display screen, an IP audio system, a switch, an optical fiber transceiver, a hard disk recorder, and an intelligent analysis server. The intelligent analysis server trains the YOLOV5 convolutional neural network provided in the intelligent analysis server through the image information collected by the camera and the license plate recognition device, and uses the trained convolutional neural network to identify the degree of sand spreading of the locomotive to be detected. The convolutional neural network can learn the feature representation of sand spreading images in different states, so as to accurately classify and identify the degree of sand spreading of the locomotive, and the system equipment only needs to be installed at a fixed position in the maintenance yard, and does not need to be installed for each locomotive, thereby reducing hardware equipment and installation costs.