1380 results about "Singular value decomposition" patented technology

The invention relates to the technical field of image data processing, and discloses a few-sample scene-oriented industrial product surface defect image analysis method, which comprises the following steps of: obtaining surface gray level image data of a product to be analyzed, calculating a structure tensor matrix and generating an anisotropy degree graph; searching similar blocks in a preset search neighborhood, and constructing a local texture data matrix; performing singular value decomposition on the local texture data matrix to extract a main subspace; constructing a projection operator and utilizing the projection operator to carry out orthogonal projection reconstruction on the local image block to generate a reconstructed background image block; according to the method, through an orthogonal subspace projection mechanism, good product textures and defect signals are separated, random noise is removed, meanwhile, high-frequency structural features are completely reserved, and the method is high in robustness, high in robustness and high in robustness. And the defect detection precision of a complex texture surface in a few-sample scene is improved.

The invention discloses an MURA defect detection method based on background reconstruction and related equipment, and relates to the field of image processing. The method comprises the following steps: acquiring an initial image displayed by a display panel; performing discrete cosine transform processing on the initial image to extract low-frequency background components; denoising the low-frequency background components based on a singular value decomposition algorithm to obtain main background features; performing fitting reconstruction on the main background features through a polynomial surface fitting algorithm to obtain a background reconstruction image; performing contrast enhancement processing on the background reconstruction image to obtain a background enhancement image; subtracting the background enhancement image from the initial image to obtain a difference image; determining an MURA defect area to be detected according to the difference image; and performing identification processing on the MURA defect to-be-detected area through a preset MURA defect detection model to obtain an MURA defect detection result of the display panel. According to the technical scheme, the MURA defect detection precision can be improved.

The invention belongs to the technical field of computer vision, and discloses a sparse point cloud completion method based on multi-view fusion, which comprises the following steps: (1) inputting point cloud data, extracting point cloud features, and obtaining an overlapping region of the point cloud data; (2) calculating a similarity matrix of inter-frame point clouds in the overlapping region, and solving inter-frame rigid transformation by using a weighted singular value decomposition algorithm; (3) realizing accurate registration of multi-view point clouds through an accumulative registration strategy; (4) performing coarse-grained completion on the point cloud by using a 3D convolutional mesh residual network module, and further fusing 3D voxel features by using a multi-layer perceptron and completing fine-grained accurate point cloud completion; and (5) optimizing the weight of the model by calculating a loss function, and improving the overall effect and precision of completion. According to the invention, the accuracy and robustness of vehicle detection are improved by fusing a multi-view registration technology and a coarse-to-fine modular point cloud completion method.

The invention relates to the technical field of data processing, and discloses an automatic noise monitoring system and method based on multi-sensor data fusion. According to the system, a calibration module carries out time synchronization processing on noise data collected by multiple sensors; the extraction module adopts a singular value decomposition algorithm to extract frequency domain and time domain features; the separation module analyzes and separates traffic, construction and industrial noise sources through attention independent components; the construction module generates noise space propagation characteristics in combination with geographic information data; the classification module identifies the type of a noise source through a space-time convolutional network, locates coordinates and allocates responsibility weight. The technical problem that the existing noise monitoring technology cannot realize multi-source noise intelligent identification and pollution source accurate traceability is solved.

The invention relates to the technical field of audio processing, and discloses an audio processing method and system for speech spectrum reconstruction, and the method comprises the steps: carrying out the preprocessing of an original speech signal collected by a vehicle-mounted microphone, and obtaining a speech signal time-frequency matrix; performing singular value decomposition on the voice signal time-frequency matrix to obtain a sparse matrix representing voice components and a low-rank matrix representing noise components, and establishing a frequency band characteristic incidence matrix through a mutual information entropy calculation method; inputting the sparse matrix into a selection forgetting extreme learning machine for voice feature prediction to obtain predicted voice feature data; and based on the predicted voice feature data and the frequency band feature incidence matrix, performing voice probability calculation and feature interpolation on each time frequency point by adopting a logarithmic frequency spectrum amplitude calculation method to obtain reconstructed frequency spectrum data, thereby improving the accuracy of voice recognition and meeting the requirement of real-time processing in a vehicle-mounted environment at the same time.

The invention belongs to the technical field of image generation, and particularly discloses a multi-modal scene fusion method based on a large diffusion model. Firstly, singular value decomposition is carried out on a weight matrix of a diffusion model backbone network to update the weight matrix of the backbone network; then, inputting the text cue word into the diffusion model to generate a text condition image; performing VAE decoding and preprocessing on the text condition image to obtain a preprocessed text condition image; inputting the preprocessed text condition image into a pre-trained visual classifier to predict the category, and obtaining a prediction label; updating loss is calculated according to the predicted label and the real label, parameters of the diffusion model are updated, and fine adjustment of the diffusion model is achieved; and finally, generating a scene fusion image by using the fine-tuned diffusion model. Scene fusion is divided into two stages, in the first stage, a combined noise image is generated by sampling a background image and a foreground image, and in the second stage, a scene fusion image is generated through iteration of the combined noise image. According to the method, image generation is guided through the image, seamless fusion of the foreground image and the background image is achieved, meanwhile, image generation is guided through control information, and controllability of scene fusion is enhanced.

The invention discloses a singular value-based pre-trained visual model parameter fine tuning method, which comprises the following steps of: for a pre-trained visual model, selecting levels which need to be subjected to parameter fine tuning from the visual model according to actual requirements, and respectively performing singular value decomposition on a weight matrix of each layer to be subjected to fine tuning; obtaining an optimal rank-r approximation matrix and a residual matrix used for reconstructing a weight matrix, forming a singular value vector by the first r singular values as a trainable parameter, freezing other model parameters, inputting an image sample in a training sample set into a visual model to update the trainable parameter, and obtaining an optimal rank-r approximation matrix and a residual matrix used for reconstructing the weight matrix; and reconstructing by adopting the updated trainable parameters to obtain a weight matrix. According to the method, singular value decomposition is adopted to initialize the low-rank matrix, learning of downstream tasks on new domain knowledge is enhanced while pre-training information is reserved, the number of trainable parameters is remarkably reduced, the knowledge and adaptability of the model are effectively reserved, and the fine adjustment effect of the model is improved.

The invention relates to the technical field of channel estimation, and discloses an OTFS sparse channel estimation method based on unitary transformation and sparse Bayesian, which comprises the following steps: S1, forming a 2D data block in a time delay Doppler domain, and dividing the time delay Doppler domain; s2, acquiring a receiving signal of a time delay Doppler domain; s3, constructing a sparse sensing matrix and performing singular value decomposition; s4, setting a judgment condition for successful channel recovery; s5, calculating posterior distribution of a transmission channel by adopting a generalized approximate message passing strategy based on unitary transformation; s6, obtaining an optimal updating rule of the hyper-parameter in the estimation channel and the hyper-parameter in the noise signal; and S7, judging whether iteration is continued or not according to a set judgment condition. According to the invention, the approximation of the channel posterior distribution information is completed by using the generalized approximation message passing framework based on unitary transformation through the information carried by the pilot frequency position, so that the calculation burden is reduced, and the calculation resource is saved.

The invention discloses a global ultra-precision strain measurement method, and relates to the technical field of strain measurement. The method comprises the following steps: continuously shooting moving images of a target area by using a high-frame-rate high-definition camera, and recording accurate time; and carrying out image correction, feature extraction and tracking and data normalization. And after Gaussian white noise is added, singular value decomposition and reconstruction are carried out. And a Lucas-Kanade optical flow method is adopted to calculate the displacement of the feature points in the reconstruction matrix. And a displacement smoothing stage: smoothing the displacement data by using an LOESS method. And carrying out global strain calculation by adopting a window sliding method, and calculating the strain through the spatial derivative of the displacement field. According to the method, the quantization effect can be relieved, the continuity of displacement calculation is improved, low-energy noise and irrelevant features are removed, main motion information is reserved, meanwhile, smoothness is enhanced, local features are captured in a high-precision mode, and measurement sensitivity is improved. The method is suitable for strain measurement of large-scale structures such as bridges and buildings.

The invention provides a civil construction monitoring intelligent sensor information detection method and system under signal mismatch, and the method comprises the steps: carrying out the time-frequency analysis of temperature and strain fundamental frequency components, selecting a wavelet basis function adaptive to the concrete setting and strength increasing process, carrying out the self-adaptive threshold processing of a wavelet coefficient, removing high-frequency noise, and carrying out the detection of the temperature and strain fundamental frequency components. Low-frequency signal characteristics reflecting concrete strength increase are reserved; performing feature extraction on the denoised temperature and strain signals by adopting a singular value decomposition algorithm, detecting internal defects of the concrete through abrupt change of a singular value spectrum, and if crack and hole defects exist, causing strain signal abrupt change by stress concentration; if it is detected that defects exist in the concrete, that is, spatial heterogeneity of temperature and stress distribution exists, temperature and strain monitoring data of multi-point arrangement and a defect detection result serve as boundary conditions and constraints, and a temperature field and a stress field in the whole concrete structure are obtained through numerical simulation.

The invention discloses a large language model training method and system based on elastic federal low-rank adaptive fine tuning, which adopts federal aggregation based on singular value decomposition (SVD) to reconstruct a local LoRA module of a client into a local increment matrix, then aggregates the local increment matrix into a global increment matrix, and generates a LoRA module of the next round through SVD, thereby effectively eliminating aggregation errors of traditional FedLoRA. According to the method, the LoRA rank is dynamically distributed for each weight matrix, and the balance between statistics and hardware efficiency is realized. In order to further reduce communication overhead and maintain model precision, a self-adaptive matrix freezing mechanism is introduced, and whether freezing is carried out or not is judged according to training stability of a weight matrix. According to the method, challenges of traditional FedLoRA in the aspects of training efficiency, communication overhead, model precision and the like are solved.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for fine-tuning diffusion-based generative neural networks in compact parameter spaces for text-to-image generation. In one aspect, a method performed by one or more computers for fine-tuning a diffusion-based generative neural network to obtain a fine-tuned version of the diffusion-based generative neural network is described. The method includes: for each of a number of neural network layers of the diffusion-based generative neural network: obtaining an initial weight matrix including a number of pre-trained weights parametrizing the neural network layer: performing a singular value decomposition on the initial weight matrix; and re-parametrizing the neural network layer with new weights that depend on spectral sifts; and training the spectral shifts of each of the number of neural network layers of the diffusion-based generative neural network to obtain the fine-tuned version of the diffusion-based generative neural network.

The invention discloses a large language model progressive field fine tuning and knowledge fusion method for shield engineering. The method comprises the following steps: constructing a layered shield training course containing a wide-area academic theory and a proprietary enterprise construction method; parallelly training a plurality of physically isolated parameter efficient adapters based on the frozen base; performing singular value decomposition on the adapter, extracting a geometric feature subspace representing knowledge distribution, and calculating a conflict correlation degree; based on this, a uniform adaptation mechanism of resource awareness is constructed. The mechanism not only can generate a static fusion model for conflict removal, but also can dynamically activate a specific rank slice of an adapter through a routing network based on real-time hardware resource budget (video memory / FLOPs) and geometry-resource signature. According to the method, multi-source knowledge is reserved, and adaptive dynamic scheduling of edge hardware resources by model reasoning is realized.

The invention provides an HPLC (High Performance Liquid Chromatography) dual-mode communication stability method, medium and system, and belongs to the technical field of dual-channel multiplexing communication, and the method comprises the following steps: firstly, collecting HPLC low-frequency / high-frequency carrier signals, working voltage / current, temperature and electromagnetic interference data in real time; stable, variable and noise components of a carrier signal are respectively obtained through empirical mode decomposition, and a correlation coefficient matrix of the variable components and interference factors is established by using multiple linear regression and singular value decomposition. And calculating a signal intensity, a signal-to-noise ratio and an error rate evaluation matrix based on the stability coefficient matrix, and constructing an abnormal feature matrix in combination with a communication protocol specification. And finally, inputting the features into a five-layer full-connection neural network, outputting fault type probability distribution through a softmax function, and taking the type corresponding to the maximum probability value as a diagnosis result. According to the invention, the technical problem of rapid and accurate diagnosis of HPLC communication stability in a complex and changeable low-voltage distribution network environment is solved.

The invention relates to the technical field of high-voltage switch cabinet partial discharge detection, and discloses a high-voltage switch cabinet partial discharge signal denoising method based on singular value decomposition and particle swarm optimization wavelet decomposition, and the method comprises the steps: firstly analyzing the characteristic parameters of a narrowband interference signal in a partial discharge signal, and extracting a partial discharge signal without narrowband interference; secondly, for the problem of white noise existing in the partial discharge signals, wavelet decomposition is carried out on the partial discharge signals after narrowband interference is eliminated, the partial discharge signals are converted into sub-band signals of different scales and frequencies, wavelet inverse transformation reconstruction is carried out after threshold processing, and denoised signals are obtained; and finally, optimizing the selection of threshold related parameters by using a particle swarm algorithm, solving the number of particles and a population optimal value by using a fitness function, and continuously iterating to finally obtain the optimal threshold related parameters. According to the method, the denoising effect is improved, and the method can be popularized to denoising of partial discharge signals of other devices in combination with performance characteristic parameters of different devices.

The invention discloses a circuit breaker fault detection optimization method and system based on SVD-XGBoost. The method comprises the steps of obtaining an original vibration signal generated when a circuit breaker is in an operation state; performing data enhancement on the original vibration signal based on an NLMS algorithm so as to adaptively filter noise, and obtaining a time domain signal after data enhancement; constructing an SVD-XGBoost model based on combination of SVD (singular value decomposition) and an XGBoost algorithm, converting an enhanced time domain signal into a frequency domain signal through fast Fourier transform, performing data dimension reduction on the frequency domain signal by adopting singular value decomposition, and extracting target features; according to the method, the extracted target features are input to the trained SVD-XGBoost model, the optimized network is adopted to carry out fault classification on the circuit breaker vibration signals, a circuit breaker fault identification result is obtained, SVD dimension reduction is introduced to convert high-dimensional features into low-dimensional features, the number of the features is reduced, the calculation efficiency is improved, and the fault classification precision and effect are improved.

The invention discloses a natural language semantic recognition method, device and equipment and a medium, and the method comprises the steps: extracting semantic features according to an input natural language, inputting the semantic features into an initial hybrid expert model, and analyzing the semantic features through a Transform model to obtain a hidden state matrix; inputting the hidden state matrix into a gating network to generate a routing weight matrix; on the basis of rank estimation of randomized singular value decomposition, generating the number K of expert networks activated by the current layer in combination with the hidden state matrix; selecting first K expert networks according to a sorting result of the routing weight matrix, generating a mask matrix, executing sparse forward calculation, and optimizing model parameters of the expert networks based on sparse gradient propagation and load balancing loss to obtain a target hybrid expert model; and the target hybrid expert model is adopted to recognize the semantic features to obtain a semantic recognition result, and the semantic recognition accuracy and efficiency are improved by adopting the semantic recognition method and device.

The invention provides a distributed photovoltaic access distribution network voltage fluctuation evaluation method, medium and system, and belongs to the technical field of distributed photovoltaic, and the method comprises the steps: firstly collecting the parameters of a photovoltaic system and a distribution network, and building a power grid response equation set considering the parameters; more parameter samples are obtained through data augmentation and expansion, and indexes such as voltage, power and harmonic waves of each node are calculated according to the response equation set. Constructing a voltage fluctuation evaluation index matrix including voltage deviation, fluctuation, harmonic waves, short-circuit capacity change and the like; and decomposing the matrix by adopting singular value decomposition to obtain a voltage stability and fluctuation evaluation matrix, and finally calculating a voltage stability index and a fluctuation index. If the voltage fluctuation evaluation index exceeds a preset threshold value, an alarm is triggered, and decision support is provided for distribution network planning and operation. According to the invention, the technical problems of short evaluation time scale and difficulty in capturing the dynamic characteristics of voltage fluctuation in the evaluation of the distributed photovoltaic access distribution network in the prior art are solved.

Provided is a method for improved hand-eye calibration method based on a structured light camera. The method includes: step 1: establishing a pinhole camera model, and using a depth camera to detect a three-dimensional (3D) coordinate to obtain a physical coordinate of each point in an image coordinate system with a known depth relative to a camera coordinate system; step 2: establishing a Denavit-Hartenberg (DH) model of a robotic arm, and moving the robotic arm to a determined coordinate using inverse kinematics; step 3: collecting n sets of point cloud data, applying depth scaling coefficients to the n sets of point cloud data to perform Singular Value Decomposition (SVD), and solving for an optimal depth scaling coefficient using a Nelder-Mead algorithm; and step 4: completing the hand-eye calibration of the robotic arm based on the solved optimal depth scaling coefficient. The method offers strong operability and robustness.

The invention belongs to the field of artificial intelligence and deep learning, and provides a small sample continuous learning method and system based on continuous knowledge protection decomposition in order to solve the problems of imbalance and poor adaptability of new and old knowledge. The small sample continuous learning method based on continuous knowledge protection decomposition comprises the following steps: based on playback sample data of each category and a continuous learning model backbone network, obtaining knowledge sensitive components and redundant capacity components through singular value decomposition; in the incremental adaptation training process, pre-training linear weight matrixes corresponding to the knowledge sensitive components are frozen; constructing a learnable adapter by using the redundant capacity component, and updating the pre-training linear weight matrix; and obtaining small sample playback data again, and carrying out singular value decomposition and incremental adaptation training operation again based on the updated pre-training linear weight matrix. Through dynamic updating of the covariance matrix, balance and efficient adaptation of new and old knowledge are realized, and the accuracy of a classification result is improved.

The invention relates to the technical field of ramie seed vigor evaluation methods, in particular to a ramie seed vigor evaluation method and system based on multispectral image analysis, and the method comprises the following steps: collecting a multispectral image and a photoautotrophic fluorescence image of a ramie seed to be detected; constructing a feature matrix of the multispectral fluorescence image, extracting a feature clustering center based on a feature clustering decomposition algorithm, and forming spectral feature data of the ramie seeds; carrying out dimensionality reduction on the feature matrix based on singular value decomposition to generate spectral principal components of the ramie seeds; projecting a spectral principal component to a low-dimensional feature matrix based on nonlinear embedded mapping; generating a spectral index matrix based on the low-dimensional feature matrix; and inputting the spectral index matrix into a depth inversion prediction model to obtain the vigor index of the ramie seeds, and effectively solving the problems of information loss and overlapping in the dimension reduction process in the traditional technology through the feature clustering decomposition and singular value decomposition technology in the processing stage of the multispectral image data.

The invention relates to the technical field of artificial intelligence, and discloses a large model continuous learning method and device based on low-rank adaptation, equipment and a medium, which are applied to a scene that a financial institution updates a credit scoring model of a user in real time, and the method comprises the following steps: when a basic large model continuous learning task is received, obtaining an initial task text, constructing training data based on the initial task text; performing singular value decomposition on the parameters of the original low-rank matrix to generate principal component parameters and residual parameters; freezing the principal component parameters, and updating the updated original low-rank matrix by adopting a singular value loss function and an orthogonal loss function based on the residual parameters of the training data to generate a basic low-rank matrix; performing iterative training on the basic large model based on the training data and the basic low-rank matrix to generate a target large language model; and predicting the to-be-predicted task data through the target large language model. According to the method, the problem of disastrous forgetting is relieved, and computing resource consumption is reduced.

The invention provides a fault diagnosis intelligent analysis method and system for industrial equipment, and a storage medium. The method comprises the following steps: segmenting monitoring time sequence data into at least one sub-sequence based on statistical characteristics and working conditions of the monitoring time sequence data; determining the length of an embedded window based on the local characteristics of the subsequences, constructing a trajectory matrix by using the length of the embedded window, and performing singular value decomposition on the trajectory matrix to obtain a primitive matrix and a corresponding singular value; forming candidate groups for the primitive matrixes of different preset combinations, calculating a separability metric value of each candidate group, and determining a component group based on the separability metric values and the singular values; performing diagonal averaging reconstruction on the selected component group to obtain signal components, and calculating preset characteristic parameters of each signal component; and according to the preset feature parameters, target signal components related to the preset fault mode are screened out, feature vectors representing the fault state are extracted from the target signal components, the feature vectors are input into a pre-configured fault diagnosis model, and fault diagnosis information is output.

The invention relates to the technical field of numerical control programming, in particular to a five-axis machining path planning method and system based on data driving, and the method comprises the following steps: obtaining real-time coordinates of each axis of a machine tool, calculating linear velocity and angular velocity components to construct a Jacobian matrix, executing singular value decomposition, and calculating a conditional number ratio by using maximum and minimum singular values; and inputting a nonlinear mapping function to calculate a dynamic penalty factor, generating a rotating shaft weighted item in combination with a rotating shaft identifier, constructing a weighted damping least square objective function, calculating a five-axis motion increment, and accumulating the five-axis motion increment with a real-time coordinate to generate a target absolute position coordinate. According to the method, the pose singularity degree is quantified by monitoring the machine tool pose condition number ratio and converted into the dynamic penalty factor to apply the self-adaptive constraint to the rotating shaft, the severe sudden change of the rotating shaft in the singularity area is inhibited, the tool nose track following error is minimized, and meanwhile smooth distribution of the motion increment is achieved; and the dynamic stability and the surface quality of five-axis linkage machining are improved.

The invention discloses a method for identifying aerodynamic parameters of an adaptive volume Kalman filter aircraft, which is used for improving a traditional Kalman filter algorithm so as to solve the common problem of negative determination of a covariance matrix in practical application. Specifically, singular value decomposition is adopted to replace traditional Cholesky decomposition to decompose a covariance matrix, and the improvement can improve the stability and precision of the algorithm, so that the aerodynamic parameters of the missile are more reliably estimated; in the filtering process, the noise variance is corrected according to real-time measurement information, so that the state estimation error is reduced; according to the method, the adaptive capacity to the dynamic change of the system can be improved, and the precision and robustness of parameter estimation are further improved.

The invention discloses a high-precision particle image velocity measurement method, and relates to the technical field of particle image velocity measurement, and the method specifically comprises the steps: 1, building a high-precision particle image measurement experiment platform, capturing continuous particle images through a laser-camera system with a synchronous time sequence, achieving the precise synchronization of laser pulses and camera exposure through an FPGA synchronous controller, and carrying out the measurement of the continuous particle images; the method comprises the following steps of: 1, acquiring a displacement image captured by a high-speed camera, and verifying synchronization precision, 2, processing the displacement image captured by the high-speed camera by using a technology combining a cross-correlation method and an optical flow method, and 3, introducing singular value decomposition in a PIV data processing process, decomposing flow field data, and separating main signals and noise. The sub-pixel-level precision of particle displacement measurement is remarkably improved, the signal-to-noise ratio of data is enhanced, and real-time error correction is achieved.

The invention discloses a high-precision dynamic calibration system and calibration method for an ultrasonic partial discharge instrument, and relates to the technical field of dynamic calibration of the ultrasonic partial discharge instrument, and the system comprises a collection and enhancement module which is used for deploying an ultrasonic sensor in a target area to collect a partial discharge signal and carrying out the denoising and enhancement of the partial discharge signal; the feature extraction module is used for extracting features according to the denoised and enhanced signals; the matrix and calculation module is used for constructing a space-time incidence matrix based on the extracted features and dynamically calculating node weights; the error calibration and verification module is used for constructing an error model according to the node weight to carry out error calibration and verification; according to the method, the anti-noise capability is remarkably enhanced through a denoising method combining sparse decomposition and a sparse dictionary matrix, global and local features of the signals are extracted through Henkel matrix construction and singular value decomposition, and high-quality input is provided for the follow-up process through optimization reconstruction and standardized fusion of the signals; and high-precision calibration of the partial discharge signal in a complex environment is realized by combining the dynamically calculated node weight and the error calibration model.

The invention relates to a large-scale multi-view tensor clustering method, device and equipment based on a linear kernel. The method comprises the following steps: acquiring multi-view data and a linear kernel matrix corresponding to each view data, and aligning the linear kernel matrixes to generate corresponding potential data representation; stacking the potential data representation corresponding to each view data into a three-dimensional tensor, and implementing tensor singular value decomposition-based nuclear norm minimization constraint on the three-dimensional tensor to construct a target function; and solving the target function by adopting a preset linear optimization strategy to obtain an optimal potential data representation of each view data, splicing the optimal potential data representations of the multi-view data to obtain a clustering representation, and clustering the clustering representation to obtain a clustering result of the multi-view data. By adopting the method, the accuracy of multi-view clustering can be improved on the premise of keeping an unsupervised learning framework.

The invention provides a federated learning backdoor defense method based on singular value decomposition and model weight amplification, and the method comprises the steps: carrying out the normalization processing of a model updating parameter locally trained by a client, and obtaining a normalized model updating parameter; on the basis of the normalized model updating parameters, model updating parameters after dimension reduction are obtained; performing clustering algorithm processing on the model updating parameters after dimension reduction to obtain clustered clusters; obtaining cluster model parameters based on the clustered clusters; combining the cluster model parameters into a cluster model parameter matrix; performing singular value decomposition on the cluster model parameter matrix to obtain a singular vector; obtaining a trust score through the singular vector; obtaining global model update parameters based on the trust score; calculating by utilizing the global model updating parameters to obtain a global model; and adding Gaussian noise to the global model through a differential privacy mechanism to obtain a final global model. According to the method, backdoor attacks can still be effectively resisted under the scene that the client data sets are non-independent and identically distributed.

The invention provides a production line product quality tendency defect determination method, medium and system, and belongs to the technical field of production digital data processing.The production line product quality tendency defect determination method comprises the steps that firstly, quality detection data is collected to construct a feature matrix, and a time sequence feature vector is obtained through singular value decomposition; and calculating a defect tendency index, and establishing association mapping in combination with a defect category vector. Then calculating the parameter utility by adopting a multi-level analysis method, and weighting the defect tendency index; a bidirectional long-short-term memory neural network is constructed as a discrimination model, a special defect screening layer is included, and dynamic identification of defect types is realized. According to the method, a defect feature database is established to store historical data, real-time monitoring and early warning of production line products are realized through model training and association rule mining, and finally early warning information is output, so that the technical problem that the quality defect tendency of the production line products cannot be accurately identified and predicted in the prior art is solved.