25results about How to "Solve the generalization problem" patented technology

The invention provides a coal mill fault diagnosis and prediction method and system based on big data analysis, and relates to the technical field of big data analysis, and the method comprises the steps: collecting a vibration signal of a coal mill bearing through a vibration sensor, carrying out the multi-scale analysis through wavelet decomposition, extracting features, constructing a deep convolutional neural network model, and carrying out the fault diagnosis and prediction of a coal mill bearing. A maximum mean value difference adaptive layer is introduced to carry out feature alignment on data of a source domain and a target domain, identification of fault types and degrees is realized, a health state evaluation index including a fault severity coefficient and a development trend coefficient is established, and the remaining service life is calculated and early warning is carried out. According to the invention, the accuracy and generalization ability of fault diagnosis are improved, accurate prediction and timely early warning of the bearing fault of the coal mill are realized, and the service life of equipment is prolonged.

The invention relates to a deep pseudo video detection method based on multi-modal comparative learning, and is suitable for the field of deep counterfeit detection. The method comprises the following steps: inputting a to-be-detected video into a trained video authentic identification model, and outputting a detection result of the to-be-detected video; the video authentic identification model comprises a visual encoder which is taken from a trained audio and video contrast learning model and is used for extracting visual features of a face sequence in a to-be-detected video; the audio encoder is taken from the trained audio and video comparison learning model and is used for extracting audio features of the to-be-detected video; the cross-modal feature fusion module is used for fusing the visual features extracted by the visual encoder and the audio features extracted by the audio encoder to obtain a feature fusion result; the spatio-temporal feature extraction module is used for extracting spatio-temporal features from the feature fusion result; the fusion features not only comprise visual features, but also comprise audio features, so that the features are richer; and the classifier is used for classifying the to-be-detected video as a real video or a forged video based on the spatial-temporal characteristics.

The present application belongs to the technical field of data processing, and discloses a gait visual feedback anti-falling safety warning method for a rehabilitation treadmill; including, converting clinical characteristic data of a patient into a network structure gating instruction set; collecting multi-modal original data in a patient's movement process in real time; inputting the multi-modal original data into a mechanism constraint engine to extract biomechanical mechanism characteristic data, neural mechanism characteristic data and cognitive mechanism characteristic data; based on the network structure gating instruction set, dynamically reconstructing internal node connection paths and data flow weights of a preset basic fusion network to form an individualized dynamic fusion model; outputting three-dimensional risk attribution data; according to the proportion distribution of each dimension in the three-dimensional risk attribution data, driving a visual feedback interface of the rehabilitation treadmill to perform differentiated anti-falling warning display of the corresponding dimension; and realizing gait visual feedback anti-falling safety warning for the rehabilitation treadmill.

The application discloses a kind of construction method and system of defect detection model, defect detection method and system, wherein specific training step is: constructing partition detector and training, obtain target partition detector, wherein partition detector has several independent detection branches corresponding to photovoltaic defect data set one by one;Based on target partition detector, obtain the classification probability of each sample picture in each independent detection branch in the verification set, obtain the corresponding inference result data;According to the mAP correlation degree matrix calculated by inference result data, determine the standard label based on mAP correlation degree matrix, and determine the mapping relationship between the heterogeneous label originally used by each data set and the standard label;Unified detector is constructed and training is carried out based on mapping relationship, and defect detection model is obtained;The application solves the label conflict and generalization problem of multi-source heterogeneous data.

The present application relates to the technical field of image sensor calibration, and particularly relates to a processing method and system of a hybrid event and frame sensor, computer equipment and a medium; the method comprises the following steps: acquiring original data of the hybrid event and frame sensor; performing a noise calibration operation on an intensity frame to generate an active pixel sensor noise model and a net intensity signal; performing an event probability coupling operation on an event stream to generate an event visual sensor event probability model; and performing unified collaborative processing on the active pixel sensor noise model and the event visual sensor event probability model to output uniformly calibrated sensor data. In this way, a unified and reproducible cross-modal calibration framework can be provided to solve the technical problems of existing calibration technologies, such as weak physical coupling, insufficient calibration granularity and black box implementation, and to provide a unified, reproducible and high-precision sensor data calibration output for high-speed and high-dynamic-range mobile perception applications.

The application relates to the field of defect identification, and discloses a rail small sample defect identification method, system, device and medium, the method comprising: constructing a feature extraction network, which comprises a convolutional neural network backbone and an adaptive feature weighting module, the module recalibrates the initial features based on a channel and space double attention mechanism, and outputs a weighted feature vector; prototype feature vectors of each category are calculated based on the weighted feature vectors of the support set samples; the distances between the weighted feature vectors of the query samples and the prototype vectors of each category are calculated, a probability distribution is obtained through softmax, and a meta-training loss is calculated based on the difference with the real label; the network parameters are optimized through gradient descent, and the feature extraction network with small sample learning ability is obtained through iterative training; in the meta-test stage, the support set and the query set containing new defect categories are input into the trained network to obtain a classification result. The application can accurately identify rail surface defects under very few labeled samples.

The application discloses a reservoir intelligent inversion method and related device, relates to the technical field of reservoir intelligent inversion, and comprises the following steps: obtaining multi-source heterogeneous data of a target area and preprocessing; constructing an initial geological model generator based on a conditional diffusion model, and optimizing hidden space parameters in a manner of seismic wave forward simulation; constructing a reservoir knowledge graph and adaptively pretraining and migrating the model online; performing task demand analysis, task decomposition and dynamic collaborative execution on a reservoir modeling task based on a thinking chain technology and a language large model technology, and obtaining iteratively optimized hidden space parameters; inputting the iteratively optimized hidden space parameters into the optimized geological model generator, outputting a three-dimensional distribution model, and extracting a confidence interval feature; incrementally updating the reservoir knowledge graph, determining a reservoir model conforming to a geological mode feature and matching a seismic waveform, and realizing reservoir intelligent inversion. The application can improve the accuracy and efficiency of reservoir inversion.

The invention relates to the technical field of large model training, in particular to a text matching model training method, device and equipment, and the method comprises the steps: obtaining a plurality of first sample sets subjected to noise enhancement processing; determining a first text of each first sample in each first sample set; combining the first sample and the corresponding first text into a second sample; performing fine tuning on the pre-training model through a plurality of second samples to obtain a text matching model; in conclusion, the robustness of the text matching model can be improved by using the first sample set subjected to noise enhancement processing; according to the method and the device, the generalization of the text matching model can be improved by using the plurality of first sample sets corresponding to the plurality of semantic scenes and determining the first text with relevance for the samples in the plurality of first sample sets; therefore, the training method of the text matching model provided by the invention gives consideration to the robustness and generalization of the text matching model.

The invention relates to the technical field of gesture interaction, and provides a 3D gesture detection method and electronic equipment, which are used for improving the accuracy and robustness of 3D gesture detection. In model training, a synthetic gesture sample subset is added to supplement complex hand gestures under the condition of lack of self-shielding of a real gesture sample subset, and a key point field adaptation technology is designed to calculate a domain difference loss value of a real gesture sample and the synthetic gesture sample. Therefore, the problem of domain difference between a real gesture sample and a synthetic gesture sample is relieved, and generalization of a model to a complex hand posture with serious self-shielding is improved through the synthetic gesture sample; meanwhile, a background sample subset without a hand is introduced, a negative sample regularization constraint item is constructed according to the background sample subset, and a negative sample distribution loss value and a positive and negative sample distribution difference loss value are calculated, so that the distribution difference between the background information and the hand key point is enlarged, the interference of the background information on the hand key point is reduced, and the user experience is improved. And the 3D gesture detection accuracy is improved.

The invention relates to a training method of a decoration element replacement model, character decoration element replacement, a character decoration element replacement device, computer equipment, a storage medium and a computer program product. The model training method comprises the following steps: acquiring a pre-training model for realizing decoration element image replacement; based on complete character head portraits respectively collected for the same sample character at different angles, constructing an incremental training sample; performing incremental training on a pre-training model by using the incremental training sample to enable the pre-training model to replace a second element image of the character decoration elements in the complete character head portrait at the second angle based on a first element image of the character decoration elements in the complete character head portrait at the first angle; and under the condition of completing incremental training by using a plurality of incremental training samples, obtaining a decoration element replacement model with angle compensation capability. Wherein the image display effect of the character decoration element changes along with the change of the angle of the character head portrait. By adopting the method, the working efficiency can be improved.

The application provides a multi-dimensional perception-oriented super-resolution imaging method and system. The method comprises the following steps: acquiring an original low-resolution polarization mosaic image collected by a defocus plane polarization camera, and calculating a first Stokes parameter and a second Stokes parameter according to light intensity components of multiple polarization directions contained in the original low-resolution polarization mosaic image; inputting the original low-resolution polarization mosaic image, the first Stokes parameter and the second Stokes parameter into a pre-trained polarization reconstruction model to obtain a high-resolution polarization image subjected to super-resolution reconstruction; wherein the pre-trained polarization reconstruction model is constructed through a multi-scale large kernel attention mechanism; the multi-scale large kernel attention mechanism is arranged at an encoder layer of the pre-trained polarization reconstruction model and is used for global feature modeling. Based on this, the super-resolution reconstruction of the polarization image is realized, which is physically faithful, efficient in feature extraction and strong in robustness.

The application provides an incremental learning method and device based on a cloud-edge collaborative architecture, equipment and a medium. The method comprises the following steps: training a cloud-end basic model based on original data collected from each edge node; training an edge model corresponding to each edge node based on self-owned data of each edge node and the cloud-end basic model; screening incremental data generated by each edge node to obtain an incremental data set, wherein the incremental data is generated by the edge node based on the edge model; and performing incremental learning training on the cloud-end basic model through the incremental data set to obtain an updated cloud-end basic model. Through the cloud-edge collaborative architecture, the application solves the problem that three important indicators, i.e., model generalization ability, model accuracy and reasoning time delay, are difficult to meet business requirements at the same time, and ensures that the cloud-end basic model can maintain good generalization ability in multiple scenarios through a cloud-end training strategy.

The application discloses a kind of high-voltage motor fault prediction method and system, to solve the problem that prior art cannot carry out high sensitivity early warning to weak, progressive early fault under the premise of not relying on fault sample.The method comprises: synchronously collecting three-phase current, three-axis vibration and stator temperature signal;Based on historical normal data, train multivariate LSTM-VAE prediction model;Real-time calculation multidimensional residual vector;Residual covariance matrix is calculated in sliding window;Generate anomaly score by structural similarity with reference covariance matrix;When anomaly score is continuously over threshold, trigger early warning.The system comprises multi-source sensor array, synchronous acquisition module, prediction model, residual calculation module, covariance analysis module and early warning decision module.The application realizes early fault high robustness early warning when signal amplitude is not over limit by monitoring multivariate residual covariance structure evolution, significantly improves predictive maintenance efficiency.

The present invention provides a chip ceramic package substrate surface defect detection method based on a multimodal large model, which relates to the technical field of ceramic chip surface defects. Multimodal data fusion and cross-domain knowledge migration are used to realize image and text data fusion of chip ceramic package substrate surface defects. Similarity and segmentation loss functions are used comprehensively for constraint optimization to realize efficient and accurate detection of surface defects of package substrates of different product types, further improve the generalization and robustness of the detection method in different scenarios, and combine visual text prompts to form a complete defect analysis report.

The invention discloses a short-term photovoltaic power prediction method and system based on data mechanism combined driving. The method comprises the following steps: acquiring the geographic latitude, the solar declination angle and the solar hour angle of a photovoltaic power station to be predicted; calculating a solar elevation angle according to the geographic latitude, the solar declination angle and the solar hour angle, and determining the projection direction of the sun in the horizontal plane to obtain a solar azimuth angle; calculating the ideal irradiance of a photovoltaic panel plane based on the solar elevation angle and the solar azimuth angle in combination with the photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted; introducing ideal irradiance of a photovoltaic panel plane to establish a photovoltaic power prediction probability neural network PNN model; and constructing a photovoltaic power prediction neural network PVFNN model by combining a long short-term memory network LSTM on the PNN model, and realizing short-term photovoltaic power prediction by using the PVFNN model. According to the method, the irradiance model in ideal weather is organically embedded into the neural network, learning in a data driving process is constrained, and a predication result with interpretability can be obtained.

The invention discloses a double-stage pre-training system for reading of an industrial inspection instrument, and belongs to the technical field of industrial visual inspection and intelligent inspection. The system comprises a structure interpretable parameterized instrument data synthesis module, a mask auto-encoder pre-training module, a multi-task joint pre-training module and an industrial deployment fine tuning module. Constructing a plurality of pointer type instrument structure templates based on a parametric modeling mode, randomizing parameters to generate high-diversity synthetic instrument data, and automatically generating corresponding structure and semantic annotations; performing unsupervised pre-training by using a mask auto-encoder to learn the relationship between the underlying structure features of the instrument image and the global space; then, multi-task joint pre-training is introduced into the shared backbone network, and multi-task collaborative optimization is carried out; in the industrial deployment stage, through multiple task heads and newly added target detection task heads, small sample fine adjustment is carried out in combination with a small amount of field data, and rapid adaptation to a specific industrial environment is realized. The method is suitable for an automatic instrument reading scene in a complex industrial environment.

The invention belongs to the technical field of radar signal processing, and provides a 94Ghz radar baseband signal resolving method and system, comprising clock signal synchronization, signal receiving and preprocessing, analog-to-digital conversion, multistage parallel signal processing, dynamic parameter adjustment, multi-target resolving and dynamic resource allocation. According to the invention, by introducing an FPGA multi-stage parallel processing architecture, the multi-target detection and tracking capability of the system in a complex scene is improved; signal processing parameters are dynamically optimized according to a real-time environment through a deep reinforcement learning algorithm, and the accurate detection capability of the system in different scenes is ensured; the real-time tracking and resolving capabilities of the system on multiple targets are enhanced through a graph neural network multi-target cooperative processing algorithm; dynamic resource allocation is realized through an energy sensing algorithm, and the power consumption of the system is greatly reduced while high-speed processing is ensured.

The invention provides a document analysis method for qualification review and similarity comparison, and relates to the field of document intelligent analysis, the method adopts an OCR, CV and NLP multi-modal fusion technology to realize bidding document deep analysis, fuses text, layout and image features through an attention mechanism, and converts unstructured data into a structured semantic map. And fusing text, image and quotation multi-dimensional similarities, and dynamically generating a comprehensive risk score through a learnable function. And qualification review adopts a rule engine and NLP double-engine cooperation, a rigid rule guarantees a review base line, flexible semantic reasoning processes fuzzy conditions, and NLP weights are dynamically adjusted according to historical behaviors of an enterprise. Manual auditing feedback is used as continuous learning data, and the model and the rule base are optimized. According to the method, full-process automation of bidding clearing is realized, the review efficiency and accuracy are improved, dynamic configuration of rules is supported, the risk of range bidding is effectively prevented, the bidding clearing cost is reduced, and continuous evolution of the system is realized.

The application belongs to the field of robot positioning, and relates to a self-adaptive layered robot positioning method and device for degenerative scenes, which comprises the following steps: monitoring the data quality obtained by a heterogeneous sensor in real time, mapping the state of the heterogeneous sensor to a unified quantization space, and outputting a unique degenerative level identifier D1-D4; for a D1 environment, the uncertainty of feature matching is quantified, and the most contributive feature subset is selected; for a D2 environment, a singular value decomposition is used to locate a degenerative subspace; for a D3 environment, an expected error of the heterogeneous sensor engine is estimated through a performance prediction network, and a continuous confidence weighting is used instead of hard switching; for a D4 environment, a pre-training-fine-tuning-online adaptation three-level architecture is used to provide a cross-platform generalized pure inertial odometer; and historical events are stored as experience data. The application realizes fine differentiation and directional processing of degenerative scenes, and builds an experience-driven closed-loop optimization mechanism.

The present application relates to a training method for a decorative element replacement model, a character decorative element replacement, an apparatus, a computer device, a storage medium and a computer program product. The model training method includes: obtaining a pre-trained model for realizing decorative element image replacement; constructing incremental training samples based on complete character portraits collected at different angles for the same sample character; using the incremental training samples to incrementally train the pre-trained model so that the pre-trained model replaces the second element image of the character decorative element in the complete character portrait at a second angle based on the first element image of the character decorative element in the complete character portrait at a first angle; when the incremental training is completed using multiple incremental training samples, a decorative element replacement model with angle compensation capability is obtained. The image display effect of the character decorative element changes with the change of the character portrait angle. The above method can improve work efficiency.

The invention discloses a bearing residual life prediction method based on a multiple working condition information fusion large model, and the method comprises the following steps: constructing a cross-modal data set, extracting time sequence features through a signal encoder, mapping the time sequence features to a large model semantic space through a projection layer, achieving the alignment with a working condition text, carrying out the fine adjustment of the large model based on an LoRA technology, and carrying out the prediction of the residual life of a bearing. Receiving a sequence fusing the instruction, the working condition and the signal characteristics for joint reasoning; in training, an asymmetric dynamic penalty mean square error loss function is adopted, and the penalty weight of optimistic prediction errors is dynamically increased. According to the method, physical signals and semantic information are effectively fused, and generalization and industrial safety of prediction results are improved.

The invention relates to an autoregression user behavior prediction system based on a large model and a method thereof. The method comprises the following steps: firstly, data preprocessing: vectorizing high-dimensional data features representing user behaviors and user portraits; secondly, feature residual quantization is carried out, feature residual quantization is carried out on the feature vectors, and a code table embeddings is generated; secondly, user behaviors are expressed through tokens, and each token corresponds to a specific behavior or event; thirdly, the token in which the embed dingings are embedded is sent to a Transform model for training, and then the token in which the embed dingings are embedded is sent to the Transform model for training; the Transform model is used for learning the mode of the token sequence and adjusting parameters to minimize a prediction error so as to form a trained Transform model; then, on the basis of the trained Transform model, predicting the next token according to the input token sequence, namely predicting the next behavior of the user; the device is reasonable in design, compact in structure and convenient to use.

The invention relates to the technical field of distribution network operation and maintenance management, and discloses a distribution network operation and maintenance management method and system based on multi-dimensional intelligent perception, and the method comprises the steps: constructing a distributed cooperative sampling perception architecture, collecting distribution network operation and maintenance data through multiple types of sensors, and carrying out the space-time alignment compensation of the distribution network operation and maintenance data through combining with a digital twin space-time coordinate system; establishing an electromagnetic interference self-adaptive filtering architecture to filter interference data, performing equipment characteristic decoupling of equipment type adaptation, performing decoding at a cloud side after encoding, and reconstructing an equipment core operation state; constructing a three-dimensional topological twin body of the distribution network, establishing a health assessment model to predict a degradation track, calculating a fault risk entropy, and substituting the fault risk entropy into a corrected life prediction result; constructing an operation and maintenance decision space in combination with the fault risk entropy and the distribution network three-dimensional topological twin state, and solving an optimal decision; and injecting the decision into the distribution network three-dimensional topology twinborn simulation verification security constraint, evaluating the fault risk entropy change, and feeding back and updating the health evaluation model and the decision engine.

The present application belongs to the technical field of radar signal processing, and provides a 94Ghz radar baseband signal solving method and system, which comprises clock signal synchronization, signal receiving and preprocessing, analog-to-digital conversion, multi-stage parallel signal processing, parameter dynamic adjustment, multi-target solving and resource dynamic allocation. The present application introduces a FPGA multi-stage parallel processing architecture, improves the multi-target detection and tracking capability of the system in complex scenes, realizes dynamic optimization of signal processing parameters according to real-time environment through a deep reinforcement learning algorithm, ensures the accurate detection capability of the system in different scenes, enhances the real-time tracking and solving capability of the system for multi-target through a graph neural network multi-target cooperative processing algorithm, and realizes dynamic resource allocation through an energy perception algorithm, thereby greatly reducing the power consumption of the system while ensuring high-speed processing.

The present invention discloses a method and system for constructing a universal ECG base large model framework. This scheme proposes an enhanced VQ-VAE model. The improved enhanced VQ-VAE model can be applied to ECG signals of different lengths, different noise levels and multiple lead conditions. A masked ECG modeling model is proposed. This model is a self-supervised pre-training framework for 1D ECG signals. Most of the existing mask modeling is developed based on language or images and cannot be directly used for ECG tasks. This scheme designs reasonable mask modeling ECG block parameters and the embedding layer dimension of the ECG encoder for ECG signals. By constructing the above-mentioned ECG self-supervised pre-training task, the data annotation cost can be effectively reduced. In addition, the model does not depend on specific tasks. It is a universal ECG framework that is widely applicable to different ECG tasks.