117results about How to "Guaranteed generalization ability" patented technology

The invention provides a robust multi-modal image segmentation method and system based on instance perception query, and the method comprises the steps: obtaining original image data and text description data, and generating training data through semantic analysis and causal graph construction; extracting image features and text features by adopting a multi-layer coding network, and establishing a feature dependency relationship through causal reasoning; generating instance clustering features based on attention calculation, and obtaining fusion features in combination with a multi-scale fusion strategy; and calculating query confidence and generating candidate segmentation masks, and outputting a final segmentation result through post-processing optimization. According to the method, by introducing causal reasoning and instance awareness query mechanisms, the cross-modal feature alignment precision and the robustness of the segmentation result are improved, and the method has better segmentation performance and generalization ability in a complex scene.

The invention relates to the technical field of seat cushion framework detection, in particular to an automobile seat cushion framework fatigue test method and system. The method comprises the following steps: acquiring information data of an automobile seat cushion framework sample; performing skeleton key part screening on the automobile cushion skeleton sample information data to obtain skeleton key part screening data; performing sensor data acquisition on the automobile seat cushion framework sample based on the framework key part screening data to obtain stress sensor acquisition data and deformation sensor acquisition data; and carrying out loading mode design on the information data of the automobile seat cushion framework sample to generate fatigue loading mode data of the automobile seat cushion framework. According to the invention, the accuracy and reliability of the fatigue test of the automobile seat cushion framework are improved by accurately screening key parts of the framework, designing a real loading mode, deeply detecting damage, predicting the service life and optimizing materials.

The invention provides a large model preference alignment method improved by using an online synchronization strategy, a new preference alignment process is provided based on a synchronization strategy optimization and online learning method, transformation is carried out from a training process, links such as synchronous sampling and online parameter updating are added, and the optimization of the large model preference alignment is realized. Reward model updating and data distribution alignment in the model training process are achieved, and finally the financial question and answer ability is effectively improved. On the basis of keeping the high efficiency of the original method, the distribution difference between the strategies can be effectively reduced, the optimization direction of the optimal strategy is ensured, meanwhile, the supervision information of preference data is updated approximately in real time, the reliability of the reward model in the high reward system is enhanced, the human preferences are better aligned, and the method is suitable for popularization and application. Therefore, the reliability and generalization in the financial service field are ensured.

The invention discloses a mine slope risk prediction method based on multi-modal deep learning, and relates to the technical field of mine slope risk prediction, and the method comprises the steps: Stepl: inputting multi-modal data, Step2: preprocessing effectiveness data, Step3: constructing a sample set, and Step4: realizing mine slope stability prediction based on a gating cycle unit GRU of a mutual information mechanism. According to the invention, a multi-modal data integration GRU network is adopted to carry out sequence modeling on fused data, more complex sequence features are extracted, slope safety event prediction analysis is completed, high-precision monitoring of a mine slope is realized, and the accuracy and comprehensiveness of a prediction result are improved; according to the method, the mutual information mechanism is applied to the GRU network to maximize the mutual information between the sample and the feature, the dependency between the feature and the label is improved, the prediction capability is guaranteed, and the generalization capability on new data is also guaranteed.

The invention discloses an image forgery detecting and positioning method based on noise auxiliary prompt learning, which introduces a contrast language image pre-training model (CLIP) and prompt learning into the field of image forgery detecting and positioning, and comprises the following steps of: 1, respectively processing an input image and noise thereof by using an image encoder of the CLIP and a visual transformer to obtain double-branch features, the forgery perception capability of the network is enhanced through multi-domain fusion and a memory mechanism; 2, creating a pair of learnable prompts as negative and positive samples to replace discrete prompts, and finely adjusting the prompts according to the features and categories of each image; and 3, completing forgery detection by calculating similarity between prompt and image features, and integrating image and text features through feature linear modulation to obtain a forgery positioning result. According to the invention, the accuracy and generalization of image forgery detection and positioning can be improved, and the false alarm problem can be reduced, so that the performance of image forgery detection and positioning can be improved.

The invention relates to the technical field of breach image processing, in particular to a breach rapid situation awareness method and system based on multi-source data. The method comprises the following steps: acquiring dam monitoring station data; performing hydrological sensor network setting on the dam monitoring station data to obtain dam hydrological sensor network data; performing ultrasonic water level change trend analysis on the dam hydrological sensor network data to generate an ultrasonic water level change trend chart; performing breach node sensing on the dam hydrological sensor network data according to the ultrasonic water level change trend chart to generate breach node data and breach prediction node data; and according to the breach node data, macroscopic disaster risk analysis is carried out on the breach area, and a breach area macroscopic disaster risk schematic diagram is generated. According to the invention, through real-time data integration, comprehensive risk assessment, unmanned aerial vehicle monitoring, dynamic prediction model and multi-party cooperative response, the accuracy and response speed of breach rapid situation awareness are improved.

The invention relates to a full-route bump feature mining and EDR estimation method based on airborne QAR data. The method comprises the following steps: forming an EDR basic data set; mining and extraction of advanced features are realized; forming an optimized feature set; obtaining a trained neural network model; and calculating the full-route EDR value of the real route. Through combination of discrete wavelet transform and a neural network model, the invention provides an accurate and reliable EDR estimation method, and compared with a traditional single-parameter method, detail description is more obvious and is completely and automatically calculated by a computer without manual reasoning. According to the method, after model training and optimization, the EDR value on a single course can be estimated in real time, and the accuracy and practicability of analyzing an airplane bumping event through flight data in the aviation field are improved.

The invention relates to a tea tender shoot identification method and device based on YOLO-TI. The method comprises the following steps: S1, constructing a data set; s2, a tea tender shoot recognition model based on YOLO-TI is built; s3, target detection is optimized in combination with an algorithm, and the detection precision is improved; s4, the improved YOLO-TI model is used for identifying and detecting the tender shoots of the tea leaves; and S5, testing and evaluating the YOLO-TI model after the training is completed. The invention relates to the technical field of tea tender shoot recognition. According to the invention, accurate identification of tea tender shoot targets in multiple scenes can be realized, and the probability of false detection and missing detection is reduced. While the detection precision of the model is improved, the generalization of the model in different scenes is enhanced, and the method has important practical significance for intelligent tea picking and development of the tea industry at present.

The invention provides a high-performance material visual screening system based on machine learning and an application method thereof. The system comprises a feature extraction module, a machine learning model training module, a prediction module and a feedback optimization module. The data preprocessing module is used for performing primary processing on the original data, wherein the primary processing comprises removal of invalid data and abnormal values; the feature extraction module is used for converting a time domain signal into a frequency domain signal by adopting Fourier transform, extracting a frequency domain feature, and performing data dimension reduction by adopting principal component analysis (PCA); the machine learning model training module is used for ensuring the generalization ability of the model by adopting a cross validation technology in the training process, and accelerating the convergence process of the model through a gradient descent method; the prediction module is used for predicting the material performance of the high-performance material to obtain a prediction result; and the feedback optimization module is used for automatically optimizing model parameters by adopting Bayesian optimization, analyzing a prediction result of the model and dynamically adjusting the model parameters so as to improve the prediction precision. Through the scheme of the embodiment of the invention, the law can be learned from a large amount of experimental data by a machine learning technology, and the material performance can be quickly and accurately predicted.

The invention discloses a fine-grained representation remote sensing image interpretation method and system based on Shearlet priori knowledge. The method comprises the following steps: performing high-frequency sub-band information extraction on preprocessed remote sensing image data by adopting Shearlet transformation; a Laplacian pyramid filter and a shearlet filter are used for obtaining a high-frequency assembly and multidirectional high-frequency sub-bands obtained after the high-frequency assembly is split in the direction; sub-bands of the same level are adaptively fused through a space attention mechanism, and a detail-enhanced feature map is generated; carrying out multi-scale and multi-level weighted fusion on high-frequency component prior extracted through Shearlet transformation and a deep residual block obtained by a deep residual network; outputting an accurate remote sensing image analysis result by using the trained detail enhancement backbone and different detection heads of a plurality of tasks; through combination of Shearlet transformation and an advanced deep learning architecture, the accuracy and efficiency of remote sensing image interpretation are significantly improved, good model generalization ability and practicability are maintained, and it is predicted that remote sensing image processing will be promoted to develop towards a more intelligent and automatic direction.

The invention relates to the technical field of database construction, in particular to a construction method and system based on an encephalopathy nursing knowledge base. The method comprises the following steps: acquiring multi-source encephalopathy data; performing structured data screening on the multi-source encephalopathy data to generate structured encephalopathy data and unstructured encephalopathy data; performing multi-modal data structured layering on the unstructured encephalopathy data to generate multi-modal structured layered data; performing data preprocessing on the multi-modal structured hierarchical data and the structured encephalopathy data to generate a standard structured encephalopathy data set; performing context analysis on the standard structured encephalopathy data set to generate an encephalopathy information semantic vector set; and performing redundant relation pruning processing on the encephalopathy information semantic vector set to generate spatial semantic optimization data. According to the method, by integrating multi-source data, structured and unstructured information, real-time dynamic mapping nursing requirements and system feedback evaluation, the construction quality and practicability of the encephalopathy nursing knowledge base are improved.

The invention provides an intelligent feeder terminal fault on-line discrimination method, which comprises the following steps: firstly, collecting electric power parameters, switch states and environment data of each feeder terminal, converting the data into feature vectors through a feature extraction algorithm, and then utilizing a pre-trained abnormal probability identification model to identify the abnormal probability of each feeder terminal. Generating an abnormal probability value and calculating a confidence interval of the abnormal probability value, if the lower limit of the confidence interval exceeds a preset threshold value, judging that the terminal is abnormal, for the abnormal terminal, determining an adjacent terminal of the abnormal terminal according to a topological structure of a power grid, and then classifying feature vectors of the abnormal terminal and the adjacent terminal by using a pre-trained fault classification model; and finally, the method comprehensively considers the fault classification data of the abnormal terminal and the fault classification data of the adjacent terminal, and generates a fault comprehensive classification result of the abnormal terminal, so that intelligent online discrimination of the feeder terminal fault is realized, and improvement of the accuracy and efficiency of power grid fault diagnosis is facilitated.

The invention discloses a robust federated learning method for processing heterogeneous noise and non-independent identically distributed data, and belongs to the technical field of federated learning. The method provides a robust learning framework of two-stage client quality perception. The method comprises the following steps of: 1, constructing a category-level loss vector and clustering by using a Gaussian mixture model, and accurately dividing a clean and noise client set; stage 2, performing differential training: performing standard training on the clean client; dual-network cooperative training, dynamic sample screening and exchange, and a self-distillation and entropy regularization mechanism are introduced to a noise client, so that robust learning is realized; in the global aggregation stage, a distance sensing weighting strategy is further adopted to dynamically suppress the influence of a noise client; according to the method, original data does not need to be shared, the robustness and generalization performance of the federated learning model in the coexistence environment of heterogeneous noise and non-independent identically distributed data can be effectively improved, and the method has wide application value in the fields of medical images, financial risk control and the like.

The invention provides a training data generation method and device for a large model, and the method comprises the steps: dividing a specific business scene of the large model, and carrying out the optimization of an initial training set of each business scene through a pre-constructed sub-capability model, and obtaining a scene training set in the business scene; according to the task learning complexity of each business scene, orderly integrating the scene training set of each business scene according to a sequence from simple to complex, and generating target training data for finely adjusting the large model, so that the large model can follow a progressive learning path; the large model subjected to fine adjustment by utilizing the target training data can inherit the training effect of each scene training set on the sub-capability model, and the large model is promoted to fully absorb the advantages of each business scene while the generalization capability of the large model is ensured, so that the training effect of the large model is optimized, and the training efficiency of the large model is improved. And the performance of the large model in each business scene is improved.

The invention discloses a transfer decision-making method and system based on a grid structure and load data, and relates to the field of power system management, and the method comprises the steps: building an initial data set through collecting real-time and historical load data, power grid topology information, environment influence factors and the like, and carrying out the preprocessing of the initial data set; and then, identifying data closely related to load prediction by adopting a feature extraction technology, and predicting a future load by utilizing an LSTM (Long Short Term Memory) model. And based on the prediction result and the power grid structure, evaluating the power transfer potential between the nodes to form an evaluation report. And then, calculating the comprehensive weight of the edges between the nodes, and further finding out an optimal power transfer and supply path set. And finally, multiple groups of power transfer and supply strategies are formulated, an optimal scheme is selected for implementation through benefit evaluation, and strategy adjustment is performed according to an actual operation effect. According to the decision-making method provided by the invention, the operation efficiency of the power system is greatly improved, a powerful decision-making support tool is provided for a power operator, and the safety and the economical efficiency of the power system are remarkably improved.

The invention discloses a property scene perception and decision-making method, device and equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing and format specification labeling of a target detection data set, and obtaining a property data set containing a plurality of images; performing label alignment on the property data set and the public data set to obtain a common label set; fusing the common label set through data screening to obtain a fused data set; converting the fusion data set into a visual question and answer data set, fusing professional knowledge of a property scene, constructing a visual-text multi-modal data set, and performing model training by using the visual-text multi-modal data set to obtain a multi-modal large model; performing fine adjustment on the multi-modal large model by adopting a pre-trained visual model; and deploying the multi-modal large model after fine tuning into a property service system. The problems that an existing multi-modal large model is high in training cost, context semantics cannot be understood and the like are solved, and the practical application value of the multi-modal large model is improved.

The invention discloses a false data injection detection method and system based on a bidirectional long short-term memory network, and belongs to the field of distributed automation system and FDIA detection in a new energy power grid, and the method comprises the steps: constructing a new energy power grid load redistribution attack model based on local information; constructing a new energy power grid false data injection detection model based on a bidirectional long short-term memory network; taking the output of the constructed new energy power grid load redistribution attack model as the input of a false data injection detection model; acquiring data of a new energy power grid; new energy power grid data are incorporated into an IEEE node system to verify the adaptive capacity of the detection model after renewable energy integration; operating data of the power system are collected in real time, preprocessed and input into the verified false data injection detection model, whether the power system is abnormal or not is judged according to a detection result output by the false data injection detection model, and corresponding response measures are taken.

The invention discloses a depression detection method based on image-text social media analysis, relates to the technical field of automatic detection of depression, and solves the technical problem in the prior art that whether a user is a depression patient or not is detected by analyzing character and picture data published by the user on a social media platform. Specifically, the method comprises four steps of data capture, post screening, image-text representation and fusion classification. Generally speaking, the method comprises the following steps: firstly, capturing a posted document or posted post of a user in a certain time period from a social media platform, including characters and picture data in the posted document; then, performing depression related symptom prediction on the posted document, and screening out the most related K posts; then, respectively carrying out text and picture representation on the screened posts; and finally, fusing the characterization vectors of the two modes for depression detection.

The invention relates to a large language model alignment optimization method based on small data volume fine tuning and a storage medium. The invention relates to the large language model alignment optimization method based on small data volume fine tuning and the storage medium. The invention aims to solve the problems that the existing large model fine tuning needs a large amount of annotation data, needs to consume a large amount of computing resources, and needs higher fine tuning cost and development time. The method comprises the steps that 1, a data set for a specific target task is constructed and divided into two parts, one part is a fine adjustment data set, and the other part is an alignment data set; meanwhile, a problem data set for a specific target task is constructed; 2, obtaining a fine tuning model; 3, obtaining an evaluation model; 4, inputting the alignment data set problem into a fine tuning model, outputting a result by the fine tuning model, and calculating the similarity between the output result of the fine tuning model and a target result; inputting an output result of the fine-tuning model into the evaluation model, and outputting an evaluation result by the evaluation model; and 5, obtaining the target model after alignment optimization.

The invention discloses a horizontal well multi-cluster fracturing crack propagation prediction and parameter optimization method and device, and the method comprises the steps: inputting the target input data of a target fracturing section into a fracturing crack propagation prediction model, and outputting a crack propagation prediction result; the target input data comprises preprocessed geological parameters and engineering parameters; the fracturing crack propagation prediction model is obtained by training according to a preset training set; the preset training set is a data set obtained by performing data enhancement on the input sample data and the corrected crack expansion sample data; based on the fracturing fracture propagation prediction model, determining the contribution degree of each geological parameter and engineering parameter in the target input data to a fracture propagation prediction result, and screening key parameters influencing multi-cluster fracture equilibrium propagation; and adjusting engineering parameters in the key parameters to serve as optimization engineering parameters by taking a crack propagation prediction result meeting a multi-cluster crack balanced propagation requirement as a target. According to the method, accurate and rapid prediction of crack propagation and targeted optimization of engineering parameters can be realized.

The invention discloses a consignment system management system and method, and relates to the technical field of time sequence prediction, and the method comprises the steps: collecting multi-source data, and carrying out the preprocessing of the data; performing feature extraction on the data, and constructing a multi-level memory unit LSTM model; future consignment material demands are predicted through a multi-level memory unit LSTM model, and inventory management is automatically adjusted; monitoring the multi-level memory unit LSTM model according to the difference between actual sales and demand prediction data, and updating the multi-level memory unit LSTM model; on the basis of inventory management adjustment, automatic warehousing and warehouse-out are carried out, related records are generated, and an inventory management center is updated; through accurate prediction of the multi-level memory unit LSTM model, inventory management can be effectively guided, the phenomena of inventory overstock and stockout are reduced, and the efficiency and accuracy of inventory management are improved.

The invention provides a baby cry feature extraction method. The method comprises the following steps: acquiring an audio signal of a baby; preprocessing the audio signal, wherein the preprocessing comprises denoising and volume standardization; analyzing a crying sound waveform in the audio signal, determining a crying sound period, and identifying an expiration stage S1, an inspiration stage S2, an excitation stage and a resting stage; and calculating and extracting the duration, intensity and frequency characteristic parameters of each cry period.

The invention discloses a deep forged audio identification method based on hybrid enhancement and a contrast loss function. The method comprises the following steps: step 1, preprocessing an original audio training data set by using a sliding window technology and a hybrid enhancement technology; step 2, for the preprocessed original audio training data set, using a contrast loss function to constrain and forge feature vector spacing of real audio training data in a high-dimensional space; and step 3, using a sliding window to divide the audio to be detected into a plurality of audio clips to be detected with the total length of 64600, and if any audio clip to be detected is detected to be forged, determining that the audio to be detected is forged. According to the method, the analysis and identification capability of the algorithm on weak and local counterfeit traces is improved, the accuracy, stability and generalization of the deep counterfeit audio identification algorithm are ensured, the extraction and analysis capability of the algorithm on essential counterfeit traces is improved, and the interference of environmental noise is reduced.

The invention relates to the technical field of intelligent driving, and particularly discloses a driver fatigue state recognition method and system and a storage medium, and the method comprises the steps: firstly collecting driving operation data under different driver fatigue states, and extracting corresponding information entropy features, so as to excavate the nonlinear features of operation behaviors, and facilitate the recognition of the driver fatigue states; then, information entropy features of all driving operations are screened, information entropy features significantly related to a fatigue state are reserved, and a key feature set is formed, so that driving behavior feature redundancy indexes are effectively removed, and the model response speed and classification performance are improved; and then, adopting a random forest model (RF) as a basic model, and introducing a sparrow search algorithm to optimize training parameters of the RF model in the process of training the RF model by adopting the key feature set to obtain a fatigue driving detection model. According to the invention, the method can achieve high accuracy in the recognition of the fatigue state of the driver.

The invention discloses an SCARA robot arm length and zero point calibration method based on support vector regression. The method comprises the following steps: establishing an SCARA robot kinematics model, recording joint angle data at a preset point of a calibration plate, constructing an overdetermined homogeneous linear equation set, and preliminarily resolving an offset value of the arm length and the zero point. Aiming at nonlinear errors which are difficult to process by a traditional calibration method, an SVR model is introduced, error modeling and compensation are carried out by utilizing a radial basis function (RBF), and dynamic error correction of the robot in different environments is realized. According to the method, the actual error data in the working space of the robot is collected in real time, the calibration precision is remarkably improved, human intervention is reduced, and the precision stability of the robot in high-speed operation and long-time operation is guaranteed. Compared with the prior art, by means of the automatic calibration and error compensation process, nonlinear errors in a complex environment are effectively processed, the industrial production efficiency is improved, the downtime is shortened, and the method has high robustness and wide application prospects.

The invention relates to the technical field of wireless communication, in particular to a wireless interference source detection method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a training data set which comprises channel state data and interference source type tags under different wireless interference sources; an interference source type identification model is established based on a Transform model, and the interference source type identification model comprises a block division layer, a feature extraction layer, a block merging layer and an output layer; and training an interference source type identification model by using the training data set, stopping training until the identification performance of the interference source type identification model reaches a preset standard, and identifying the type of the wireless interference source in the target wireless interference environment by using the interference source type identification model. Therefore, the problem that the interference source cannot be identified in the prior art is solved.

The application discloses a tunnel or mine gushing water space-time prediction method and system coupled with a water power numerical model, and the method comprises the following steps: based on the identified and verified underground water numerical model, outputting multi-source data, complementing the missing measured data, quantifying the difference between fault and normal stratum permeability characteristics and coupling to the data system, and incorporating the tunnel or mine excavation space data. An LSTM-isolation forest-K nearest neighbor regression coupled model is constructed, and a multifunctional module is configured to realize common training of multi-scene data. The pretreated multivariate time series data is divided into a training set and a test set, hidden features are extracted through the coupled model, abnormal detection results are fused, a residual correction model is trained synchronously, and the hyperparameters and weights are adaptively optimized according to the multi-project prediction error feedback. Based on the trained coupled model, a window rolling strategy is adopted to carry out synchronous gushing water space-time prediction, residual correction is combined, and prediction data meeting the engineering precision is output. Reliable technical support is provided for engineering construction safety control.

The invention relates to an image pre-division method based on a multi-layer multi-motion search result, and belongs to the technical field of video coding. The method comprises the steps that an original image and a prediction search image are acquired, and according to the original image and the prediction search image, motion information of the original image is obtained through multi-layer multi-time motion search; obtaining a video sequence and extracting to obtain an image frame set, coding and extracting features of the image frame set to obtain an image pre-division data set; training and testing according to the image pre-division data set through a decision tree algorithm to obtain an image pre-division decision tree; and obtaining a current image frame, extracting current image motion information through the multi-layer multi-time motion search, and obtaining a current image pre-division state through the image pre-division decision tree according to the current image motion information, thereby assisting an encoder in encoding, accelerating the process of determining the encoding block division form, and improving the encoding efficiency. The calculation consumption of the encoder is reduced, and efficient and accurate image pre-division is realized.

The invention relates to an audio classification method, a model training method, a system, equipment and a medium. The training method comprises the following steps: acquiring an extended audio set and an audio tag set corresponding to the extended audio set; inputting the extended audio set into the trained original model, and extracting generated reference original category labels and original features; inputting the extended audio set into an extended model, and extracting a generated original category label and / or a newly added category label and an extended feature generated by the same network layer as the original model; and updating parameters of an expansion model based on the difference degree between the original category label and the reference original category label, the difference degree between the expansion feature and the original feature, and the difference degree between the original category label, the newly added category label and the corresponding audio label in the audio label set, and taking the trained expansion model as an audio classification model. According to the method, the problem of knowledge forgetting possibly occurring when the extension model learns a new category is solved, and meanwhile, the learning ability of the extension model to a new category is enhanced.

The invention relates to a data classification method applied to a government affair platform, belongs to the technical field of data classification, and solves the problem that data classification of government affair information cannot be efficiently and accurately realized in the prior art. Obtaining to-be-classified government affair data, inputting the to-be-classified government affair data into a target feature extraction model to obtain a target feature vector, and inputting the target feature vector into a target feature dimension reduction model to obtain a target feature dimension reduction vector, and inputting the target feature dimension reduction vector into a target classifier model to obtain a category corresponding to the to-be-classified government affair data. The method is suitable for complex scenes of government affair data, heterogeneous and complex government affair data can be effectively processed, and efficient and accurate data classification is achieved.