1598 results about "Training set" patented technology

The application discloses an infrared long-distance space adjacent target super-resolution method, wherein the method comprises the following steps: S1, analyzing the imaging characteristics of a long-distance target optical detection system and the imaging characteristics of space adjacent multiple targets, modeling point target imaging, and generating a simulation image dataset; S2, initializing the input simulation image based on linear mapping; S3, realizing infrared space adjacent target super-resolution modeling by adopting a sparse reconstruction algorithm, and constructing a deep unfolding network framework; S4, sequentially passing the initialized image through each stage of the deep unfolding network; S5, constructing a GPU training environment, setting a data loader, a model and an evaluator configuration file, training the model, applying the trained model to a test set, generating a high-resolution image, extracting target coordinate information by post-processing and inputting the target coordinate information into the evaluator, obtaining evaluation indexes based on infrared long-distance space adjacent targets, and calculating an average detection rate.

The application discloses a government affair text auxiliary writing system and method based on a multi-modal large model, and relates to the technical field of artificial intelligence and informationization; comprising: step 1: converting heterogeneous government affair data into unified semantic representation, realizing cross-modal deep alignment through fine-grained contrast learning; encoding text, image and voice features by using an encoder respectively, and constructing a training set containing 100,000 pairs of government affair image-text samples; weighting and fusing the text, image and voice features, dynamically adjusting the weight through an attention mechanism, training a large model, step 2: performing lightweight deployment of the large model, and step 3: using the large model to understand and identify user intention, and outputting government affair text.

The application discloses a differential pressure sticking probability prediction method and system based on a Bayesian belief network, and comprises the following steps: collecting historical drilling data of a target oil reservoir block, preprocessing sample data, and creating a sample training set and a test set; determining characteristic input variables and characteristic output variables from the sample data; using the sample training set to establish a differential pressure sticking training model based on the Bayesian belief network; and verifying the differential pressure sticking probability result after calculation by using the verification set. The differential pressure sticking probability prediction model based on the Bayesian belief network is used to input the fact drilling characteristic input parameter value of the target oil reservoir development area into the model for prediction analysis, so that the differential pressure sticking probability of instant drilling is obtained, drilling guidance is provided for field engineering and technical personnel, a decision scheme is timely adjusted, the differential pressure sticking probability is reduced, and unnecessary operation time and economic loss are reduced.

This invention discloses a monitoring method for the genus *Fagus* based on environmental simulation enhancement and attention guidance, belonging to the field of image processing technology. The method includes the following steps: S1, acquiring RGB images using a high-resolution camera mounted on a UAV and preprocessing them to generate a training set; S2, generating an expanded robust training sample set; S3, constructing an attention context guidance network; S4, training the attention context guidance network using the expanded robust training sample set; S5, inputting the images to be monitored into the trained attention context guidance network to generate several indicators. This invention, through the synergistic innovation of the environmental simulation enhancement framework and the attention context guidance network, has achieved significant technical progress in the monitoring of *Fagus* communities; it demonstrates outstanding advantages in segmentation accuracy, generalization ability, and practicality.

The application provides an extreme weather wind power prediction method based on reinforcement learning adaptive sampling, and relates to the field of wind power prediction. The method comprises the following steps: obtaining meteorological time series data and wind power data of a wind power station site, constructing a training set, a validation set and a test set, and respectively extracting a training subset, a validation subset and a test subset corresponding to extreme weather; designing a training framework based on reinforcement learning to obtain a Markov decision process component, build a parameterized sampling strategy network and a wind power prediction model; iteratively performing a cooperative optimization process of the sampling strategy network and the prediction model until the training converges, and outputting an optimized target prediction model and a target sampling strategy network. Through the reinforcement learning adaptive sampling method, the sampling strategy network and the prediction model form an optimized closed loop, effectively improving the accuracy of wind power prediction under extreme weather, and ensuring the prediction effect under normal weather.

This invention relates to the fields of artificial intelligence and image recognition technology, specifically to an image recognition method and system for identifying missing key components in discarded switchgear. The method comprises the following steps: collecting image data of discarded switchgear; labeling the images; dividing the labeled image dataset into training, validation, and test sets; performing illumination normalization and geometric adaptive alignment on the original images of the discarded switchgear to obtain adaptively normalized images; enhancing the texture of the adaptively normalized images and fusing them with occlusions to output an occlusion-synthesized enhanced image; constructing and training a component missing identification model; calculating the model's loss function and iteratively training the model to obtain a trained model; and inputting newly collected data, after processing in steps S2 and S3, into the trained model to obtain the final missing component identification results. This invention can achieve accurate identification of missing key components in discarded switchgear.

The application discloses an ophthalmic disease image classification system based on a big data model, which comprises a data set construction module, an image segmentation module and a model training module.The data set construction module is used for collecting eye images of patients with different ophthalmic diseases in an ophthalmic clinic and performing image marking processing to construct training and test data sets.The image segmentation module adopts YOLOv7 to perform image segmentation on the images taken by the smart phone to obtain eye images containing only the part below the eyebrows and above the zygomatic arch.The model training module adopts a five-fold cross-validation method to train the model on the images, and applies data enhancement technology, white balance adjustment and a transfer learning algorithm.The application relates to the technical field of ophthalmic disease image classification.The ophthalmic disease image classification system based on the big data model has high classification accuracy for diseases such as cataracts, keratitis and pterygium in the development stage test set and different clinical evaluation stages.

A system receives staging test results associated with a software product from a staging testing unit. The staging test results are obtained by testing the software product with a test set that imitates a real production environment. The test set includes associations of a set of customer-facing service (CFS) features and a set of network-facing service (NFS) features, and the staging test results include identified anomalies arising from the associations of the set of CFS features and the set of NFS features. The system processes, using an artificial intelligence model, the staging test results to produce a confidence value indicating whether the identified anomalies have a likelihood of causing a failure of the software product when deployed in the real production environment, and determines, using the confidence value, whether to deploy the software product in the real production environment.

The present application belongs to the technical field of image processing, and discloses a primary alpha phase efficient segmentation method for small sample data titanium alloy microstructure, comprising the following steps: S1, collecting and preprocessing the titanium alloy microstructure image, forming a training set, a verification set and a test set through artificial labeling and data enhancement, and constructing a microstructure segmentation data set; S2, building a MaterialAlphaSAM segmentation model, which is based on SAM, combined with a field adaptation module and a geometric constraint prompt prior module, so that the model can effectively locate the target area under the condition of few samples, and enhance the segmentation precision and stability; S3, model training and performance verification based on MaterialAlphaSAM; S4, inputting the test set image into the trained model, outputting the pixel-level segmentation result of the primary alpha phase, and further performing quantitative analysis. Under the premise of not introducing large-scale parameters and labeling costs, the present application adaptively generates high-quality prompts consistent with the microstructure semantics and topographic features, and effectively improves the segmentation performance in the few sample scene.

This invention relates to an adaptive severe weather image restoration method based on continuous testing, belonging to the field of computer vision and image processing technology. It includes the following steps: constructing a severe weather dataset and dividing it into training and testing sets; constructing a severe weather image restoration framework, including a pre-trained model DA-CLIP, a student model, and a teacher model; inputting the publicly available dataset into the pre-trained model DA-CLIP for training, and initializing the student and teacher models using the parameters of the trained DA-CLIP model; inputting the training set of the severe weather dataset into the student and teacher models respectively for model training and optimization, obtaining a trained severe weather image restoration framework; inputting the test set images of the severe weather dataset into the trained severe weather image restoration framework to obtain the image restoration result. This invention can improve the model's image restoration performance under complex and realistic weather conditions.

The application discloses a source code annotation automatic generation method and system based on word-level retrieval and belongs to the field of natural language processing text generation. A training set composed of code function text, code abstract syntax tree and code annotation text is used to train an encoding-decoding network; an overall feature vector of each annotation word in the code annotation text is obtained to construct a near-neighbor word database; for the code function text to be annotated and the abstract syntax tree thereof, a model-based target word probability distribution and an overall feature vector of a target word at a current time step are generated in a self-recurrent manner; K near-neighbor words with the highest similarity to the overall feature vector of the target word are searched in the near-neighbor word database to generate a target word probability distribution based on the near-neighbor words; and the two target word probability distributions are fused to take the target word with the maximum probability as an annotation word generated at the current time step. The application can greatly improve the annotation generation quality of an original model and also improve the generation probability of low-frequency words in code annotation.

The application provides a ship identification model training method, a cloud server and a ship identification system. The method comprises the following steps: inputting sample images of ships of multiple models as a first training set into an initial ship identification model, and determining a first loss value of the initial ship identification model under initial model parameters; inputting the first loss value into the initial ship identification model in reverse, adjusting the initial model parameters and iteratively executing until the first loss value output by the initial ship identification model meets a preset condition, and obtaining a target ship identification model; inputting multiple new sample images of a new ship as a second training set, performing incremental learning and updating of the target ship identification model according to the second training set and a pre-training language model, obtaining an updated ship identification model, and sending the updated ship identification model to each edge node device, so that each edge node device performs ship identification based on the updated ship identification model, and the ship identification precision is improved.

The application relates to an intelligent metasurface security transmission optimization method based on a supervised deep neural network. The method comprises the following steps: acquiring a signal of a transmitting integrated RIS, modeling an RIS-to-legal user channel, combining the signal and the channel, and calculating a legal user received signal-to-interference-and-noise ratio and an achievable rate; meanwhile, corresponding indexes at an eavesdropper and an achievable rate of an eavesdropping link are calculated. Then, a multicast system security capacity is designed according to a minimum value of the achievable rate of the legal user and the achievable rate of the eavesdropping link. In order to achieve the target, an optimization model is constructed by combining the achievable rate of the legal user, RIS power and a perceived signal-to-noise ratio constraint. A supervised deep neural network and a training set are constructed, an offline training, online inference and parameter updating are carried out by using an Adam optimizer according to a preset loss function, and a trained network is obtained. Finally, the optimization model is solved by using the network, and RIS phase shift and power allocation coefficients are acquired. The method can reduce the calculation complexity and improve the real-time performance.

This invention relates to the field of new energy technology and discloses a short-term wind speed prediction method based on LSTM neural networks. The method first collects multi-dimensional historical wind speed data from wind farms and performs standardized preprocessing to construct a high-quality training set. Then, it designs and trains an LSTM neural network model that can effectively capture temporal dependencies. Finally, it obtains a high-precision prediction model through validation and optimization. In application, the preprocessed real-time data is input into the model to achieve quantitative prediction of wind speed for the next few hours. This method establishes an online performance evaluation and automatic update triggering mechanism based on the inherent volatility of the data, forming a complete prediction-evaluation-optimization closed loop. This allows the model to self-perceive performance degradation and autonomously trigger retraining, thereby overcoming the problem of performance degradation of traditional static models due to environmental changes after long-term deployment. Ultimately, this ensures the accuracy, adaptability, and industrial application value of the prediction model in long-term operation.

The application provides a read cache optimization method and device in an AI model training scene, comprising: during AI model training, setting each Epoch reading training set data process as two rounds of iterations; in the first round of iteration, each Mini-Batch queries S*N items of data to the cache in random order; S items of data are read from the data in the cache after the query for the current Mini-Batch, and the remaining data is marked; after all the training set data is traversed once, the first round of iteration ends, the second round of iteration starts, and the marked data in the first round of iteration is traversed; each Mini-Batch queries S items of data to the cache, the data is directly read in the cache, or the data is read from a remote storage node. The application improves the read cache hit rate when loading the training set, thereby reducing the time delay of reading data, improving the utilization rate of GPU and other scarce computing power devices, saving computing power, and improving the model training efficiency.

The application relates to the technical field of artificial intelligence, and provides an encoder training method, device and equipment and a storage medium, which can be used for encoder training before obtaining a financial bill in the financial field, the trained encoder obtained can identify a fake face, and does not give permission to obtain the financial bill, so that the security of a financial system is improved.The method comprises the following steps: obtaining a training set; inputting a training video subset into a video encoder to perform video representation; inputting a training audio subset into an audio encoder to perform audio representation; inputting a training image subset into an image encoder to perform image representation, constructing a first loss function according to a video feature and an audio feature, and constructing a second loss function according to an image feature and an audio feature; training the video encoder according to the first loss function, and training the image encoder according to the second loss function.The method can guarantee the reliability of identity verification through face recognition, and makes the process of obtaining the financial bill have strong confidentiality.

A kind of thin plate part numerical control machining quality prediction method based on multi-task transfer learning, first form the original machining feature data set set D of numerical control machining, the pre-processing of each data set in set D is carried out;For the feature data set set D' after pre-processing, the source domain and target domain with small distribution difference are selected for transfer, and the training set and test set division of the selected source domain and target domain data set is carried out;Then construct the shared layer network E of multi-task feature extraction;Then construct the task-specific layer network L based on dynamic distribution self-adaption;Finally, the weight of each task total loss is dynamically adjusted, and the multi-task transfer learning model R construction is completed;The present application realizes the quality prediction under unknown working condition and parallel output multiple quality index prediction results, and the result precision of multiple quality index parallel output is high, and the transfer effect of quality prediction is high in reliability.

Embodiments of the present application provide an unsupervised visual detection method for assembly abnormalities of industrial structural parts and related equipment, relating to the technical field of industrial visual detection. The method comprises: extracting local patch features and constructing a memory bank through a normal sample training set, reconstructing features based on the memory bank, generating a memory index histogram descriptor for representing local statistical structures, simultaneously constructing a feature similarity graph and performing graph enhancement processing to obtain a graph-enhanced reconstructed feature representation, and then constructing an associated memory index histogram memory bank and a global feature memory bank; performing the same processing on the image to be detected, obtaining the candidate normal sample most similar to the structure of the image to be detected through two-stage candidate sample screening, performing similarity matching based on the global feature memory bank to obtain a logical abnormality score, and fusing it with the local abnormality score obtained based on the local patch memory bank to obtain the final abnormality detection result. The present application has the advantages of high detection accuracy, strong robustness, and no need for abnormal samples.

The application discloses a bearing surface defect detection method and system based on improved YOLO-LMSE, S1, collecting bearing defect images; S2, performing data enhancement and preprocessing on the original image set to obtain standardized image data; S3, dividing the standardized image data into a training set, a test set and a verification set; S4, constructing a YOLO-LMSE bearing defect detection model; S5, inputting the bearing defect data set into YOLO-LMSE for training; S6, performing comparative testing on the trained YOLO-LMSE and an original YOLOv11m model on the test set, and verifying the performance difference of the model in bearing defect position positioning, category identification and confidence output; S7, performing multi-dimensional comparison between YOLO-LMSE and mainstream target detection algorithms, comprehensively evaluating the detection performance, calculation efficiency and parameter scale of the model, and determining the industrial deployment applicability of the model. The application can effectively adapt to the deployment requirements of resource-limited industrial scenes, improve the efficiency and reliability of bearing defect detection, and provide a powerful guarantee for the stability of industrial production.

The application relates to a multi-modal sensory self-adaptive adjustment method and device based on physiological characteristic feedback, a computer device and a storage medium. The method comprises the following steps: determining a target physiological index to be adjusted, defining a sensory parameter space composed of multi-modal sensory control parameters and a control parameter vector; generating an initial control parameter vector based on a preset sampling strategy and converting the initial control parameter vector into a physical control instruction for execution, collecting corresponding physiological signals to generate a physiological characteristic vector; constructing a training set by combining the control parameter vector and the physiological characteristic vector into a data pair, establishing a probability agent model; performing parameter optimization iteration based on the model and the training set, selecting a target characteristic vector from a non-dominant physiological characteristic vector set and executing a physical control instruction corresponding to the target characteristic vector, and calculating an average physiological characteristic vector in a sliding time window to perform real-time monitoring and dynamically adjusting the physical control instruction. The application can realize personalized real-time adjustment of multi-modal environmental stimuli such as vision, hearing and olfaction, and effectively improve the user state.

The application provides a battery missing data recovery method based on CVAE and meta learning, relates to the field of data processing, and comprises the following steps: a CVAE model is constructed and trained by using a Reptile meta learning algorithm; for a charge-discharge cycle with missing data, corresponding battery working condition temperature and maximum capacity are taken as condition variables and input into the trained CVAE model to generate a fixed-length data sequence; the maximum capacity and the total cycle time corresponding to each charge-discharge cycle in the training set are extracted as a data group, a polynomial function is fitted based on the data group by using a least square method, and a mapping relationship between the maximum capacity and the total cycle time is established; based on the mapping relationship, the total cycle time of the charge-discharge cycle with missing data is predicted according to the condition variables, a linear time sequence corresponding to the total cycle time is generated, the data sequence is aligned with the linear time sequence, and recovered data is obtained. The application can generate battery data with high fidelity based on a small amount of samples.

This application provides a method and related equipment for preprocessing game trajectory data. The method includes: converting game trajectory data into a first grid sequence; merging high-frequency adjacent grids in the first grid sequence to calculate a second grid sequence; extracting embedding vectors from the second grid sequence to calculate an embedding vector sequence; and using the embedding vector sequence as a training set input to a deep learning model. This application's embodiment, by converting game trajectory data into a first grid sequence, merging high-frequency adjacent grids in the first grid sequence, and then extracting embedding vectors from the second grid sequence, can capture the prior influence of environmental factors on trajectory distribution and the potential contextual semantic information in the trajectory, achieving effective compression of long trajectory data.

This invention provides a method for high-speed train bearing fault diagnosis and selection based on multi-model comprehensive evaluation. The method first obtains bearing fault feature vectors, then divides the training and test sets using 8:2 stratified sampling, mitigating sample imbalance bias through a sample weighting strategy. Next, it constructs a diagnostic system comprising five models, including random forest and support vector machine, and employs optimal hyperparameters for parallel training. Subsequently, a multi-dimensional evaluation system for accuracy and efficiency is established, and model reliability is verified using a confusion matrix. Finally, model selection is tailored to specific scenarios: multilayer perceptrons are used for real-time monitoring, gradient boosting trees for high-precision scenarios, and random forests or K-nearest neighbors for lightweight deployment. This achieves scientific multi-model selection, balancing diagnostic accuracy, real-time performance, and stability, adapting to the maintenance needs of high-speed trains, and demonstrating strong engineering practicality.

The application discloses a coal rock hydraulic fracturing signal identification method based on an MLP neural network model, microseismic characteristic parameters capable of representing different coal rock lithologies are screened out first, then microseismic characteristic parameters corresponding to different types of coal rock lithologies are obtained, and the above data is divided into a training set and a verification set; then an MLP neural network model with a specific structure is constructed, the model is trained and learned by using the training set, the weights and bias terms of the MLP neural network model are optimized and updated by using a BP algorithm, after the optimization training is finally completed, all microseismic data in the hydraulic fracturing process are input into the model, the model can accurately output the coal rock lithology labels corresponding to each group of data after identification, and the coal rock hydraulic fracturing signal distinguishing process is completed; thereby, the lithology and horizon information of the hydraulic fracturing coal rock rupture are identified, the search range of microseismic positioning inversion is reduced, and finally the efficiency and accuracy of the microseismic positioning inversion are effectively improved.

The application relates to the field of radiation source identification, and specifically provides a high-explainability radiation source identification method and system, a training set radar pulse signal is acquired, a multi-domain joint time-varying fingerprint feature is extracted based on physical layer modeling, the fingerprint feature is input into a time-frequency double-branch feature fusion network to obtain time-frequency fusion features; the time-frequency fusion features are subjected to signal-level physical consistency enhancement and feature-level mixed enhancement; based on the enhanced features, an end-to-end training is conducted on the time-frequency double-branch feature fusion network through a hierarchical contrast learning framework containing intra-class contrast and inter-class prototype contrast, a radiation source identification model is obtained, and the trained radiation source identification model is used to identify the individual class of a radar pulse signal to be identified. Through the introduction of the physical layer explainable time-varying fingerprint feature, the adaptive double enhancement mechanism and the hierarchical contrast learning framework, the application realizes high-credibility, strong-adaptation and stable-identification radiation source individual identification.

The application discloses a small sample magnetic anomaly data enhancement and boundary identification method and system based on StyleGAN2-ADA, and relates to the field of geophysical exploration.The method comprises the following steps: establishing an initial small sample set; constructing a generative adversarial network based on adaptive discriminator augmentation (ADA), training a generator by using the initial small sample set, and generating a large amount of high-fidelity magnetic anomaly data; calculating boundary labels for the generated data by using a physical operator fusion strategy, and constructing a large-scale labeled training set; constructing a U-Net boundary identification network, pre-training the U-Net boundary identification network by using the large-scale labeled training set, fine-tuning the U-Net boundary identification network by using measured data, and finally outputting the magnetic anomaly boundary position of a target to-be-measured area.The application solves the problems of a lack of measured samples and high labeling costs in deep learning in magnetic exploration, realizes high-quality expansion of data under small samples, and improves the generalization ability and noise resistance of the boundary identification network to complex geological features.