81 results about "Information loss" patented technology

This invention relates to the field of artificial intelligence technology, addressing the technical problems of existing technologies that easily lead to information fragmentation, lack of complex spatiotemporal dependencies, and lack of sensitivity to dynamic data changes, resulting in bandwidth waste or loss of critical information. Specifically, it relates to a method and system for real-time synchronization processing of interactive data from intelligent devices. This method includes the following steps: deploying various sensors and monitoring devices on campus, collecting raw sensor data and monitoring video data, and preprocessing the raw sensor data and monitoring video data to obtain a standardized dataset. This invention integrates multimodal data by constructing a spatiotemporal heterogeneous graph, significantly reducing redundant data transmission, achieving efficient edge-cloud collaborative computing under low bandwidth, improving deep coupling and dynamic data optimization, and enhancing the real-time performance of campus security and the monitoring efficiency of sensitive campus scenarios.

The present application relates to a kind of time series prediction methods under multi-scale double clustering, the present application can simultaneously capture the data pattern and relevance under different time scales by multi-scale input mechanism, improve the modeling ability to complex real-world signal.Effective decoupling and modeling TDS and CCT: by parallel patch intra-domain clustering module and mask attention module, respectively, special modeling is carried out on two core challenges of time distribution drift and channel correlation conversion, avoids causal confusion, and improves the robustness and generalization ability of the model.The model carries out twice dynamic, hybrid expert-based dependency filtering at patch level, can accurately capture instantaneous and time-varying correlation, avoids the information loss or noise interference caused by traditional coarse-grained or static dependency modeling.

This invention discloses a remote sensing image description evaluation method based on cycle consistency, comprising: generating fine-grained text descriptions of the original remote sensing image using a remote sensing multimodal large model; inputting the generated long text descriptions into a text-to-image model for image reconstruction, and introducing a weighted basis image strategy to compensate for information loss during the text-to-image conversion process; using a multimodal quality assessment model to compare and score the cycle consistency of the original image and the reconstructed image from dimensions such as information sufficiency, scene consistency, and spatial layout, and performing quantitative evaluation according to a preset six-level alignment standard, followed by normalization processing to obtain the final description quality evaluation result. This method solves the problem that existing remote sensing image description evaluation indicators fail in long text and fine-grained scenarios, achieving objective and reliable automatic evaluation without relying on reference text. It is suitable for applications such as remote sensing multimodal model performance evaluation, large-scale dataset quality screening, and intelligent geographic information perception.

The application discloses a kind of based on multi-granularity dynamic receptive field electric energy meter abnormality detection system and method, belong to electric energy measurement technical field.The system of the present application, comprising: multi-granularity dynamic receptive field module, for the patch block data of electric energy meter is iterated and handled, to output receptive field data;Multi-dimensional time series codec module is used to reconstruct the receptive field data that the multi-granularity dynamic receptive field module outputs, and output reconstruction data;Double-layer abnormality detection module is used to calculate the abnormality score of reconstruction data that the multi-dimensional time series codec module outputs, determines the abnormality of electric energy meter based on the abnormality score.The present application can identify the anomaly of data by receptive field and data reconstruction, to determine abnormality score, and solves the problem that information loss or semantic feature mining is insufficient when existing reconstruction method is handled low information density multi-dimensional time series data.

The application discloses a kind of connection terminal fault arc multilevel feature selection and detection method and system, the method of the present application includes collecting the original current waveform data sample of connection terminal under a variety of load conditions, the feature of the feature domain of the sample in sample set is extracted to obtain original feature set;The quality of the feature in original feature set is evaluated and the feature that quality does not meet the requirement is eliminated;According to the quality of the feature in each feature domain, the feature domain weight of each feature domain is determined;The global importance evaluation score of each feature in combination feature set is obtained;The feature in combination feature set is screened based on target dimension hard constraint and information loss soft constraint Execution screening decision finally generates optimization feature subset;Based on the feature in optimization feature subset, the fault detection of connection terminal is carried out.The present application aims at realizing the independent evaluation of feature effectiveness, dynamic optimization and fault arc detection under the premise of not depending on specific identification model.

This invention relates to the field of industrial data processing technology, and discloses a method and system for processing vibration data and shock data. The method includes: a sensor determining a dynamic floating threshold based on neighbor vibration summaries; initiating a local consensus challenge when the shock exceeds the threshold; if the challenge fails, querying the survival status of physically non-adjacent witness sensors, and performing final arbitration and differentiated data routing based on the query results to assess the source reliability of the shock event. This invention, by establishing a pre-emptive source reliability arbitration process at the data source, transforms the basis for judging data reliability from passive trust in individual sensors to active verification of sensor community consensus and external witnesses. This not only suppresses false data caused by individual faults but also prevents the erroneous filtering of real shocks due to local cluster failures, avoiding the risks of data contamination and loss of critical information.

The present application relates to a kind of structure inference, automatic analysis and standardized output of multi-variant BOM table, belong to electronic manufacturing data processing, table analysis and structured data cleaning field.The present application constructs a set of analysis architecture for multi-variant BOM table by the modular processing link of "title row detection-column head normalization-multi-column identification-variant reconstruction-segment cleaning-intelligent inheritance-standardized output", robustly locates complex table head by strong and weak evidence weighting and verification under-probing mechanism, realizes the split and identification of multiple quantity columns using stop mark and semi-finished product number analysis, and introduces cross-row backfilling and position number inheritance rules under paragraph constraint, effectively solves the analysis problem caused by template difference, multi-column variant and information loss, realizes the automation, precision and standardization of BOM data from messy input to engineering level standardized output, significantly reduces manual intervention and improves data quality.

This invention discloses a spatial target semantic segmentation method and apparatus based on ISAR echoes, belonging to the field of radar target recognition technology. The method first uses a trainable domain alignment module (DAM) to directly perform a learnable orthogonal transformation on the original ISAR echo without imaging transformation. While maintaining the same amount of information, the echo domain of the ISAR echo is mapped to a feature domain compatible with the mask domain of the semantic segmentation generation mask, obtaining a first feature. This first feature is then fed into a complex domain encoder for scattering feature extraction. The scattering feature is further processed by a complex domain decoder to achieve the recognition of the semantic segmentation mask. This invention differs from the traditional "image first, segment later" process, avoiding potential information loss and time consumption during the imaging stage, and achieving a direct mapping from the original ISAR echo to pixel-level segmentation.

This invention discloses a method, system, and computer device for extracting outlines from long documents, belonging to the field of document processing technology. The method includes: S100, reading the document to be processed; S200, processing the document to obtain text blocks and recording the positions of the text blocks within the document; S300, inputting the text blocks into a large language model to generate corresponding result blocks, where each result block is the outline of the current text block; S400, repeating step S300 until all result blocks corresponding to all text blocks are obtained, and then concatenating all result blocks to obtain the document's outline. This invention employs a multi-level processing strategy, segmenting at the paragraph level, and, when encountering extremely long paragraphs, performing precise segmentation at the sentence level. This avoids truncation and information loss due to excessively long text, and also prevents resource waste caused by excessively short text blocks.

The application discloses a neighborhood density minimum uncertainty sample selection method based on an improved shadow set, and relates to the technical field of data preprocessing.The method comprises the following steps: calculating the neighborhood density of each sample in a data set; applying an improved shadow set balance factor optimization algorithm to determine an optimal threshold by establishing a target function and minimizing the function, wherein the target function quantifies the information loss of the neighborhood density in the division process, and introduces an adjustable balance factor to balance the loss of the determined area and the uncertain area; dividing the data set into core samples and boundary samples based on the optimal threshold, removing the core samples, and retaining the boundary samples; and training a classifier model using the retained boundary samples.The application replaces the traditional membership degree with the neighborhood density, optimizes the shadow set threshold, and constructs a sample selection framework, accurately retains key boundary samples, efficiently removes redundant data, significantly improves the training efficiency of the classifier, the generalization performance, and reduces the calculation and storage overhead.

The application discloses a kind of based on IDC-Net network's ceramic tile product surface defect classification method and system, belong to ceramic tile surface defect classification technical field, including.The application designs the basic unit IDC-Block with feature enhancement capability, the module integrates dynamic large kernel convolution DLK and inherits the advantage of efficient feature extraction of visual state space block VSSBlock IDC-Layer is constituted, on the basis of IDC-Layer, dense connection design is used to constitute IDC-Block by multilevel structure stacking, gradually enhanced feature representation ability;Further design adaptive gaussian pooling module AGP, reduce the information loss in down-sampling process, significantly improve the classification performance of model to ceramic tile small target defect;Dynamic feature fusion module DYFF is designed, adaptive fusion of different resolution and semantic level features is realized, and the robustness of feature expression is enhanced;With EfficientNetV2-s classification network framework as auxiliary branch, local and global information is integrated to constitute double-path classification network IDC-Net by DYFF module, and the classification accuracy is further improved.

The present application belongs to the technical field of computer vision, in particular to a psychological condition evaluation method based on channel attention and variable-length sequence processing, comprising: obtaining and processing psychological evaluation video data of a subject to construct an original feature tensor; constructing an adaptive multi-feature spatiotemporal attention network to obtain aligned features; constructing an adaptive aligned channel attention module to obtain enhanced features; inputting the enhanced features into a regression prediction head network, and performing end-to-end training on the network based on a joint loss function to evaluate the subject's psychology; and the present application solves the technical problems of large differences in actual psychological evaluation task duration, significant sequence length changes, and low processing efficiency and large information loss of variable-length sequences in the prior art.

The application discloses a content security protection method, an electronic device and a computer readable storage medium, relates to the technical field of artificial intelligence security, and comprises the following steps: dividing to-be-protected content into at least one slice of binding metadata, solving the problem of extensive pollution positioning, and realizing accurate positioning at the slice level; determining the comprehensive judgment result of the slice through multi-factor quantitative scoring and a multiple injection detection mechanism, solving the problem of high detection missing rate, and realizing quantifiable and accurate detection; performing reservation, marking or isolation processing on the slice according to the comprehensive judgment result, extracting factual elements from the isolated slice to construct a safe context, solving the problems of detection interruption and information loss, and realizing the isolation of pollution without interrupting the task; while continuing to execute the task under the safe context, implementing gating on tool calling, output content and memory writing, solving the problems of pollution cross-round diffusion and tool chain risk amplification, and realizing adaptive protection of risk perception.

This invention discloses a real-time segmentation modeling method based on a multi-scale constraint mechanism, comprising: extracting features from an image and adding them to a segmentation output head for compression processing; performing an X-fold upsampling operation on the compressed features to calculate loss1; performing a full-size upsampling operation on the compressed features and extracting their boundary features to calculate loss2; designing a hybrid loss function based on loss1 and loss2, calculating the total loss L, connecting it to the segmentation output head, and outputting the semantic segmentation result. This invention can achieve differential edge detection, avoiding information loss caused by traditional post-processing, and improving boundary segmentation accuracy by 15.6%. In addition, a dynamic weight allocation strategy is adopted to achieve automatic balancing at different training stages. The final calculated total loss comprehensively considers global classification and local boundary optimization problems, achieving 82.3% mIoU on the Cityscapes dataset, ensuring more comprehensive and complete data capture.

The application belongs to the technical field of image compressive sensing reconstruction, and discloses an image compressive sensing reconstruction method and system based on optimization algorithm expansion. In the sampling stage, a convolution sampling method is used instead of a traditional random matrix sampling. In the reconstruction stage, a generalized iterative threshold shrinkage algorithm is expanded into a deep network, and a jump information connection structure is designed in a reconstruction submodule R. Residual modules are used to connect the feature information before and after the modules, so that the inherent information loss in the deep expansion network is avoided. Furthermore, a double-scale denoising module is designed at the back end of the reconstruction submodule R, and different scale features are combined to denoise the image. The application not only realizes the application of the algorithm expansion method in the image compressive sensing, but also improves the reconstruction effect by using the jump connection structure and the double-scale denoising module. The application has higher accuracy and better robustness.

The application discloses a graph processing method applied to the field of artificial intelligence and used for training a graph neural network. The graph processing method determines the influence of simplifying each edge on information propagation in a network structure graph by calculating the simplification sensitivity of each edge in the network structure graph, so that edges with less influence on information propagation are selected for simplification, thereby reducing the size of the network structure graph and reducing information loss caused by the size reduction of the network structure graph. Moreover, the simplification of the network structure graph is realized based on the simplification sensitivity of the edges in the present scheme, so that the information in the network structure graph can be preserved to the greatest extent, thereby guaranteeing the training effect of the graph neural network trained based on the network structure graph.

The application discloses a kind of drainage pipeline defect identification method based on improved YOLOv10, belong to defect detection field and image recognition field.The method includes: collecting drainage pipeline defect image, carries out pre-processing operation to the data set collected, obtains the data set as training model;YOLOv10 model's main network and head network are improved, and the improved YOLOv10 model is obtained;Using the data set obtained as training model, the improved YOLOv10 model is trained and verified, obtains the optimal YOLOv10 model of drainage pipeline defect, completes drainage pipeline defect identification.The application provides more smooth category transition basis, reduces information loss and improves the accuracy of feature extraction, optimizes the processing capacity of model to small object and low resolution image.For the drainage pipeline defect image collected, there are still problems such as uneven image illumination and low resolution of picture after frame processing, improve the detection efficiency and detection precision of model.

With the development of big data, the privacy security of users is attracting more and more attention. K-anonymity algorithm is an effective and low-complexity privacy protection algorithm, which can protect the privacy of users by anonymizing the data. However, the algorithm currently only focuses on improving user privacy and ignores data availability. In addition, due to the influence of quasi-identifier attributes on sensitive attributes, the availability of processed data is reduced in statistical analysis. Based on this, a new K-anonymity algorithm is proposed. First, an information loss function is designed based on the amount of information and the influence degree of each quasi-identifier attribute. Second, an initial centroid selection algorithm is designed to improve the 2-means clustering effect. Finally, an improved 2-means clustering and a greedy algorithm are designed to design a K-anonymity algorithm. The algorithm can solve the privacy security problem in the context of big data, while improving the availability of data.

The application discloses a kind of multi-modal fusion target detection method and system based on global semantic condensation perception and local optimal transmission alignment, propose global semantic condensation perception fusion mechanism, by spatial condensation and serialization modeling to cross-modal feature, realize efficient, low-redundancy global semantic fusion;Introduce local optimal transmission alignment module, formalize cross-modal local feature alignment problem as optimal transport problem with constraint, improve the geometric consistency and semantic discriminability of feature alignment;Build the two-stage collaborative fusion architecture of "global fusion+local alignment", give consideration to high-level semantic complementarity and local structure accuracy, effectively alleviate the fusion misalignment and information loss problem in existing method, under the premise of ensuring high reasoning speed and low computational complexity, significantly improve the discriminability and robustness of fusion feature, more suitable for real-time multi-modal target detection task in complex industrial scene.

A method for repairing highly reflective areas combining digital twins and transfer learning is proposed. The specific implementation includes the following steps: First, the camera and projector of the real 3D measurement system are calibrated to obtain their intrinsic and extrinsic parameters. Second, a digital twin measurement system corresponding to the real system is constructed using the simulation software Blender, generating a large amount of virtual data. Overexposed images are used as label values, and unsaturated images are used as the network output. A U-net network model is trained using the dataset to obtain a pre-trained network, giving it initial capabilities for repairing overexposed areas. Finally, transfer learning is introduced, and the pre-trained network is fine-tuned using a small amount of real data to improve the model's generalization performance in real-world scenes. This method provides a new approach to solving the problem of stripe information loss caused by high reflectivity, significantly reducing data acquisition costs and time, and can be applied to the reconstruction of highly reflective objects.

This invention discloses a RecycleNet2 model and data processing method based on information loss, belonging to the field of computer technology. The model includes K-1 sequentially cascaded RecycleNet2 modules, K feature transformation modules, a fully connected layer, a concatenation and fusion layer, and an output layer. The main branch of the RecycleNet2 module uses 1×3 and 3×1 asymmetric convolutions to extract preliminary features, while the side branch uses a subtraction layer to subtract the input from the preliminary feature map, extracting and recovering lost details to generate recovered feature maps. The K recovered feature maps are then subjected to secondary feature mining via 1×1 convolutions built into the feature transformation modules before being input into the corresponding fully connected layers. Finally, the concatenation and fusion layer performs unified fusion of all feature output representations, and the output layer outputs the classification result. This invention effectively solves the problem of information loss during convolution, improving the depth representation of network features and classification accuracy.

The application discloses a cross-platform CAD drawing intelligent collaborative management method and system and a medium, relates to the related technical field of data management, and comprises the following steps: uploading original CAD files and management tasks; developing a cross-platform intermediate engine and establishing information interaction of the cross-platform intermediate engine with each target platform; scanning the original files and the management tasks, driving a coder to perform platform classification coding based on CAD element characteristics, and driving a decoder to perform optimal expression collapse based on platform migration operators; issuing N decoding operators to perform platform graph reconstruction and storage; performing graph matching calling and uniform template integration, and visualizing as real-time project graphs. The application solves the technical problems of information loss, semantic attenuation, low collaborative efficiency and difficulty in realizing real-time synchronization and visual management of design elements in a heterogeneous environment during cross-platform data exchange in the prior art, and achieves the technical effects of improving the fidelity, intelligent interoperability and real-time collaborative management of cross-platform CAD design data.

The application discloses a kind of three-dimensional pipeline design drawing generation systems based on MBD size marking, it is related to engineering graphics automation field;The system includes: MBD semantic analysis module is used to extract the size, tolerance and technical requirements in three-dimensional model and establishes topological correlation;View projection planning module, the main extension direction of pipeline is calculated using principal component analysis, and the projection plane is rotated to minimize the overlap rate;Intelligent layout optimization module, construct the cost function including repulsion and compact weight, and use simulated annealing algorithm to iteratively optimize the marking position to eliminate occlusion;Drawing rendering module, output the engineering drawing and interactive SVG file in line with national standard.In addition, the system also has version change automatic cloud line marking and local magnified drawing automatic generation function.The application effectively solves the problems of information loss, layout confusion and change tracking difficulty in the process of three-dimensional to two-dimensional, and realizes the automated production of high-quality pipeline engineering drawing.

This invention discloses a lightweight target detection method and device for power operation scenarios, belonging to the field of computer vision and target detection technology. Addressing the challenges of small-scale targets, complex backgrounds, and limited computing power of edge devices in power inspection scenarios, this invention proposes a lightweight detection model based on YOLOv8. The method introduces an SPD-Conv module into the backbone network to reduce downsampling information loss, and a C2f-PA module (containing PConv and the dependency fusion module ARFM) into the neck network to enhance the expression of relationships between the main target (worker) and its associated fine-grained targets. GhostConv is also introduced into the detection head to reduce computational overhead. Simultaneously, a multi-level distributed perception knowledge distillation framework is constructed to distill and train the student model. This method improves the detection accuracy for small and fine-grained targets while reducing the number of model parameters and computational complexity, making it suitable for real-time detection of UAV-borne platforms and other resource-constrained edge devices.

This application provides a transformer fault alarm and location method and system, belonging to the field of intelligent monitoring and fault diagnosis of power equipment, to solve the problems of information loss from multidimensional heterogeneous data, delayed fault warning, and ambiguous physical location in related technologies. The method embeds the multidimensional data tensors acquired by a sensor array into a low-dimensional smooth manifold to construct a neural differential manifold model to characterize the continuous evolution of states, and identifies fault attractors at the macroscopic level based on causal emergence theory. It determines the key directions of state evolution by calculating the gradient of the holographic attention field on the manifold, and maps it back to the original data space to generate a difference field, thereby achieving accurate physical location and early alarm of the fault source. The system includes a holographic sensing array implementing the above method, a manifold computing processing unit, and an interactive terminal.

This invention discloses a GPS interference / spoofing detection method and system based on a lightweight large language model. The method encapsulates GPS time-series data and key features into structured text with complete semantics, constructs a Prompt, and generates JSON tags containing multi-dimensional information. The lightweight model is fine-tuned using the Prompt-JSON data, and optimized and compressed using pruning, quantization, and distillation techniques. The optimized model is deployed to an embedded device, and inference is performed through a lightweight inference engine to output a structured early warning protocol. This invention solves the problems of information loss, "black box" issues, and large model edge deployment challenges of traditional methods, achieving efficient and accurate detection of GPS attacks in resource-constrained environments.

This invention discloses a sparse attention modeling method and system for long texts based on four-dimensional spatiotemporal coordinates, belonging to the field of large-scale model long text optimization and sparse attention technology. Existing full-scale attention mechanisms for large models suffer from high memory consumption, large inference latency in long texts, attention diffusion, and failure of long-range dependencies. Various sparse attention algorithms are mostly based on window truncation and fixed-interval sampling, which are empirical simplification strategies that easily lose key long-range information and disrupt the text's logical temporal structure. This invention utilizes four-dimensional topological constraints to construct an explicit sparse attention mask, abandoning the fixed-window sampling mode to achieve spatiotemporal topological adaptive sparse attention modeling. This invention preserves long-range temporal correlations through temporal causal constraints, isolates invalid cross-cycle interference through contextual branch constraints, preserves key dependencies in the inference chain through logical hierarchy constraints, isolates mixed information from multiple subjects through identity coordinates, and dynamically weights key token attention weights based on cognitive memory strength. This invention significantly reduces the computational power and memory overhead of long texts while fully preserving long-range causality, logic, and contextual correlations, completely solving the core pain points of degradation and information loss in traditional sparse attention for long texts, and significantly improving the stability and logical consistency of inference for millions of ultra-long texts.

This invention discloses a stock price trend prediction method based on multi-level dynamic graph fusion, relating to the fields of financial technology and artificial intelligence data analysis. The method mainly comprises three parts: multi-channel temporal feature extraction, collaborative modeling of macro and micro spatial structures, and multi-level feature fusion prediction. The steps include: First, constructing a multi-channel technical indicator sequence based on historical stock trading data, and using a multi-channel attention pooling GRU network to extract differentiated temporal features in parallel; Second, constructing a stock association hypergraph based on industry and concept labels, introducing graph information loss (GIL) as a feedback signal to drive an adaptive hyperedge reconstruction mechanism, dynamically adjusting hyperedge weights to filter noise and capture macro market hotspots; Simultaneously, constructing a simple graph within the hyperedge and combining contrastive learning constraints to semantically align stocks with similar technical patterns to enhance the discriminative power of micro local features; Finally, integrating macro hypergraph features and micro simple graph features through a cross-graph fusion module, concatenating them with temporal features, and inputting them into the prediction layer to complete the prediction of the stock price trend at the next moment. This method overcomes the limitations of existing static graph models in dynamically capturing high-order stock correlations and fine-grained technical pattern resonances, providing a new approach for financial spatiotemporal data mining.