477 results about "Information loss" patented technology

The invention provides an RAG knowledge base construction method and system based on hierarchical semantic indexes. The method belongs to the cross technical field of artificial intelligence and information retrieval. The method comprises the following steps: performing multi-level semantic analysis on an input original document set to generate document semantic hierarchical structure data; constructing a hierarchical semantic index tree based on the document semantic hierarchical structure data; performing dynamic knowledge graph initialization according to the hierarchical semantic index tree to generate an initial dynamic cognitive graph; and collecting real-time interaction data through a user feedback interface and a new data access module, and performing incremental updating on the initial dynamic cognitive map to form a knowledge representation system supporting life cycle evolution. Through multi-level semantic analysis and construction of a hierarchical semantic index tree (HSIT), deep semantic analysis can be performed on an original document set, the context integrity of knowledge is ensured, structured storage is realized, the knowledge can be expressed and stored more accurately, and information loss or semantic ambiguity is avoided.

The invention discloses a document knowledge base LLM intelligent question and answer method, device and equipment and a storage medium. The method comprises the steps that dynamic partitioning is conducted on a to-be-processed document, and vectorization and index storage are conducted on knowledge blocks; vector retrieval and keyword retrieval are carried out according to the natural language question, a vector retrieval result and a keyword retrieval result are fused, an answer is obtained through LLM, and the answer is fed back to a user after being safely filtered; when it is detected that the to-be-processed document is updated, the vector library and the index are synchronously updated, and the cue word template and the partitioning strategy are periodically optimized, so that the problem of form picture information loss can be solved, the integrity of document information analysis is guaranteed, the situation that the partitioning strategy is single is avoided, the multi-hop recall rate is increased, and the problem of semantic missing is avoided; a partitioning mechanism is reasonable, retrieval precision is improved, updating cost is reduced, data security is improved, implementation is convenient, universality is good, and the speed and efficiency of LLM intelligent question answering of the document knowledge base are improved.

The invention discloses a pavement throwing detection method and device based on deep learning, electronic equipment and a program product. The method is realized through a trained detection model, the model adopts an LMSADet detection head, a multi-scale feature extraction and space attention mechanism is introduced into a task branch, and multi-scale modeling is decoupled from a backbone network and a neck network and integrated to the detection head so as to fit a detection task to directly optimize local details and scale differences of a throwing target. In order to suppress background interference and improve the recognition effect of fuzzy boundaries, the neck network is added into an MSHA module so as to efficiently capture the semantic relation between the thrown object and the background and enhance the regional understanding ability. A C3ESP module is introduced into the backbone network, deep features are extracted through stacking depth separable convolution, and information loss is avoided in combination with residual optimization fusion; meanwhile, a PEMA attention mechanism is introduced, the importance of different receptive field features is dynamically adjusted, the model focuses on key features, data information is captured more comprehensively, and therefore the detection performance is remarkably improved.

The invention discloses a fireman occupational health risk dynamic prediction method based on multi-modal data fusion, relates to the technical field of occupational health risk assessment, and aims to solve the problems of information loss, difficulty in capturing cross-modal complex dependence, lack of interpretability of prediction results and the like when unstructured data is processed in the prior art. The method comprises the following steps: intelligently analyzing a multi-modal document, and analyzing unstructured physical examination information into structured data; performing semantic standardization and knowledge graph dual verification of domain knowledge enhancement; the time-space-static cross-modal attention deep fusion network is used for modeling physiological, dynamic and situational data; and carrying out interpretable risk prediction and attribution. By adopting the technical scheme, the method can realize early, accurate, dynamic and high-credibility prediction of the occupational health risk of the firefighter, provides quantitative attribution, and improves the transparency and application value of the system.

The invention discloses a retrieval enhancement generation method and system based on a composite knowledge base, and belongs to the technical field of retrieval enhancement generation. Obtaining an original document, dividing the original document into text blocks according to paragraphs, and storing the text blocks into a vector database to form a text vector library; constructing a knowledge hypergraph based on the text blocks, and storing the knowledge hypergraph into a vector database to form a knowledge hypergraph database; obtaining cases and storing the cases into a vector database to form a case library; and obtaining a question input by a user, and generating a final answer by using a large language model based on a composite knowledge base formed by the text vector base, the knowledge hypergraph database and the case base. By constructing a composite knowledge base, the problems of context breakage, information loss, lack of professional expression and the like existing when an existing retrieval enhancement generation method is used for processing long documents are solved.

The invention discloses a multi-modal multi-scale retrieval enhancement generation method, system and device applied to external knowledge questions and answers and a medium. The method comprises question perception, multi-modal multi-scale query fusion coding, dense recall and answer generation. Analyzing a key query phrase from the question through a fine-tuned instruction language model, and accurately positioning a region of interest corresponding to the phrase in an image by using an open set visual positioning model; multi-source information is compressed and distilled into an optimal query vector through a deep fusion network integrating multi-head self-attention and an information bottleneck theory; executing a maximum inner product search to recall related knowledge; guiding the large language model to synthesize all information to generate a final answer; the system, the equipment and the medium directly perform feature fusion in the vector space based on the method, so that challenges such as information loss and cascading errors caused by a traditional normal form can be effectively dealt with, high correlation and high accuracy of retrieval knowledge are ensured, and accurate and reliable image-text questions and answers are realized.

The invention discloses a software requirement modeling method and device based on a natural language, equipment and a medium, and relates to the field of software requirement modeling. According to the software demand modeling method and device based on the natural language, the equipment and the medium, through automatic demand analysis and model generation, the manual workload is greatly reduced, and meanwhile, errors caused by manual understanding deviation are reduced; close connection among the steps is ensured by sharing intermediate representation and a unified data model, output of the previous step provides structured information for input of the next step, and information loss or distortion is avoided; a demand provider is allowed to participate in a model confirmation process through an iterative demand acquisition and modeling framework of man-machine cooperation, and the quality of a demand model is continuously improved; by means of the model consistency analysis technology based on the data flow dependency relationship, the problem of inconsistency between the requirement specification and the software model can be automatically detected, and the software product quality is improved.

The invention discloses a static database self-adaptive fuzzification desensitization implementation method for highly sensitive data in the power industry. The method comprises the following steps: a data preprocessing module, a sensitivity analysis module, a fuzzification algorithm module, a self-adaptive mechanism module and an effect evaluation and verification module. Compared with the prior art, the method has the advantages that 1, self-adaptive dynamic adjustment is performed: a desensitization strategy is automatically optimized according to data characteristics and use scenes, and privacy protection indexes (such as k-anonymity values) are improved by 30%-50%; 2, multi-dimensional balance: the information loss rate is reduced by 10%-20%, the query accuracy is improved by 5%-10%, and safety and service requirements are considered; 3, system expansibility: a plug-in architecture supports dynamic loading of new algorithms and rules, and a configuration management module realizes strategy real-time adjustment; and 4, compliance guarantee: through a law and regulation rule base and a compliance check algorithm, the compliance is ensured to meet the standard.

The invention belongs to the technical field of target detection, and discloses an unmanned aerial vehicle image target detection method based on an improved RTDETR model, and the method introduces an FAPPA module in a backbone network, and effectively alleviates the information loss problem of small target features in a deep network. According to the method, a feedforward neural network of an AIFI module in an encoder is replaced by an EDFFN module, and high-frequency features such as edges and textures are selectively enhanced by using a patchwise FFT strategy and learnable frequency domain weights. Besides, according to the method disclosed by the invention, adaptive alignment of multi-scale features is realized by utilizing a Zoomcat module, and local details and global semantics are fused by utilizing a convolution attention double-path architecture of a CAFMFusion module, so that complementary information of different levels is fully utilized. According to the method, the unmanned aerial vehicle image target detection capability can be remarkably improved, and small target features in a complex background can be identified.

The invention discloses a multi-model game method for multi-scene intelligent safety management. The method comprises the following steps: acquiring multi-domain original data and generating information structure features; the heterogeneous indexes are uniformly mapped into information loss components based on the features, and stable routing of the game template is carried out in combination with sequential statistics and hysteretic decision; performing game payment-oriented decision sensitive subspace compression on the unified semantic state output after routing; a dynamic risk-to-go (dynamic risk surplus) time difference method is adopted to generate step risk signals meeting total amount conservation, and the step risk signals are injected into a compression space; and through online estimation of the spectral radius of the Jacobian matrix of the optimal response operator, an equilibrium vulnerability index is generated, and stability management and closed-loop adjustment of an equilibrium point are realized. According to the invention, the decision adaptability, stability and robustness of the intelligent system in a dynamic confrontation environment can be improved.

The invention relates to the technical field of artificial intelligence, in particular to an electric energy quality disturbance denoising method based on artificial intelligence. The invention discloses a power quality disturbance denoising method based on artificial intelligence. The method comprises the following specific steps: S1, collecting and marking signal data; s2, time-frequency domain weighted fusion preprocessing including time domain normalization processing, frequency domain wavelet transform, adaptive weight calculation and time-frequency fusion representation is carried out; comprising a dual-path feature encoder module, a self-adaptive feature fusion module, a gating jump connection decoder module, a denoising signal purification module, calculation of purification constraint terms, calculation of a total loss function and model iterative training and parameter updating, and S4, power quality disturbance denoising. The problems that in the prior art, information loss is likely to happen, feature extraction is insufficient, the denoising effect is unstable, and the precision of electric energy quality analysis is limited are solved.

The invention discloses a drainage pipeline defect identification method based on improved YOLOv10, and belongs to the field of defect detection and the field of image identification. The method comprises the following steps: collecting a drainage pipeline defect image, and executing preprocessing operation on a collected data set to obtain a data set serving as a training model; a backbone network and a head network of the YOLOv10 model are improved, and an improved YOLOv10 model is obtained; and training and verifying the improved YOLOv10 model by using the obtained data set serving as a training model to obtain an optimal YOLOv10 model of the drainage pipeline defect, and completing drainage pipeline defect identification. According to the method, a smoother category transformation basis is provided, information loss is reduced, the accuracy of feature extraction is improved, and the processing capacity of the model on small objects and low-resolution images is optimized. Aiming at the problems that the collected drainage pipeline defect image is not uniform in image illumination, the image resolution after framing processing is not high and the like, the detection efficiency and the detection precision of the model are improved.

The invention discloses a multi-modal image fusion model construction method, and particularly relates to the technical field of image fusion model construction. Interactive modeling of a modal contribution imbalance coefficient and a confidence coefficient estimation anomaly coefficient is introduced in a multi-modal image fusion process; a causal association between confidence prediction fluctuation and modal weight extreme is converted into a quantifiable cross-modal complementarity degeneration index, and the cross-modal complementarity degeneration index is used as a core adjustment mechanism to dynamically optimize a fusion weight and training regularization, so that the problem that a single mode is excessively amplified or mistakenly weakened is effectively inhibited on a fusion strategy level; according to the method, the structure retention capability of the model in edge details and texture regions is remarkably improved, and artifacts and information loss caused by modal complementarity degradation are avoided.

The invention provides a hybrid AI adaptive optimization system based on agent context engineering. The hybrid AI adaptive optimization system comprises a context knowledge base, a generation module, a feedback analysis module and a knowledge integration module. According to the method, the problems that in an existing context optimization technology based on a large language model (LLM), context Bias is simplified, Context Collapse is caused by overall rewriting, and the learning process is high in delay, high in cost, unstable and the like are solved.

The invention relates to the field of machine learning, particularly discloses an artificial intelligence-based power system fault analysis and diagnosis method, and effectively solves the problem of information loss caused by neglecting a key waveform form in a transient signal in the prior art through a local feature extraction and serialization module. An original signal is converted into a local feature sequence with more characterization significance. Aiming at the averaging bottleneck of an existing model in an information aggregation stage, a traditional feature compression method is abandoned, and a sequence information aggregation and decision-making mechanism is provided. According to the mechanism, a context sensing sequence is regarded as a probability event, and modeling is carried out on the sequence from three orthogonal dimensions of a content center, time sequence dispersion and distribution uncertainty by calculating feature expectation, time sequence variance and information entropy of the context sensing sequence. The method can deeply insight and quantify the essential difference of different events in the time sequence dynamic evolution mode, thereby fundamentally solving the problem of misjudgment caused by feature confusion.

The invention discloses a multi-source heterogeneous time series data compression and motor fault prediction method and system, and the compression method achieves the intelligent compression and feature enhancement of time series data through the cooperation of multi-scale feature extraction, SE weighted fusion, mixed pooling and aggregator collapse. According to the architecture, reasonable reduction of the sequence length and effective improvement of the feature depth can be completed at the same time, on the premise that key fault features are reserved, the technical problem of multi-source heterogeneous sensor data fusion in an industrial scene is solved, and the limitation of serious information loss of a traditional dimension reduction method is overcome.

The invention is suitable for the technical field of computer vision, and provides a polyp image segmentation method based on a basic model and feature guidance. Two innovative lightweight modules are specifically introduced into an SAM image encoder to compensate for information loss: 1) a long-distance context injection module (LCI) is introduced behind a local window self-attention layer, and global context information is efficiently injected for local features by using a shared learnable token (FeatureToken); and 2) introducing a frequency-guided attention module (FGA) behind the global self-attention layer, and guiding an attention mechanism by the module by combining high-frequency component (HFC) features of the image with a low-rank (LoRA) learnable semantic token so as to supplement key edge and texture details for global features. According to the method provided by the invention, the leading segmentation performance can be obtained on a plurality of public polyp segmentation reference data sets only by finely adjusting a small number of parameters, and the method is obviously superior to the prior art.

The invention relates to the technical field of automatic driving, in particular to a multi-modal three-dimensional target detection method, which comprises the following steps of: in a first stage, firstly acquiring neighbor depth information of point cloud to enhance image features, and then further enhancing image edges and reducing semantic confusion by utilizing wavelet transform; and then a cross attention mechanism is introduced to realize effective fusion of the enhanced image features and the point cloud, an initial region suggestion is obtained, and bounding box classification and prediction in the first stage are realized. And in the second stage, designing a double-attention module based on grid features to supplement more geometric detail information, acquiring more context information and position information by utilizing the initially suggested spatial features and channel features generated in the first stage, and then introducing a self-attention mechanism to dynamically distribute interaction weights of the grid features, so as to realize the self-attention interaction of the grid features. Rich context information is effectively captured, a local geometric structure is better coded, information loss is reduced, and the detection precision of the model is improved.

The invention belongs to the technical field of underground structure prediction, and particularly discloses a tunnel settlement prediction method and system based on experience information and data driving, and the method comprises the steps: receiving influence factor data of multiple aspects of tunnel settlement, and carrying out the sampling based on the influence factor data, and obtaining a tunnel settlement data set; constructing a neural network model, and fusing the settlement development rule as a constraint condition into the loss function to obtain a total loss function fusing empirical information loss and data loss; performing hyper-parameter optimization on the neural network model by adopting a Bayesian optimization algorithm to determine an optimal hyper-parameter; training the neural network model, and updating the neural network model by using the total loss function to obtain a trained prediction model; and performing tunnel settlement prediction performance evaluation on the prediction model, and performing interpretability evaluation on the prediction model by adopting an SHAP interpretation method. According to the invention, the accuracy and efficiency of the model prediction result can be improved.

The invention provides a PDF (Portable Document Format) optimized translation method based on intelligent text detection, which comprises the following steps of: 1, analyzing a PDF document, respectively identifying a text region and an image region in the PDF document, and extracting a native text block of the text region in the PDF document and structured position information of the native text block; the recognition result and the native text block are combined into a unified structured data set; 3, performing machine translation by adopting a window translation strategy in combination with the context, and extracting a target translation; adjacent text blocks are dynamically associated during translation, and a target text range is defined through boundary markers; and 4, dynamically optimizing the text layout according to the target translation length, and backfilling a translation result to an original position. According to the method, the translated text can be accurately and seamlessly backfilled to the corresponding position of the original image, meanwhile, the content of the original image is ensured not to be shielded, and visual disorder or information loss caused by text coverage is avoided.

The invention discloses a medical multi-mode distributed machine learning system based on a prototype, and relates to the field of machine learning. According to the method, data of the missing mode is compensated through the mode prototype. The modal prototype is generated by aggregating data of all clients and contains rich semantic information. In the training process, when data of a certain mode is missing, a mode prototype can be used as a mask, and global priori knowledge is introduced into the missing mode, so that information loss caused by mode missing is compensated. According to the invention, the representation of the client and the global prototype are aligned through modal comparison training. In each communication round, the client receives the global model and the global prototype, and aggregates the global model and the global prototype with the local model and the local prototype respectively to form a personalized model and a personalized prototype. By comparing the local representation with the personalized prototype, the consistency between the local representation and the global prototype can be enhanced, thereby reducing the heterogeneity between clients.

The invention provides a multiple-input-multiple-output soft measurement method and device and a medium. The method comprises the steps that historical multi-sampling-rate process variable data and historical multi-quality variable data are acquired; obtaining a plurality of historical task specific features from the historical multi-sampling-rate process variable data; inputting the plurality of historical task specific features to obtain historical cross-task sharing interaction features; training according to the historical cross-task sharing interaction features, the plurality of historical task specific features and historical multi-quality variable data to obtain a trained feature interaction hybrid expert network model; and obtaining a plurality of online prediction results according to the trained feature interaction hybrid expert network model. According to the method, the device and the medium, the problems that information loss, time alignment errors and task negative migration are easily caused when different variable sampling frequencies are inconsistent in an existing multi-input multi-output soft measurement method, an irregular sampling phenomenon exists, and the prediction precision of multiple quality indexes is not high can be solved.

The invention discloses a semi-supervised prognosis prediction system based on irregular sampling medical data pre-training, and the system comprises a clinical electronic medical record data collection and preprocessing module which automatically collects original clinical electronic medical record EHR data from a medical database; and the pre-training module is used for receiving the patient feature vector sequence output by the clinical electronic medical record data acquisition and preprocessing module and carrying out feature representation learning on the preprocessed clinical time sequence data by utilizing a combined multi-task self-supervised learning mechanism. And the classifier fine tuning and pseudo-label iterative optimization module is used for training the pre-trained model through fine tuning of samples with labels and carrying out iterative optimization through guidance of pseudo-labels to obtain a prediction result. And the application display module is used for displaying and outputting a post-hospital-admission vital sign sequence and a prediction result. According to the method, irregular sampling and missing data are effectively processed, information loss is avoided, and the prediction capability of the model and the performance of the model under the conditions of data imbalance and label scarcity are improved.

The invention relates to a real-time semantic segmentation method based on hierarchical feature fusion and channel attention enhancement, and belongs to the technical field of computer vision. The method comprises a training stage and a reasoning stage. In a training stage, firstly, data enhancement is performed on an input image through an image enhancement strategy, then a semantic segmentation model taking hierarchical feature fusion and channel attention enhancement as a core is constructed, and training is performed by adopting a deep supervision technology to obtain an optimal weight. In the reasoning stage, the optimal weight and the model are loaded, and real-time semantic segmentation is carried out on the image by utilizing a GPU (Graphics Processing Unit). The semantic segmentation model can effectively solve the problems of local detail information loss and intra-class semantic tag inconsistency, and has higher reasoning speed and higher segmentation precision compared with other similar methods.

The invention provides an occupied grid prediction method and device, electronic equipment and a readable storage medium. The method comprises the following steps: acquiring continuously acquired multiple frames of visual images and laser point cloud data; converting the visual image to a point cloud scene to obtain image point cloud data; fusing the image point cloud data and the laser point cloud data to obtain fused voxel features; performing motion perception prediction on the dynamic voxels in the fused voxel features through the occupancy flow to obtain motion perception features corresponding to the dynamic voxels; and performing state prediction based on the motion perception features to obtain an occupation prediction result. Visual images are converted into image point cloud data, and the image point cloud data and laser point cloud data are fused in a 3D space to obtain voxel features, so that information loss caused by multi-mode mutual projection can be avoided, and meanwhile, motion perception is directly performed on fused voxels through an occupation stream, so that complete space information can be effectively captured and utilized, and the accuracy of the motion perception is improved. Therefore, more accurate and comprehensive occupied grid prediction is realized.

The invention relates to the field of sensor data processing, and discloses an MEMS drift correction method based on data dynamic sampling and multi-source fusion, and the method comprises the following steps: S1, collecting MEMS three-axis acceleration original data at a first sampling frequency, and synchronously collecting environment temperature, humidity and air pressure data at a second sampling frequency, the first sampling frequency is higher than the second sampling frequency, the basic drift correction model is established by using the low-frequency data to capture the macroscopic and slow-varying relationship between the environmental factors and the drift, then the residual refinement model is established for the high-frequency data, and the high-frequency residual is specially processed; high-frequency drift components related to the environmental change rate are captured, through the two-stage fusion strategy, the accuracy of the model in the macroscopic trend is guaranteed, high-frequency data are fully utilized to improve the instantaneous correction precision, information loss caused by simple downsampling in a traditional method is avoided, and false signals introduced by rough interpolation are also avoided.

The invention provides a data reasoning method and a data reasoning device, and the method comprises the steps: obtaining to-be-processed data, inputting the to-be-processed data into a pre-training encoder, and obtaining a first implicit feature outputted by the pre-training encoder; wherein the pre-training encoder is obtained by jointly training a data distribution transformation model and a discrimination model, and the first implicit feature obeys first continuous probability distribution; and inputting the first implicit feature into a decoder so as to restore the first implicit feature through the decoder to obtain first generated data. According to the method and the device, the traditional Mel spectrum is no longer used as an intermediate representation form, and the implicit features are used, so that the feature information of the original data can be better reserved, the generation capability of the model is further improved, the higher-precision and more diversified generation effects are realized, and the problem of information loss in the traditional generation model is solved.

The invention discloses a building construction safety monitoring method and system based on multi-source sensing fusion, and the method comprises the steps: firstly carrying out the precise space-time reference alignment of an original video stream and a UWB positioning data package, so as to solve the inherent asynchronism and conflict problems of heterogeneous sensor data, and guarantee the accuracy of subsequent data fusion; then, parallel feature extraction and preliminary entity association are carried out on the aligned data, and scene constraint-based dynamic confidence evaluation is carried out on a preliminary fusion object by innovatively utilizing BIM semantic data, so that priori knowledge of a building environment is fully utilized, and the reliability of sensor data is dynamically optimized; therefore, the defect of information loss caused by environmental influence, shielding or signal interference of a single sensor is overcome. And finally, the system performs adaptive fusion decision and state optimization according to the obtained confidence score, and outputs a high-confidence object state. In this way, the accuracy, the real-time performance and the reliability of building construction safety monitoring are remarkably improved.

The invention is suitable for the technical field of computer vision, and provides a self-supervised monocular depth estimation method based on wavelet feature enhancement, and the method comprises the steps: firstly obtaining a two-dimensional target image and an adjacent image frame, and then generating a multi-scale first feature map and a multi-scale second feature map based on a main encoder and a wavelet feature extractor, the method comprises the following steps of: constructing a multi-scale second feature map, constructing a wavelet guide feature fusion module, injecting and enhancing high-frequency detail information of the multi-scale second feature map to a multi-scale first feature map based on the wavelet guide feature fusion module, generating a plurality of third feature maps, generating a multi-scale depth estimation map based on a decoder, and finally obtaining a multi-scale depth estimation map based on relative pose information, the multi-scale depth estimation map and a luminosity consistency error. And carrying out self-supervised training on the to-be-trained deep learning model. According to the method, the limitation of high-frequency information loss in a sampling process in a traditional method can be overcome, structural details and boundary information in depth estimation are effectively enhanced and supplemented, and a depth map with higher quality is obtained.

The invention discloses a fusion method for balancing medical text positive and negative sample training, belongs to the field of medical text classification crossing natural language processing and medical informatics, and solves the problems of long text information loss, labeling noise interference and positive and negative sample imbalance in the prior art. The method comprises the steps of preprocessing a medical text, defining five types of medical entities such as disease diagnosis and the like, and labeling and establishing entity association; segmenting a long text by using a sliding window, extracting text features based on BioBERT, adding a gradient inversion layer, and setting a classifier and a discriminator; in positive and negative training, a weighted cross entropy loss function is used for learning real label mapping of a text in positive training, and a noise suppression loss function is used for reducing the influence of noise on model learning in negative training. According to the method, the delirium-symptom-containing medical record can be accurately identified, and the accuracy and robustness of medical text classification are improved.