33 results about "Simple Features" patented technology

The invention discloses a double-branch edge detection method and system based on CNN and Transform deep interactive fusion, and relates to the technical field of computer vision and deep learning, and the method comprises the steps: extracting multi-scale local features through a CNN-based fine semantic edge branch by employing an OfficientNet-B2 backbone network, and carrying out the deep supervision in combination with a side output structure; meanwhile, through a global context branch based on lightweight Transform, an efficient local attention mechanism is adopted to capture a long-distance dependency relationship of the image; the core of the invention lies in that a cross-attention module is designed and realized, local features extracted by a CNN branch are used as a Key and a Value, and features of a Transform branch are used as Query, so that deep interaction and fusion of the two branches on the feature level are realized, instead of simple feature splicing; in addition, the method also adopts a uniform loss function for balancing dynamic weighted combination of binary cross entropy loss, focus loss, Dice loss and the like to perform model optimization.

The invention discloses a network anomaly detection method and system for STN equipment. The method comprises the following steps: collecting network traffic data of the STN equipment; performing multi-dimensional feature extraction on the network flow data to obtain a time sequence feature set, a statistical feature set and a protocol feature set; performing cross-domain information fusion on the time sequence feature set, the statistical feature set and the protocol feature set to generate a fusion feature vector; and inputting the fusion feature vector into a pre-trained network anomaly detection model, and outputting a network anomaly detection result. Through cooperative extraction of a time sequence feature set, a statistical feature set and a protocol feature set, the limitation of the expression ability of single-dimensional entropy features in the prior art is overcome. A feature fusion mechanism is introduced, contribution weights of different feature domains to anomaly detection are dynamically learned, and compared with simple feature splicing in the prior art, the detection accuracy is improved. And the fusion feature vector is identified based on a network anomaly detection model, so that the robustness is improved.

The invention relates to a semi-supervised medical image segmentation method based on double-path feature alignment, and belongs to the field of medical image processing. Aiming at the problems that medical image annotation is scarce, high-frequency information is sensitive and feature fusion is not accurate in an existing method, a three-branch network architecture is provided, a main network extracts original image features, and an auxiliary network processes high-frequency and low-frequency images subjected to discrete wavelet decomposition respectively; cross-branch feature deformation correction and fusion are realized through a wavelet feature alignment module (encoder) and a simple feature alignment module (decoder); and unsupervised loss dynamic training is carried out in combination with supervised loss and pseudo-label, so that label dependence is relieved. According to the method, the segmentation precision and the boundary detail retention capability are effectively improved, the noise robustness is enhanced, the annotation data dependence is remarkably reduced, meanwhile, the cross-branch fusion efficiency is improved through a feature alignment mechanism, and the method has high clinical application value.

The invention relates to the field of image content understanding, and discloses an image geographic positioning system and method based on geographic feature extraction, and the system comprises a feature extraction module which is used for extracting a visual feature vector, a GPS feature vector, a text position description feature vector and a text scene description feature vector from multi-modal input; the comparative learning module is used for realizing multi-modal feature alignment through comparative learning of the visual features, the GPS features, the text position description features and the text scene description features; and the data set construction module is used for fusing the multi-modal features to generate geographic feature vectors and constructing a retrieval vector data set. By adopting the technical scheme of multi-modal feature fusion and cross-modal comparative learning, the technical effect of improving the geographic positioning precision and generalization ability is achieved. Compared with a scheme depending on a single mode or simple feature splicing in the prior art, the problem of semantic gaps caused by mode information splitting in a traditional method is solved.

The invention discloses a multi-modal self-adaptive online shopping comment authenticity and score credibility evaluation system, which belongs to the technical field of online shopping comment analysis and comprises an enhanced text semantic understanding module, an image authenticity verification module, a user behavior deep modeling module, a self-adaptive weight distribution module and a two-dimensional prediction module. According to the multi-modal self-adaptive online shopping comment authenticity and score credibility evaluation system, the false comment recognition capability is greatly improved, the misjudgment rate is effectively reduced, comment authenticity and score two-dimensional evaluation is achieved, the multi-modal fusion effect is better than that of simple feature splicing, high-concurrency processing is supported, complex scenes can be covered, and the online shopping comment authenticity and score credibility evaluation method is suitable for large-scale popularization and application. And the engineering practicability is high.

The invention relates to the technical field of building energy conservation and intelligent buildings, in particular to a load simulation calculation method and system for building energy efficiency, and the method comprises the steps: collecting multi-source data, and constructing a multi-dimensional data set; building load data is deconstructed into three kinds of heterogeneous modalities including a class sequence, a class image and a class video; calculating a dynamic correction coefficient based on the heat storage effect of the enclosure structure; respectively processing three types of heterogeneous modes through a BiGRU network, an STNN network and a 3DCNN network, and fusing and outputting an initial load by adopting Stacking ensemble learning; optimizing and outputting a final load prediction value through a multivariate feature recurrent neural network; the system integrates a multi-source data acquisition module, a multi-modal feature processing module, a physical correction module, a hybrid prediction module and a time sequence optimization module, and is deployed at an edge node through a knowledge distillation compression model. According to the method, a physical model and a data driving method are fused, and the problems that in the prior art, data processing is insufficient, feature construction is simple, a model fusion shallow layer is insufficient, and calculation architecture and timeliness are insufficient are effectively solved.

The invention provides a scene perception depth completion method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of environment perception optimization. The depth completion method comprises the following steps: acquiring image data of a target scene and millimeter wave radar point cloud data; based on ResNet and PointNet, feature extraction is carried out on the image data and the millimeter wave radar point cloud data, and a multi-modal feature map of the target scene is generated; and based on the U-Net and the CSPN, performing depth completion on the multi-modal feature map to obtain a depth completion map of the target scene. The method and the device are used for at least solving the problems that in the prior art, single-mode or simple feature splicing is still dependent, the expression ability of different sensor data and the potential of internal feature fusion cannot be fully mined, and the depth completion performance and the application range are poor, so that a depth completion graph generated in a severe environment is difficult to provide data support. The method is suitable for environment perception optimization scenes.

The invention discloses a motion and appearance decoupling multi-scale interaction and semantic calibration video abstraction method, and belongs to the technical field of computer vision. The method comprises the following steps: respectively extracting multi-scale motion features and appearance features from an input video frame sequence to obtain corresponding feature pyramids; inputting the motion feature pyramid and the appearance feature pyramid into a video abstract model, and predicting an initial frame-level importance score sequence through the model; then, interframe time sequence dependence is captured through bidirectional LSTM, scores of key action boundaries are strengthened, redundant frames are reduced through diversity punishment, and an optimized frame-level importance score sequence is output; and finally, optimizing frame distribution by applying a time interval constraint, and selecting a key frame through a greedy algorithm to generate a video abstract. According to the method, the problems of motion and appearance feature coupling, insufficient multi-scale characterization, simple feature fusion mechanism and modal weight imbalance in the existing video abstraction method are effectively solved, and the effectiveness of the method is proved by experimental results on a SumMe and TVSum reference data set.

The invention provides a feature fusion method, system and device for multi-modal data and a medium, and the method comprises the steps: obtaining text data and video data, carrying out the feature extraction of the text data and the video data, and obtaining a text feature and an image feature; dividing the text features and the image features into a plurality of text segments and a plurality of image segments respectively; constructing a memory unit, and generating historical information of each text segment or image segment by using the memory unit; inputting all the text segments and the image segments and the corresponding historical information into a feature extraction model for feature extraction to obtain depth text features and depth image features; and fusing the depth text features and the depth image features to obtain fused features. According to the method, the problems that in the prior art, dynamic features with long-distance dependence and obvious spatial-temporal features are difficult to extract, and key information is lost due to simple feature splicing or weighted average during multi-modal feature fusion are solved.

The application relates to the field of alloy defect detection, and specifically discloses a praseodymium-neodymium alloy nondestructive detection method and system based on acoustic feature analysis, which comprehensively captures defect information contained in an original probe signal from two complementary physical perspectives of instantaneous dynamic characteristics and frequency band energy distribution by simultaneously adopting Hilbert-Huang transform and wavelet packet transform. Further, the scheme discards simple feature splicing, and instead utilizes canonical correlation analysis as an information decoupling tool to online decompose two groups of original feature vectors into a shared part describing defect commonality and unique information parts respectively representing the unique resolution capabilities of HHT and wavelet packet. Finally, the three decoupled components are structurally recombined to form a fusion feature vector which can effectively eliminate redundancy, amplify differences and has higher information density, thereby providing a clear structure and highly refined input for a subsequent classification model.

The application discloses a kind of multi-variable DC fault arc adaptive detection method based on similarity measure and transfer learning, to the system output current of different time period applies wavelet packet decomposition and coefficient reconstruction processing after extraction feature, and input to state identification model to carry out multi-period fault arc judgment, to current working condition through machine learning evaluation feature similarity measure value D and model output probability distribution P joint distribution mode, according to this, the extracted feature is distinguished into unidentifiable working condition type data and identifiable working condition type data, using field adaptive strategy to the simplest feature group of unidentifiable working condition type data carries out transfer learning, to update state identification model, to realize real-time accurate detection of DC fault arc under complex and variable working condition in this way.

This invention belongs to the field of artificial intelligence technology and discloses a method, apparatus, device, and storage medium for generating features for network field strength prediction. The method includes: acquiring network field strength prediction samples; performing label interpolation on the network field strength prediction samples to generate rasterized label feature values; determining basic feature samples based on the network field strength prediction samples and generating basic feature values ​​based on the basic feature samples; generating network field strength prediction features based on the rasterized label feature values ​​and the basic feature values, wherein the network field strength prediction features are used to train a network field strength prediction model. Through the above method, multiple feature values ​​are applied to feature vector extraction to generate features for mobile network field strength prediction, which are then input into the neural network for model training. Compared to a simple feature extraction process, the prediction accuracy is significantly improved, and the model training effect is better.

The invention provides a foggy day smoke detection method. The foggy day smoke detection method comprises the following steps: step 1, collecting a plurality of modal characteristic patterns; step 2, mapping each modal feature pattern into a unified scale; step 3, performing spatial enhancement on each modal feature pattern through channel attention and spatial attention; step 4, fusing each modal feature pattern after space enhancement to obtain a fused feature pattern; and step 5, performing classification detection based on the fused feature map, and outputting a smoke detection result. According to the foggy day smoke detection method, various modal data are integrated, the robustness bottleneck of single-modal detection is broken through, and the foggy day smoke detection method is more suitable for detection in a dense fog scene; compared with simple feature splicing of an RGB-HSV-dark channel feature fusion technology, the method has the advantages that the obvious features of the smoke can be adaptively enhanced through dynamic cross-modal attention, fog interference is inhibited, the feature distinction degree of the fog and the smoke is improved, and a high-quality feature basis is provided for subsequent detection.

According to the natural language to structured query language model training method and device provided by the invention, the sample data set is obtained by performing synonym replacement on the initial sample data set and / or adjusting the initial database mode, so that targeted adversarial disturbance is introduced to the source of model training, and the training efficiency is improved. And the fixed literal correspondence between the natural language query keyword and the database mode is broken. Training is carried out based on the sample data set rich in disturbance features, so that a natural language to structured query language model is forced to jump out of simple feature matching logic, and a deep semantic alignment relationship between the natural language and a database architecture is learned; therefore, the robustness and generalization ability of the model when facing complex scenes such as synonymous rewriting in user query, structural change or disorder of a database mode and the like are remarkably improved, and high accuracy and usability of the generated structured query language are ensured.

The invention provides an image restoration detection method and system based on cross-domain representation modeling, and electronic equipment. The method comprises the following steps of: 1, preprocessing an image by using a minimum filling strategy; step 2, extracting three kinds of noise information of Noiseprint + +, SRM filtering and Bayar convolution from the image, and inputting the noise information to a noise information gating unit for splicing and fusion to obtain fused noise information; 3, inputting the image preprocessed in the step 1 and the noise information fused in the step 2 into a mask double-flow encoder to generate a feature map; and 4, inputting the feature map generated in the step 3 into a simple feature pyramid to generate a multi-scale feature map, fusing all features, outputting final prediction, and calculating a loss function. Compared with the prior art, the method has higher detection capability and robustness.

The present invention relates to the field of building energy conservation and intelligent building technology, and in particular to a load simulation calculation method and system for building energy efficiency, comprising: collecting multi-source data and constructing a multidimensional data set; deconstructing building load data into three types of heterogeneous modes: sequence-like, image-like, and video-like; calculating a dynamic correction coefficient based on the thermal storage effect of the enclosure structure; processing the three types of heterogeneous modes through a BiGRU network, an STNN network, and a 3DCNN network, respectively, and adopting Stacking ensemble learning to fuse and output the initial load; optimizing and outputting the final load prediction value through a multi-feature recurrent neural network; the system integrates multi-source data collection, multi-modal feature processing, physical correction, hybrid prediction, and timing optimization modules, and deploys the system on a local edge node through a knowledge distillation compression model. The present invention integrates physical models and data-driven methods, effectively solving the problems of insufficient data processing, simple feature construction, shallow model fusion, and insufficient computing architecture and timeliness in the prior art.

The invention belongs to the technical field of data annotation, and particularly relates to a semi-automatic annotation method suitable for a single object, which adopts the technical scheme that simple feature annotation is performed on a target object in an image, data and morphological features of a color center value of the target object are extracted, and the objects in the image are automatically classified through a clustering algorithm; matching a potential annotation object according to the color center value, performing denoising processing in combination with image morphological operation and morphological characteristics of a target object, exporting a pre-annotation file in a specified format, and finally correcting the pre-annotation file through annotation software. According to the method, through man-machine interaction, the pre-labeling of the target objects in all the images can be automatically, quickly and accurately completed only by performing simple feature labeling on one image in the data set, and compared with manual labeling, the method can greatly shorten the data labeling time and reduce the data labeling cost, and is suitable for data labeling of a single object.

The invention relates to the technical field of spatial sequence data prediction, in particular to a spatial sequence data long-term prediction method based on fusion of CNN and RNN, and the method comprises the steps: constructing a plurality of simple feature generators through an original sequence to form a multi-dimensional feature generator, and achieving the data generation thought of inputting a time step serial number and outputting a corresponding time step multi-dimensional feature sequence. Therefore, the monitoring sequence is prevented from being directly input into the prediction model, and the multi-dimensional feature sequence of the monitoring sequence is adopted as prediction input data. For spatial sequence data, a CNN module is added into a prediction model to construct spatial dimension feature extraction and a time dimension feature extractor based on RNN to construct a time dependency relationship, and a single-model one-time prediction output sequence is simplified into a single-model multi-time prediction combination which outputs a result of one time step each time to form a sequence.

The invention relates to the field of alloy defect detection, and particularly discloses a praseodymium-neodymium alloy nondestructive testing method and system based on acoustic characteristic analysis. Defect information contained in an original probe signal is comprehensively captured from two complementary physical perspectives of instantaneous dynamic characteristics and frequency band energy distribution. Furthermore, according to the scheme, simple feature splicing is abandoned, the canonical correlation analysis is used as an information decoupling tool, and two groups of original feature vectors are decomposed into a shared part for describing defect generality and a unique information part for respectively representing the unique resolution capability of the HHT and the wavelet packet on line. And finally, carrying out structured recombination on the three decoupled components to form a fusion feature vector which can effectively eliminate redundancy, amplify differences and has higher information density, and providing input with a clear structure and high refining for a subsequent classification model.

The invention relates to the technical field of intelligent traffic management systems, and discloses an expressway event detection method and device based on long video semantic analysis, and the method comprises the following steps: S1, collecting and preprocessing long video data of an expressway scene, and generating an event-level vector, a segment-level vector and a key frame vector based on the video data; the semantic vector is used for subsequent event detection and discrimination; and S2, based on the obtained semantic vector samples, establishing corresponding probability density models for abnormal event samples and non-event samples in different environment domains, and constructing a calibration cache based on the non-event samples. The long video data of the expressway scene is collected and preprocessed, and then the event-level vector, the segment-level vector and the key frame vector are generated, so that the problems that the traditional video data processing method mostly depends on a simple feature extraction mode, and multi-level feature information in the video is not fully considered, so that the video data processing efficiency is greatly improved are solved. Therefore, the problem of low detection precision is solved.

This invention discloses a method for segmenting melanoma images, belonging to the field of medical artificial intelligence. The method includes the following steps: S1, acquiring image data and preprocessing the image; S2, extracting features from the preprocessed image using an encoding module; S3, controlling the contribution of feature information at each stage through a gate structure, fusing deep and shallow features; S4, concatenating the output of the gate structure with the output of the decoding module by channel, and upsampling to obtain a multi-channel feature map; S5, sending the data to a segmentation head and outputting a binary black-and-white segmentation image. This invention uses a feature fusion gate structure to filter and fuse the feature information output by the encoding module, avoiding feature information redundancy caused by simple feature fusion. It adopts a bottom-up, deep-to-shallow fusion order, which ensures mutual compensation of effective information while reducing computational complexity, offering the advantages of semantically rich feature fusion and high efficiency.

The application discloses a kind of multi-scale interaction and semantic calibration video abstract method of motion and appearance decoupling, belong to computer vision technical field.The method includes: respectively extracting multi-scale motion feature and appearance feature to input video frame sequence, obtain corresponding feature pyramid;Motion feature pyramid and appearance feature pyramid are input video abstract model, and initial frame level importance score sequence is predicted by model;Again, through bidirectional LSTM, the score of key action boundary is strengthened by capturing interframe timing dependence, and the output optimized frame level importance score sequence is reduced by diversity penalty Redundant frame;Finally, by applying time interval constraint optimization frame distribution, video abstract is generated by selecting key frame through greedy algorithm.The method effectively solves the problems of motion and appearance feature coupling, insufficient multi-scale representation, simple feature fusion mechanism and modal weight imbalance in existing video abstract methods, and the experimental results on SumMe and TVSum benchmark datasets prove its effectiveness.

The invention discloses an improved end-to-end target detection system and method, and the method comprises the steps: constructing an end-to-end target detection network which comprises an encoder structure and a decoder structure, the encoder structure comprises a ConvNeXt-Tiny backbone network, a bridge module and three cascaded dynamic heads, and carrying out the shallow feature extraction, simple feature enhancement and deep feature extraction respectively; training the constructed end-to-end target detection network to obtain an end-to-end target detection model, evaluating the end-to-end target detection model, and adjusting model training parameters to obtain an optimized and improved end-to-end target detection model; and selecting an improved end-to-end target detection model with optimal performance to perform target detection on the dense pedestrian image to obtain a target detection result image. According to the method, the detection precision and robustness are enhanced, the feature extraction capability is enhanced, the long iteration structure is optimized, and the parameter quantity and calculation complexity of the Sparse R-CNN are reduced while the detection precision is improved.

The invention discloses a seawater intrusion real-time prediction method based on physical mechanism guidance feature selection, and the method comprises the steps: building a quantitative and fixed-source model through combining an evolution law of an underground water system, obtaining the contribution of a seawater intrusion source, taking the contribution as a weight of a physical mechanism, and calculating the contribution value of each feature to each sample through employing an SHAP method based on an XGBoost model, and carrying out weighted summation on the sample contribution values by taking the physical mechanism weight as a weighting coefficient so as to obtain a weighted overall feature importance ranking. And screening out a simple feature subset with mechanism reliability according to a ranking result, and training and tuning an XGBoost model based on the optimal feature subset. According to the method, the samples with high contribution degree of the seawater intrusion source are preferentially endowed with higher weights, so that feature misselection caused by neglecting of physical reliability of the samples in the prior art is effectively avoided, the consistency of model logic and a physical mechanism is ensured, and a technical scheme with a reliable mechanism is provided for low-cost and real-time early warning of seawater intrusion.

The invention provides a picture element extraction method and device fusing a large language model and a neural network, equipment and a storage medium. Relates to the field of artificial intelligence and computer vision. The method comprises the following steps: detecting and positioning a human face in a picture by adopting a multi-task cascade convolutional network, and identifying characters in the picture by using an optical character identification module to extract simple elements; on the basis of a language chain framework, in combination with a locally deployed multi-modal large language model, complex elements in the picture are extracted; a large language model is adopted to verify complex elements, and output content is verified from the aspects of semantics and structures; and outputting the simple elements and the complex elements in a structured form. The understanding depth and richness of the image content can be improved, and the requirements of application scenes such as intelligent recommendation, content auditing and visual search can be met.

The embodiment of the invention provides a data matching method, electronic equipment and a storage medium. The data matching method comprises the following steps: receiving a test case identifier, querying from a second server according to the test case identifier to obtain environment request data, performing feature extraction on the environment request data to obtain a plurality of environment request features, and matching the plurality of environment request features with a plurality of environment resource features in the environment resource model one by one according to a target matching rule until a target test environment matched with the target test case is found, generating a data matching result, and sending the data matching result to the user terminal. Based on this, according to the embodiment of the invention, the matching of the complex data is realized by adopting the mode of extracting the features, the matching problem among the complex object data is changed into the matching problem among the simple features, the matching complexity of the complex object data is reduced, the matching time among the multiple instance data can be shortened, and thus the data matching efficiency is improved.

This invention relates to hyperspectral and LiDAR remote sensing image fusion and recognition technology. Considering the natural complementarity between hyperspectral images and LiDAR data in terms of spectral features and elevation information, and the problems of existing methods introducing redundant interference through simple feature stitching, this invention provides a collaborative recognition method for hyperspectral and LiDAR images based on Mamba attention fusion. This invention constructs a dual-branch convolutional feature extractor to extract shallow spatial-spectral features of both modalities; introduces a spatial Mamba module for multi-scale global-local long-range dependency feature mining; and designs a dual-attention interactive guided fusion module to achieve deep interaction and fine fusion of cross-modal complementary information through a parallel mechanism of channel attention and spatial attention. This invention effectively suppresses redundant information interference, significantly improves the accuracy and robustness of ground feature fusion classification, and provides high-quality ground feature classification support for remote sensing applications such as urban planning.

The invention relates to the technical field of rolling bearing diagnosis, and discloses a rolling bearing fault diagnosis method based on an SDP image, and the method comprises the following steps: S1, a multi-source data collaborative collection stage; s2, an SDP image generation and enhancement stage; s3, a dual-path feature extraction and model training stage; s4, a real-time diagnosis and closed-loop optimization stage; s5, a fault traceability and report generation stage; the objective of the invention is to solve the problems that an existing fault diagnosis method mostly depends on single-type operation data, complex features of bearing faults are difficult to comprehensively capture, and missed judgment of slight faults or composite faults is easily caused by data dimension limitation; meanwhile, a traditional method mostly adopts a single feature extraction mode, either only simple features of a shallow layer can be excavated, or the stability of basic features is ignored due to excessive dependence on deep features, the identification requirements of different fault types cannot be considered, and the adaptability to operation condition changes is poor. And fault type misjudgment or inaccurate severity evaluation is easy to occur.

The invention discloses a multi-modal image fusion-based coronary plaque stability evaluation system and method, and particularly relates to the technical field of modal image fusion analysis, and the method comprises the steps: extracting depth feature vectors from registered coronary CT, intravascular ultrasound and optical coherence tomography images; constructing a heterogeneous graph by taking the depth feature vectors as nodes, aggregating neighbor node information through a graph convolutional network, and generating enhanced feature representation fused with cross-modal context semantics; a cross-modal attention mechanism is applied to the enhanced features to optimize feature weights, and refined feature vectors are obtained; and finally, outputting a stability classification result by using a classifier, and generating a feature heat map for visual interpretation. According to the method, the multi-modal features are fused by adopting the graph structure, and the decision basis visualization is provided, so that the problems of insufficient fusion, low evaluation accuracy and unexplainable result caused by simple feature splicing in the existing method are solved.

The invention relates to a high-energy-consumption industrial load pattern recognition method and device and a storage medium, and is applied to the technical field of power load clustering, and the method comprises the steps: carrying out the feature weighting through obtaining the SHAP value of each feature, precisely quantifying the influence of the production, environment and other factors on the power consumption behavior, enabling the model decision-making process to be transparent, and improving the accuracy of the model decision-making process. Compared with a simple feature processing mode in the prior art, the precision of pattern recognition and the credibility of a result are remarkably improved; by adopting the adaptive density clustering, the defect that the clustering number is manually set in the prior art is overcome, the load mode of any shape can be automatically and objectively found from complex data, noise can be eliminated, and the robustness and accuracy of mode division are improved; the improved particle swarm optimization algorithm is utilized to perform hyper-parameter automatic optimization on the long-short-term memory network model, so that high performance of the classification model is ensured, a complete process from offline clustering to online classification is constructed, and rapid and accurate identification of the newly increased load is realized.