27results about How to "Rich input" patented technology

The invention relates to the technical field of traffic safety management, in particular to a vehicle abnormal driving recognition and early warning method and system. The method comprises the following steps: firstly, acquiring vehicle position, speed, direction and driver operation data in real time, then correcting time sequence data by adopting a smooth denoising method of adaptive segmented clustering, generating an information confirmation code through standardization, and then carrying out weighted fusion on multi-source data to obtain a multi-source data fusion model; the strategy decision-making module is used for extracting vehicle speed change rate, lane deviation degree and driver behavior characteristic key indexes, identifying and grading abnormal driving behaviors by setting a threshold value, calculating a comprehensive abnormal score, and generating a grading intervention strategy for slight, medium and serious anomalies according to a comparison result of the abnormal score and the set threshold value. And finally, combining automatic intervention in the vehicle with global coordinated intervention of a road manager. According to the invention, the intervention strategy is generated based on multi-source data acquisition, processing, analysis and decision, so that the vehicle driving safety and the road management efficiency are improved.

In the breathing state evaluation model training and breathing state evaluation method and system, breathing-related physiological parameter data is collected; the physiological parameter data comprises a blood oxygen saturation numerical value, a heart rate numerical value, a breathing carbon dioxide concentration numerical value and a breathing rate numerical value; interpreting the physiological parameter data, and giving a breathing state evaluation label corresponding to the physiological parameter data; the breathing state evaluation label comprises an IPI index; inputting the physiological parameter data and the breathing state evaluation label into an artificial intelligence training device, and training by the artificial intelligence training device to obtain a breathing state evaluation feature data set; the breathing state evaluation feature data set is matched with a corresponding artificial intelligence algorithm and is used for breathing state evaluation. The breathing state comprehensive evaluation can be directly carried out based on the complete waveform signal, and the IPI index is closer to reality.

The invention discloses an AI-based viral pneumonia auxiliary diagnosis method and system, and relates to the field of pneumonia diagnos.The method comprises the steps that abnormal state monitoring is conducted on target people in a target area, and when it is detected that a target human body is in the abnormal state within a preset time period and the occurrence frequency of abnormal behaviors exceeds a preset threshold value, the target human body is judged to be in the abnormal state; mapping the occurrence position, the moving track and the real-time people flow density into a three-dimensional building model to obtain space environment associated data; inputting the space environment associated data into a propagation prediction model to obtain a propagation risk probability of each target sub-region in the target region; and marking the target sub-region of which the propagation risk probability is greater than a preset risk threshold as a target risk region, and generating first early warning information containing the target risk region and the target human body. By implementing the method, the accuracy of the viral pneumonia early warning information can be improved.

The invention relates to a stress injury assessment opportunity prediction method based on a Braden scoring system and an RNN model. The method comprises the following steps: collecting a Braden index and other related data of a patient; preprocessing the data; performing feature extraction on the collected data, and constructing a feature vector; an RNN model is trained through the feature vectors, and a stress damage risk prediction model is formed; inputting the real-time data of the patient into the trained stress injury risk prediction model, predicting the risk probability of the patient, and automatically determining the optimal time of the next evaluation according to a preset threshold value; and result output and feedback: feeding back the prediction result and the suggested evaluation opportunity to nursing personnel, and providing personalized nursing suggestions. According to the method, the risk change of the pressure injury of the patient can be automatically judged, the optimal evaluation and intervention opportunity is determined, the nursing quality is improved, and the occurrence rate of the pressure injury is reduced.

The embodiment of the invention provides a graphic typesetting method which comprises the steps that target typesetting elements and target typesetting prompt information are obtained, and the target typesetting elements comprise information of at least one mode; identifying the target typesetting element to obtain element identification information; and inputting the target typesetting prompt information and the element identification information into a target large language model, and performing graphic typesetting by utilizing the target large language model according to the typesetting prompt information and the element identification information. According to the technical scheme, the adaptability and flexibility of typesetting can be improved, rapid adjustment and adaptation can be achieved in the face of constantly changing requirements, and the efficiency in large-scale design tasks is improved.

The present application discloses a data processing method, including: obtaining the first feature data of a subsequence for inputting into a neural network model; the subsequence is a data subset obtained by performing frame processing on a target sequence; obtaining short-term memory cache data corresponding to the first feature data of the subsequence; the short-term memory cache data is part of the first feature data of the previous subsequence of the subsequence in time order; inputting the first feature data of the subsequence and the corresponding short-term memory cache data into the first convolutional layer of the neural network model for noise reduction processing to obtain second feature data; the second feature data is the first feature data with some noise removed. The above method is used to solve the problem of slow operation and large delay of large models when the neural network model is used to perform noise reduction processing on input data in the prior art.

The invention provides a user behavior data processing method and device, electronic equipment and a computer program product, and belongs to the field of advertisement data processing. The method comprises the following steps: acquiring and integrating heterogeneous data from a plurality of data sources; according to performance indexes of entities related to advertisement putting in a predefined evaluation period, the entities are distributed to different entity sets, and the entity sets comprise a main entity set and an aggregation entity set; taking the entity set as an operation granularity, generating a combined feature set based on the integrated heterogeneous data, the combined feature set comprising a first type of features obtained based on behavior data and attribute data of all users in the corresponding entity set in the preset time period, obtaining a second type of features based on the historical time sequence data of the corresponding entity set; and inputting the combined feature set into a trained machine learning model to obtain a target index related to the entity set.

The invention relates to the technical field of oil quality analysis, in particular to an aviation oil property analysis method and system based on spectral oblique projection. According to the method, when the oblique projection operator is constructed, the limitation of a traditional oblique projection operator is broken through through dynamic multi-component cooperation and self-adaptive residual error separation, key information which is in complex association with properties in the aviation oil spectrum is extracted more accurately, the pertinence and robustness of feature extraction are improved, and the accuracy of feature extraction is improved. Higher-quality input variables are provided for an aviation oil property prediction model, and the method is a core technical support for realizing high-precision aviation oil property analysis.

The invention provides a trailer hinge angle real-time estimation method and system based on feature point tracking. The method comprises the steps of obtaining a first image frame of a target trailer at a current moment and a second image frame of the target trailer at a previous moment; determining a first feature point list of the first image frame according to a second feature point list in the second image frame; according to the first feature point list and a preset feature point template, calculating to obtain a visual estimation angle and confidence; a first Kalman filter at the current moment is updated according to the visual estimation angle and the confidence coefficient, an updated second Kalman filter is obtained, and a hinge angle estimation value at the current moment is output based on the second Kalman filter; and within a preset time period after the current time, according to the motion data of the target trailer, outputting the hinged angle estimated value of each time within the preset time period through the second Kalman filter, thereby improving the accuracy and real-time performance of real-time estimation of the hinged angle of the trailer.

The invention discloses a snake wound classification and recognition model construction method based on serum marker electrochemical detection and deep learning, and belongs to the technical field of biosensing and intelligent diagnosis. The method comprises the following steps: firstly, screening based on proteomics to obtain multiple serum protein biomarkers related to snake bite types, constructing a multi-electrode electrochemical sensing platform, and collecting electrochemical signals through cyclic voltammetry (CV) to obtain electrochemical fingerprint characteristics of a serum sample; and after the obtained signal is subjected to normalization and data enhancement processing, the signal is input into a one-dimensional convolutional neural network (1D-CNN) model for feature learning and classification identification. Through combination of multi-index detection of biomarkers and an end-to-end deep learning algorithm, the method realizes rapid and accurate identification of three different snake wound types of agkistrodon acutus, cobra and trimeresurus remeresurus. The invention constructs a high-sensitivity and rapid diagnosis method for snake wounds, and the method has good popularization and clinical transformation potential.

The invention relates to the technical field of sports health monitoring, and particularly discloses a real-time running fatigue monitoring method and system, electronic equipment and a storage medium, and the method comprises the following steps: receiving real-time heart rate data in a running process; extracting a plurality of heart rate variability characteristics from the real-time heart rate data; inputting the plurality of heart rate variability characteristics into a fatigue monitoring model; wherein the fatigue monitoring model is obtained in advance based on indoor running data training, and is obtained by carrying out cross-scene optimization by adopting transfer learning and incremental training strategies and utilizing outdoor running data; the multiple heart rate variability characteristics are processed through the fatigue monitoring model, and a real-time fatigue quantification result is generated; and outputting fatigue state information according to the real-time fatigue quantification result. The technical problems that subjective evaluation is poor in real-time performance, the application of an existing sensing technology is limited, and the dynamic adaptability of a model is insufficient are solved, and the running fatigue state real-time monitoring capacity and the cross-scene adaptability are improved.

The present invention provides an algorithm and device for estimating the cost of online car-hailing, which relates to the technical field of network prediction. The present invention collects traffic data T of the area where the taxi passenger is located. d 、Weather Information d 、Urban activity data E d and historical order data H d Then, we construct a price prediction formula P based on the recurrent neural network (RNN) and obtain data R of several online taxi services that can be accepted within the user’s taxi range. d , and the user's waiting time expectation data W e , based on online car-hailing service data R d and the expected waiting time W e Generate the predicted price calibration index C i , obtain the adjusted forecast price formula and combine it with the real-time collected data and forecast price calibration index C i And user feedback data, to calibrate the cost forecast, allowing users to have certain choices or voting rights on the travel cost, and finally combine multi-source data such as traffic data, weather forecasts, urban activities, etc. to provide more comprehensive input for the estimation algorithm and improve the accuracy of the estimation.

The application belongs to the technical field of automobile electronic testing, and relates to an automobile intelligent cockpit display and automatic test system based on deep learning, aiming to solve problems such as low efficiency of traditional manual testing, poor adaptability of automatic tools, insufficient test coverage, lack of full-process closed loop, and difficulty in model deployment, and realize intelligent, high-precision and automatic test of cockpit display functions. The system constructs a special cockpit display dataset, collects images of multiple vehicle instrument screens and central control screens, covers fault icons, ADAS prompt texts and various lighting environments, and improves the model generalization ability through data enhancement. In view of the small target icon detection and text recognition efficiency problems, the YOLOv8 model is improved, the detection precision is improved through optimization of anchor frame; the SVTR text recognition model is improved, a lightweight network structure is introduced, the parameter quantity is reduced and the inference speed is improved. The system integrates data collection, signal simulation, visual recognition, result evaluation and report generation modules, realizes automatic analysis of test cases, CAN signal interaction, display response detection and automatic generation of test reports.

This invention discloses a multi-factor driven logical sequence generation method and apparatus, comprising: a feature input module, a multi-factor fusion module, a dynamic constraint module, a logical generation module, a reordering and evaluation module, and a feedback learning module. The method involves receiving text information from multiple sources, performing structured parsing and normalization to obtain a standardized set of text feature vectors; extracting multi-factors and constructing multi-factor constraints; generating a multi-factor driven logical sequence based on the set of multi-factor constraints; performing dynamic information updates and output constraint adjustments; generating candidate logical sequences; and designing multiple-dimensional indicators for re-evaluation. This invention significantly improves the accuracy, adaptability, and optimality of logical text sequences by addressing the problems of insufficient input information processing, lack of dynamic adaptability, and insufficient post-optimization mechanisms in the logical text sequence generation process.

The invention discloses an infectious disease trend prediction method based on hybrid decomposition and a Time2Vec-BiLSTM-SA model. The method comprises the following steps: directly constructing a first feature matrix according to a sample sliding window width by using daily newly-added definite case number data after smoothing processing; performing L types of decomposition on the smoothed daily newly-added definite diagnosis case number data by adopting a hybrid decomposition algorithm to obtain L types of decomposed daily newly-added definite diagnosis case number component data; constructing a second feature matrix for the L types of daily newly-added definite case number component data according to the width of a sample sliding window; constructing a multi-branch input network prediction model; and taking the first feature matrix and the L types of second feature matrixes as input of a multi-branch input network prediction model, and outputting a daily newly-added definite case number prediction value of single-step prediction by using the multi-branch input network prediction model. According to the method, the hybrid decomposition algorithm is ingeniously introduced, so that richer feature information is provided for the prediction model; the prediction model can greatly improve the precision and reliability of epidemic situation trend prediction.

The invention discloses a cross-modal pedestrian re-identification method, system and device based on modal supplementation and feature collaboration, and a medium, and the method comprises the steps: obtaining a visible light modal image and an infrared modal image under a same pedestrian label, and carrying out the modal supplementation of the visible light modal image and the infrared modal image, obtaining a first intermediate modal image corresponding to the visible light modal image and a second intermediate modal image corresponding to the infrared modal image; constructing a cross-modal pedestrian re-identification model; adopting the visible light modal image, the infrared modal image, the first intermediate modal image and the second intermediate modal image to perform feature extraction and feature cooperative training on the cross-modal pedestrian re-identification model to obtain a trained model; and inputting a visible light modal image and / or an infrared modal image to be re-identified into the trained cross-modal pedestrian re-identification model for cross-modal re-identification. According to the invention, high-accuracy pedestrian cross-modal re-identification of the visible light-infrared image can be realized.

The application provides a power transmission tower reinforcing process parameter optimization method and system, and relates to the technical field of power transmission tower reinforcement. The method comprises the following steps: obtaining image data of a power transmission tower to-be-sprayed base material; obtaining heterogeneity characteristic quantitative information of the to-be-sprayed base material by processing the image data; dividing the surface of the to-be-sprayed base material into multiple heterogeneity regions according to the heterogeneity characteristic quantitative information; determining the spraying reinforcement parameters of each heterogeneity region according to the heterogeneity characteristic quantitative information corresponding to the heterogeneity region; and controlling the spraying equipment to spray and reinforce each heterogeneity region according to the spraying reinforcement parameters. The method of the application aims to solve the problems of poor coating performance and the inability to realize real-time adaptive optimization of the base material caused by the uneven surface of the base material in the prior art, and improves the overall quality and reliability of the reinforcing coating.

The invention discloses a lake water quality parameter inversion method based on a remote sensing image, a medium, equipment and a product, and relates to the field of water quality parameter inversion, and the method comprises the steps: obtaining a Sentinel-2 remote sensing image and water quality sampling data, re-sampling a wave band in the remote sensing image to a uniform resolution, and carrying out the time-space matching of the wave band and the water quality sampling data, selecting a dual-band combination, a three-band combination and a four-band combination related to the water quality sampling data from the matched resampling bands, and constructing a training set; a deep learning inversion model is constructed, the deep learning inversion model comprises three feature extraction sub-networks, an attention mechanism and a fusion main network, wave band combinations are input into the three feature extraction sub-networks respectively, obtained three feature vectors are subjected to weighted fusion through the attention mechanism and then input into the fusion main network, and predicted water quality parameter concentration is output; and training the model by using the training set, and performing lake water quality parameter inversion by using the trained model. According to the invention, the inversion precision of water quality parameters is improved.

The invention provides a texture generation method and system for a three-dimensional figure model and a storage medium, and relates to the technical field of image processing, and the method comprises the steps: carrying out the multi-angle image collection of a target figure, and obtaining a multi-angle figure image set meeting a preset image quality constraint; performing visible texture feature detection according to the predetermined aiming frame and the visible texture detection channel to obtain multi-angle visible texture feature distribution; performing three-dimensional modeling to generate a first character texture model; performing invisible texture feature compensation to generate a second character texture model; performing multi-angle UV expansion inspection to obtain a multi-angle UV expansion inspection result; and correcting and generating a third texture model of the figure. The technical problems that in traditional three-dimensional character model generation, due to limitation of the collection view angle, texture information of partial areas of the surface of the model is lost, and consequently the texture of the generated model lacks details and is low in authenticity are solved.

The invention discloses a business personnel seat number identification method and device, electronic equipment and a program product, relates to the field of artificial intelligence, and is applied to the field of financial science and technology, and the method comprises the steps: carrying out the preprocessing of collection information of a transaction liquidation request when the transaction liquidation request is detected, and obtaining a preprocessed text; extracting a text semantic feature, a character space feature and a business logic feature from the preprocessed text, and performing feature fusion to obtain a fused feature vector; and inputting the fusion feature vector into a target recognition model, and outputting a business personnel seat number recognized from the collection information, the target recognition model being obtained by training a preset recognition model by using historical collection information, and the preset recognition model at least comprising: a bidirectional long short-term memory network, a converter encoder, and a conditional random field model. Through the method and the device, the problem that the seat number of the business personnel in the collection information is difficult to identify based on a traditional keyword matching method in related technologies is solved.

The invention discloses a pulse neural network communication system identification method and system based on time-frequency feature fusion, electronic equipment and a storage medium, and the method comprises the steps: obtaining communication system data from Internet of Things equipment, and making the communication system data into an initial communication system data set, dividing the initial communication system data set into training set data, verification set data and test set data according to a proportion; importing the divided training set into a time-frequency domain feature fusion module to obtain sample features of comprehensive time-frequency domain information; inputting the sample features of the comprehensive time-frequency domain information into a pulse neural network model to obtain an initial model predicted value and a model weight; substituting the model prediction value and the label value into a classification loss function for calculation to obtain a classification loss value; updating parameters of the spiking neural network model by using the classification loss value until an optimal spiking neural network model is obtained; according to the method, the communication system data detection efficiency and recognition precision in a complex real environment are improved, and meanwhile, the recognition energy consumption problem in a resource-limited scene is reduced.

The invention provides an image-based question and answer method and device, equipment and a storage medium, and relates to the technical field of image-text processing. The method comprises the following steps: acquiring a to-be-processed image and a corresponding problem text, and inputting the to-be-processed image and the problem text into an image-text large model comprising a visual expert module, a large language sub-model and a plurality of small models of different task types; wherein the input of each small model is a to-be-processed image and a question text, and the output of each small model is a first sub-image feature and a first sub-answer text. Next, the visual expert module determines a target task type according to the question text, and endows the output of the small model corresponding to the target task type with weight confidence higher than that of other small models; and finally, the large language sub-model generates an answer text according to the to-be-processed image, the question text and the output endowed with the weight confidence of each small model. According to the method provided by the invention, the image-text understanding capability of the image-text large model is improved, so that accurate answers can be output.

The invention discloses a fault positioning method, device and system for an optical fiber line, and relates to the technical field of optical fiber communication, and the method comprises the steps: controlling an optical time domain reflectometer of each station to transmit an optical pulse signal to the optical fiber line, and receiving an optical pulse reflection signal; performing time-frequency transformation on the light pulse reflection signal to obtain a reflection signal two-dimensional time-frequency diagram set; performing feature extraction on the reflected signal two-dimensional time-frequency diagrams in the reflected signal two-dimensional time-frequency diagram set to obtain a reflected signal time-frequency feature vector set; extracting a reflected signal time-frequency feature vector corresponding to the station to be evaluated from the reflected signal time-frequency feature vector set as a query feature vector; performing fine-grained decoupling and feature matching on the query feature vector and other reflected signal time-frequency feature vectors to obtain a matching result representation feature matrix; and optimizing the matching result representation feature matrix, and judging whether the to-be-evaluated site is a fault site or not. According to the invention, the accuracy of optical fiber line fault positioning can be improved.

The invention relates to the field of image processing and financial science and technology, and provides an interactive vehicle loss assessment method and device based on multi-model fusion, computer equipment and a computer readable storage medium, and the method comprises the steps: receiving multimedia data uploaded by a user through terminal equipment; inputting the multimedia data into a trained vehicle loss assessment judgment model, wherein the vehicle loss assessment judgment model comprises a part maintenance machine learning model, a damage form maintenance machine learning model and an independent maintenance machine learning model; and carrying out accident vehicle loss assessment according to maintenance schemes output by the component maintenance machine learning model, the damage form maintenance machine learning model and the independent maintenance machine learning model. By adopting the method and the device, the technical problem of low accuracy of an output result of a single automatic loss assessment model is solved, and the accuracy of a vehicle loss assessment result is improved.

The specification discloses a spatio-temporal knowledge graph construction method, system and device for ecological risk prediction, which accesses spatio-temporal data through a platform data interface, structurally processes the spatio-temporal entity model data containing geometric attributes, time attributes and state attributes, constructs a hybrid spatio-temporal index combining spatial index and time index, performs spatial range retrieval on an entity model library according to a given target entity, determines a candidate entity set, combines the state attributes, calculates the correlation strength between entities through a preset quantitative calculation model, generates an enhanced relationship edge containing a weight value and a quantitative indicator, maps the spatio-temporal entity and the enhanced relationship edge to a graph database, and obtains a spatio-temporal knowledge graph; receives a query request containing spatio-temporal constraint conditions and attribute filtering conditions, performs graph traversal and screening based on the weight value and the quantitative indicator of the enhanced relationship edge, and outputs a query result; and updates the spatio-temporal knowledge graph according to a preset data cycle.

The embodiment of the invention discloses a vehicle deceleration control method, a medium, equipment and a product. The method comprises the steps that a front vehicle image sequence for a front vehicle and the real-time deceleration of a target vehicle are acquired; the front vehicle is in front of the target vehicle; performing image analysis on the front vehicle image sequence to obtain a front vehicle lamp state; determining a front vehicle driving intention according to the front vehicle lamp state; determining a target deceleration and a deceleration change rate according to the driving intention of the front vehicle and / or the real-time deceleration; and generating a control instruction for controlling the target vehicle to decelerate based on the target deceleration and the deceleration change rate. According to the embodiment of the invention, the vehicle lamp state of the front vehicle can be accurately identified, the driving intention of the front vehicle is judged based on the vehicle lamp state, the differential deceleration and the change rate are formulated accordingly, smooth and timely automatic deceleration is realized, the braking distance is remarkably shortened, and the rear-end collision risk is effectively reduced and the driving safety is comprehensively enhanced while the comfort is improved.

The application provides a robot dynamic error compensation method based on a capsule neural network, adopts Newton-Euler method to establish a robot dynamics model, and performs linearization and dynamics parameter identification; uses joint motion data as input to obtain predicted torque for model training by using the linearized dynamics model after the dynamics parameter identification, and calculates error between the predicted torque and actual torque; regards normalized parameters as vectors in a vector space, calculates inner products of the vectors to form a Gram matrix and converts the Gram matrix into a two-dimensional image; learns error between the predicted torque and the actual torque by using a capsule neural network, and obtains a trained capsule neural network; obtains real-time prediction error by using the trained capsule neural network, and compensates real-time predicted torque obtained by the linearized dynamics model after the parameter identification by using the real-time prediction error. The application realizes high-precision and adaptive dynamic error compensation.