773 results about "Background information" patented technology

The invention relates to the technical field of image target detection, and discloses a complex scene small target detection system and method based on mask attention and context feature optimization, and the system comprises a backbone network integrating a mask attention mechanism, a backbone network in which an original image is input into the integrated mask attention mechanism, and a background network in which the original image is input into the integrated mask attention mechanism. Redundant background information is filtered through a mask mechanism, and a multi-scale feature map is output; the context-aware enhanced feature refining encoder is used for carrying out multi-branch processing on the multi-scale feature map, and finally weighted fusion is carried out to generate a multi-scale fusion feature sequence; and the deformable Transform decoder inputs the multi-scale fusion feature sequence, processes the multi-scale fusion feature sequence through a cross attention and self-attention module, constructs a target query, and generates a detection result containing bounding box coordinates, category labels and confidence scores. According to the method, the performance of small target detection is improved by introducing efficient feature extraction, a fusion mechanism and a sampling and attention strategy.

According to the medical image segmentation method and device based on spatial perception and frequency domain information, the precision and robustness of medical image segmentation are effectively improved by combining frequency domain information guidance and a multi-head state spatial perception technology. Frequency domain transformation is performed on a medical image, high-frequency and low-frequency components of the image are separated by using a multi-scale decomposition technology, and low-frequency features are extracted to obtain global information. By introducing a learnable noise filtering mechanism, noise and irrelevant background information in a frequency domain are suppressed, so that the model can be focused on a lesion area more accurately. A multi-head perception visual state space module is designed on a bottleneck layer, lesion features of different scales are captured through a multi-scale adaptive feature fusion mechanism, and the capability of segmenting small-size lesions and complex structures is enhanced. A context focusing attention mechanism is introduced into jump connection, fusion of global information and local details is further enhanced, and the accuracy of a segmentation result is ensured; and finally, recovering a high-resolution segmented image through a decoder.

The invention discloses a supply chain risk identification method and system based on a knowledge graph, belongs to the technical field of supply chain management and artificial intelligence crossing, and aims to solve the technical problem of how to realize dynamic monitoring, accurate identification and active early warning of supply chain risks, improve full star, real-time performance and interpretability of supply chain risk identification, and improve the risk identification efficiency. According to the technical scheme, the method comprises the following steps: collecting and treating multi-source data: collecting static background information and dynamic risk information of a supplier, and carrying out highly intelligent data treatment on the collected static background information and dynamic risk information of the supplier through a data treatment engine to ensure data quality and consistency; constructing a dynamic knowledge graph; intelligent risk identification: based on a graph topological structure and dynamic attributes, identifying key risk nodes and communities, tracing in time, marking risks, and performing early warning; decision support and visualization are carried out; and dynamically optimizing and feeding back.

The invention relates to the technical field of big data, and discloses an audio and video monitoring and early warning method and system based on multi-modal model driving, and the method comprises the steps: S1, collecting audio and video signals in a monitoring scene, and extracting the energy distribution characteristics of the audio and video signals, wherein the energy distribution characteristics at least comprise the frequency energy distribution of the audio signal and the brightness and color change rate of the video signal; s2, multi-modal feature extraction of energy distribution features is carried out based on audio and video signals, and environment background feature vectors are constructed in combination with environment background information; s3, dynamically adjusting anomaly detection thresholds of the audio signal and the video signal based on the environmental background feature vector; s4, comparing the audio and video signal energy distribution characteristics extracted in real time with an anomaly detection threshold value, and when the audio and video signal energy distribution characteristics exceed the range of the anomaly detection threshold value, determining an abnormal event and obtaining early warning information; and S5, sending the early warning information from the edge equipment to a monitoring center through a low-bandwidth communication protocol.

The invention discloses a doctor-patient negotiation method and device fusing LLM and a multi-agent architecture, equipment and a medium, and relates to the technical field of doctor-patient negotiation. The doctor-patient negotiation method comprises the following steps: acquiring doctor-patient background information, and constructing a doctor agent, a patient agent and a coordination agent based on a large language model. The doctor agent generates a preliminary treatment regimen. The patient agent generates feedback to the preliminary treatment regimen. The coordination agent parses the feedback, generating a structured acceptance score and a semantic vector. The dynamic belief is updated based on the semantic vector doctor agent. And performing fuzzy utility evaluation on the current treatment scheme, and judging whether strategy adjustment is triggered or not. And the coordination agent calculates the adjustment amount of each issue based on the acceptance difference value, the patient yielding willingness coefficient and the maximum yielding amplitude, and obtains a new quotation scheme. And triggering a belief calibration and strategy reconstruction process at a preset stage node of multiple rounds of negotiation to obtain a new strategy. And when the negotiation round reaches a preset maximum round number, outputting a final treatment scheme.

The invention provides a wound surface grading model construction method and a wound surface self-analysis system, and the method comprises the steps: processing an obtained wound surface image, clinical text description data and structured background information data through a specific feature extraction strategy, and obtaining an image feature vector; clinical text description data and structured background information data are subjected to cleaning, word segmentation, entity labeling and other operations, a wound semantic evolution graph is constructed, related vectors are fused to obtain text embedding, wound background context vectors are generated according to the structured background information data, dynamic weighting, splicing and fusion are performed on image and text feature vectors, and the image and the text feature vectors are subjected to image fusion. According to the method, the fusion feature vector is obtained, the corresponding grading model is constructed according to the fusion feature vector, more accurate, comprehensive and explainable grading evaluation can be performed on different types of wounds, and a self-analysis system designed for the constructed model can quickly and accurately judge the wound image condition shot by a user.

According to the small sample target detection method based on prototype network feature enhancement, firstly, a foreground feature aggregation module is adopted, redundant background information is removed in the category prototype construction process, and purer category features are extracted; and a condition information coupling module is utilized, and the category prototype is dynamically adjusted in combination with the characteristics of the query image, so that the adaptability is better. And then, dynamically supplementing the most similar support sample through a support sample expansion module in the training process, improving the expression ability of the category prototype, and enhancing the generalization performance of the model. Finally, an optimization strategy based on meta learning is adopted, and a joint optimization mode of classification loss, regression loss and meta loss is combined, so that the matching precision of the category prototypes is improved. The method has the advantages that the detection precision and generalization ability of small sample target detection are improved under the condition that extra data labeling cost is not increased, and the method is suitable for various application scenes.

The invention discloses a DeepLabV < 3 + >-based magnetotelluric two-dimensional inversion method, and belongs to the field of geophysical inversion methods, and the method comprises the steps: constructing a plurality of underground resistivity distribution models, and obtaining corresponding apparent resistivity and phase data through forward modeling calculation; designing a deep learning network based on a DeepLabV < 3 + > architecture, and enhancing the capability of capturing multi-scale geological features by using a cavity convolution and improved cavity spatial pyramid pooling (ASPP) module of the deep learning network; di ce Loss is adopted to replace a traditional cross entropy loss function, and the method is more suitable for the characteristics of anomaly and background information imbalance in geophysical observation data; according to the method, the network is trained and verified through the training sample set, and the optimal network parameters are obtained; the trained network is utilized to directly map the apparent resistivity and phase data of a test set into underground resistivity distribution, the end-to-end inversion process is realized, and the multi-scale feature extraction and semantic segmentation capabilities of the DeepLabV3 + network are utilized, so that compared with a traditional inversion method, the method has the advantages of being independent of an initial model, capable of obtaining a global optimal solution and the like.

The invention discloses a sports video commentary evaluation method and system based on a multi-mode large language model, and the method comprises the steps: obtaining a data set which comprises a data pair composed of a video segment of sports commentary and text commentary; semantic classification is carried out on the data set, semantic tags are determined, and the semantic tags divide the data into at least one of key event description, technical detail analysis, background information interpretation, tactical analysis, competition condition interpretation and emotion expression; constructing a multi-modal large language model, calling the data set to train the multi-modal large language model, and determining a sports video explanation model; and scoring the sports video explanation model, and determining an evaluation result. According to the method, the performance of the model in a sports explanation task can be more comprehensively reflected through a multi-dimensional evaluation method, and the limitation that fine-grained professional details, time dynamics and human emotion cannot be captured by traditional indexes is overcome.

The invention provides an RAG method based on semantic precise blocking and precise background information generation, and belongs to the technical field of artificial intelligence. The method comprises the following steps: in a preparation stage before reasoning, preprocessing a large number of related knowledge texts, and segmenting the texts into coarse-grained text blocks according to the number of words; dividing the coarse-grained text block into a plurality of fine-grained text blocks according to semantics through a large language model, and generating a context abstract and an association problem for each fine-grained text block; splicing the fine-grained text block and the context abstract into a final text block, and taking the final text block as a block containing accurate background information; and storing the final text blocks and association problems in different vector databases after vectorization. In the reasoning stage, when facing a user question, the system firstly vectorizes a user question text, then retrieves the most relevant items from the different vector databases for merging, and then optimizes and sorts the retrieved results through a reordering model. And selecting the top-ranked text blocks for the large language model to generate answers. According to the method, the problems of inaccurate blocking, context information missing and inaccurate retrieval in the traditional RAG method are effectively improved.

The invention discloses a video fact and viewpoint alignment traceability method, which relates to the technical field of information retrieval and verification, and comprises the following steps of: performing frame analysis on a video by using a computer vision technology, extracting scenes, objects, dynamic characteristics and background information in the video, extracting audios from the video by using a voice recognition technology, converting the audios into text information, and storing the text information in a database; utilizing an event extraction technology to extract event elements from the video; the event elements are identified through scenes, objects, dynamic features and background information, and event facts are obtained; and based on the text information, using a natural language processing technology to calculate semantic similarity between the viewpoints and the event facts in the video, and automatically aligning the viewpoints and the event facts in the video according to the semantic similarity. According to the method, high-dimensional semantic modeling is carried out on the text, the similarity is calculated, the internal relation between viewpoints and facts can be accurately recognized, and therefore automatic matching of the viewpoints and the facts is achieved.

The invention belongs to the technical field of intelligent medical treatment, and discloses a chronic disease information management system based on a behavior interaction model. Comprising the following steps: acquiring background information, internal motivation and cognitive evaluation; based on the background information, a health education content library is constructed, and health education content is pushed in combination with cognitive evaluation; fusing the background information and the internal motivation to generate a multi-dimensional treatment scheme matrix; acquiring willingness information of a patient, performing quantitative evaluation on acceptance degrees of different treatment schemes, and screening out an optimal treatment scheme; acquiring and analyzing real-time interaction data, and dynamically formulating an emotion support strategy; patient health data are integrated, and the interaction effect is evaluated; according to the interaction effect, the health education content, the optimal treatment scheme and the emotion support strategy are intelligently optimized in sequence; behavior interaction is taken as the core, and accurate management of the whole life cycle of the chronic disease patient is realized, so that the health management effect of the patient is remarkably improved, and the complication risk is reduced.

The invention provides an atmospheric pollution process analysis method, electronic equipment and a storage medium, and relates to the technical field of data processing, and the method comprises the following steps: receiving an atmospheric pollution problem sent by a user, and respectively carrying out vector database retrieval and knowledge graph retrieval based on the atmospheric pollution problem to obtain a first retrieval result and a second retrieval result; wherein the vector database retrieval comprises the following steps: calculating cosine similarity between background information vectors in a vector database and atmospheric pollution problem vectors, and determining multiple pieces of background information of which the cosine similarity accords with a preset threshold value as a first retrieval result; the knowledge graph retrieval comprises the following steps: performing entity recognition and standardization on the atmospheric pollution problem, and generating a target knowledge triple as a second retrieval result; and fusing and inputting the atmospheric pollution problem, the first retrieval result and the second retrieval result into a large language model, and outputting an analysis result of the atmospheric pollution problem. Through the method provided by the invention, the illusion problem of a general large language model in the atmosphere field is relieved.

The invention discloses an infrared image denoising method and system based on artificial intelligence, and relates to the technical field of image denoising, and the method comprises the steps: collecting infrared image data, extracting frequency domain spatial features, carrying out Gaussian filtering, restoring the features to an image space, determining a high-frequency feature map, extracting image background information, and determining a low-frequency feature map. And splicing the high-frequency feature map through a generator, and carrying out two-dimensional transpose convolution operation by adopting an encoder based on a convolution transpose self-attention mechanism. According to the method, the frequency domain of an infrared image is converted into high-frequency and low-frequency characteristic decomposition, high-frequency characteristics are extracted based on a Gaussian high-pass filter, space structure information such as edges and textures can be effectively reserved, the importance of the high-frequency characteristics can be weighted through an attention mechanism, exploration of noise can be enhanced, and for the background part of the image, the resolution of the image is improved. The attention module can identify an area with excessive brightness fluctuation, and can synchronize transmission of secondary features in a noise removal process through residual features.

The invention provides a system and a method for evaluating the independent living ability of old people. The method comprises the following steps: acquiring multi-modal original data (including physiology, behavior and interaction data) of target old people; performing modal coding, query vector interactive fusion (including missing information completion) and feature compression on the multi-modal original data to obtain a standardized feature vector; a first-layer model evaluates behavior and cognitive ability indexes, and a second-layer model combines individual background information to predict a living ability grade and a degradation risk; optimizing the first-layer model through a feature feedback mechanism; and finally, outputting and visualizing a structured evaluation result. The method breaks through the limitation of traditional manual evaluation, achieves the deep fusion of multi-modal data, improves the objectivity, robustness and interpretability of evaluation, and can meet the precise evaluation demands of home care, community monitoring and smart medical scenes.

The invention belongs to the technical field of biomedical image processing and analysis, and particularly discloses a chromosome abnormality detection method based on image analysis, which comprises the following steps: obtaining a user chromosome microscopic image set and extracting a multi-modal feature data set for processing to obtain abnormal chromosome feature data of an abnormal tracking area; and performing collection processing to obtain an abnormal risk assessment index, performing correction in combination with abnormal identification coefficients of a plurality of historical similar abnormal samples to obtain an abnormal risk correction index, and performing assessment processing according to an obtained clinical detection data set corresponding to the abnormal tracking area in combination with genetic background feature data to obtain an abnormal tracking area genetic risk judgment index. And comparing the abnormal risk correction index with the abnormal risk correction index to obtain an abnormal evaluation matching degree coefficient, and verifying an abnormal evaluation result through threshold comparison, so that efficient identification and accurate evaluation of various complex chromosome abnormality types are realized, and comprehensive risk evaluation is performed in combination with genetic background information and clinical data to improve the reliability of a detection result.

The invention discloses a digital animation role dynamic generation method based on a generative model. The method comprises the following steps: inputting a role original image and scene background information, extracting skeleton key points and scene features, analyzing an action sequence, predicting a motion track, calculating an optimal position of a role in a picture, and carrying out dynamic adjustment. And according to the role position, obtaining morphological feature data, evaluating the quality grade, and optimizing the coordination of the role image. And finally, comprehensively scoring by adopting a multi-dimensional quality evaluation system, and determining the visual presentation quality of the role. According to the method, deep fusion of role actions, positions and scenes is realized, the dynamic expressive force and visual coordination of animation pictures are improved, and an efficient solution is provided for generating high-quality animation contents.

The invention discloses a lightweight brain tumor segmentation network based on CNN-Mama parallel coding, and belongs to the technical field of neural networks and Mama architectures. The lightweight network adopts a U-shaped network architecture, and a network main body consists of four layers of encoders and decoders, a bottleneck layer and a jump connection part; wherein each layer of encoder and decoder is composed of a PMamba module and a corresponding sampling module, and the jump connection part is composed of an MDA module. According to the lightweight network, local features are efficiently extracted by using a PMamba module, the modeling capability of the network on a remote dependency relationship between brain tumor images is enhanced, and the utilization efficiency of global features is effectively improved. Meanwhile, redundant background information is filtered by focusing associated feature information of the same layer of the encoder and the decoder from three view directions of the 3D brain tumor image through an MDA module. According to the network, the segmentation precision of each region of the brain tumor is effectively improved while relatively low calculation cost is maintained.

The invention provides a small target detection method based on a superpixel mask and a dynamic kernel. The small target detection method comprises the following implementation steps: acquiring a training and testing sample set; constructing a target detection network model based on the superpixel mask and the dynamic kernel, and carrying out iterative training on the target detection network model; and obtaining a small target detection result. The greedy slicing module carries out initial image block segmentation according to the center point and the scale, the defect that space correlation features are segmented due to segmentation is avoided, the dynamic kernel module carries out multi-scale feature fusion on each image block, the receptive field can be dynamically adjusted, wide-area context information of targets of different scales can be flexibly adapted, and the image segmentation efficiency is improved. The detection precision is effectively improved; a superpixel mask generator generates a foreground region mask irrelevant to the category for each extracted feature image, and a dynamic kernel module performs multi-scale feature fusion on each foreground image block containing a target cluster obtained by a greedy slicing module; and the influence of processing useless background information on calculation overhead when the multi-scale features are obtained is avoided.

The invention discloses an artificial intelligence home management system, which relates to the technical field of smart home and comprises a data acquisition module, a behavior analysis module, an environment sensing module, a cooperative control module and an energy consumption optimization module. According to the artificial intelligence home management system provided by the invention, through multi-source data fusion and cross-equipment collaborative decision, the intelligent management level of a home scene is remarkably improved, user behavior data can be deeply analyzed, a dynamic behavior model can be constructed, and a user habit mode can be accurately identified and predicted, so that the initiative and adaptability of equipment control are realized, and the user experience is improved. The system comprehensively senses environmental parameters, constructs an environmental state map, provides accurate environmental background information for a cooperative control strategy, ensures the accuracy of an equipment control strategy, combines equipment operation data with a user behavior scene, generates an energy consumption optimization scheme, ensures the comfort of a user, realizes high efficiency and energy conservation, and improves the user experience. And user privacy leakage is effectively prevented, and potential faults are early warned.

The invention provides a large model test case generation method, system and device based on multi-Agent collaboration. The method comprises the steps of obtaining product information of target product demand information; if the product demand information is incomplete, performing structured supplement on the extracted product demand information according to a slot filling technology to obtain complete product demand information; performing assembly according to the complete product demand information and the background information, and determining a first prompt word used for requesting the large model to generate a test case; and generating a first edition test case according to the first prompt word by using the plurality of agents, reviewing the first edition test case, generating an optimization suggestion, and fusing the first edition test case based on the optimization suggestion to generate a final test case. Through knowledge injection and structured requirements, the test case not only covers functional requirements, but also improves the accuracy of the test case; dynamic verification and correction of the use cases are achieved through cooperation of the multiple agents, the generation efficiency of the test use cases is improved, and the coverage rate is increased.

The invention discloses a high development potential academic entity identification and prediction method based on a heterogeneous graph Transform, and the method comprises the following steps: firstly constructing a heterogeneous information graph, representing multiple types of entities in an academic network through nodes, and representing the relation between the entities through edges; fusing structure embedding and semantic embedding of the nodes to generate a multi-modal initial feature vector; iteratively aggregating multi-hop neighbor information through a multi-layer heterogeneous graph Transform layer, dynamically learning weight and updating node representation to realize global feature aggregation; fusing multiple types of neighbor information of a target node based on cross-layer attention weighting, generating refined context features, and completing local background information aggregation; and finally, splicing the global node representation and the local context feature, and outputting the probability that the target entity becomes a high-development potential entity through a classifier. According to the method, the complex isomerism of the academic network can be effectively processed, the prediction precision is improved, and support is provided for scientific research management and the like.

The invention discloses a deep learning-based benthic organism detection method, system and device and a medium, and belongs to the technical field of computer vision, and the method comprises the steps: obtaining a benthic organism data set picture, and generating a training data set according to the benthic organism data set picture; on the basis of the training data set, a coordinate attention mechanism and loss function calculation are combined, and a frame of a benthic organism detection model is built; the loss function is used for calculating the positioning loss of the actual target frame and the predicted target frame; training the frame of the benthic organism detection model to obtain the benthic organism detection model; and determining a benthic organism detection result based on the benthic organism detection model. According to the method, the deep learning algorithm is utilized, the morphological characteristics of the benthos and the environmental background information are fused, accurate detection and dynamic tracking of the seabed target in the complex underwater environment can be achieved, the accuracy and efficiency of benthos detection are improved, and powerful support is provided for protection and management of marine resources.

The invention discloses a video face changing consistency enhancement method based on multi-source feature collaboration, which belongs to the technical field of video processing, and comprises the following steps: S1, video input: inputting an original video, and extracting a frame of image; s2, constructing an expression network; s3, face network construction: extracting face 2D key point picture information, and injecting the face 2D key point picture information into a face network; s4, background network construction: extracting face background information of the video, and injecting the face background information into a background network; s5, constructing an illumination network; s6, feature fusion: splicing outputs of the expression network, the face network, the background network and the illumination network together, and entering a feature extraction network; and S7, video output: injecting the picture information of the reference face and the features extracted by the four networks into a video generation large model, and generating a face changing video by adopting the video generation large model and inputting the face changing video. According to the method, various information of the face in the video, such as illumination, background, face key points, expressions and the like, is comprehensively extracted, so that the quality of the video generated in video replacement is improved.

The invention relates to the technical field of finance, and provides a time series data prediction method and system based on an auto-reflection mechanism large language model, and the method comprises the steps: firstly collecting structured transaction data and public unstructured text information in a preset historical time window, inputting a first large language model, and generating structured market background information in combination with a preset cue word; inputting the data and preset risk preference data into a second large language model to generate first causal reasoning training data; and inputting a third language model for reflection correction, checking whether the argument can be verified in the background information, and if so, reserving and strengthening to obtain second causal reasoning training data. And constructing a prediction large language model, optimizing parameters by means of a group relative strategy optimization algorithm, and then outputting the parameters. The system and the method depend on an auto-reflection mechanism, the model considers risk preferences in financial time sequence prediction to generate various risk strategies, manual annotation dependence is greatly reduced, the cost is reduced, and the risk preference module is independent and configurable and can be flexibly expanded.

The invention discloses a design method of a star simulator three-dimensional divertor target plate, and belongs to the technical field of magnetic confinement fusion. The method comprises the following steps: firstly, deducing a circumferential thermal load uniformity theoretical model, then, calculating background information required by target plate generation, and finally, carrying out target plate generation by using the obtained background information and the circumferential thermal load uniformity theoretical model. The target plate is generated based on parameter control, the target plate capable of achieving uniform distribution of the circumferential thermal load is provided, the circumferential uniform distribution of the thermal load of the divertor target plate can be effectively improved, and the peak value of the thermal load of the divertor target plate can be effectively reduced. The target plate generated by the method conforms to standard output formats among different analog calculations. Data can be conveniently transmitted between integrated simulation platforms, and the working process is simplified. The divertor target plate is generated on the basis of the basic physical assumption that the magnetic lines of the magnetic field and the heat flow are attenuated along the radial position, and the method is suitable for all star simulator devices with three-dimensional magnetic island structures and has universal applicability.

The invention discloses an invoice information automatic filling method and system based on a finance and taxation big language model. The method comprises the steps of obtaining a related finance and taxation data set; pre-marking, checking and correcting the finance and taxation data set to complete pre-training and fine tuning of a finance and taxation big language model; common voice information input by a user is converted into text information through voice recognition; and on the basis of the finance and taxation large language model, according to the prompt word text of the finance and taxation large language model, combining the text information with background information and invoicing information of a user to complete invoice information automatic filling. The invoice information is automatically extracted, so that the requirement of manual input is reduced, the labor cost is reduced, and meanwhile, the extra cost caused by errors can be saved.

The invention provides a blueberry picking system and method and a computer medium, an image acquisition module of the system obtains image information of at least two angles of a target fruit tree, a target detection module analyzes mature blueberry fruits in the fruit tree according to the image information by using a deep learning algorithm, and a control module sets a picking path and strategy of the blueberry fruits. The picking execution module executes the picking action of the blueberry fruits; the target detection module comprises an image division sub-module for dividing a blueberry image into a plurality of sub-regions, a first analysis sub-module for respectively capturing feature vectors of the plurality of sub-regions and determining a target sub-region where a fruit tree is located according to the feature vectors, and a second analysis sub-module for analyzing and outputting the position, size and maturity of the blueberry fruit in the target sub-region. According to the method, each sub-region is independently analyzed, the target region with the fruits in the acquired image is focused, the background part is weakened and removed, and the interference of background information in the pattern is reduced, so that the accuracy of blueberry fruit recognition is improved.

The invention relates to an open-pit mine area vehicle trajectory prediction method based on a multi-attention mechanism, and belongs to the technical field of open-pit mine area unmanned driving. The open-pit mine area vehicle trajectory prediction method comprises the following steps: extracting spatial features of a grid map based on a convolutional neural network by using the grid map of an open-pit mine area and acquired data of a target vehicle and surrounding vehicles; the method comprises the following steps: extracting features of an interactive vehicle based on a feature pyramid, fusing extracted grid map space features and interactive vehicle time sequence information features based on a multi-head attention mechanism, and finally realizing decoding output of a multi-modal prediction trajectory of a target vehicle. According to the method, the background information of the scene and the interaction information between the vehicles are fully fused, the relation between the vehicles and the relation between the vehicles and the environment are better captured, compared with a traditional physics-based method, higher prediction precision can be kept within a long-time prediction range, and powerful support is provided for intelligent mine construction.

The invention discloses an image sensitive content auditing method and system, and the method comprises the steps: obtaining the original description of a to-be-audited image, and the original description comprises the text summarization of the visual content of the image and the text content in the image; extracting an entity area in the to-be-audited image, retrieving related external background information based on the to-be-audited image and the entity area, eliminating interference information by using a label of the entity area, and generating a retrieval enhancement result; semantic correlation screening based on sensitive topics is carried out on the retrieval enhancement results, the screened retrieval enhancement results are fused into the original description in an iteration mode, and final fusion description is generated; and constructing a training data set containing positive and negative judgment sample pairs based on the fusion description, performing fine tuning on the visual language model, and performing sensitive content auditing on a new image by using the fine-tuned visual language model. The method can effectively improve the recognition accuracy and reliability of the sensitive content of the image.