423 results about "Global consistency" patented technology

The invention provides an intelligent traffic vehicle behavior modeling method and system based on collaborative learning, and relates to the technical field of unmanned driving, and the method comprises the steps: obtaining traffic data in a target region; performing data calibration processing according to the traffic data to obtain a collaborative perception data set; a vehicle space-time trajectory function is generated according to the collaborative perception data set, and a conflict domain space-time distribution diagram is obtained; according to the vehicle space-time trajectory function and the conflict domain space-time distribution diagram, probability distribution characteristics of the vehicle driving intention and a multi-workshop competition cooperation relation matrix are extracted, and a multi-dimensional interaction characteristic vector is obtained; and behavior modeling is carried out according to the multi-dimensional interaction feature vectors, and a vehicle behavior model representing interaction behavior rules and decision logic between vehicles is obtained. According to the method, through multi-vehicle cooperative sensing data calibration and dynamic fusion of road geometric constraints, the problems of single-vehicle sensor shielding and cross-vehicle data heterogeneity are effectively solved, and the global consistency of vehicle position and motion state estimation is remarkably improved.

The invention discloses a trackless rubber-tyred vehicle multi-terminal data fusion system, relates to the technical field of data fusion, and is used for solving the problem of poor trackless rubber-tyred vehicle multi-terminal data fusion. Comprising a path quantification module, an equipment health classification module, an alternative optimization module and a consistency verification module. According to the method, multi-terminal data fusion is carried out on a trackless rubber-tyred vehicle, a sensor dependency graph is dynamically constructed, dependency weights are calculated, a critical path is optimized, health indexes are defined according to a signal-to-noise ratio, output fluctuation and response delay, multi-dimensional abnormity is detected in combination with a sliding window and a mahalanobis distance, and a health score is generated to monitor the state of the sensor. A fault sensor is isolated, substitute data is generated through a redundant data source, the substitute data is incorporated into a fusion process after verification, a dependency graph is adjusted, a path is optimized, global consistency verification analysis is performed, different strategy management modules are triggered, a data fusion rule and task allocation are optimized, and the data fusion reliability of the trackless rubber-tyred vehicle in a complex operation environment is improved.

The invention discloses a laser SLAM (Simultaneous Localization and Mapping) method in a dynamic environment of a well mine, which relates to data processing and comprises the following steps of: performing mapping by utilizing a Cartograph SLAM algorithm; segmenting the PCD point cloud map by adopting an octree-based segmentation method according to a preset movement or deformation frequency threshold value of an object in each region in the well mining environment; adjusting the map resolution of each area in the PCD point cloud map according to the segmentation result; a change detection algorithm is used to carry out change detection on an area with object movement or deformation frequency exceeding a threshold value in the segmented PCD point cloud map, and new point cloud data of the change area is updated to the original PCD point cloud map in an increment mode; carrying out registration and fusion on the PCD point cloud map after local updating and an original PCD point cloud map; the global consistency is improved by aiming at the poor global consistency existing in the laser SLAM mapping of the well industry and mining in the prior art.

The invention discloses a digital production plan scheduling method and system, and belongs to the technical field of optimal scheduling, and the method comprises the steps: constructing a distributed storage architecture based on edge computing nodes; a central coordinator is adopted to realize cross-node data synchronization through an improved Raft consensus algorithm, multi-version concurrency control is realized based on a vector clock, and a global consistent data view is established; a visual scheduling platform is built based on a Vue3 framework, and man-machine interaction is realized by adopting a Canvas and WebGL collaborative rendering framework; establishing a dynamic coordinate conversion model based on bilinear interpolation, designing a space mapping function containing distortion compensation, establishing a multi-thread coordinate service based on WebWorker, and realizing submillimeter-level bidirectional mapping of pixel coordinates and physical coordinates; constructing a three-dimensional space-time analysis model fused with the multi-dimensional features; and all the units are subjected to feature fusion through residual connection, and finally a scheduling scheme with a confidence coefficient weight is output. The method and the device have the effect of meeting various scheduling requirements.

The invention discloses a visual time sequence feature network implementation method based on multi-module feature fusion. The system comprises a visual time sequence feature encoder VTFE, a multi-scale self-attention fusion module MSAF, an efficient feature reconstruction engine EFRE and an adaptive feature refinement processor AFRP. The VTFE module fuses a convolutional neural network and an extended long-short-term memory network, and realizes high-precision capture of spatial-temporal features. The MSAF module processes characteristic frequency domain information through fast Fourier transform, and the recognition capability of complex boundaries and noise is enhanced. The EFRE module optimizes feature reconstruction through depth separable convolution and jump connection to maintain global consistency. The AFRP module integrates multi-level features and improves the detail expression ability through deep convolution and an attention mechanism. The method is excellent in performance on a large-scale OCT data set, the segmentation precision of the retina edema and the recognition capability of a fine structure are remarkably improved, and an efficient and non-invasive auxiliary tool is provided for early diagnosis of ophthalmic diseases.

The invention belongs to the technical field of unmanned driving, and particularly discloses a synchronous positioning and mapping method based on heterogeneous multi-modal data fusion. By means of data fusion of the millimeter-wave radar and the laser radar, filtering of dynamic objects, enhancement of loopback detection, global consistency optimization and the like, the problems existing when a traditional positioning and mapping method is applied to complex environments such as a tunnel are solved; the invention provides a positioning and mapping method which is more accurate, more stable and high in adaptability. The method can be applied to positioning and mapping of autonomous mobile devices such as unmanned aerial vehicles and robots in dynamic environments such as tunnels, and can provide stable positioning and mapping capabilities in the presence of dynamic obstacles (such as construction equipment and vehicles) and under low light conditions. And the positioning precision, the map construction capability and the system stability of the unmanned aerial vehicle or the robot in a complex dynamic environment are remarkably improved.

The invention discloses a personalized federated learning optimization method and system, and mainly solves the problem of poor performance of an existing personalized federated learning model. The method comprises the following steps: establishing a communication link between a client and a server; each client receives a current global sharing model parameter broadcasted by the server, loads the current global sharing model parameter to a local model, and introduces a total loss function of a dynamic alignment strength definition model; training and optimizing local model parameters, and updating shared parameters by using the parameters; the client side calculates a self-adaptive aggregation weight based on the parameter updating quantity norm, the data volume weight and the synchronous frequency weight of the client side, and uploads the self-adaptive aggregation weight and the updated shared parameters to the server; and the server receives the parameter and weight information uploaded by the server, executes global model aggregation to obtain an updated global model, and outputs the global model reaching accuracy convergence or a preset training round on the verification set. According to the method, local personalization and global consistency can be balanced, the robustness and efficiency of global aggregation are improved, and the method can be used for processing scenes of high data isomerism and dynamic change of client participation states.

The invention belongs to the technical field of three-dimensional scene reconstruction, and particularly discloses a multi-view three-dimensional Gaussian densification method and system for adaptive density control, and the method comprises the following steps: collecting a multi-view original image, and carrying out the preprocessing of the multi-view original image; complexity features are extracted, a pixel-level complexity heat map is generated, and a globally unified three-dimensional complexity field is constructed; performing back projection on the reconstruction residual error, high-frequency inconsistency and depth / geometric consistency cost of each view angle, generating three-dimensional error popularity, determining a candidate newly-added set and a candidate pruned set, generating a weak label to train a lightweight multilayer perceptron classifier, outputting a ternary probability corresponding to newly-added / pruned / maintained, and obtaining a new / pruned / maintained three-dimensional perceptron classifier; and performing Gaussian densification operation on the newly added region. By adopting the technical scheme, fine point adding is carried out on the complex area, effective pruning is carried out on the simple area, and meanwhile, the synthesis quality, the global consistency and the calculation efficiency of the new view angle are improved.

The invention discloses a distributed energy system source load coordinated optimization method based on a quasi-potential game method, and the method comprises the steps: constructing a distributed energy system model, inputting system parameters, and predicting renewable energy power generation and initial load demands. In a source side optimization stage, a leader layer potential function of a quasi-potential game is established by taking minimization of source side cost # imgabs0 # as a target, and an initial power generation plan is generated by comprehensively considering economical efficiency and carbon emission constraints; and then, based on a scheduling result, calculating a carbon potential epsilon t of each node and a dynamic carbon emission factor # imgabs1 # of each stage, optimizing a load side response based on an LCDR scheme, constructing a local potential function of a follower layer, reflecting a relationship between a user profit maximization target and carbon emission, and adjusting user behaviors through a distributed decision. And the updated load demand is fed back to the source side, and the source side optimizes the output plan of each unit based on the updated load, so that an iterative process of source side potential function optimization-load side equilibrium response is formed, and an optimal scheduling strategy and scheduling result of the energy supply side are obtained. According to the framework, the global consistency requirement of a traditional potential game is relaxed, independent optimization of source-load two sides under the guidance of respective potential functions is allowed, and a Nash equilibrium state is finally achieved only by ensuring monotonous convergence of total potential energy of a system in an iteration process. According to the method, the convergence advantage of the potential game is reserved, the method is also adapted to the characteristics of a source-load heterogeneous decision subject, efficient consumption of renewable energy and collaborative optimization of carbon emission are realized through bidirectional transmission of the carbon potential signal, and the overall efficiency of the system is remarkably improved.

The invention discloses a heterogeneous computing system, a cache consistency maintenance method and device, equipment and a medium, and relates to the technical field of heterogeneous computing. The method comprises the steps that when a data operation request is detected, if a receiving party and a transmitting party are a cluster cache layer and a global cache layer, the protocol format of the data operation request is converted according to the mapping relation between a group consistency protocol and a global consistency protocol. And when the data is forwarded to the local cache layer, executing corresponding operation based on the memory consistency model of the calculation unit to which the data belongs, and maintaining the consistency of the local cache. And when the data is forwarded to a cluster cache layer, corresponding operation is executed based on the cluster consistency protocol, and the consistency of the cluster shared cache is maintained. And when the data is forwarded to a global cache layer, maintaining the consistency of the shared caches of the clusters based on a global consistency protocol. The method and the device can solve the problem of poor compatibility and expansibility of related technologies, can be compatible with various computing units and different memory consistency models thereof, and are easy to expand.

The invention discloses a dynamic object suppression video image stabilization method based on depth information and global consistency, and the method comprises the steps: extracting key points in a video, carrying out the optical flow estimation, and obtaining the depth information of the key points; a static weight and a motion vector residual error are calculated based on the motion information, the depth information and the global homography matrix of each key point, a dynamic weight is obtained based on an attention mechanism, the dynamic weight and the static weight are fused to obtain a comprehensive weight, and a weighted motion residual error is calculated based on the comprehensive weight and the motion vector residual error; spreading the motion information to grid vertexes, obtaining dense residual motion based on weighted motion residual prediction, and accumulating the dense residual motion to obtain a track; and carrying out smoothing and motion compensation on the track and outputting a stable video sequence. The problem that in a complex scene, motion of a dynamic object causes distortion and inaccuracy of a stabilization result is solved, interference of the dynamic object can be effectively suppressed, and the effect and robustness of video image stabilization are improved.

The invention relates to the technical field of medical data processing, and discloses a medical data source real-time processing and quality control method based on distributed computing. The method comprises the steps of obtaining and classifying multi-source heterogeneous medical data streams, constructing a distributed parallel processing framework, setting an initial constraint condition, executing distributed streaming computation to generate a primary processing strategy, and optimizing and generating a global consistency processing strategy by using a federated learning algorithm. The system comprises a plurality of modules such as a data access and classification module and a processing framework construction module. According to the method, multi-source heterogeneous medical data can be efficiently processed, real-time calculation is realized through a multi-dimensional processing space and dynamic task allocation, and the data quality and privacy are guaranteed by utilizing a federal learning optimization strategy. And equipment abnormity can be monitored, quality risks can be predicted, task scheduling can be managed, the accuracy, reliability and availability of medical data processing can be improved, and powerful support can be provided for medical decision making and the like.

The invention discloses an intelligent segmentation method and device for ceramic matrix composite CT image defects, and belongs to the technical field of image detection. The method comprises the following steps: inputting preprocessed CT image data into a preset double-branch collaborative segmentation model, and respectively outputting to obtain a local texture feature map and a global relation feature map; performing three-branch material perception fusion processing on the local texture feature map and the global relation feature map according to a preset perception fusion model to obtain a fusion feature map containing local detail accuracy and global consistency; sequentially carrying out progressive decoding and refined reconstruction processing on the fused feature graph to obtain a defect probability graph; and calculating a loss function of the defect probability graph and a manually pre-labeled real defect segmentation graph, and dynamically adjusting the loss weight of the initial model according to a calculation result to obtain a CT image defect segmentation model meeting a preset convergence condition. The method can meet the detection requirements of industrial scenes on the defects of the ceramic-based composite material.

The invention relates to the technical field of video analysis, in particular to a video editing method based on grid layout alternate diffusion and multi-attention control, and the method comprises the steps: segmenting an original video frame sequence into a plurality of grids, each grid comprising a plurality of pixel space video frames which are continuously arranged, and forming grid data; mapping the gridding data to a low-dimensional submerged space through an encoder, and generating initial submerged space feature data; the initial submerged space feature data are edited, the editing process comprises a diffusion process and a sampling process, the diffusion process is based on a pre-trained stable diffusion model, and a time attention module is embedded in the diffusion process; in the sampling process, executing an odd-even time step alternate replacement strategy on the grid layout to promote cross-grid global consistency, and dynamically fusing attention maps of a reconstruction branch and an editing branch according to a timestamp threshold to generate de-noised data; and decoding, splitting and recombining the de-noised data through a decoder to generate an edited continuous video frame sequence.

The invention discloses a distribution line load prediction and optimal scheduling method and system, and relates to the technical field of intelligent scheduling of power systems, and the method comprises the steps: generating a time-aligned multi-source fusion input data set; constructing a mixed time sequence load prediction model, and introducing a weighted quantile loss function in a model training process; constructing a joint probability distribution model of the renewable energy output and demand response participation rate, and sampling joint probability distribution; constructing a rolling time domain power distribution network optimization scheduling model; a two-layer mixed strategy is adopted to deal with uncertainty, an optimization problem is decomposed into a plurality of sub-problems, and an alternating direction multiplier method with adaptive penalty parameters is used for distributed solution. According to the method, a multi-objective optimization scheduling model is established in a rolling time domain, and dynamic closed-loop optimization is realized; and by introducing a two-layer hybrid solving strategy and an ADMM distributed algorithm with an adaptive penalty parameter, the calculation efficiency and expandability are remarkably improved while the global consistency is ensured.

The invention discloses an autonomous decision-making system of a multi-level intelligent agent in an urban area and an operation method thereof. Each end-side decision-making unit broadcasts an equipment state variable to a side-side decision-making unit; the side decision-making unit determines a global reference state and generates a control instruction of each end decision-making unit; after the equipment executes the control instruction, each end-side decision-making unit updates an equipment state variable according to the real-time operation data of the equipment, and when an abnormal event is judged to occur, a side decision-making unit predicts a decision-making variable of each piece of equipment based on a collaborative optimization model of an equipment target and an urban area target; the cloud regulation and control platform predicts a global consistency variable based on the collaborative optimization model of all the side decision units; predicting a regulation and control instruction of each side decision-making unit according to a global dynamic optimization target under a system operation constraint; and the side decision-making unit decomposes the regulation and control instruction into the control instruction of each end decision-making unit, so that the problems that the global cooperative capability of the urban regional integrated energy system is weak and cooperative scheduling is not timely under extreme events are solved.

The invention provides a multi-granularity open vocabulary query method based on an object-level lossless Gaussian field, and the method comprises the steps: introducing an object-level Gaussian field with a global consistency codebook, rendering a learnable semantic tag vector in the Gaussian field back to a corresponding object tag; the direct mapping between the label and the corresponding uncompressed high-dimensional feature is established through the code book, so that the semantic feature of any dimension is supported, additional compression is not needed, and the understanding capability on an object is remarkably improved; according to the method, wide quantitative and qualitative evaluation is carried out in a plurality of scenes, excellent performance in the aspects of object level zero sample segmentation and open vocabulary understanding is shown, the highest precision is particularly achieved in object-component hierarchical retrieval, and meanwhile multi-granularity scene editing is supported.

The invention provides a burn wound area segmentation method and system, and the method comprises the steps: obtaining a to-be-segmented image through constructing an anatomy database, and employing a lightweight CNN classifier to extract a feature vector; determining a healing stage of the burn wound according to the feature vector, and adjusting boundary parameters according to the healing stage; inputting an image to be segmented into the improved U-Net encoder network, and outputting a segmentation mask and a boundary probability graph; determining an interaction area according to the boundary probability graph, carrying out interaction in the interaction area by referring to the segmentation mask, and generating a smooth boundary according to interaction data; and post-processing optimization fused with prior constraints is performed on the smooth boundary to obtain a target boundary so as to complete burn wound area segmentation, specific and prior knowledge provides basic constraints for segmentation, interactive operation pertinently corrects a complex area, post-processing realizes global consistency optimization, and a'constraint-correction-optimization 'closed-loop processing flow is formed.

The invention discloses an underground pipeline three-dimensional deformation quantitative inspection method, which comprises the following steps: processing a multi-source sensing sequence to generate a baseline data packet; the method comprises the following steps of: performing first-stage on-line correction, acquiring a medium state agent quantity by applying micro-excitation sampling combined by a micro-amplitude angle and double power to laser, and performing joint inversion to obtain section geometric data subjected to medium correction by utilizing a forward model with endogenous refraction and scattering correction items; second-stage scale robustness processing is carried out, in a unified optimization framework, motor current-based dynamic consistency constraint, geometric conservation constraint and wellhead external reference strong constraint are fused, and a globally consistent pose track is solved; and performing three-dimensional reconstruction and deformation quantification according to the corrected section data and the pose track. According to the invention, the accuracy of geometric measurement of the pipeline and the global consistency of the pose track in a complex environment such as a humid environment can be improved.

The invention discloses a personalized federated learning method based on a kernel distance between clients, a framework and an application, and belongs to federated learning and an application technology thereof. The method aims at double challenges of data isomerism and privacy protection, and is based on a federated learning framework of a conditional policy network and differential privacy. The CPN adaptively balances the conflict between the global consistency and the local personalized characteristics by dynamically generating the weights of the personalized characteristics and the global characteristics, so that the influence caused by data isomerism is effectively relieved. Meanwhile, the differential privacy technology protects the sensitive data characteristics of the client and reduces the risk of privacy disclosure by injecting Gaussian noise in gradient updating. The personalized federal learning method provided by the invention not only is superior to the existing mainstream method in global model performance and fairness between clients, but also realizes good balance between privacy protection and model performance.

The invention relates to a blade-like curved surface global consistent geometric design method based on a PHT spline, and the method comprises the steps: carrying out the high-precision fitting of a blade-like complex curved surface and the geometric features of the blade-like complex curved surface through employing a PHT-spline curved surface. According to the main technical scheme, the method comprises the following steps: splicing and cutting a parameter domain into a globally consistent complete single parameter domain as a definition domain of a PHT spline through a boundary consistency re-parameterization process; geometric features and boundary information are comprehensively considered, a to-be-fitted data point set is obtained through reasonable sampling, and corresponding high-order geometric information is calculated; and performing high-precision fitting by controlling hierarchical subdivision of a grid to obtain a blade curved surface represented based on a PHT spline. According to the method, the problem of discontinuity caused by cutting and splicing in a traditional design method is solved, the global consistency of the overall design is ensured, and meanwhile, the design precision and efficiency are improved.

The invention discloses a method and system for constructing a city information model, and belongs to the field of city intelligent management, and the method comprises the steps: carrying out the spatial registration and semantic alignment of future multi-source heterogeneous data through a unified coordinate reference, and constructing a multi-dimensional incidence relation between a spatial entity and the attribute of the spatial entity through a graph database; performing time serialization processing on the city information model based on the real-time sensing data, and performing real-time correction on the geometric state and the attribute of the model by adopting an incremental modeling algorithm; performing automatic calibration on the model data in combination with rule reasoning and probability correction methods; predictive calculation is carried out on multiple scenes, and parameterized optimization is carried out on the city information model according to a calculation result; and carrying out adaptive hierarchical abstraction on the city information model, and automatically generating model subsets with different precision levels. According to the method, spatial registration and semantic integration are performed on the data, so that the global consistency of the city information model is realized, and the problems of data inconsistency, repetition and redundancy in a traditional method are solved.

The invention discloses a video generation method and device, computer equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the steps of inputting a to-be-processed image into an image processing model to obtain a first image feature; inputting the to-be-processed image into a global encoder to obtain a second image feature; combining the first image feature and the second image feature to obtain a combined image feature; based on the combined image features, using a first video generation model to generate an initial video corresponding to the to-be-processed image; obtaining video text description, and encoding the video text description to obtain text encoding features; and inputting the initial video and the text coding feature into a second video generation model to obtain a target video corresponding to the to-be-processed image. The invention also relates to the technical field of block chains, and the to-be-processed image is stored in a block chain network. According to the method, visual details can be reproduced, global consistency can be kept, and the visual quality and semantic consistency of the video are effectively improved.

The invention provides a brain tumor classification method based on deep learning, and the method comprises the steps: uniformly segmenting a multi-mode magnetic resonance imaging image into a plurality of small blocks, generating an enhanced puzzle through random geometric transformation, inputting the enhanced puzzle into a ResNet50 backbone network, extracting primary local features, and optimizing feature distribution; multi-scale feature maps of different stages are obtained, lightweight convolution blocks are constructed, and local pathological relevance is enhanced; performing full-connection layer splicing on the multi-scale feature maps in different stages, introducing a learnable weight matrix to dynamically allocate weights in each stage, and generating fusion features; a progressive parameter unfreezing mechanism is adopted to ensure orderly learning of the model from local to global; the sample weight is dynamically adjusted according to the category frequency, and the local discriminant force and the global consistency are balanced in combination with multi-stage classifier loss weighted fusion; and a final classification result is generated through multi-stage prediction probability weighted average and dynamic threshold adjustment, and clinical availability is improved in combination with a sigmoid calibration module.

The present invention relates to the technical field of wireless sensor networks, and to an average consistency time synchronization method based on a multi-hop virtual link and event triggering. According to the method, in the case that a communication delay is present in a wireless network, a least squares estimator is used to restrain the effect of the delay in the relative frequency offset estimation process; clock information of nodes is fully called by means of the multi-hop virtual link for update, thereby effectively increasing the clock synchronization speed; an event triggering policy is used to reduce a transmission frequency between the nodes, thereby effectively reducing communication overhead; and a consistency algorithm is used to update logic clock parameters of the nodes, thereby realizing global consistency of clocks of the nodes in a wireless sensor network. According to the present invention, the effect of the communication delay in the wireless sensor network is taken into consideration, an iterative least squares estimator is used to obtain a high-precision relative frequency offset value to improve the synchronization precision, and by using an interaction mechanism of the multi-hop virtual link and event triggering, the communication overhead is effectively reduced while increasing the convergence speed of a clock synchronization error.

The invention discloses a real scenic spot cloud data and BIM point cloud data fusion method and device, and belongs to the technical field of point cloud registration, and the method comprises the steps: obtaining real scenic spot cloud data of a railway scene; building a building information model BIM of the railway scene; based on the BIM of the railway scene, acquiring BIM point cloud data of the railway scene; performing multiple registration on the BIM point cloud data and the real spot cloud data; converting the registered BIM point cloud data and the real scene point cloud data into a global coordinate system to obtain global point cloud data; and correcting the global point cloud data by taking the geometric consistency function and the global consistency function as constraints to obtain a fusion result of the real scenic spot cloud data and the BIM point cloud data. According to the method, high-precision fusion of the real spot cloud and the BIM point cloud can be realized.

The invention discloses a multi-task image processing method, system and device, and relates to the technical field of image processing, and the method comprises the following steps: constructing a multi-task network model; inputting the image data set into a multi-task network model to respectively obtain an infrared enhanced image, an infrared super-resolution image, an infrared and visible light image fusion image and an infrared image colorization image; and calculating a loss function between different images output by the multi-task network model and the infrared image and the visible light image in the input image data set, and carrying out loop training on the multi-task network model. According to the technical scheme, through unified training of the multi-task model, local details are effectively extracted, global consistency is kept, the processing efficiency and the image quality are greatly improved, redundant steps in the processing process are reduced, and the robustness and adaptability of the system are improved.

The invention provides a three-dimensional point cloud reconstruction method and application thereof in target detection. The point cloud reconstruction method comprises the following steps: acquiring an incomplete original point cloud; the original point cloud is input into a trained point cloud completion model, a complete point cloud is obtained, and the point cloud completion model comprises a feature extraction module, a feature generation module, an encoder, a query generator, a decoder and a reconstruction head. According to the method, automatic completion of incomplete point clouds is realized by constructing a point cloud completion model, global consistency and local detail expression are considered during feature extraction and generation, and the multi-scale representation capability and generalization performance of the point clouds are remarkably enhanced; the feature of the prediction center point is quickly generated through an encoder-decoder structure, the missing point cloud is predicted based on the reconstruction head, all the point clouds are fused, and the point clouds can be complemented and predicted more accurately through the method.

The invention belongs to the technical field of artificial intelligence, and particularly relates to an AI-based snowfield trajectory and extreme environment intelligent identification method, which comprises the following steps: a multi-modal data acquisition module; a fusion type deep learning model; a confrontation generation data enhancement unit; an adaptive control subsystem; and an edge calculation node deploys a lightweight model inference engine to realize low-delay trajectory feature extraction and classification decision. According to the method, a multi-scale feature pyramid structure and a cross-modal attention mechanism are adopted, the convolutional neural network and the recurrent neural network are combined, and spatial features and modeling time sequence dynamics can be extracted at the same time. The synthetic data of the extreme snowfield scene is generated by using the conditional generative adversarial network, and the physical constraint generator and the multi-scale discriminator are combined, so that the real snow particle motion trail can be simulated, and the global consistency and the local texture authenticity of the generated data are ensured.

The present invention relates to the field of computer vision technology, and specifically to an infrared and visible light image fusion method with adaptive feature extraction, comprising: using a single-modal spatial adaptive feature mixing module to perform multi-scale spatial adaptive feature extraction on two modal images; inputting the extracted features into a dual-branch feature extraction module to obtain low-frequency global features and high-frequency local features; then splicing the dual-modal features, and reconstructing the fused image using a dual-modal spatial adaptive feature fusion module. In order to solve the imbalance problem between modality-specific features and shared features, decomposition loss is introduced into the loss function to further optimize the fusion model. The present invention adopts a dual-branch feature extraction module to establish local and global contextual dependencies, and extracts potential features of the modalities at multiple scales to generate a fused image with rich details and global consistency, which can perform well even in harsh environments and improve the effects of downstream visual tasks.