4790 results about "Network parameter" patented technology

The invention relates to the technical field of twinborn model simulation, and discloses a twinborn model simulation method and system for hot working of large forgings. The method comprises the following steps: collecting and preprocessing process parameters, quality data and environment information in a multi-source manner, and obtaining hot working characteristic data; performing correlation analysis to construct a process knowledge graph; calculating the distribution of a temperature field, a stress field and an organization field by using a self-sensing multi-field coupling var value neural network; comparing and analyzing to obtain deviation data and correction parameters; adjusting a network parameter optimization prediction result; and executing multi-objective optimization calculation, and generating a whole-process technological parameter and a control instruction. According to the method, full-process multi-physics field coupling calculation from smelting, casting, forging and pressing to heat treatment can be achieved, model parameters are dynamically adjusted according to real-time production data, the optimal process scheme is generated, and therefore the manufacturing quality and efficiency of large forgings are improved.

A system and method for adaptive edge-cloud data processing dynamically distributes computational tasks between edge devices and cloud infrastructure in response to changing conditions. The system continuously monitors resource availability, network parameters, and workload characteristics while predicting future conditions using hierarchical forecasting models. A multi-objective optimization approach determines optimal task distribution, balancing processing latency, energy consumption, bandwidth utilization, and result quality. The system implements a partitionable processing pipeline that enables seamless task migration through state synchronization protocols and checkpoint mechanisms. During migration, the system preserves processing continuity by establishing dependencies, creating execution checkpoints, and verifying successful state transfer. Performance metrics may be continuously collected and analyzed to improve future decision-making. The system maintains operational resilience during connectivity disruptions through local decision-making capabilities and eventual consistency protocols, making it suitable for diverse applications including industrial IoT, connected vehicles, healthcare wearables, and smart city infrastructure.

The invention provides a decision model construction method based on a big data environment. The method belongs to the economic data decision field. The method comprises the following steps: firstly, acquiring multi-source heterogeneous data including sensor time sequence data, an expert rule base and equipment causal priori knowledge, and constructing a standardized training data set; then, extracting an explicit causal relationship from the historical data by utilizing a causal discovery algorithm, and generating an interpretable knowledge graph in combination with priori knowledge of an equipment manual; thirdly, constructing a neural symbol joint model; neural network parameters and rule confidence coefficients are synchronously adjusted, and collaborative learning of data driving and knowledge driving is realized. And then, carrying out multi-dimensional logic verification on the trained model, including rule conflict detection, anti-factual reasoning and decision path traceability verification. The method can effectively improve the accuracy and interpretability of the decision model, and is suitable for processing decision tasks in a big data environment.

The invention provides an igniter robot production line control method and system based on deep reinforcement learning, and relates to the technical field of robot production line control, and the method comprises the steps: collecting the real-time state data of a production line station, extracting features through a multi-scale residual neural network and a spatial pyramid pooling module, and carrying out the feature extraction; channel attention and space attention modules are combined to perform feature enhancement, a cross-modal encoder is used to perform time sequence modeling, task decomposition is performed based on hierarchical reinforcement learning, soft constraint trajectory optimization and adaptive impedance control are used to generate a station control instruction, and a multi-target Pareto evaluation and priority dual experience pool is used to perform sample screening. Network parameters are optimized and updated through a near-end strategy, and a production line optimization decision scheme is output, so that the production line control precision can be improved, the system robustness can be enhanced, and the optimization of the production efficiency is realized.

The invention discloses an artificial intelligence-based engineering resource allocation optimization method and system, and relates to the technical field of artificial intelligence and resource management crossing. The problems of resource configuration dynamic change and resource conflict coordination are solved. And processing the multi-modal heterogeneous data in the target engineering scene through space-time alignment and semantic coding to generate a dynamic data stream. A task, resource and environment node relationship is constructed based on a dynamic heterogeneous graph network, and a node dependency weight is updated through an event-driven mechanism. Two-stage collaborative optimization is executed, in the first stage, a baseline scheme is generated through Monte Carlo sampling, and in the second stage, resource conflicts are eliminated through back propagation negotiation. The resource dynamic entropy is monitored in real time, a rebalance algorithm is triggered during local overload, and a distribution scheme is adjusted in combination with security constraints. And finally, outputting the optimization scheme to an engineering management system for execution, dynamically returning updated graph network parameters, forming a data acquisition, optimization decision and feedback closed loop, and improving the intelligence and robustness of resource allocation.

The invention discloses an unmanned aerial vehicle trajectory planning and tracking method, system and device based on deep reinforcement learning and adaptive nonlinear model predictive control, and a medium. The method comprises the following steps: constructing various static multi-obstacle and dynamic multi-obstacle simulation environments; constructing a kinetic model of the unmanned aerial vehicle; constructing an adaptive nonlinear model predictive control (ANMPC) algorithm; constructing a reward function of the tracking performance of the unmanned aerial vehicle to the reference trajectory generated by the adaptive nonlinear model predictive control algorithm; constructing a network framework based on deep reinforcement learning and an adaptive nonlinear model predictive control algorithm; setting network parameters; training a network framework, and selecting an optimal weight file; outputting a test result; the system, the device and the medium are used for realizing the unmanned aerial vehicle trajectory planning and tracking method. The method can effectively cope with changes of targets and environments, shows strong obstacle avoidance capability and anti-interference performance when facing dynamic obstacles and wind noise interference, and embodies a high intelligent decision-making level.

The invention discloses a semi-supervised medical image segmentation method based on eye movement guide mixed data enhancement. The method comprises the following steps: collecting fixation position information of a doctor when reading a medical image through eye movement tracking equipment; using a teacher-student network model and adopting a two-stage training mode to train the model; the method comprises the following steps: taking GazeMix as a data enhancement strategy in training, extracting a fixation area in a labeled or unlabeled image, and pasting the fixation area to the background of other unlabeled images to generate a mixed image data pair; in the first stage, a GazeMix data enhancement strategy is combined to pre-train a teacher network; in the second stage, a student network is trained, GazeMix generates diversified data, and meanwhile, a teacher network is pre-trained for pseudo tag generation; multi-scale gaze perception is adopted, and model perception information is extracted on different scales and used for perception loss calculation; and the optimizer iteratively optimizes the student network parameters, and updates the teacher network by using an index moving average method.

The invention discloses an access anomaly analysis method and system based on multi-dimensional features and user behaviors, and relates to the technical field of dynamic access anomaly detection, and the method comprises the steps: based on a dynamic hypergraph structure, extracting high-order correlation features of the user behaviors through a multilayer hypergraph convolutional network, and generating a high-order feature matrix; based on the high-order feature matrix, generating an authority approval threshold through a causal reinforcement learning framework, constructing a user behavior causal graph to generate strategy network parameters, and storing the strategy network parameters to distributed nodes of a regional data center; based on strategy network parameters stored by distributed nodes, security multi-party computing is adopted, cross-node collaborative optimization is carried out, and global defense strategy parameters are generated through a security aggregation algorithm. According to the method, security multi-party computing is adopted, cross-node collaborative optimization is performed, and the global defense strategy parameters are generated in combination with homomorphic encryption and a block chain fragmentation technology, so that the collaboration efficiency and strategy consistency among distributed nodes are improved on the premise of ensuring data privacy.

The embodiment of the invention provides a chip platform hardware abstraction layer construction method, which comprises the following steps: scanning chip platform hardware characteristic information, constructing a hardware characteristic analysis model, carrying out classified labeling on the hardware characteristic information to extract equipment interfaces, resource attributes and operation characteristics, establishing a characteristic vector library of hardware components, and constructing a hardware abstraction layer of the hardware components, grouping the hardware components by adopting a hierarchical clustering algorithm to form a function category tree; training an adaptive abstract model based on the function category tree, mapping hardware components with similar functions to a unified interface space, and calculating target Q value update strategy network parameters based on a time sequence difference algorithm to generate a hardware abstract rule base; loading a hardware abstraction rule base in a hardware abstraction layer engine, dynamically monitoring a resource use state, updating an allocation strategy, and returning a hardware operation result to an application layer through a uniform interface; according to the method, hardware features can be deeply understood, an intelligent interface packaging strategy is constructed, and efficient scheduling of resources is realized through technologies such as reinforcement learning.

The invention discloses a multi-agent robot cooperative control method based on distributed reinforcement learning, and the method comprises the steps: building a multi-agent reinforcement learning algorithm environment, and defining a state space, an action space and a reward function; constructing a distributed reinforcement learning training framework which comprises a sampling module, a training module and a scheduling management and storage module; based on an actor-commentator architecture, constructing an independent actor network and a commentator network for each robot agent; through a distributed training framework, sample data sharing of the sampling module and neural network parameter sharing and cooperative training of the training module are realized; and finally, loading the trained neural network into a multi-agent robot system, and carrying out cooperative control according to environment changes and task requirements. Therefore, the robot intelligent body of each node can share experience, and cooperative training is shared through neural network parameters, so that a cooperative control strategy is realized, and the overall performance and robustness of the system are improved.

The invention discloses an online reinforcement learning data increasing and expanding method based on a diffusion model. The method comprises the following steps: processing original trajectory data to generate enhanced feature data; extracting global structure representation and local detail representation; carrying out fusion by adopting a hierarchical integration method to obtain unified hierarchical representation data; generating constraint coding data based on pre-stored physical constraint parameters and task constraint parameters, and fusing to generate constraint perception representation data; generating an evaluation result based on the current strategy parameter; performing adaptive adjustment on the constraint perception representation data to obtain adjusted representation data; modal analysis is carried out, diversified trajectory data is generated, and a candidate trajectory set is obtained; and carrying out quality evaluation, carrying out trajectory screening through a multi-objective optimization method, and updating strategy network parameters to obtain updated strategy parameters. According to the invention, high-quality training data can be automatically generated, and the training efficiency and performance of a reinforcement learning system are improved.

The invention relates to the technical field of information synchronization, in particular to a cross-platform electronic communication synchronization system, which comprises a network construction and path optimization module for collecting basic network parameters of communication equipment of a plurality of platform networks, obtaining cross-platform network data according to the basic network parameters, selecting an optimal transmission path according to the cross-platform network data, and transmitting the optimal transmission path to a server; and generating real-time optimized path information. In the invention, the basic network parameters of the communication equipment of the plurality of platform networks are collected, the cross-platform network data are acquired, the edge weight is dynamically updated according to the delay and the bandwidth, and the optimal transmission path is selected based on the network state, so that the effect of efficiently utilizing network resources is achieved, and the real-time performance and the accuracy of synchronous data transmission are remarkably improved. The to-be-synchronized data is classified and subjected to priority setting, and the data processing sequence is adjusted in combination with the real-time network data, so that accurate scheduling of the synchronization task is realized, the delay and conflict problems in data transmission are reduced, and the efficiency and reliability of data synchronization are improved.

The invention discloses an intelligent flow arrangement method based on fusion expert network and deep reinforcement learning, which comprises the following steps: collecting network node and link state data in real time, and constructing a time sequence input vector and a topological graph structure; a time sequence neural network and a graph neural network are used for extracting traffic spatial-temporal features and node topological features respectively, future traffic is predicted through a classification network after fusion, and coarse-grained arrangement of network slices of different service levels is completed; modeling resource scheduling into a multi-agent Markov decision process, and designing a state space, an action space and a reward function; a deep reinforcement learning agent is initialized, and training is carried out through interaction experience; fusing a pre-trained expert strategy network, and constructing a total loss function to optimize network parameters; and finally generating an intelligent strategy capable of dynamically optimizing the flow path and resource allocation according to the real-time state. According to the invention, efficient resource scheduling under multi-service differentiation service quality requirements can be realized.

The invention provides a factory dynamic production scheduling optimization method and system based on deep reinforcement learning, and the method comprises the steps: firstly obtaining a real-time production monitoring data set of a target factory, including a plurality of production link monitoring sequences which are composed of parameters such as an equipment operation state, an order execution progress and environment interference; extracting features of the real-time production monitoring data set to obtain a target production feature set containing equipment state dynamic and other features, calling a pre-trained deep reinforcement learning strategy network, performing dynamic constraint matching processing to generate dynamic production constraint features, and performing iterative optimization on the basis of the dynamic production constraint features and a preset production scheduling optimization objective function strategy to obtain a real-time production monitoring data set; and generating a real-time production scheduling optimization strategy set containing sequences such as equipment scheduling priorities and the like, finally inputting the real-time production scheduling optimization strategy set into a production scheduling control system, triggering an instruction dynamic adjustment operation, updating deep reinforcement learning strategy network parameters according to feedback data, and realizing dynamic production scheduling optimization of the factory.

The invention belongs to the technical field of artificial intelligence chip acceleration, and provides an acceleration method for an expert hybrid model to execute an operation task and related equipment. The method comprises the steps that a to-be-processed operation task is acquired, a target expert network used for executing the to-be-processed operation task in an expert hybrid model is determined, and network parameters of all first expert networks in the expert hybrid model are stored in an on-chip memory of an accelerator chip, network parameters of each second expert network in the expert hybrid model are stored in an off-chip memory of the accelerator chip; calling a first network parameter of a first expert network in the target expert network from the on-chip memory, and calling a second network parameter of a second expert network in the target expert network from the off-chip memory; and executing the to-be-processed operation task through a target expert network in the expert hybrid model based on the first network parameter and the second network parameter. Through the technical scheme provided by the invention, the acceleration efficiency of executing the operation task by the expert hybrid model can be improved.

The invention provides a complex scene-oriented AI large model lightweight deployment method, and relates to the technical field of edge computing, and the method comprises the steps: carrying out the structured pruning of a pre-trained Transform network based on the attention head importance score, carrying out the dynamic sparsification of the activation state of a feedforward network according to the input tensor entropy value, employing the dynamic mixing precision quantization, and carrying out the reconstruction of an AI large model. Obtaining network parameters after pruning quantization; deploying the pruned and quantized network parameters to an edge computing device, distributing a feature extraction operator to a neural network processor through a heterogeneous computing scheduler, and unloading a classification operator to a multi-core central processing unit; and managing an on-chip memory in combination with a virtual memory paging mechanism, realizing zero-copy data transmission by utilizing a direct memory access controller, and outputting a reasoning result tensor. According to the method, efficient and reliable operation of the large model at the resource-constrained edge node is realized.

The invention discloses a non-impact grid-connected recovery method, system and device for a coal mine isolated network power supply system and a medium, and belongs to the technical field of non-impact grid-connected recovery of power supply systems, and the method comprises the steps: monitoring the operation state of an isolated network, and obtaining voltage, frequency and phase parameters; preparation before grid connection is executed, and main network parameter confirmation and equipment verification are completed; performing parameter synchronous adjustment based on the controller to complete dynamic matching of voltage, frequency and phase; pre-synchronization control is implemented to realize phase prediction and power pre-balance; executing closing operation and performing impact suppression; carrying out stability regulation and control after grid connection, wherein the stability regulation and control comprise oscillation suppression and smooth load loading; and when grid connection is abnormal or fails, an emergency processing branch is entered, and standby power supply switching or island splitting control is triggered. Through multi-dimensional synchronous control, impedance reconstruction and hierarchical coordination mechanisms, high-precision and low-impact grid connection of multiple energy units under extreme working conditions is realized, and the grid connection stability and success rate of a coal mine isolated network power supply system are remarkably improved.

The invention discloses an AI model adaptive tuning method based on deep reinforcement learning, and belongs to the technical field of industrial process control, the method comprises the following steps: firstly, constructing a deep reinforcement learning environment comprising a state space, an action space and a reward function, the state space comprising a process variable measurement value, a controller parameter and a performance index; the action space is a parameter adjustment quantity vector, and the reward function comprehensively considers the stability, rapidity and accuracy of the system; then, based on a deep Q learning network of a dual-network structure, training data are stored and sampled by adopting an experience playback mechanism, and network parameters are iteratively optimized through a deep Q learning algorithm; and finally, realizing online adaptive adjustment of controller parameters by using the optimized deep Q network. The method has the advantages of being high in adaptability, good in optimization performance, high in practicability and the like, intelligent optimization control over an industrial process control system can be achieved, and the overall performance of the system is improved.

The invention provides a short-time rainfall prediction method based on an MSPTS-Net network model. The short-time rainfall prediction method comprises the following steps: acquiring a radar echo data set formed by radar echo images; a network model is constructed based on MSPTS-Net, a spatial-temporal feature extraction and fusion mechanism is embedded in a framework of the MSPTS-Net, and an initial short-term rainfall prediction model is generated; training the initial short-time rainfall prediction model by using the radar echo data set, and carrying out reverse adjustment and optimization on network parameters through a combined loss function to obtain a final short-time rainfall prediction model; and inputting a radar echo image to be measured into the final short-time rainfall prediction model to obtain a short-time rainfall prediction result. Through the MSPTS-Net model, spatio-temporal change information in radar echo data can be more comprehensively captured, multi-scale feature extraction and a self-attention fusion mechanism are utilized, data changes of different time scales and spatial scales are processed, a combined loss function is combined, the model training process is more efficient, convergence to an optimal solution can be faster, and the method is suitable for being applied to the field of radar imaging. And the precision of short-time rainfall prediction is improved.

The invention relates to the technical field of power system dispatching, and discloses a multi-agent reinforcement learning active power distribution network area coordination method and system, and the method comprises the steps: dividing a power distribution network into a plurality of areas, and each area is managed by an agent; collecting observation information; inputting the observation information into an upper reinforcement learning strategy network, and outputting control parameters; inputting the control parameters into a target function of the lower-layer local physical optimization model, and solving an output setting point of the equipment under the condition of meeting the safety operation constraint; constructing a De-POMDP problem, and obtaining a reward signal of each agent; a sequential updating mechanism is introduced, global network parameters are optimized, and corresponding decisions are obtained; and inputting the multi-agent decision into the global active power distribution network model to obtain the total operation cost, feeding back the total operation cost as an award to the reinforcement learning strategy network, updating global network parameters, and converging to obtain an optimal decision. According to the invention, regional wind-solar-storage multi-energy scheduling can be effectively optimized, and energy balance in the region is realized.

The invention provides a task allocation and optimal scheduling method and system for a multi-robot collaborative industrial assembly line, and relates to the technical field of robots, comprising the steps of constructing a task allocation strategy network through deep reinforcement learning, encoding workpiece types and attribute information into a state vector input network, a multi-objective optimization function including total assembly time minimization, load balance degree maximization and energy consumption minimization is designed, a near-end strategy optimization algorithm is adopted to train network parameters, a robot task mapping matrix is output as a distribution scheme, and the network parameters are dynamically optimized and updated in an online learning mode. The task allocation efficiency of the multi-robot collaborative assembly production line can be remarkably improved, the dynamic optimization scheduling of the production line is realized, the system energy consumption is reduced, and the assembly efficiency is improved.

The invention provides an intelligent drilling speed prediction method based on physical feature guidance and multi-source information fusion, and relates to the technical field of intelligent drilling speed prediction, and the method specifically comprises the following steps: collecting multi-source heterogeneous data from a drilling real-time database, a logging system, a logging system and a geological database; constructing a dual-channel deep learning prediction model, wherein the dual-channel deep learning prediction model comprises a dual-channel convolution feature extraction module, a feature fusion module, a time sequence fusion module, a time sequence modeling module and a full connection layer which are connected in sequence; obtaining a predicted drilling speed by using a dual-channel deep learning prediction model; a joint loss function is constructed by considering a data driving error and a physical constraint error, an error is calculated according to the joint loss function, and network parameters are updated through back propagation; carrying out loop iteration training until convergence; and the trained dual-channel deep learning prediction model is used for drilling speed prediction. According to the technical scheme, the problems that in the prior art, a mechanism model is insufficient in precision, and a data driving model is poor in reliability are solved.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical health and the like, and discloses a natural language driven table processing method, device, equipment and medium. And generating and executing an operation action by utilizing the reinforcement learning strategy network, generating a reward signal according to execution result data and user feedback information, storing the state vector representation, the target table operation action and the reward signal into an experience playback queue, and updating network parameters of the reinforcement learning strategy network based on historical data in the experience playback queue. According to the method, the state vector is constructed through the reinforcement learning strategy network in combination with the natural language instruction and the table context, and the operation strategy is continuously optimized according to the operation result and the user feedback, so that intelligent understanding and action planning of the spreadsheet operation intention are realized, and the data processing efficiency and the interaction intelligence level of non-professional users are improved.

The invention discloses a ship path planning method based on deep reinforcement learning. The ship path planning method is suitable for a complex sea area environment. Through environment modeling and dynamic obstacle identification, a detailed state space is constructed, and perception of the environment is enhanced by using an artificial potential field. An ID3QN-PER algorithm is adopted to train a model, an adaptive exploration strategy and a network parameter updating mechanism are designed, and navigation safety and efficiency in a dynamic environment are ensured. The method further comprises the steps of discretizing an action space, designing an obstacle avoidance strategy in combination with a COLREGS specification, smoothing a route through a B-spline technology, adjusting a navigation strategy according to real-time environment data, and improving the real-time performance and the global optimization performance of route planning.

Disclosed in the present invention is a lightweight defect detection method based on hybrid multi-scale knowledge distillation. The method comprises: constructing a dataset; constructing a teacher network model and a lightweight student network model; using the dataset to train the teacher network model, and saving a weight file of the trained teacher network model; and loading into the teacher network model the saved weight file of the teacher network model, inputting defect images in the dataset into the teacher network model and the student network model to respectively obtain first multi-scale features and second multi-scale features, respectively inputting the first multi-scale features and the second multi-scale features into a cascaded knowledge blending module to obtain final deeply fused first multi-scale features and final deeply fused second multi-scale features, then calculating a hybrid multi-scale knowledge loss, and in combination with the prediction loss of the student network model, using a backpropagation algorithm to update network parameters, so as to obtain a trained lightweight student network model for implementing defect detection of intelligent manufacturing products. The cognitive ability and recognition performance for defects of different scales are improved.

The invention provides a resource scheduling control method and system for a big data server, and the method comprises the steps: constructing a multi-dimensional resource portrait module, collecting the CPU, memory, network, storage I / O load and task queue length of each node in real time, and predicting a resource demand trend through a time sequence algorithm; extracting characteristics such as calculation intensity, data dependence, memory requirements, network transmission quantity and the like; adjusting the weight coefficients of the resource utilization rate, the task completion time and the energy consumption efficiency according to the system load and the historical effect; establishing a bipartite graph model by taking a resource trend as a node feature and a task vector as an edge feature, and calculating a matching score through graph convolution and a multi-objective optimization function; the scheduling scheme is synchronized by adopting a consistency algorithm; automatic rollback and reallocation are carried out when resources are detected to be insufficient; and optimizing a weight coefficient and a network parameter through reinforcement learning. Through the method, the system resource utilization rate can be improved, the task execution efficiency is improved, the overall scheduling effect stability is improved, and the system fault recovery time is shortened.

The invention relates to the technical field of deep learning, and particularly provides a slope deformation real-time early warning algorithm and system based on deep reinforcement learning, and the method comprises the steps: generating a standardized state representation vector, and building a state quality evaluation index; receiving the state representation vector as input, constructing a dual-mode strategy architecture, generating a preliminary decision and confidence evaluation thereof, and forming a quantitative index of decision reliability; forming a final early warning instruction and a strategy updating suggestion, and generating a strategy performance evaluation report; continuously tracking the change trend of state feature distribution, constructing a strategy degradation early warning system, and triggering a model reconstruction mechanism; the updated policy network parameters are fed back to the dual-mode policy architecture, and the optimized feature extraction suggestions are fed back to the establishment of the state quality evaluation indexes. The system comprises an index establishment module, a report generation module and a reconstruction mechanism module. According to the slope deformation real-time early warning algorithm, a monitoring-decision-optimization end-to-end closed-loop control system is realized through a deep reinforcement learning framework.

The invention discloses a CNN (Convolutional Neural Network)-oriented mobile edge computing dynamic unloading method and equipment medium, and the method comprises the steps: obtaining image task data, equipment state data and network state data in a communication, sensing and computing integrated system, and forming a data set; the method comprises the following steps of: constructing a general-inductance-calculation integrated system model, and converting the general-inductance-calculation integrated system model into a calculation task segmentation and resource allocation problem for a calculation task; a Markov decision model is constructed; a joint optimization strategy based on a multi-agent deep reinforcement learning algorithm is adopted, and a neural network parameter model meeting a termination condition is obtained through training image task data and network state data; deploying the trained neural network parameter model into the system; and generating a calculation task segmentation decision and a calculation resource allocation decision according to the real-time data, and transmitting decision result information to each node to realize unloading of the calculation task. According to the invention, smooth execution of the calculation task is realized under the condition of shortage of network environment and network resources.

The invention discloses a video semantic segmentation method based on a time sequence cross attention mechanism, and belongs to the field of computer vision and the field of material detection.The video semantic segmentation method comprises the steps that firstly, a video used for training is preprocessed, a frame sequence is extracted, and then a coding-decoding network for multi-level feature extraction and fusion is constructed; according to the method, feature extraction is enhanced through a time sequence cross attention module, network parameters are optimized through weighted IoU loss and binary cross entropy BCE loss, then frame-by-frame prediction segmentation is carried out on a target video by using a trained model, and a multi-classification segmentation result is exported. According to the method, a time sequence cross attention mechanism is integrated into the SAMUNet network, the segmentation precision is effectively improved for image data with time sequences, the time cost and the labor cost of material video processing are greatly reduced, the method can be widely applied to the field of industrial detection, and the product quality and the production efficiency are improved.

The invention discloses a scene topology understanding method and device, a storage medium and a program product, and relates to the field of computer systems based on a specific calculation model, and the method comprises the steps: inputting a multi-view environment image set into a backbone network, and generating bird's-eye view features corresponding to the environment image set; calculating a spatial transformation matrix of the aerial view features at the current moment and the aerial view features of the previous K frames, and performing space-time alignment on the obtained K + 1 frames of aerial view features to obtain multi-frame fusion features; inputting the multi-frame fusion features into a map prior model to obtain aerial view correction features; decoding the aerial view correction features based on a topological decoder, and generating a lane topological graph; comparing the lane topological graph with the annotation data, calculating an error loss function, and optimizing network parameters based on the error loss function; and generating an optimized topological graph, and determining a scene topology result based on the optimized topological graph. By implementing the method, the environment topology understanding capability in a complex scene can be improved, and the generation precision of the lane topological graph is optimized.