792 results about "Global information" patented technology

The invention belongs to the technical field of natural language processing and multi-modal information extraction, and particularly relates to a multi-modal named entity recognition method based on semantic alignment and cross-modal graph fusion, which comprises the following steps: S1, acquiring a data sample containing a text sequence and image content; s2, encoding the text and the image into vectors respectively; s3, similarity is calculated through a trainable bilinear function, and optimization is carried out through loss comparison; s4, cross-modal attention is used to enhance association information between modals; s5, determining the proportion of reserved image information through a modal matching module; s6, introducing a gating mechanism to dynamically fuse visual and text features; s7, realizing local and global information complementation by a cross-modal graph fusion model; and S8, inputting the fused representation into the CRF layer to predict the entity type. According to the method, fine semantic alignment can be realized in a weak image-text correlation context, and balance between local entity recognition and global semantic understanding can be achieved.

The invention discloses a heterogeneous double-flow fusion method and system for diabetic retinopathy grading. The method comprises the following steps: obtaining an output result of diabetic retinopathy grading by utilizing a heterogeneous double-flow architecture; processing an input fundus image into images with different resolutions; extracting global context features from the low-resolution image by using a lightweight visual Transform model distilled by composite knowledge, and extracting local focus features from the high-resolution image by using a convolutional neural network model; performing interactive fusion on the global context features and the local focus features of the double-branch architecture through a symmetric bidirectional cross attention fusion module to obtain enhanced fusion feature representation; and finally, inputting the fusion features into a classifier, and outputting a severity grading result of the lesion. The method aims at improving the accuracy and robustness of hierarchical diagnosis through deep analysis of global information and local details, and can be applied to the medical fields of clinical computer-aided diagnosis, eye image analysis and the like.

The invention discloses a human body posture estimation method, system and device and a medium, and relates to the technical field of computer vision, and the method comprises the steps: obtaining a to-be-estimated picture containing a human body; the method comprises the following steps of: performing feature extraction and down-sampling on a to-be-estimated picture, performing dimension raising, depth separable convolution operation, channel aggregation operation and dimension reduction on a feature map with down-sampling resolution in an output branch after down-sampling to obtain a local feature map, performing up-sampling reconstruction on the local feature map by adopting dynamic weight interpolation, and fusing an output branch which is not down-sampled to obtain a local feature map; obtaining a first fusion feature; taking the first fusion feature as an initial feature, repeating the step of obtaining the first fusion feature, obtaining a second fusion feature, carrying out fusion to obtain a dual-scale fusion feature, extracting a depth perception feature in the dual-scale fusion feature, carrying out human body posture estimation through the depth perception feature, and obtaining a human body key point heat map. According to the invention, multi-scale and global information is obtained through a lightweight structure, and accurate key point positioning is obtained.

The invention relates to the technical field of computer vision, and discloses an infrared light and visible light image fusion method, system and device based on edge information guidance and a medium, and the method comprises the steps: inputting the obtained infrared light and visible light images into a shallow feature encoder in parallel, and extracting initial multi-modal shallow features; inputting the initial multi-modal shallow layer features into an edge guiding branch, acquiring comprehensive edge representation by using an average pooling layer and a maximum pooling layer, and extracting enhanced edge features through a global perception injection module; inputting the initial multi-modal shallow layer features into depth feature branches, extracting depth multi-modal features through a double-branch depth feature encoder, and fusing the depth multi-modal features through a cross-domain global fusion module to obtain depth fusion features; and inputting the enhanced edge features and the deep fusion features into a decoder for reconstruction, and generating a final fusion image. According to the method, the fusion image which is excellent in marginal definition, detail texture retention and global information expression can be generated.

The invention discloses a rice leaf tip small target detection and counting method based on global information enhancement, and the method comprises the steps: firstly building a field rice leaf tip image data set as a basis, and constructing an improved YOLO-DCL model; the core innovation of the method lies in that a global context information enhancement module, an inverted residual-cascade grouping attention module and a re-parameterization shared convolution detection head are introduced, key feature expression is enhanced and detection efficiency is optimized through effective fusion of global information, the recognition precision and processing speed of the model on a tiny rice leaf tip target are significantly improved, and the method is suitable for large-scale popularization and application. After training parameters are set, an optimization model is trained, then system performance verification and module optimization are carried out on the model, and a version with the optimal rice leaf tip detection performance is screened out. The technical bottlenecks that small target leaf tips are prone to missing detection and false detection and insufficient in real-time performance under the complex farmland background are effectively overcome, and reliable and efficient technical support is provided for rice growth early-stage yield prediction.

The invention provides a real-time command and control strategy optimization method and device based on multi-agent reinforcement learning, equipment and a storage medium, and the method comprises the steps: deploying a pre-trained cooperative intention network on edge equipment of each combat unit, and coding local observation data and a historical sequence thereof into a low-dimensional local cooperative intention vector; replacing original high-dimensional data as inter-unit communication content; whether broadcasting is carried out or not is dynamically determined according to observation uncertainty and a channel state by combining a self-adaptive intention broadcasting mechanism, and communication resources are distributed according to needs; after each unit receives an adjacent collaborative intention vector, a lightweight space-time attention module in a pre-trained local strategy execution network carries out space weighting on adjacent intentions and fuses time features of own historical intentions, and a collaborative and consistent real-time command and control instruction is generated under the condition that global information convergence is not needed.

The invention relates to the technical field of adaptive filtering, discloses a variable-step robust adaptive filter and a filter network, and aims to solve the technical problem that a traditional adaptive filter cannot effectively consider convergence speed, steady-state precision and impulse noise interference resistance. A non-linear error factor is introduced, large errors caused by impulse noise are effectively suppressed through a generalized function form of the non-linear error factor, and the robustness of the filtering process is remarkably improved; meanwhile, a variable step length mechanism is provided, the theoretically optimal candidate step length is calculated on line by minimizing the mean square deviation of the next moment, and a self-adaptive target variable step length is generated through truncation and time smoothing processing, so that dynamic balance is realized between rapid convergence and low-steady-state maladjustment. According to the invention, the filter is expanded to the distributed network, each node follows a diffusion strategy of first updating and then combining, and global information sharing and performance collaborative optimization are realized by using a combined coefficient.

The invention relates to the technical field of computer vision research, and discloses an infrared light and visible light image fusion method, system and device based on double-branch cross-domain feature fusion and a medium, and the method comprises the steps: inputting an obtained infrared light image and an obtained visible light image into a shallow feature encoder in parallel, and extracting the initial multi-modal shallow features of the corresponding images; respectively inputting the initial multi-modal shallow layer features of the corresponding images into a double-branch depth feature encoder, and extracting depth multi-modal features; respectively fusing the deep multi-modal features output by the corresponding branches through a cross-domain global fusion module to obtain deep fusion features; and inputting the deep fusion feature into a decoder for decoding operation to generate a reconstructed fusion image. According to the method, the fusion image which is excellent in global information expression, detail texture fidelity and marginal definition can be generated, and an innovative technical scheme is provided for realizing high-quality and multi-modal image fusion in a complex scene.

The invention discloses a multi-agent reinforcement learning method. The method comprises the following steps: step 1, system initialization and edge device modeling; 2, generating a data source and constructing a task package; step 3, edge equipment trajectory planning and movement acquisition; 4, local task execution and unloading strategy decision making; 5, communication link modeling and bandwidth resource allocation are carried out; step 6, scheduling optimization driven by an information age and depreciation mechanism; step 7, multi-agent strategy optimization based on Actor-Critic is carried out; and step 8, reward function design and reinforcement learning process. The technical problems that an existing centralized scheduling or heuristic algorithm cannot perform efficient learning and scheduling in a resource heterogeneous and data dynamic environment, and cannot still have good learning ability and generalization ability under the condition of lack of global information are solved; the method is suitable for system modeling and optimization work of efficient strategy learning and collaborative decision making of multiple agents in a complete cooperation task.

The invention discloses a spine image key point detection algorithm based on multi-task learning. The algorithm comprises the following steps: constructing a multi-task deep learning network model comprising a segmentation branch and a key point positioning branch; a feature aggregation module based on multi-scale cavity convolution is embedded in the segmented branches, and the multi-scale cone feature extraction capability of the model is enhanced through splicing fusion of multiple cavity rate convolution branches and global pooling branches; a cross-task attention fusion module is introduced between the two branches, and bidirectional dynamic interaction and complementation between segmentation features and key point features are realized by generating and fusing first-order and second-order context attention maps; semantic alignment loss is designed in a training stage, collaborative optimization of two tasks is promoted by constraining the consistency of segmentation masks and key point heat maps in a high-level feature space, global information of segmentation and local information of key point detection are fully utilized, and the accuracy and stability of spine centrum key point positioning and the accuracy of a segmentation result are improved.

The invention discloses a formation riding danger information sharing method and system for intelligent riding helmets, and relates to the technical field of information sharing, and the method comprises the steps: carrying out the deep fusion of the adjacent beacons of all intelligent helmets, the state information of a vehicle, and the multi-source heterogeneous data of an external sensor, vehicle-road cooperation and the like; and constructing and dynamically maintaining a topological graph reflecting the space-time relationship among the formation members in real time. On the basis, intelligent source end propagation strategy decision is carried out on perceived semantic danger information, and topology-based directional relay and propagation are carried out. And finally, the decided propagable information is combined with the state of the rider, so that a highly personalized early warning instruction is generated. In this way, undifferentiated global information broadcast can be effectively converted into accurate risk announcement based on individual context, and the efficiency of dangerous information sharing in formation riding and the cooperative safety capability of the whole system are remarkably improved.

The invention discloses a distillate oil property prediction method based on deep learning feature extraction and partial least squares regression. The method comprises the following steps: firstly, carrying out classification training on a near infrared spectrum through a convolution-attention double-branch fusion network, and extracting high-dimensional spectral features with local and global information; then, historical samples are retrieved from a database based on prediction categories, a plurality of most similar samples are selected by adopting cosine similarity measurement to construct a correction set, and the spectral features and property labels are subjected to standardization processing; and finally, carrying out partial least squares regression modeling on the correction set, extracting latent variables to maximize covariance between spectral features and physicochemical properties, and inputting feature vectors of an oil sample to be detected into the trained PLS model to obtain a corresponding property prediction result. According to the method, the modeling requirement and the category specificity characteristics under the small sample condition are considered while the prediction precision is guaranteed, and the method is suitable for rapid property detection and intelligent analysis in the refining process.

The invention discloses a landslide segmentation model based on multi-loss function fusion, particularly relates to the field of landslide monitoring, is used for solving the problems of precise segmentation of a landslide area and insufficient identification precision in a complex scene, realizes precise segmentation of a landslide image through multi-level feature extraction and adaptive optimization, and has remarkable beneficial effects. The detail identification of the landslide area is enhanced by utilizing self-adaptive channel response and multi-scale feature fusion, so that features of different scales are fully expressed; weighted combination and residual connection of global and local features ensure continuous transmission and integration of high-level and low-level features, and detail and global information are completely reserved. The spatial resolution is recovered through up-sampling and channel splicing, and it is ensured that a segmentation result is consistent with an original image; the multi-level nonlinear loss function improves the robustness of the model under the condition that the samples are unbalanced, samples difficult to classify are effectively concerned, and the model shows a high-precision segmentation effect in a complex scene.

The invention discloses a multi-unmanned aerial vehicle cooperative inspection trajectory control method for optimizing medical area coverage and service efficiency. The method comprises the following steps: firstly, constructing a medical multi-unmanned aerial vehicle auxiliary inspection mobile edge computing system model, defining an unmanned aerial vehicle and mobile user set, and establishing a communication model containing A2G and A2A links, an unmanned aerial vehicle mobile model and an energy consumption model; then taking a joint function of a coverage score, a system throughput and an emergency task completion rate as an optimization target, under energy and communication connectivity constraints, proposing an LT-MADDPG algorithm, adopting a CTDE framework, processing a time sequence state by an actor network integrated with LSTM, fusing global information by a commentator network embedded with Transform through multi-head attention, and finally obtaining an emergency task. And modeling a cooperative relationship between the unmanned aerial vehicles and a medical task priority. Experiments show that the method is superior to a traditional algorithm in the aspects of coverage, service fairness and system throughput, and the medical inspection efficiency and reliability are effectively improved.

The invention discloses an FTU-based power distribution network fault positioning method and system, and belongs to the technical field of power distribution automation, and the method comprises the steps: constructing a space-time correlation feature matrix according to a transient current sequence and a voltage drop sequence during a fault period, and extracting the convolution features of a graph to obtain a fault feature graph containing the fault correlation degree between nodes; according to a static topological structure in the power distribution information model, virtual impedance is calculated based on the fault feature graph, and network equivalent topology is dynamically identified to obtain a dynamic virtual topological structure graph; and based on the dynamic virtual topological structure diagram structure and the fault feature diagram, performing fault section confidence competing decision through the intelligent agent unit corresponding to each FTU based on an incomplete information game, and outputting a fault section positioning result. The power distribution network fault positioning method solves the problems that a traditional power distribution network fault positioning method is insufficient in positioning accuracy and poor in fault tolerance and excessively depends on centralized processing and global information synchronization when information is incomplete, fault features are complex and network topology dynamically changes.

The invention discloses an edge computing node collaborative task unloading method for guaranteeing a low-delay service, and relates to the field of edge computing, each edge computing node generates a collaborative view comprising the edge computing node and a neighbor edge computing node through a local LSTM prediction model and federated learning, and the collaborative view comprises a predicted resource state and a predicted network state; according to the invention, the local LSTM prediction model and federated learning are combined to generate the collaborative view, so that accurate prediction and global information sharing of edge node resources and network states are realized, and the accuracy of decision making is improved; the tasks are analyzed into a dependency graph with key path marks, so that priority scheduling of the key tasks is ensured, and the overall task time delay is reduced; based on a weighted voting consensus mechanism of node credibility and resource adequacy, the efficiency and reliability of decision consensus among nodes are improved; the task unloading efficiency and the service quality of the low-delay service are effectively improved, and the stability and the reliability of the edge computing system are enhanced.

The invention discloses a low-voltage series arc fault detection method, system and device, and belongs to the technical field of low-voltage series arc fault detection, and the method comprises the steps: obtaining an original current signal, and carrying out the preprocessing through sliding window segmentation and instance normalization; performing multi-scale feature fusion on the preprocessed analysis unit, and generating fusion features through parallel feature extraction and an attention mechanism; performing context modeling on the fused features through an encoder, inputting a self-adaptive bottleneck layer containing an expert hybrid network, and routing the features to the most appropriate expert network by context sensing gating according to global information; the decoder reconstructs the signal and calculates an error, and generates a dense abnormal fraction sequence; gaussian position weighted aggregation abnormal scores are adopted, and fault judgment is carried out in combination with a self-adaptive threshold decision mechanism based on K-Means clustering. The method can be trained without a fault sample, can dynamically adapt to complex current modes under different loads, gets rid of dependence on the fault sample, and accurately detects the arc fault.

The invention provides a radar signal classification method and system based on multi-modal feature adaptive fusion. The method comprises the following steps: performing compressed sensing processing on a time-frequency image by using a pre-constructed sparse sampling matrix to generate observation data; inputting the observation data into a multi-branch feature extraction network, and extracting local texture features and global semantic features through the multi-branch feature extraction network; calculating a global information theory feature tensor based on the time-frequency image, and generating a gating weight matrix based on the global information theory feature tensor; and performing adaptive fusion on the local texture features and the global semantic features by using the gating weight matrix to generate adaptive fusion features, and generating a classification result of the radar signals according to the adaptive fusion features. According to the technical scheme provided by the invention, the data dimension is reduced through compressed sensing, the complementary features are extracted by using the multi-branch network, and the precision and robustness of radar signal classification are effectively improved in combination with an adaptive fusion mechanism guided by an information theory.

The invention relates to the technical field of target detection, and particularly provides a detection method for boundary enhancement and small target perception of a problem map. The method comprises the steps that a double-branch attention convolution module fusing convolution and an attention mechanism is constructed, and the boundary sensing ability is enhanced through a global-local feature collaborative enhancement mechanism; designing a multi-path feature aggregation mechanism, and dynamically reconstructing multi-scale feature representation through bidirectional path interaction and adaptive weighted fusion; constructing a global context fusion module to balance global information and local information; according to the method, the target query is refined through the prediction head, and the category label and the bounding box are generated, so that the overall boundary perception capability is enhanced, the sufficiency and effectiveness of feature aggregation are ensured, and the positioning precision of the target and the accuracy of national boundary region identification are improved.

The invention discloses an unmanned cluster dynamic task game confrontation method based on MAPPO deep reinforcement learning of Actor-Critic. An MAPPO deep reinforcement learning strategy of Actor centralized training and Critic distributed execution is adopted as a training algorithm of unmanned cluster dynamic task game confrontation. According to the method, centralized training is carried out by using global information in a training stage, and distributed execution is carried out by only using local observation information and historical information in an execution stage, so that the task can be completed through deep cooperation among multiple agents; the problem that the task completion efficiency is low due to the fact that the flight path drifts and jitters and is prone to falling into a local optimal solution during training of a traditional algorithm is solved, multi-aircraft cooperative path planning can be effectively achieved, and the learning efficiency and stability are improved in a complex multi-agent scene.

The invention provides a day-ahead prediction method and device for wind power cluster power, equipment and a storage medium. Relates to the technical field of wind power generation prediction. The method comprises the steps that clustering is completed based on historical multivariate information of all stations in a wind power cluster, cluster virtual nodes are set, and the visual angle of global information is expanded for cluster modeling; and information propagation characteristics between nodes are explored based on a cluster prevailing wind direction. Constructing an adjacent matrix of the dynamic graph, and quantifying the dynamic evolution process of the cluster topology; and finally, mining cluster dynamic information by using a frequency domain channel attention mechanism and STGCN, and completing short-term power prediction of the wind power cluster. The method is a prediction method which considers multivariate information space-time diversity and is suitable for the day-ahead wind power cluster power, and is simple in calculation, high in prediction performance, clear in physical significance, effective in prediction result and high in practicability.

The invention provides a multivariate time sequence prediction method and system and a medium, an input time sequence is adaptively decomposed into an approximate coefficient sequence and a detail coefficient sequence by adopting an adaptive discrete wavelet transform and inverse transform mode, and compared with the previous wavelet transform depending on a predefined wavelet basis function, the multivariate time sequence prediction method and system have the advantages that the time sequence prediction efficiency is improved; the mode is more flexible, and filtering kernels of discrete wavelet transform and inverse transform can be updated in a data driving mode in the training process according to the characteristics of the time sequence, so that the filtering kernels are more suitable for the current sequence. Wavelet transformation realized by carrying out convolution operation on a time dimension lacks extraction and modeling of correlation between global information and different variables in a multivariate time sequence; therefore, a coefficient mixing module and a wavelet domain attention channel enhancement module are provided to carry out supplementary modeling on an approximation coefficient and detail coefficient sequence obtained after wavelet transformation. The method can meet the prediction requirements of the time series in actual production such as power load.

The invention provides a real-time scheduling method for a wireless network based on a large language model, the wireless network comprises a scheduling node and a plurality of user nodes, and the scheduling node is provided with a scheduling agent, a historical memory pool, an reflection agent and a suggestion buffer area. The method comprises the following steps executed in each time slot: step S1, the scheduling node acquires global information of a wireless network under the current time slot and calculates states of all data streams under the current time slot according to a preset calculation rule; and S2, the scheduling agent generates a scheduling decision of the current time slot according to a preset scheduling prompt word by using a large language model configured by the scheduling agent based on the states of all data streams under the current time slot and decision suggestions stored in the suggestion buffer area and by taking a maximized reward as a target. According to the technical scheme of the invention, by introducing the large language model, the problem of insufficient adaptability in a dynamic network environment in the prior art is solved.

According to the cooperative control method and system of the non-linear multi-agent system, under the condition that communication link faults are considered, a non-linear dynamic model of each single-connecting-rod mechanical arm is established, and unknown uncertain non-linear dynamic states in the single-connecting-rod mechanical arms are fuzzified through a fuzzy logic system; a mixed event triggering mechanism is introduced, so that state information of a leader is estimated, and the communication frequency is effectively reduced; and constructing a virtual controller and an actual controller by utilizing a backstepping method based on the estimated state information, so as to control the position of the following mechanical arm to be synchronous with the expected trajectory of the leading mechanical arm under the condition of being influenced by the communication link fault. According to the method, the state of the multi-single-connecting-rod mechanical arm system can be kept consistent and stable tracking control can be achieved under the condition that communication link faults and nonlinear uncertainty exist without global information.

The invention discloses a local-global information aggregation remote sensing image change detection method based on Version Mama, and belongs to the technical field of remote sensing image change detection. The method adopts a Dual-LGNet network model for prediction, and comprises the following steps: inputting a dual-time-phase remote sensing image map before and after change into a twin encoder to obtain a feature map aggregating global features and local features; inputting the feature map aggregating the global features and the local features into a feature fusion module to obtain a fused difference feature map; and inputting the difference feature map into a progressive decoder to obtain a final prediction result. Wherein the LG-SS2D is introduced into the encoder to carry out feature extraction, and an LG-SS2D block comprises a parallel convolution branch and a Mama-based global branch. Through combination of local-global information aggregation and boundary enhancement, the recognition and detection precision of a remote sensing change area can be improved on the premise of keeping linear complexity, and especially the detection precision of a large building and a small target change area can be improved.

The invention discloses a typical power equipment fault identification method based on cross-source image fusion of YOLOv11, which is used for power equipment fault detection. The specific principle is as follows: firstly, denoising, normalizing, enhancing and other preprocessing are performed on a multi-source image, and accurate registration is performed to eliminate spatial distribution difference; deep features of the registered image are extracted through a CSFT model, efficient fusion of global information among sources is realized by using a multi-head attention mechanism, and complementarity and correlation of multi-source features are enhanced in combination with source alignment loss and a position coding technology in fusion; and finally, the fused multi-source features are transmitted to a YOLOv11 network, and accurate positioning and classification of a fault target are realized by using an optimized feature pyramid structure and a decoupling detection head module of the YOLOv11 network. The method provides reference for combination of power equipment fault detection, cross-source data fusion and target detection, and supports safe operation of an intelligent power grid.

The invention provides a long text processing method and device, electronic equipment and a storage medium, and the method comprises the steps: inputting long text data into a long text processing model, and obtaining a long text processing result; the long text processing model is constructed by adopting a hierarchical mixed attention strategy on the basis of a large language model; according to the hierarchical mixed attention strategy, a first attention mechanism in each network structure of a large language model is replaced by an attention mechanism with a global information retrieval capability, and other attention mechanisms are replaced by attention mechanisms with a local information modeling capability. The hierarchical mixed attention architecture can greatly improve the reasoning efficiency on a long text while ensuring the reasoning effect of the model, reduces the consumption of computing resources, and approaches the modeling effect when only global attention is used.

The invention discloses a storage robot cluster management system and method, and belongs to the technical field of robot scheduling. The system comprises a central management server and an autonomous mobile robot cluster. The central management server comprises a congestion prediction and global path planning module; the congestion prediction and global path planning module comprises a congestion prediction sub-module and an A * path planning sub-module with space-time cost; the autonomous mobile robot is internally provided with a multi-sensor sensing and communication module and a local decision module, and the off-line centralized training platform is used for optimizing the local decision module in each autonomous mobile robot through a centralized evaluation network with global information in a simulation environment. The invention provides a hybrid control architecture integrating global path planning, dynamic task allocation and local real-time decision based on multi-agent reinforcement learning, and efficient and smooth cluster collaborative operation can be realized.

The invention discloses an electric vehicle charging management method based on an HA-SAC algorithm, and the method comprises the steps: firstly, carrying out the modeling of an electric vehicle charging management optimization model through fully considering the life loss of a transformer and the charging preference of a user; modeling a charging management optimization problem as Markov Game (MG for short), taking each electric vehicle as an independent intelligent agent, constructing a state space according to information such as the charging state, the electricity price and the user preference, and taking the charging power as an action variable; then, on the basis of an HA-SAC algorithm, a local action network and a centralized evaluation network are constructed for each agent, and centralized training is carried out through global information; and finally, the trained strategy is deployed to each electric vehicle agent node, so that the electric vehicle agent node can autonomously decide a charging behavior only by depending on local information in actual operation, and distributed and privacy-protected cooperative charging control is realized.

The invention discloses an EAGLE-Net remote sensing image segmentation method, which comprises the steps of extracting multi-scale features of an input image, and obtaining low-level features of spatial details and high-level features of semantic information; the low-level features of the space details and the high-level features of the semantic information are input into an attention gating module, space-channel two-dimensional attention weights are generated, and weighting processing is conducted on the low-level features of the space details and the high-level features of the semantic information; inputting the weighted high-level features into a dynamic void space pyramid, predicting multiple groups of void rates based on global information, and generating enhanced semantic features through multi-scale void convolution fusion; splicing and decoding the weighted low-layer features and the enhanced semantic features to obtain a segmented prediction map; and performing edge detection on the segmented prediction map to generate an edge prediction map. According to the method, noise can be suppressed, key signals can be enhanced, large targets and small targets can be adaptively covered, and object boundaries can be accurately recovered by means of edge supervision.