47 results about "Semantic integration" patented technology

The invention discloses a multi-modal data fusion method and system based on a large model, and relates to the technical field of data fusion, and the method comprises the steps: receiving multi-modal data, and carrying out the noise layering filtering and time-space alignment; mapping data to a large model hidden space through each modal lightweight encoder, extracting initial features by using a single-modal pre-training model, and converting the initial features through an adapter to generate a hidden space vector; each modal hidden space vector and a task cue word are received, a fusion feature matrix is dynamically aggregated and generated through a self-attention mechanism and a cross attention mechanism of a large model, and semantic integration of multi-modal information is realized; taking the fusion feature matrix as a soft label, and learning a cross-modal semantic mapping capability through knowledge distillation; the trained model is deployed to the edge through dynamic quantization and pruning optimization, an optimized feature matrix is input into a task-customized lightweight head network, and a final task result is output in combination with multi-modal context information. The method can break through the semantic gap between modals, and improves the fusion efficiency.

The invention relates to the technical field of data processing, and discloses an old people health state assessment data processing system based on multi-modal data, and the system comprises a medical data integration module which obtains electronic medical record data and a medical examination report through an FHIR interface, and extracts a structured health index; the cross-modal causal fusion processing module is used for fusing the monitoring data and the medical text through an image, text and image interlayer architecture; the health state evolution modeling module is used for mapping the health feature vectors into physiological function, cognitive level and athletic ability three-dimensional state indexes; an evaluation report backtracking module; and a decision output module. Through an image, text and image interlayer architecture, deep semantic fusion of multi-modal features is realized under the constraint of medical pathology rules, feature weight adaptive distribution is dynamically guided based on a medical causal atlas, a high-dimensional fusion vector retaining key pathology information is generated, the semantic integration ability of health data is remarkably improved, and the health data fusion efficiency is improved. And the reliability of discrimination and decision making is 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 an intelligent retrieval method and system fusing text and image semantic features. The method comprises the following steps: S1, carrying out quality detection and preprocessing on an image scanning copy of an electronic file and a case simultaneous recording video; s2, constructing a structured electronic file directory; s3, extracting a text semantic feature, a file image semantic feature and a video image semantic feature as multi-modal features of the text and the image; s4, carrying out feature fusion and alignment on the multi-modal features of the text and the image through a multi-modal large model, and generating a cross-modal unified feature vector with semantic consistency; s5, automatically constructing a case knowledge graph, and realizing structured and semantic integration of legal information; and S6, performing semantic analysis and multi-hop reasoning based on natural language query and the case knowledge graph, and generating and presenting a retrieval result in a structured or question and answer form. According to the method, the semantic features of the text and the image are fused, so that deep knowledge mining and efficient intelligent retrieval of the electronic file are realized.

The invention relates to a power load spatio-temporal dynamic knowledge graph construction and load prediction method, which comprises the following steps of: constructing a text and digital sequence hybrid vector coding module, providing a hierarchical entity relationship joint extraction framework oriented to power system load data, constructing a Multi-Encoder-Bi-GRU-CRF power load entity recognition model, and constructing a power load entity model. Constructing a power load spatio-temporal dynamic knowledge graph in combination with a predefined relation rule base; meanwhile, time-space sub-graphs are divided, a space-time coupling self-adaptive adjacency matrix is constructed, and the space-time dependency relationship between nodes is quantified; and finally, combining the knowledge graph node embedded vector and the adjacency relation embedded vector, and jointly extracting the spatial feature and the time feature of the power load by adopting a space-time diagram convolutional neural network. Therefore, the load prediction algorithm provided by the invention not only can give full play to the advantages of multi-modal semantic integration and space-time modeling capability of the knowledge graph, but also can improve the load prediction precision, assist in realizing refined energy management of the power system and assist in making an optimal scheduling strategy, and has a good engineering application prospect.

The invention relates to the technical field of electric digital data processing, and discloses an intelligent data processing method for clinical research. The method comprises the following steps: constructing a basic clinical knowledge graph fused with an international medical ontology; the method comprises the following steps: performing deep learning driven entity recognition and knowledge graph linking on original data from heterogeneous sources such as an electronic medical record and an inspection system; executing cross-source entity alignment and knowledge fusion based on the graph attention network; performing automatic data quality verification and restoration according to clinical logic rules embedded in the atlas; and finally, extracting and generating a standardized analysis ready data set from the enhanced atlas according to research requirements. According to the technical scheme, high-quality, automatic and semantic integration and flexible delivery of clinical research data are realized, and the data processing efficiency and research reliability are improved.

The invention discloses a medical image segmentation method and system based on feature extraction optimization. The medical image segmentation method comprises the following steps: performing target region segmentation on a segmented image through a constructed image segmentation model; an image segmentation model in the invention adopts a U-Net structure and comprises an encoder and a decoder; a Haar wavelet transform and KAN convolution module is introduced into the encoder to carry out feature extraction on the input feature map; meanwhile, a channel cross attention module and a space weighting module are connected in series on jump connection of the U-Net structure, the channel cross attention module captures complex dependence of cross-stage features through channel attention, and the space weighting module generates dynamic weights by using KAN and space attention, optimizes a space-channel relation of the features, and obtains the cross-stage features. The two are combined to relieve the problems of multi-scale information loss and semantic inconsistency, and the global semantic integration capability is remarkably improved.

The invention provides a wide remote sensing image semantic segmentation method based on multi-scale image pyramid cutting, and aims to realize pixel-level fine interpretation of a wide-range remote sensing image. The method comprises the following four core processes: 1, multi-scale pyramid cutting of a wide remote sensing image; 2, performing long-distance context dependency representation and modeling; 3, performing cross-scale feature efficient semantic integration; and 4, integrating local reasoning results. The wide-range remote sensing image semantic segmentation method overcomes the bottleneck that a traditional wide-range remote sensing image semantic segmentation model is difficult to capture long-distance context semantic feature association, effectively solves the problems that an existing wide-range remote sensing segmentation method is high in calculation complexity, large in video memory occupation and insufficient in precision, has the advantages of being high in segmentation precision and high in training reasoning efficiency, and is suitable for large-scale popularization and application. The method can be widely applied to the fields of territorial space planning, ecological environment monitoring, agricultural resource investigation, natural disaster assessment and the like.

The invention relates to the technical field of data management, in particular to a data weaving semantic integration method based on a semantic network and a knowledge graph, which comprises the following steps: acquiring multi-source heterogeneous data and preprocessing to obtain standardized data, the multi-source heterogeneous data comprises structured data, semi-structured data and unstructured data from different business scenes; performing rule injection on the standardized data to generate semantic web ontology data; extracting entities and attribute values thereof in the standardized data to generate structured knowledge graph data of the instance layer; and establishing a semantic association mapping network of the semantic network ontology data and the structured knowledge graph data, and generating a semantic integration result of data weaving. According to the method, the basic differences of the multi-source data in the aspects of formats, codes and the like are eliminated, a unified semantic standard is provided, and conflicts caused by non-unified semantics are reduced.

The application relates to the technical field of data processing, and discloses an old person health state evaluation data processing system based on multi-modal data, which comprises a medical data integration module, an across-modal causal fusion processing module, a health state evolution modeling module, a backtracking evaluation report module and a decision output module. The medical data integration module acquires electronic medical record data and medical examination reports through a FHIR interface and extracts structured health indexes. The across-modal causal fusion processing module fuses monitoring data and medical texts through an image, a text and an image sandwich architecture. The health state evolution modeling module maps a health feature vector into three-dimensional state indexes of physiological functions, cognitive levels and motor abilities. The backtracking evaluation report module and the decision output module are used for backtracking evaluation and decision output. Through the image, the text and the image sandwich architecture, deep semantic fusion of multi-modal features is realized under the constraint of medical pathology rules, feature weight adaptive distribution is dynamically guided based on a medical causal graph, a high-dimensional fusion vector retaining key pathological information is generated, the semantic integration capability of health data is significantly improved, and the reliability of discrimination and decision is improved.

The invention discloses a multi-modal three-dimensional point cloud semantic segmentation method based on adaptive noise filtering, and belongs to the technical field of automatic driving environment perception. The method comprises the following steps: firstly, acquiring data by using a laser radar and a monocular camera, and constructing a multi-modal panoramic feature tensor containing a geometric structure and color textures through projection and mapping; extracting shallow geometric distribution features through a residual context module, and extracting multi-scale environment features through an expanded residual encoder; performing global context aggregation by using a self-attention mechanism of a Transform architecture, and establishing a full-image pixel dependency relationship to make up for a convolution locality defect; in the decoding stage, a channel cross fusion attention module (CCA) is adopted to process deep semantic features and shallow jump connection features, and a channel weight mask is dynamically generated to adaptively screen effective features and suppress high-frequency noise; and finally, outputting a two-dimensional semantic segmentation result and back-projecting the result to a three-dimensional space to obtain a semantic point cloud. According to the method, through global semantic integration and local detail screening, the problems of terrain misjudgment and noise interference of severe weather (such as rain, snow and dust) in a cross-country scene are effectively solved, and the robustness and precision of automatic driving perception are remarkably improved.

The invention relates to the technical field of artificial intelligence and multi-agent cooperative control, in particular to an agent autonomous cooperation method and system based on group hierarchical cooperation, and the method comprises the steps: responding to a natural language task request input by a user; generating a preliminary task execution plan based on the task description data, selecting expert agents matched with sub-tasks in the task execution plan from the group, and generating a task allocation instruction to perform task cooperation; and each expert agent receives the task allocation instruction to execute the corresponding sub-task, writes an execution state, a stage result and intermediate data into a group memory space in real time in an execution process, and performs semantic integration and induction after all the sub-tasks are executed to generate a unified task output result. Under the unified command of the super agent, each expert agent can realize autonomous cooperation and real-time information sharing based on the group memory space, and realize adaptive dynamic re-planning when the task is abnormal or the environment is changed.

The invention provides a multi-personality decision simulation and game method for military artificial intelligence. Simulation of an AI opponent model on complex human behaviors is enhanced through a DIKWP semantic framework. The system firstly configures personality parameters based on a psychological model (such as MBTI, big five personalities), and maps the personality parameters to an AI cognitive model to construct a behavior preference model; then, a noise injection algorithm is adopted to simulate emotion and cognitive deviation, and unique deviation of different personalities in decision making is reproduced; the multi-agent self-gaming engine enables AIs with different personality characteristics to repeatedly game in a red-blue confrontation environment, optimizes a strategy and evaluates winning rate distribution. And the decision fusion analysis module summarizes the data, generates personality performance comparison and behavior pedigree diagrams under each tactical situation, and predicts opponent behaviors. The DIKWP semantic integration interface ensures that a personality simulation result can be seamlessly connected with a knowledge graph and a strategic intention module of an existing AI system. According to the method, AI opponent modeling is enriched on the semantic level, the decision simulation degree of a real commander is improved, and as a supplement of an existing AI simulation system, the ability of military AI in the aspects of strategic pre-judgment and cognitive game is enhanced.

This application provides a method, apparatus, computer-readable medium, and electronic device for low-light image enhancement based on semantic integration. The method includes: acquiring a low-light image to be enhanced; the low-light image includes images captured when the light intensity is below a set threshold; extracting local detail information and global structural information from the low-light image based on its spatial information; and generating an enhanced image by performing image convolution and transformation through an image signal processing pipeline based on the detail information and the structural information. This application's solution can efficiently and effectively enhance low-light images while preserving rich semantic information, significantly improving perceptual quality.

The invention discloses a traffic over-limit early warning method and system under multi-source data fusion, and relates to the related field of traffic monitoring, and the method comprises the steps: carrying out the multi-source data collection of a target vehicle through the interaction of a multi-source data linkage collection system, and obtaining a multi-source data set; performing key anchor point data screening on the multi-source data set by using a cross-modal attention mechanism, and determining a key anchor point data set and an associated multi-source data set; respectively carrying out interactive semantic integration analysis on the key anchor point data set and the associated multi-source data set, and determining multi-source integration interaction features; and pre-constructing a traffic over-limit multi-dimensional label library, carrying out matching identification on the multi-source integrated interaction features and the traffic over-limit multi-dimensional label library, and generating a traffic over-limit early warning instruction according to an identification result. The technical problems of low detection accuracy and high false alarm rate of the existing traffic over-limit early warning are solved, and the technical effects of improving the detection accuracy and reducing the false alarm rate are achieved.

The invention relates to a deep learning-based breast cancer rehabilitation question and answer method, apparatus and device, and a medium. According to the method, patient demands are accurately captured through multi-modal feature analysis to generate knowledge demand vectors, and medical knowledge sub-graphs and general knowledge fragments are retrieved in parallel to form a knowledge basis; based on demand vector parameter dynamic weighting fusion of the two types of knowledge, a collaborative knowledge unit is constructed, and then importance distribution is performed on the medical knowledge by using a graph attention neural network to generate attention distribution; semantic integration is performed in combination with attention distribution and general knowledge to form fusion representation, and conflicts are arbitrated through a rule engine to ensure safety and reliability; finally, the conflict-free reasoning path and the patient demand vector are combined to generate professional, accurate and living personalized rehabilitation guidance, and seamless cooperation of medical preciseness and life practicability is achieved.

The invention provides a multi-modal three-dimensional brain tumor segmentation method based on a lightweight network, and relates to the field of image processing, and the method comprises the steps: obtaining a brain tumor MRI image with an annotation, and carrying out the preprocessing of the image; the method comprises the following steps: constructing an LE-BTS model based on a three-dimensional dynamic wavelet transform module DyHWT, a multi-scale space coordination module MSSC and a cross-scale aggregation module CSAFormer; training an LE-BTS model through the preprocessed brain tumor MRI image; and obtaining a brain tumor MRI image to be segmented and inputting the brain tumor MRI image into the trained LE-BTS model to obtain a brain tumor high-resolution segmentation result. The invention provides a light and efficient brain tumor segmentation model. According to the model, redundant features are inhibited through dynamic wavelet transformation, tumor boundary expression is enhanced through a multi-scale space coordination module and a self-adaptive multi-dimensional attention mechanism, semantic integration is enhanced through a cross-scale aggregation module, and collaborative optimization of precision and efficiency is achieved.

The application discloses a kind of intelligent retrieval method and system of fusing text and image semantic features, the method includes: S1.to the quality detection and pretreatment of the image scan of electronic dossier and case simultaneous recording video;S2.structural electronic dossier catalog is built;S3.text semantic features, dossier image semantic features and video image semantic features are extracted as the multimodal feature of text and image;S4.the multimodal feature of text and image is fused and aligned by multimodal big model, and cross-modal uniform feature vector with semantic consistency is generated;S5.case knowledge graph is automatically constructed, and the structured and semantic integration of legal information is realized;S6.based on natural language query and case knowledge graph, semantic analysis and multi-hop reasoning are carried out, and the retrieval result in the form of structured or question and answer is generated and presented.The application realizes deep knowledge mining and efficient intelligent retrieval of electronic dossier by fusing the semantic features of text and image.

The invention provides a feature reconstruction and classification method combining multi-mask reconstruction and a U-shaped integrated network. According to the method, firstly, images with different shielding visual angles are generated through multiple groups of masks, multi-mask features are extracted by an independent encoder, and multi-angle reconstruction is performed by a shared decoder after splicing and channel compression, so that maskless fusion features are obtained. On the basis, the U-shaped integrated network performs semantic integration and interlayer fusion on the multi-scale features to realize more efficient feature extraction. According to the method, a high-fidelity reconstructed image and a category prediction result are output at the same time, and joint optimization of reconstruction and classification is realized. Experimental results show that the method has excellent performance in image classification and reconstruction tasks, and has strong generalization ability and application value.

The invention relates to the field of music information processing, and particularly discloses a folk music classification system combining cultural semantics. The system aims at solving the problems that culture semantic integration is insufficient, precision is limited and deep connotation is difficult to reflect in folk music classification. The system comprises a multi-modal data acquisition module, an audio feature extraction module, a culture semantic knowledge graph construction module, a culture semantic feature vectorization module, a multi-modal feature fusion module, an adaptive classification model training and reasoning module and a classification result interpretation and feedback module. By fusing audio features and culture semantics, the system can comprehensively process multi-modal data, deeply mine culture semantic information, improve classification precision and culture sensitivity, and more accurately reflect culture attributes and values of folk music.

The invention relates to an accurate medication time recommendation method and system based on a multi-organization rhythm map. The method comprises the following steps: constructing a drug database and a rhythm expression database; acquiring a drug identifier input by a user, performing retrieval in a database, performing rhythm feature analysis based on a retrieval result, and generating a drug-target rhythm analysis report; based on the drug information queried by the two-stage serial query architecture, querying all biological pathways and related genes in which the drug participates, carrying out standardized conversion on gene identifiers, carrying out rhythm feature analysis, and generating a drug-pathway rhythm analysis report; and taking the generated analysis report as input, carrying out semantic integration and reasoning through a locally deployed large language model, and automatically generating an optimal medication time suggestion. Compared with the prior art, the method has the advantages that batch rhythm characteristic analysis of drug target genes and related pathway genes is realized, the analysis efficiency and repeatability are greatly improved, and the decision intelligence level is improved.

The invention discloses a method for automatically generating a spaceflight field quality problem return-to-zero report based on a large model. The method comprises the following steps: acquiring a spaceflight product fault phenomenon and reason analysis text input by a user; screening a historical fault candidate set from the historical fault data, and obtaining a historical fault association set from the historical fault knowledge graph based on the candidate set; performing report analysis and information extraction on the fault phenomenon and reason analysis input by the user and the historical fault association set to obtain a structured knowledge triple, and generating a fault tree fusing a product physical structure; semantic integration is carried out based on the fault tree, and a problem positioning text and a mechanism analysis text are generated; respectively generating a corrective measure text, a preventive measure text and a first-third measure text; and finally outputting a return-to-zero report conforming to the format specification. The method aims at overcoming the defects of an existing spaceflight field quality problem return-to-zero report generation mode, and rapid, accurate and standard generation of the quality problem return-to-zero report is achieved.

The invention discloses a BIM (Building Information Modeling) and GIS (Geographic Information System)-based rail transit digital model multi-source data conversion and integration method, which comprises the following steps: constructing a B / S (Browser / Server) architecture based on a Cesium open source map engine, and the B / S architecture comprises a data layer, a business logic layer and a presentation layer. According to the system, topographic data collected by an oblique photogrammetry technology, vector data collected by an oblique model and BIM data generated according to rail transit construction requirements are integrated, the data from different sources have respective characteristics and advantages, and a spatial semantic integration algorithm in the conversion integration unit is utilized; according to the method, data in a BIM model is converted and integrated into a GIS model, consistency in geometry, semantics and precision is ensured, the problems of semantic loss, geometric deformation and the like existing in BIM and GIS data conversion in a traditional method are solved through a spatial semantic integration mode, data seamless integration in the true sense is achieved, and the method is suitable for large-scale popularization and application. Reliable guarantee is provided for subsequent data analysis and application, and the requirements of workers are met.

This application discloses a data processing method, apparatus, device, storage medium, and program product, relating to the field of artificial intelligence technology. The data processing method includes acquiring service dialogue data between customer service representatives and users, wherein the service dialogue data includes at least two consecutive messages from the same role in both the customer service representative and the user role; inputting the service dialogue data and a first prompt word into a first large language model to obtain a standard dialogue corpus output by the first large language model; using the first prompt word to constrain the first large language model to semantically integrate the at least two consecutive messages from the same role, forming a single-sentence dialogue text of alternating question-and-answer between the two roles; and constructing a training dialogue corpus based on the results of business topic verification and semantic structure verification in the standard dialogue corpus. This enables the rapid construction of a training dialogue corpus, solving the problem of low training efficiency.

The application provides a multi-source heterogeneous data interaction and fusion method and system for a shield scene, and relates to the technical field of data fusion.The application comprises the following steps: integrating multi-source heterogeneous data sources from different systems by using a network service Webservice technology, establishing databases under different construction scenes, and forming a multi-source database; presetting a micro-service granularity, and performing data transfer on data of the multi-source database by using multiple micro-service components; performing unified semantic query calling on output data after transfer by using a unified data interaction service gateway, and completing multi-source heterogeneous data interaction and fusion.The application divides the whole system into multiple micro-services, and simultaneously communicates between the micro-services by using a unified RestAPI interface, thereby improving the expansibility of the system and reducing the difficulty of operation and maintenance; the core data integration micro-service integrates multi-source heterogeneous data sources by using Webservice and ontology technology, thereby effectively improving the semantic integration degree of data and data interaction.

The invention belongs to the technical field of computer vision, and particularly relates to a multi-scale-based anaphora video target segmentation method, which comprises the following steps of S1, sending input text information into a text encoder to obtain text embedding; the input video frame sequence is sent to a video encoder, and visual features are extracted; s2, inputting the text embedding and visual features into a multi-modal fusion module to generate multi-modal fusion features; s3, inputting the multi-modal visual fusion features into a semantic integration module, and outputting enhanced object features; s4, performing multi-scale fusion on the enhanced object features to obtain high-resolution semantic perception features, inputting the features into a classification head, a bounding box prediction head and a mask prediction head at the same time, and performing classification, bounding box position prediction and pixel-level mask prediction respectively; and finally, matching and integrating prediction results, and outputting a segmentation result of the target object. The problem that an existing reference video target segmentation method is low in multi-scale target segmentation accuracy is solved.

The application discloses a text classification method based on label semantic learning and attention adjustment mechanism, and mainly comprises the following steps: preprocessing text data, extracting text semantic features, text label graph embedding, using a multi-head adjustment attention mechanism to measure the semantic relationship between words and labels, then multi semantic integration and network training, thereby realizing multi-label text classification, training the model, and then using the trained model to predict the category of a text. The application proposes a multi-head adjustment attention hybrid BERT model for a multi-label text classification framework, which can effectively extract useful features from text content, establish semantic connection between labels and words, obtain label-specific word representation, and thus improve the performance of multi-label text classification.

The invention discloses a multi-modal 3D target detection method and device based on unified voxel-semantic integration, and the method comprises the steps: S1, constructing a multi-modal 3D target detection data set, carrying out the voxelization processing of laser radar point cloud data, and carrying out the preprocessing and data enhancement of an RGB image; s2, constructing a unified double-flow perception encoder, extracting high-fidelity geometric features through geometric decoupling convolution, and realizing decoupling purification of the geometric features and semantic features; s3, constructing a cross-modal layered feature fusion module to realize bidirectional enhancement of point cloud and image features; s4, constructing a semantic perception global enhancement module, and actively extracting information; and S5, designing a composite multi-task loss function. According to the invention, a high-precision and high-robustness 3D target detection solution is provided for application scenes such as automatic driving.

The invention provides a multi-modal fused high-precision semantic map construction and centimeter-level positioning navigation method, which comprises the following steps of: firstly, constructing a'structure-texture-semantics' integrated high-precision map by fusing multi-source data, and then constructing a multi-modal fused high-precision semantic map based on a multi-technology fused centimeter-level positioning technology. The method comprises the following steps: realizing centimeter-level positioning in a complex health-care scene through'high-frequency odometer + low-frequency absolute observation 'double-level data acquisition and in combination with a multi-positioning fusion algorithm, and finally establishing an end-to-end navigation large model: taking multi-source sensing data as input, outputting an executable local path through'coding-fusion-generation' end-to-end process, and finally establishing an end-to-end navigation model; the core logic is combined with a vision-language-action fusion thought of TrackVLA to establish an end-to-end navigation large model. According to the method, a multi-modal semantic map is constructed, a multi-technology positioning scheme is fused, real-time path re-planning and personnel behavior intention prediction are realized, and the method adapts to layouts and people flow rules of different health-care areas.

The invention relates to an unmanned aerial vehicle view angle crowd counting method and a computer readable storage medium, an established unmanned aerial vehicle view angle crowd counting model takes FasterNet as a backbone network, a cross-layer feature pyramid module and a scale perception fusion module are introduced into a neck network, and a density map regression head is introduced into a head network; the cross-layer feature pyramid module realizes depth interaction of space and channel dimensions among different scales, so that the model has stronger structure alignment and semantic integration capabilities; meanwhile, the scale perception fusion module breaks through the bottlenecks of high calculation complexity and low ectopic information utilization rate in traditional multi-scale fusion through a guide type attention mechanism with a dominant scale as a core, and the selectivity and adaptability of feature fusion are improved; according to the model, on the premise that the structure is simple and the calculation efficiency is high, multi-scale feature expression and a global attention mechanism are effectively fused, and the modeling capability of dense small targets is remarkably improved.