327 results about "Spatial perception" patented technology

The invention relates to the field of airline planning, and discloses a digital twin system for low-altitude airline planning and a construction method, and the method comprises the steps: obtaining a spatial perception data set of a preset sensor in a target airspace, carrying out the fusion of the spatial perception data set, obtaining the fusion perception data, obtaining the historical flight hidden danger data of the target airspace, and obtaining the historical flight hidden danger data of the target airspace; performing multi-layer convolution on the fusion perception data by using the historical flight hidden danger data to obtain a risk analysis result, performing fuzzy fusion on preset risk factors, generating a dynamic obstacle risk field according to the risk factors after fuzzy fusion and the risk analysis result, performing flight simulation on the dynamic obstacle risk field, and obtaining a flight simulation result; and collecting flight simulation real-time data in the flight simulation process, generating an air route space cost map according to the flight simulation real-time data, and performing optimal path search on the air route space cost map to obtain a low-altitude air route scheme. In the face of airspace actual risks, the searching efficiency of the optimal path in low-altitude route planning is improved.

The invention belongs to the technical field of computer vision and artificial intelligence, belongs to the cross technical field of target detection, deep learning and image processing, and particularly relates to a YOLOv8 algorithm improvement method based on an unmanned aerial vehicle aerial image small target detection model, which comprises the following steps of: introducing a user-defined feature enhancement module into a YOLOv8 backbone network, a neck part and a detection head part; the self-defined feature enhancement module comprises a context guide self-adaptive fusion module introduced into a backbone network so as to replace part of traditional convolution operation; a space edge sensing feature up-sampling module and a space sensing enhanced convolution module are adopted in the neck fusion network; a fine-grained dynamic pruning detection head is introduced into a detection head detection network. According to the method, the performance of the model in a small target detection scene is effectively enhanced, and the accuracy, robustness and real-time response capability of a detection system are remarkably improved.

The invention discloses a remote sensing image segmentation method fusing frequency modulation and spatial perception, and the method comprises the steps: obtaining and preprocessing an original remote sensing image, and generating a standardized input image; the image is input into a multi-scale frequency domain enhanced feature extraction network, features are extracted step by step according to a plurality of feature levels, each level realizes frequency adaptive semantic enhancement through frequency domain modulation transformation and spatial feature fusion, and deep feature expression is enhanced through feedforward neural network modeling and residual connection output and cross-level residual fusion introduction; the final multi-scale features are decoded through a decoding module, the spatial resolution is recovered, and a pixel-level segmentation result is generated; and constructing a composite loss function containing classification errors, boundary perception and frequency consistency items, and carrying out optimization training on the network. According to the method, semantic complementarity of a remote sensing image in a frequency domain and a space domain is fully mined, so that segmentation precision and robustness of a ground object target in a complex scene are improved, and the method has good generalization ability and engineering practicability.

A system for three-dimensional spatial perception underwater that includes a primary submersible vehicle. Frontward facing projectors are used to scan beams of light across a selected field of underwater view. Further, event cameras with pixels triggered by events are used to detect corresponding spots of the scanned beams that are reflected from underwater objects or surfaces. Circuitry is configured to determine positions of the objects based on the pixels that are triggered. Additionally, pilot submersible vehicles the projectors and event cameras are tethered to the primary submersible vehicle to provide extend detection of spots corresponding to reflections from underwater objects or surfaces.

The invention relates to the technical field of civil engineering construction, in particular to a precise positioning civil engineering pile foundation construction system which comprises a multi-source space sensing and positioning module, a space-time multi-dimensional data fusion module, a three-dimensional guiding and correcting module, a hydraulic control module and a closed-loop feedback optimization module. The multi-source space sensing and positioning module obtains pile machine position and attitude information through the Beidou / GNSS high-precision positioning unit, the inertia high-precision navigation unit and the laser horizontal measurement unit, and the space-time multi-dimensional data fusion module processes and fuses the information to generate unified space position data. The three-dimensional guiding and correcting module compares the data with preset pile position coordinates, if the deviation exceeds a threshold value, a correcting instruction is generated, the hydraulic control module executes the instruction and adjusts the position and the posture of a pile machine, the closed-loop feedback optimization module collects the data in the correcting process and optimizes a correcting strategy, all-weather centimeter-level precision space positioning is achieved, and the positioning precision is improved. And the accuracy and efficiency of pile foundation construction are improved.

The invention discloses a digital twin interaction control system based on robot vision and tactile fusion, and the system comprises a data collection module which is used for collecting visual image data and tactile sensing data; the preprocessing module is used for preprocessing; the multi-mode perception modeling module is used for executing spatial perception modeling and tactile state modeling; the fusion state vector construction module is used for generating a fusion state vector; the digital twinborn mapping module is used for constructing a digital twinborn body; the simulation interaction control module is used for executing interactive action control and generating a control instruction parameter group; the action execution and feedback module is used for controlling the robot to execute the actual action and updating the fusion state vector; and the dynamic evaluation and self-adaptive adjustment module is used for dynamically optimizing and adjusting the fusion state vector. Visual and touch information of the robot is fused, a digital twinning synchronous control mechanism is constructed, and the interactive optimization method which is fast in operation response, high in execution precision and stable in control process is achieved.

The invention discloses a road disease detection method based on an unmanned aerial vehicle, electronic equipment and a program product. The method is realized based on a trained road disease detection model, a C3k2-MDDSC module is introduced into a backbone network of the model, the feature multiplexing capability is enhanced through gradient shunting and multi-scale fusion, the gradient disappearance problem is relieved, and the robustness of the model is improved by means of jump connection and packet convolution. An ACFP module is introduced into the tail end of the backbone network, dynamic fusion of local and global features is realized by using multi-scale cavity convolution and a channel-space attention mechanism, and the complex scene modeling capability is improved. And the neck network is integrated with an SGF module, so that the spatial perception of the model to a tiny target can be improved. Besides, the ES-FPN proposed based on the SGF module not only can enhance the utilization of shallow spatial information, but also can optimize the complementarity of cross-level features. During training, regression loss, namely fast high-quality intersection-to-union ratio loss, is proposed, and angle punishment is introduced to improve the alignment precision of the rotating frame and the convergence speed of the model.

The invention discloses a zero sample anomaly detection method and system based on a triple perception learning enhanced visual language model, and relates to the field of computer vision, and the method comprises the steps: extracting global and local visual features from an input image; in the visual coding process, local features in a deep network are corrected through a spatial perception attention enhancement module, and fine-grained attribute text description is generated for abnormal visual features; performing deep semantic alignment on the attribute text description and the general text prompt through an attribute perception guide module; calculating the similarity between the enhanced visual features and the optimized text features, and generating a pixel-level abnormal segmentation map; in the inference stage, the segmented image is converted into a space attention weight through an anomaly perception reconstruction module, the space attention weight is fed back to a visual encoder to generate final global feature representation, and an anomaly score is calculated. According to the method, under the condition that a target domain training sample is not needed, the anomaly detection and positioning accuracy and generalization ability of the model under the scenes of industrial defect detection and the like are remarkably improved.

The invention is applicable to the technical field of medical image processing and mixed reality, and provides a laparoscopic surgery mixed reality navigation method based on deep learning and dynamic point tracking, which comprises the following steps: dynamically registering a three-dimensional model containing kidney, tumor and vessel with an initial frame of a laparoscope through a mixed reality alignment technology; the method comprises the following steps: constructing an operating forceps motion sensing model based on a time sequence deep neural network, realizing real-time control and parameter locking of a three-dimensional model pose, dynamically updating a two-dimensional feature point set by adopting a multi-feature-point combined tracker, constructing a candidate feature combination through a cross-quadrant sampling strategy, and constructing an operating forceps motion sensing model; a candidate feature combination is generated through a four-quadrant division and cross-regional sampling strategy, an optimal camera pose parameter is generated in combination with a re-projection error and pose continuity constraint, and an operation video is dynamically covered with a semitransparent three-dimensional model. The method can significantly enhance the spatial perception capability of the kidney anatomical structure, reduce the registration error of the kidney in the three-dimensional integrated kidney structure model and the laparoscope video, and improve the navigation precision.

The present disclosure provides a method for calibrating a humanoid robot, comprising obtaining a humanoid robot with original kinematic biasing values, controlling the humanoid robot through predetermined poses, capturing image data of body parts using vision sensors mounted on the humanoid robot while moving through the poses, determining revised kinematic biasing values by processing the image data using a bipedal spatial perception model trained using synthetic image data containing keypoints, and replacing the original kinematic biasing values with the revised kinematic biasing values. The bipedal spatial perception model processes captured image data to generate observed keypoint locations on robot components, which are compared with kinematic-based locations from joint encoder measurements to minimize discrepancies through optimization algorithms.

The invention discloses a traffic flow prediction method based on multi-scale dual hypergraph fusion. Aiming at the problem that an existing traffic flow prediction model is difficult to capture a multi-scale high-order spatial dependency relationship of a traffic road network, the method comprises the following steps of: firstly, constructing an urban road dual hypergraph with three scales of microscopic individual travel intention, mesoscopic community commuting interaction and macroscopic area flow conduction, and then designing a traffic flow prediction method of space-time perception based on the urban road dual hypergraph. Wherein the spatial perception module extracts and fuses a dependency relationship between a high-order local spatial feature and a global spatial feature in a multi-scale dual hypergraph; the time sensing module captures short-term fluctuation and long-term trend of traffic flow; and residual connection is introduced to enhance spatio-temporal feature fusion, and finally multi-step traffic flow prediction is realized. The invention provides a traffic flow prediction model with multi-scale high-order space perception, which adaptively fuses high-order space features of three scales of microcosmic, mesoscopic and macroscopic, and significantly improves the precision and generalization ability of traffic flow prediction.

The invention discloses a visual language action large model design method for enhancing spatial perception ability, which comprises the following steps of: acquiring a plurality of frames of task scene images, inputting the images into a two-dimensional image encoder and a visual geometry guidance Transformer VGGT encoder, extracting image features and performing time sequence spatial embedding, encoding a natural language instruction into language embedding representation, and extracting a visual language action large model. Image features and language embedding are fused through a cross attention mechanism, a pre-training large model is input to generate cross-modal representation, the fusion features are input into an action expert module to output an action control sequence in combination with robot body state information, and a robot is driven to execute an operation task. Therefore, damage of additional information to an original pre-training model is avoided, and compared with an original combined structure of a pre-training visual language model and an action expert, utilization of multi-view picture information is enhanced, so that higher understanding ability on space depth is achieved in the task execution process, the task success rate is increased, and the task execution efficiency is improved. And a more efficient and more robust robot sensing and decision-making integrated system is realized.

The invention discloses a bridge inclination and settlement monitoring method and system based on multi-sensor fusion, and belongs to the field of bridge structure monitoring, and the method comprises the steps: obtaining the data of a multi-source heterogeneous sensor; the sensor data is converted into space-time diagram data, and the space-time diagram data comprises the steps that each sensor is mapped into nodes in a diagram, edges between the nodes are defined according to the physical connection relation of the bridge structure, and multi-source heterogeneous sensor data are unified into dynamic feature vectors with the same dimension on the nodes through learnable feature mapping; inputting the time-space diagram data into a preset neural network model, performing spatial feature aggregation on the dynamic feature vector through a diagram attention mechanism, performing time feature extraction through a time convolutional network, and generating a hidden state vector fused with time-space information; and generating a monitoring state value of the bridge based on the hidden state vector. According to the invention, depth feature fusion of spatial perception is realized, and the sensitivity of anomaly recognition is improved.

The invention discloses a Bluetooth hearing-aid earphone optimization method and system for adaptive hearing compensation. The method comprises the following steps: S1, acquiring a personalized hearing curve of a user and background audio data of left and right ears in each typical scene, and respectively preprocessing the background audio data; s2, extracting time-frequency fusion features of each typical scene, generating scene feature vectors, and constructing a scene feature vector library; s3, calculating a scene perception conversion vector, an abrupt change risk vector and a smooth gain vector; s4, calculating a binaural gain change vector, a binaural synchronous entropy vector, a phase alignment factor vector and a left and right ear cooperative gain vector; and S5, according to the left and right ear cooperative gain vectors, processing the background audio data after left and right ear preprocessing in the current scene, and outputting the processed background audio data to the Bluetooth hearing-aid earphone. According to the invention, the problem of gain abrupt change generated during scene switching of the existing Bluetooth hearing-aid earphone and the problem of spatial perception distortion caused by binaural coordination imbalance can be solved.

The invention relates to a remote sensing image target detection method and device based on an adaptive auxiliary head structure. According to the method, a remote sensing image target detection network based on a self-adaptive auxiliary head structure is designed, and the network is obtained by adding the self-adaptive auxiliary head structure between a YOLO classical network structure neck and a detection head and replacing an original neck structure with a context aggregation bidirectional connection structure; a spatial perception attention module, convolution recalibration multi-scale feature fusion, a context aggregation bidirectional connection structure and a self-adaptive auxiliary head structure are designed; the network has good spatial feature aggregation capability to retain key feature information, an adaptive weight mechanism is integrated, information loss caused by fusion of different scales is reduced, feature refinement is performed on a to-be-detected image, and the accuracy and effectiveness of remote sensing target detection are improved.

The invention relates to the field of robot navigation, and particularly discloses a spatial perception observation information quality evaluation method and system, a terminal and a medium, and the method comprises the steps: converting the measurement information of each dimension into a normalization probability according to the measurement target of each dimension; according to the concept of entropy in the information theory, the normalized probability of each dimension is converted into a computable entropy value; a downstream task weight coefficient is introduced for the entropy value of each dimension, the multi-dimensional entropy value after the downstream task weight coefficient is introduced is mapped into a physical interpretable confidence coefficient in a [0, 1] interval through an exponential attenuation function, and the confidence coefficient explicitly represents the uncertainty of observation information of each dimension. According to the method, a space perception quality evaluation system with mathematics interpretability, scene adaptation and real-time calculation is formed, multi-source uncertainty is mapped into executable confidence, and multi-dimensional accurate evaluation of intelligent agent space perception observation information is achieved.

The application discloses a kind of based on triple perception learning enhanced visual language model's zero sample exception detection method and system, it is related to computer vision field, method includes: extracting global and local visual features from input image;Visual coding process is corrected local feature in deep network by spatial perception attention enhancement module, and fine-grained attribute text description is generated for abnormal visual feature;Through attribute perception guide module, attribute text description and general text prompt are deeply semantically aligned;The similarity of enhanced visual feature and optimized text feature is calculated, and pixel-level exception segmentation map is generated;Inference stage converts segmentation map into spatial attention weight by exception perception reconstruction module, and feedback is generated to visual encoder final global feature representation and calculates exception score.The method of the application significantly improves the accuracy and generalization ability of model in industrial defect detection and other scenarios without target domain training samples.

The invention discloses an interactive commodity selection method and system based on augmented reality, and relates to the technical field of augmented reality interaction, and the method comprises the steps: obtaining the commodity three-dimensional information of a target commodity and the commodity attribute information corresponding to the commodity three-dimensional information; determining scene reference coordinates of the augmented reality scene and ambient light features matched with the scene reference coordinates according to the spatial perception data of the terminal device, and generating an augmented reality presentation object of the target commodity in the augmented reality scene; obtaining interaction behavior data according to the interaction action of the user in the augmented reality scene; performing association matching on the interaction behavior data and the commodity attribute information to generate a candidate commodity selection feature set; and according to a preset fusion criterion, performing fusion on the candidate product selection feature set to obtain a target product selection decision, and outputting interaction feedback information corresponding to the target product selection decision. According to the application, robust fusion and reliable decision of multi-modal interaction information are realized, and the accuracy and interaction consistency of an augmented reality product selection process are improved.

The invention discloses a light-weight photovoltaic module infrared defect detection method, which comprises the following steps of: 1, acquiring a private infrared thermal image data set by matching an unmanned aerial vehicle with an infrared camera, and acquiring a public infrared thermal image data set on a Roboflow website; step 2, marking defects for the private infrared thermal image data set, performing data enhancement, and dividing the private infrared thermal image data set into a training set, a verification set and a test set; 3, introducing a spatial perception and local enhancement convolution module (SLConv), a spatial attention and dynamic modulation convolution module (SDConv) and an image enhancement mask module (IEM) on the basis of the YOLOv11; 4, obtaining a lightweight high-precision model through private data set training; and step 5, verifying the robustness on the public data set, and deploying a fixed infrared acquisition detection system to verify the generalization of the model so as to realize real-time intelligent monitoring of the defects of the photovoltaic power station.

The invention provides a needle beam trolley construction digital management method, and relates to the technical field of digital management, and the method comprises the steps: employing a laser scanner to scan the wall surface of a to-be-constructed tunnel; acquiring position information of a plurality of working windows of the needle beam trolley; carrying out operation sequence enumeration by taking the maximum number of cooperative work windows as a constraint, and generating a plurality of work window operation sequences; based on the tunnel wall face three-dimensional model, with the requirement for the expected construction quality as the constraint and the requirement for maximizing the construction efficiency as the target, according to the multiple working window operation sequences, casting parameter and vibration parameter optimization is conducted, an optimal construction scheme is output, and digital construction management is conducted on the needle beam trolley. The technical problem of low construction efficiency caused by unreasonable operation sequence and non-uniform resource distribution in the construction process is solved, spatial perception is performed on the to-be-constructed tunnel through three-dimensional modeling, multi-aspect optimization is realized in combination with digital management, and the construction efficiency and precision of the needle beam trolley are improved.

The invention relates to the technical field of deep learning, in particular to a small sample fine-grained image classification method based on multilayer feature dynamic interactive fusion, which comprises the following steps: inputting a support set and a query set into an image classification model, and outputting a prediction category of a query sample; the image classification model processing steps are as follows: S201, extracting to obtain multi-scale features; s202, obtaining interaction features after interaction optimization of each support sample through a multi-layer feature interaction module; s203, obtaining fusion features of each support sample through a dynamic feature fusion module; s204, projecting the fusion features to a corresponding feature space; s205, calculating category prototype representation of each category in each feature space; s206, calculating the cosine similarity between the query sample and the category prototype representation of each category in each feature space; and S207, determining a prediction category of the query sample based on the cosine similarity. According to the method, efficient cross-layer feature aggregation can be realized, and the capability of capturing fine-grained differences is enhanced through an attention mechanism of spatial perception.

The present disclosure relates to the technical field of audio control. Provided are a computer-implemented method for audio control, and an electronic device and a storage medium. The computer-implemented method for audio control comprises: on the basis of a specified measurement signal, performing acoustic effect testing on a target space, in order to obtain environmental response information, wherein the environmental response information comprises impulse response information and post-separation impulse response information; analyzing the environmental response information, in order to obtain acoustic characteristic information of the target space; on the basis of the acoustic characteristic information, performing compensation processing on audio to be processed, in order to obtain regulated audio; regulating gains of multiple channels corresponding to the regulated audio, in order to obtain re-regulated audio; and performing multi-channel collaborative optimization processing on the re-regulated audio, in order to obtain audio having a target effect. The method achieves precise compensation for spatial acoustic defects, resolves inter-channel crosstalk, improves the acoustic image localization and spatial perception, significantly improves the perceptual audio quality, and optimizes an output.

The invention provides an under-mine target detection method based on multi-scale feature fusion, and the method comprises the steps: obtaining an under-mine image, carrying out the shape preprocessing of the under-mine image, obtaining an image with a fixed size, and taking the image as an input image I; constructing a feature extraction network based on deformable convolution and a cross attention mechanism, and performing feature extraction on the input image I to obtain four output feature maps P2, P3, P4 and P5 of different scales; constructing a feature aggregation network based on attention and scale consistency, and performing feature aggregation on the feature maps P2, P3, P4 and P5 to obtain three feature maps F3, F4 and F5; constructing an IoU loss function based on spatial perception, and performing detection frame regression on the feature maps F3, F4 and F5 to obtain a plurality of candidate detection frame sets B = {B1,..., Bi,... BN}; and based on linear attenuation, performing target positioning in a candidate box screening stage to obtain category information and position coordinates of the miner and the wear thereof, and performing equal-proportion coordinate change on the original image according to the position coordinates to obtain actual coordinate values.

The invention provides a self-service zooming method of a camera system based on depth-guided image quality evaluation. The method comprises three core links of multi-scale depth perception, depth-guided image quality evaluation and joint-driven intelligent focusing control. By fusing lightweight feature coding and a multi-scale optimization strategy, accurate depth information is generated, and a spatial perception basis is provided for subsequent processing; the depth features and the image texture features are combined, a focusing fuzzy evaluation model is constructed, and reliable feedback of the imaging quality is achieved; a key focusing area is identified based on depth and quality information, a target focal plane is decided, and a lens is driven through an intelligent control algorithm to complete rapid and accurate focusing. Through collaborative fusion of depth perception and image quality evaluation, intelligent zoom control of the camera system in a dynamic scene is realized, the imaging consistency and focusing precision are effectively improved, the method is suitable for multiple camera devices such as smart phones, security monitoring and automatic driving, and the imaging experience of users is improved.

The invention discloses a method for realizing three-dimensional point cloud classification based on an NCFMama network, relates to the technical field of three-dimensional data processing, and solves the problems that an existing point cloud classification method is insufficient in local feature extraction, difficult in consideration of calculation complexity and global modeling capability and the like. According to the method, three dimensions of spatial distribution, feature similarity and direction consistency of a local neighborhood of the point cloud are comprehensively analyzed through a neighborhood consistency feature modulator, adaptive feature aggregation of different geometric regions is realized, and meanwhile, a bidirectional spatial perception fusion module is provided. The module uses point cloud three-dimensional space distance information to guide feature interaction of a bidirectional state space model, information exchange between adjacent points of a point cloud space is enhanced, and high-precision and high-efficiency point cloud classification is realized. According to the method, the advantages of calculation efficiency, classification precision and robustness are combined, and a new technical path is provided for application of the Mama architecture in point cloud processing.

The invention relates to the technical field of image analysis, in particular to a steel tube tower assembly intelligent welding visual identification robot cooperative control system, which comprises a groove surface pollutant identification module, a groove surface pollutant identification module, a linear polarization component calculation module and a control module, meanwhile, the original light intensity image is converted into a CIELAB color space, an a channel value is extracted, the polarization degree value of each pixel point is fused with the corresponding a channel value, and a pollutant pixel mask is established. According to the method, in the image analysis process, the multi-angle polarization image is introduced and the color channel value is fused, so that precise recognition of a tiny pollution area is achieved, and the space sensing capacity of groove pollutants in a complex welding environment is enhanced; the space coordinates of the pollutant area and the welding seam edge are compared and screened point by point, the overlapped part of the pollutant area and the welding path is positioned, and the pertinence and accuracy of pollution interference identification are improved.

A humanoid robot system comprises vision sensors for capturing image data, a computing architecture with processing hardware and memory, and a bipedal spatial perception model. The model includes a feature extractor that extracts hierarchical feature maps from input images, a robot data module that detects robot parts, and a robot vector data module that calculates three-dimensional spatial position and orientation data for each detected robot part. The feature extractor uses a feature pyramid network generating multi-scale feature maps through bottom-up and top-down pathways with lateral connections. The robot vector data module predicts 2D-to-3D point correspondences and solves perspective-n-point problems to obtain final position and orientation vectors, enabling real-time robot self-awareness and closed-loop visual servoing for precise object interaction.

The present invention relates to the field of audio processing technology, and discloses a wireless microphone intelligent control system and method, wherein a wireless microphone intelligent control method includes: collecting multi-point acoustic data through a distributed microphone network to construct a spatial acoustic topology model; constructing an adaptive notch filter group to achieve howling suppression based on the topology analysis results; monitoring environmental changes in real time through a distributed acoustic sensor network to dynamically optimize suppression parameters; integrating a lightweight convolutional neural network on the microphone device to identify acoustic events; and adopting a layered edge computing architecture to achieve low-latency acoustic event processing and howling prevention. The present invention effectively solves the technical problems of unstable howling suppression, weak environmental adaptability and lack of spatial perception in complex environments.

The invention discloses an urban space unit digital segmentation method, and belongs to the technical field of urban planning, and the method comprises the steps: inputting site data, simplifying and integrating the plane contour of a building in a site, and obtaining a site space and the integrated plane contour of the first layer of the building; on the basis of the integrated plane contour of the first floor of the building, according to a site space relation, connecting and screening building end points to nearest points of the building contours to obtain preliminary division control lines; performing constraint denaulay triangulation based on the preliminary division control line, the building contour and the site plane contour to generate a triangulation network unit; and merging the triangulation network units according to the maximum edge merging and concave angle avoiding principle to obtain an urban space division result. According to the method, space division can be carried out on a public space in a relatively complex urban plot, a space unit division result relatively conforming to actual space perception of people is formed, and meanwhile, the calculation efficiency and the calculation accuracy are improved.

The invention provides a deep learning channel estimation method based on spatial perception interpolation, and belongs to the technical field of wireless communication and artificial intelligence fusion. The method comprises the following steps: firstly, constructing a pilot frequency index set according to sparse distribution of pilot frequency points, and mapping the pilot frequency index set to a two-dimensional subcarrier-symbol grid; then generating a centrosymmetric and edge-attenuated space weighting matrix alpha (x, y), and adjusting interpolation weight distribution by introducing a position-related Gaussian weighting function; complex field interpolation is carried out on the pilot frequency observation value in combination with a space weighting radial basis function, and a complete channel initial estimation matrix is constructed; and then decomposing the interpolation result into a real part, an imaginary part and a spatial weighting coefficient, constructing a three-channel tensor as input, sending the three-channel tensor into a convolutional neural network for refined estimation, and outputting a final complex channel estimation result. In a preferred embodiment, the convolutional neural network adopts an SRCNN structure, and inputs a real part, an imaginary part and a spatial weighting matrix alpha (x, y) including an interpolation channel to enhance the spatial perception ability of the model. Simulation results show that under a VehA standard channel model, the method is always superior to a traditional LS method under the sparse pilot frequency condition, the performance of the method is close to that of an MMSE method, higher estimation precision and higher robustness are shown, and the method is suitable for a channel recovery task in a high-speed dynamic wireless communication system.