79 results about "Distance mapping" patented technology

The invention relates to the technical field of power systems, and discloses a distribution line fault positioning method and system based on traveling wave distance measurement. The method comprises the following steps: injecting a zero-mode signal through an electrified pulse coupler, carrying out wave velocity hierarchical analysis on a mixed line impedance node to obtain a propagation parameter matrix, generating an empty-mode echo by utilizing mode conversion of a fault asymmetric point, screening by adopting a frequency domain fingerprint identification algorithm to obtain a pure fault signal, and carrying out fault detection on the pure fault signal. And three-dimensional fault coordinates are obtained based on reverse time-distance mapping processing. According to the invention, the problem of accurate positioning of the fault point in the live-line operation state of the hybrid distribution line is solved, and the fault positioning accuracy and the power supply reliability are improved.

The invention discloses a city three-dimensional model rendering method, which comprises the following steps of: establishing a multi-level detail model system to create a plurality of three-dimensional models with different levels of detail for each building object in a city scene so as to form a continuous hierarchical structure from high precision to low precision; spatial distance, view angle factors and object importance are comprehensively considered, effective distance measurement from a view point to a target object is calculated, a continuous LOD switching function is established according to the effective distance measurement, and the distance is mapped into continuous detail level values; a smooth interpolation algorithm based on time is applied, progressive switching is carried out between adjacent LOD layers, and visual jump during switching is eliminated; visual jump of LOD switching in urban three-dimensional model rendering is remarkably reduced, the stability of rendering performance is improved, and smooth and continuous visual experience is achieved.

The invention relates to a multi-modal perception fusion target following method and system. The method comprises the following steps: adopting an improved KCF algorithm to realize a closed-loop process of target tracking, multi-scale space construction, feature fusion, response calculation, optimal scale decision and model updating; designing a four-level shielding processing mechanism fusing motion prediction and depth verification, and realizing tracking recovery in a shielding scene through shielding judgment, motion prediction, fine search and template protection; constructing a target distance mapping function fusing geometric distortion correction and attitude compensation, and realizing high-precision distance estimation based on monocular vision; laser radar point cloud information is integrated, an obstacle threat degree model is constructed, and cooperative path planning of following and obstacle avoidance is realized in combination with an improved TEB algorithm; and designing a linear velocity control law and an angular velocity control law based on the distance deviation and the azimuth angle deviation, and driving the robot to complete target following motion. According to the invention, high-precision and robust following of the robot to the target can be realized.

The invention provides a power transmission line icing detection method and system. The method comprises the following steps: collecting an icing image, temperature distribution data and sparse point cloud data; performing alignment processing on the sparse point cloud data to generate a distance mapping graph; inputting the icing image into a SegFormer model, and adjusting the expansion rate of a grouping depth separable convolutional layer through a distance mapping graph to obtain a first icing segmentation feature; generating an icing area pseudo label based on the temperature distribution data, and supervising the icing image to be segmented into a second icing segmentation feature through comparative learning; fusing and outputting the first icing segmentation feature and the second icing segmentation feature to obtain an icing feature; and according to the icing characteristics, extracting and complementing the corresponding sparse point cloud data, and constructing a three-dimensional model of power transmission line icing to obtain a power transmission line icing detection result, so that the technical problem of low accuracy of video detection of the icing condition of the power transmission line under the condition of severe weather in the prior art can be solved.

The embodiment of the invention discloses an RTK-fused laser inertial navigation odometer mapping method and system, and the method comprises the steps: carrying out the time synchronization and space calibration of a laser radar, an inertial measurement unit and a double-mushroom-head antenna RTK system, and achieving the coordinate system alignment through lever arm compensation; multi-source constraints are fused in pose map optimization, and point cloud ghosting is eliminated by associating poses of a current frame and multiple historical frames; designing a dynamic fusion strategy for RTK positioning data, detecting the height of an environmental obstacle by using a laser radar, dynamically adjusting an RTK constraint weight, adding constraint factors in stages, carrying out intensive accumulation in an initialization stage to realize rapid alignment of a global coordinate system, and adding a suppression accumulative error in a subsequent low frequency; and a GTSAM optimizer is adopted to execute pose map optimization, and the optimized key frame point cloud is converted to a global coordinate system to generate a down-sampling map, so that the map consistency, precision and real-time performance of long-distance mapping are remarkably improved.

The invention discloses a train positioning method and device based on ground UWB base stations, and the method comprises the steps: obtaining a first distance between each ground UWB base station and a vehicle-mounted UWB base station for at least three ground UWB base stations, mapping the first distance measured by each ground UWB base station at the moment to a track plane for each moment, and obtaining a first projection distance; determining the position information of the vehicle-mounted UWB base station at the moment according to the position information of each ground UWB base station and each first projection distance; and determining the position information of the train at the moment. According to the running speed information of the train at the moment and a first time difference between the moment and the current moment, determining a first running distance of the train, and further determining first position information corresponding to the train at the moment; and finally, according to the determined first position information corresponding to the train at each moment, determining target position information of the train at the current moment. And the accuracy of train positioning is improved.

The invention discloses a multi-attribute decision-making method and system based on normal swing hesitant fuzzy information and a foreground theory. The method comprises the following steps: constructing an initial decision-making matrix and carrying out normalization processing; generating a multi-dimensional hesitant fuzzy decision matrix by calculating the Hamming distance, the chi-square distance and the trigonometric function distance between each scheme and the positive ideal solution, and converting the multi-dimensional hesitant fuzzy decision matrix into a normal swing hesitant fuzzy decision matrix; the comprehensive hesitant fuzzy entropy method and the logarithm percentage change driven target weighting method are fused to determine the comprehensive weight of the index; mapping the negative ideal solution distance into an income value and mapping the positive ideal solution distance into a loss value in combination with a foreground theory, and generating a foreground value decision matrix and normalizing the foreground value decision matrix; and calculating the correlation degree between the scheme and the dynamic ideal solution through a normal swing hesitant fuzzy weighting correlation coefficient, thereby realizing scheme sorting. According to the method, various kinds of uncertainty information and incomplete psychological characteristics of decision makers can be fully considered, and more accurate index weights and decision results are provided.

The present invention belongs to the technical field of high-precision map quality evaluation for autonomous driving. The present invention provides a crowdsourced map evaluation method and device based on lane lines, wherein the method includes: obtaining a sequence of lane line points based on crowdsourced data, using a custom distance calculation method between lane lines, hierarchically clustering the lane lines collected multiple times to obtain multiple clusters; for each cluster, performing polynomial fitting on the lane lines within the cluster, calculating the distance from each lane line to the fitted lane line, and mapping the distance to between 0 and 1 using a custom function to obtain the cluster-internal confidence; and calculating the global confidence of the map based on the confidence within each cluster and the total length of the lane lines. In the absence of a true map value, the present invention performs relevant calculations and processing on the lane lines to obtain the local and global confidence of the crowdsourced map, thereby respectively evaluating the local and global quality of the crowdsourced map.

The invention belongs to the field of deep learning and computer vision, and discloses an adaptive feature fusion method and device based on optimal transmission, and the method comprises the steps: obtaining a to-be-fused first image feature map and a to-be-fused second image feature map; flattening the two image feature maps in a spatial dimension to obtain a first feature vector set and a second feature vector set; constructing a transmission cost matrix based on the pairing distance between the feature vectors in the two feature vector sets; performing iterative optimization on the transmission cost matrix until an optimal transmission scheme matrix is obtained; obtaining a Warisstein distance through the transmission cost matrix and the optimal transmission scheme matrix, and mapping the Warisstein distance into an adaptive fusion gating weight; and fusing the image feature map through the self-adaptive fusion gating weight. According to the adaptive feature fusion method based on the optimal transmission theory, a principle, distribution perception and more intelligent feature fusion mechanism is realized, and the accuracy of the fused image feature map is improved.

The application provides an image target matching method and device based on local homography, and relates to the field of image processing; the method adopts a registration algorithm based on image local features to pre-acquire natural feature point pairs between background image pairs; secondly, the source background image is evenly divided into multiple rectangular grids; a moving direct linear transformation algorithm is used to calculate the local homography matrix of each rectangular grid; and the mapping point of the image target to be matched in the source image in the target image is calculated by using the local homography matrix; and the image target uniquely existing in the mapping area or the image target closest to the mapping point in the Euclidean distance is taken as the matching point. The application uses the local homography matrix to weaken the mapping error introduced by the parallax change, improve the image target matching precision, reduce the mis-matching probability of the cluster target with similar features and dense distribution in a complex environment, and significantly improve the image target matching speed.

The application discloses a mineral prediction method based on spatial intelligence, relates to the technical field of spatial intelligence, and comprises the following steps: identifying the spatial relationship of a spatial entity element list, constructing a topological connected graph, extracting connected structure attributes, executing semantic relationship reasoning, and generating an object-level spatial intelligent scene graph; based on the object-level spatial intelligent scene graph, constructing an anisotropic cost field, calculating a minimum cost channel distance, mapping into a channel weight expression, and generating a channel weight layer; performing spatial alignment and multi-modal spatial correlation feature fusion on a multi-source mineralization evidence data set, the object-level spatial intelligent scene graph and the channel weight layer, executing mineral inference through a mineral prediction machine learning model, and generating a mineral prediction layer. The application realizes dynamic modeling of a mineralization path, and improves prediction accuracy and engineering practicability.

The invention discloses an algorithm, equipment and system for identifying abnormal operation of a coal mine scraper, relates to the technical field of safety monitoring of coal mine equipment, and aims to solve the problems of lagging and low accuracy of traditional manual inspection. According to the algorithm, videos are collected through an intrinsically-safe camera (1920 * 1080 pixels and 25fps) above the middle of the scraper conveyor, identification points are set to construct an ROI so as to establish pixel-actual distance mapping, scraper blades are recognized through a YOLOv8 model, the actual distance (normal 1.0-1.4 m) and the scraper blade inclination slope (the threshold value is larger than or equal to-0.15, and the abnormality lasts for more than 5 seconds) of the adjacent scraper blades are calculated, and if the threshold value is exceeded, it is judged that the scraper blades are abnormal. The device comprises an image acquisition module, an ROI construction module and the like. The system comprises an underground acquisition terminal, an edge server and an AI platform, and realizes local sound-light alarm and remote monitoring. According to the invention, real-time monitoring can be realized, the abnormal response is less than 30 seconds, the accuracy rate is 94.6%, equipment shutdown can be reduced by 35%, and the labor cost is saved.

The invention discloses an optimization decision-making system and method based on energy production scheduling, and belongs to the technical field of energy management and production scheduling, and the method comprises the steps: obtaining the operation parameters of various types of energy equipment, and constructing an initial operation state set; edge response characteristic parameters under the high-frequency variable load condition are extracted; constructing a graph neural network model, calculating a relative response distance between a current state and an edge state of the equipment, and constructing an edge critical judgment function based on the classification model; mapping the relative response distance into a scheduling priority index, and generating an equipment sorting list; according to the real-time load and the predicted fluctuation range, selecting a response equipment combination in combination with the priority, and performing risk correction on the scheduling instruction; the robustness of the scheduling scheme under the disturbance condition is verified through simulation, and if the requirement is not met, feedback optimization is carried out. According to the method, dynamic perception, self-adaptive scheduling and risk avoidance under the multi-source energy system can be realized, and the scheduling stability and intelligence are improved.

The application discloses a digital phase-shifting phase recovery method of ESPI fringe pattern based on distance mapping, which comprises the following steps: 1) extracting the direction of each pixel point of the fringe pattern, and then calculating the normal curve of the fringe according to the fringe direction to obtain the moving direction of the fringe; 2) extracting the center line of the dark fringe, calculating the width of the fringe according to the center line of the fringe, and obtaining the moving distance of the fringe through distance mapping; 3) comprehensively moving the direction and the distance, and performing digital phase-shifting on the initial fringe pattern point by point, and adopting the bilinear gray interpolation to obtain the fringe pattern after digital phase-shifting; 4) calculating the wrapped phase corresponding to the fringe pattern; and 5) performing unwrapping processing on the wrapped phase to recover the full-field phase of the fringe pattern. The application extracts the phase through the digital image processing technology on the single fringe pattern, avoids the problem of inaccurate phase-shifting existing in the traditional physical phase-shifting technology in dynamic measurement, and can effectively improve the accuracy of the phase recovery of the single fringe pattern.

The application discloses a kind of based on energy production scheduling optimization decision system and method, belong to energy management and production scheduling technical field, obtain the operating parameter of various energy equipment, construct initial operating state set;Extract the edge response characteristic parameter under high-frequency variable load condition;Graph neural network model is constructed, the relative response distance of current state and edge state of equipment is calculated, and edge criticality judging function is constructed based on classification model;Relative response distance is mapped as scheduling priority index, and equipment ranking list is generated;According to real-time load and predicted fluctuation range, combined with priority selection response equipment combination, and risk correction is carried out to scheduling instruction;The robustness of scheduling scheme under disturbance condition is verified by simulation, if it does not meet the requirements, then feedback optimization.Edge characteristic parameters are updated after each scheduling cycle, and the application can realize dynamic perception, adaptive scheduling and risk avoidance under multi-source energy system, improve scheduling stability and intelligence.

The present invention provides a system for three-dimensional distance mapping of an object, the system comprising a light detection and ranging (LIDAR) system, the LIDAR system including an array of beam emitters, at least one detector element, and a computing unit, the computing unit being configured to: instruct the beam emitters to simultaneously emit emission beams; embed ranging information in the emission beams; use unique orthogonal waveforms to identify the respective emission beams; cause the unique orthogonal waveforms in each of the reflected beams received at each detector element to be automatically associated with the unique orthogonal waveforms in the emission beams so as to provide emission and reflected beam pairs; determine the time of flight of each emission and reflected beam pair; and determine the distance from the time of flight.

This application discloses a radar ranging method, apparatus, device, and storage medium. A reflector is installed at one end of the elevator shaft, and a radar is installed in a car moving within the shaft. The method includes: controlling the radar to emit electromagnetic wave signals to the reflector and receiving a first echo signal reflected from the shaft; extracting a first spectral feature of the shaft from the first echo signal; matching the first spectral feature with a second spectral feature pre-learned from the shaft, where each second spectral feature is labeled with a reflector; if the match is successful, measuring the car's position in the shaft based on the distance mapping of the labeled reflectors. By matching the first and second spectral features, the position of the car in the shaft can be measured, reducing interference from highly reflective obstacles such as metal objects, ensuring positional accuracy, and guaranteeing the safe operation of the elevator regardless of where the radar is powered on.

The invention provides an intention recognition and dangerous action evaluation method based on three-dimensional hand posture reconstruction, and relates to the field of action recognition, and the method comprises the steps: obtaining an RGB image, and carrying out the recognition, and obtaining a hand type and an object type; mapping the RGB image to obtain three-dimensional human hand key points; identifying a basic gesture category through the three-dimensional human hand key points; obtaining a basic gesture action through the basic gesture type, the hand type and the object type; performing pose estimation on an object in the RGB image to obtain three-dimensional object key points; establishing a contact dynamic model through the three-dimensional hand key points and the three-dimensional object key points, and mapping the distance into contact force; the contact force is injected into a Transform model, and a fine hand action intention is obtained; according to the object type, the hand action intention, the contact force and the three-dimensional hand key point, a weighted confidence decision tree is set, a danger alarm level is generated, and hand dangerous action evaluation is completed.

A human interface device includes a touch panel and a processing module that is electrically coupled to the touch panel. The touch panel has a first setting region and defines a second setting region in the first setting region. The processing module has a coordinate detection mechanism and a coordinate conversion mechanism. The coordinate detection mechanism defines a first longitudinal coordinate axis having a distance mapping to a length of the first setting region. The coordinate conversion mechanism defines a second longitudinal coordinate axis having a distance mapping to a length of the second setting region. When the second setting region is operated, a touch point of the second setting region has a coordinate of the first longitudinal coordinate axis that is obtained through the coordinate detection mechanism and that is converted to a second longitudinal coordinate of the second longitudinal coordinate axis through the coordinate conversion mechanism.

The present disclosure relates to a method and apparatus, electronic device, and storage medium for determining a touch screen compensation point. The method comprises: obtaining a first touch point, a second touch point, and a third touch point that are not on the same straight line and are collected at a preset frequency; obtaining a first distance between a first intersection point and a fourth point when a target compensation point to be added is located between the second touch point and the third touch point; determining a second distance that has a preset distance mapping relationship with the first distance, wherein the preset distance mapping relationship makes the second distance positively correlated with the first distance, and the magnitude of the increase in the second distance is negatively correlated with the first distance; and determining a point on the vertical line that is the second distance away from the fourth point as the target compensation point. Using this method to determine the target compensation point can improve the efficiency of determining the target compensation point, thereby making the arc of simulated motion estimation more uniform and reducing the probability of burrs, insufficient curvature, and reverse curvature.

This application proposes a Bluetooth ranging method, control device, and ranging equipment, relating to the field of communications technology. The Bluetooth ranging method includes setting a first Bluetooth as a reference position Bluetooth, setting the first end of the position to be measured as a reference point, and calibrating the relative spatial position and relative spatial distance between the first Bluetooth and the second Bluetooth; after the relative spatial position and relative spatial distance calibration are completed, installing the second Bluetooth to the second end of the position to be measured, and controlling the second Bluetooth to send a ranging request signal to the first Bluetooth; based on the ranging request signal, controlling the first Bluetooth to receive and generate a signal transmission time; and obtaining a first distance corresponding to the signal transmission time based on the signal transmission time and a preset time-distance mapping relationship. This application can achieve high-precision and stable indoor ranging.

The invention discloses a two-point voltage sampling type inductance distance measuring sensor and a control method. According to the method, voltage sampling is carried out at two key time points, the equivalent resistance and the equivalent inductance of an inductance measuring head are calculated based on the sampled voltage, and then matching with a pre-stored high-precision inductance-distance mapping table is carried out through an error minimization algorithm, so that the actual distance of a measured object is obtained. The sensor comprises an inductance distance measuring circuit, a time sequence control module, a signal acquisition module and a data processing module which cooperatively realize high-precision distance measurement. Through time sequence innovation and intelligent calibration, the accuracy and robustness of a distance measurement result are remarkably improved, and the method is particularly suitable for precise positioning and mechanical detection scenes in industrial automation.

The invention provides an old man getting-up safety magnetic attraction alarm system based on pressure induction and delay triggering, and relates to the technical field of monitoring alarm, and the system comprises a serial number processing module which obtains the track of a monitored person; the active state detection and threshold comparison module is used for setting an initial delay trigger period and outputting a judgment result of whether the independent area triggers reset alarm or not; an alarm state resetting and updating module; the spatial distance mapping and extension multiplying power calculation module is used for calculating an alarm extension multiplying power for reflecting the time required by the monitored person to return to the pressure sensing point; and the dynamic delay synthesis triggering module is used for outputting alarm triggering final time. According to the invention, intelligent trajectory tracking and self-adaptive risk response capabilities which cannot be achieved by a conventional system are realized, and the system not only can reset alarm countdown every time an old person enters a new room, but also can automatically prolong a safety time window according to the distance between the room and a getting-up point; and the passing difficulty caused by the ground material and the barrier density is additionally fused.

The invention discloses an age estimation network and method based on an elastic law and a non-uniformly distributed X-ray human skeleton image, and mainly solves the problem of insufficient optimization effect on a sample distribution imbalance task in medical image age estimation in the prior art. The network construction comprises the following steps: establishing a feature extractor for extracting batch features from an input image; introducing a feature queue, and using supervision force, distance elasticity and midpoint elasticity to do work on the samples so as to carry out comparative learning between the samples; defining the basic age estimation loss by using the supervision force, and obtaining the total loss of network iteration training; establishing a monotonicity constraint residual block for mapping the sample pair feature distance into a partial monotone increasing distance feature between samples; and establishing mutually independent age estimation regression heads for outputting an estimation result. According to the method, the performance of the automatic age estimation model on few sample data is improved, and the method has a good aging change characterization effect and can be used for disease risk assessment, individualized health management and medical research.

This invention belongs to the field of quantitative loading technology. It proposes a quantitative loading assistance method and system based on radar scanning and deep learning. The method involves synchronously acquiring the original point cloud data of the vehicle compartment using forward and backward scanning radars arranged along the vehicle's movement direction. After denoising, coarse registration, and ICP fine registration, a high-precision registered point cloud is generated. The point cloud is projected onto the Y-O-Z plane and enhanced, then input into a deep learning model to identify the pixel coordinates of the rear baffle, front baffle, and unloading port. Combined with the actual dimensions of the vehicle compartment, a pixel-to-actual distance mapping coefficient is calculated. The vehicle's movement distance is calculated by weighting the point cloud coordinate displacement and image pixel displacement. Furthermore, based on the vehicle compartment's geometry, dynamic decay of loading progress, and loading deviation, a dynamic correction coefficient for quantitative loading is constructed, and the unloading speed and vehicle movement speed are adjusted in real time accordingly. This achieves multi-vehicle adaptive, high-precision, and fully automated quantitative loading.

This invention relates to the field of image defect detection technology, specifically to a semiconductor wafer defect detection method and system based on machine vision. The method includes acquiring an original wafer image and preprocessing it; generating candidate regions for suspected defects using the Canny edge detection algorithm based on the preprocessed wafer image; and extracting wafer surface feature vectors based on the candidate regions by constructing a gray-level co-occurrence matrix (GLCM). A reflection compensation mechanism is introduced during the GLCM construction process to reduce the interference of light spots on the image through a reflection compensation module. This invention parameterizes the edge blurring characteristics of different process layers and constructs an adaptive gradient kernel and optimal pixel distance mapping mechanism accordingly, enabling Canny edge detection and GLCM texture analysis to dynamically adapt to surface morphology differences caused by various manufacturing processes.

The application discloses a kind of based on graph manifold learning's literature relationship discovery method and system, comprising: according to literature search result, obtain first target literature and the neighbor literature of first target literature;With literature search result as node set, by obtaining the first reference relationship between first target literature and neighbor literature as edge set, and according to the same first feature label of first target literature and neighbor literature, construct attribute graph;Based on graph epidemic learning technology, obtain the graph geodetic distance and graph geodetic similarity of attribute graph, by mapping graph geodetic distance to graph geodetic similarity, and with Briggman divergence as loss function, construct graph convolution network model, for by identifying the mapping relationship between first target literature and neighbor literature, obtain second target literature with the same mapping relationship with neighbor literature, supplement to literature search result, provide more detailed and more closely related literature search result for user.

The application provides a microscope fundus surgery image position adjustment method and system, and relates to the technical field of computers, and comprises the following steps: acquiring an original surgery video image sequence collected by a camera in real time; performing surgery scene classification according to the original surgery video image sequence to obtain a judgment signal; performing action region detection processing according to the judgment signal to obtain preliminary action region position information; performing region extraction according to the preliminary action region position information to obtain circumscribed rectangle position information of a surgery region contour; performing adjustment judgment according to the circumscribed rectangle position information to obtain an adjustment direction instruction; and performing position mapping and output according to the adjustment direction instruction, wherein the direction and distance of moving the surgery region to the center in the image plane are mapped into parameters for deflecting an optical reflection element on a two-dimensional plane to obtain adjustment parameters for driving the optical element. The application realizes automatic tracking and centering of the surgery region in the camera imaging picture.

The invention discloses a pure machine vision distance measurement method and system, and relates to the technical field of environmental perception. The method comprises the following steps: a calibration stage: a plurality of concentric arc marks are preset on the flat ground, the marks take the projection of a preset reference point on the ground as the circle center, and the radius is the known distance; a first camera and a second camera are installed on a support with pitch angles capable of being synchronously adjusted at a fixed non-zero relative pitch angle difference, calibration image pairs are synchronously collected at a plurality of preset overall pitch angles, center lines of arc marks in the images are extracted, and a pixel-distance mapping relation pair library in one-to-one correspondence with the overall pitch angles is established. In the distance measuring stage, a first camera and a second camera are fixedly installed on a carrier according to the fixed non-zero relative pitch angle difference, a to-be-measured image pair is collected in real time, the contour of a target object is detected, all preset overall pitch angles are traversed, and a first mapping relation and a second mapping relation under the corresponding overall pitch angles are obtained from a mapping library respectively; and querying the minimum distance of the contour pixels to obtain a first candidate distance and a second candidate distance, selecting the overall pitch angle of the target according to the consistency of the first candidate distance and the second candidate distance, and determining a final distance measurement result based on the two candidate distances corresponding to the overall pitch angle of the target. Through double-view multi-overall pitch angle calibration and consistency matching, the method is adaptive to complex three-dimensional terrains, does not need camera internal reference, does not depend on target priori, does not need any active sensor, and realizes high-precision, low-cost and low-calculation-amount real-time distance measurement.

This invention belongs to the field of deep learning and computer vision, and discloses an adaptive feature fusion method and apparatus based on optimal transmission. The method includes: acquiring a first image feature map and a second image feature map to be fused; flattening the two image feature maps in spatial dimensions to obtain a first feature vector set and a second feature vector set; constructing a transmission cost matrix based on the pairing distance between feature vectors in the two feature vector sets; iteratively optimizing the transmission cost matrix until an optimal transmission scheme matrix is ​​obtained; obtaining the Wasserstein distance through the transmission cost matrix and the optimal transmission scheme matrix, and mapping the Wasserstein distance to adaptive fusion gating weights; and fusing the image feature maps using the adaptive fusion gating weights. This invention, based on optimal transmission theory, provides an adaptive feature fusion method to achieve a principled, distribution-aware, and more intelligent feature fusion mechanism, improving the accuracy of the fused image feature map.