865 results about "Image registration" patented technology

The invention discloses a microscopic image seamless splicing and enhanced reconstruction method based on deep learning, and the method comprises the following steps: S1, collecting a plurality of original images with overlapped regions, and recording the spatial position information and imaging parameters of the original images; s2, preprocessing the original image to generate a standardized image sequence; s3, inputting the standardized image into a structure perception feature extraction network, and extracting a feature map fusing textures and structures; s4, inputting the feature image and the original image into an image registration module; s5, inputting the registration image into the boundary attention splicing network; s6, inputting the seamless image into the residual hierarchy reconstruction network, and enhancing image details through hole convolution and multi-scale branches; s7, image quality evaluation is executed, and the structural similarity, the signal-to-noise ratio and the edge retention rate are calculated; and S8, constructing a training set and carrying out end-to-end training optimization based on a joint loss function. According to the method, a multi-module deep network is fused, and seamless splicing and high-quality enhanced reconstruction of microscopic images are realized.

The invention discloses a small target recognition method and system for a multi-modal fusion image in a complex environment, and belongs to the technical field of computer vision and image recognition, and the method comprises the steps: obtaining a visible light image, an infrared image and environment sensor data; image registration is carried out on visible light and infrared images, and a multi-scale image feature pyramid is constructed. And respectively extracting visible light and infrared image features to obtain visible light and infrared imaging feature data. And performing multi-modal data fusion on the visible light and infrared imaging feature data based on a cross-modal attention mechanism, and adaptively adjusting a fusion weight based on environmental sensor data to generate fusion features. And performing space-time enhancement processing on the fusion feature to obtain an enhanced fusion feature. And performing target tracking detection on the small target, and outputting position and category information of the small target. According to the method, the small target recognition capability in a severe environment is remarkably improved, and high precision and robustness can still be kept in a foggy, low-visibility and dark scene.

The invention discloses an unmanned aerial vehicle infrared thermal imaging method and device for cold leakage detection of a low-temperature storage tank, and relates to the field of automatic inspection and detection of storage tanks, and the method comprises the steps: determining the inspection range and detection distance of an unmanned aerial vehicle based on geometric parameters, a cold leakage frequency region and environment parameters of a site; calculating the distance between the shooting points and the vertical coverage height of single-circle flight, further dividing flight elevation layering, and calculating the number of the shooting points of each circle of flight; establishing a three-dimensional model; collecting a visible light image and an infrared thermal imaging image of each shooting point; the method comprises the following steps: performing multi-dimensional correction and temperature image conversion on an acquired infrared thermal imaging image, performing image registration on a temperature image with a temperature scale and a visible light image, performing cold leakage area identification to obtain a cold leakage area, performing quantitative calculation to obtain three-dimensional positioning of a cold leakage position and a cold leakage area, and performing cold leakage area identification on the cold leakage area. And generating a visual detection result, a detection report and a maintenance suggestion. According to the invention, automatic detection and accurate analysis of the cold leakage area of the low-temperature storage tank can be realized.

The invention discloses an ultra-thin glass defect detection system and positioning method based on line scanning and phase deflection, and the system comprises a line scanning camera module, a distortionless phase deflection detection module, a high-magnification microscope module and a central processing unit. The line scanning camera module is used for collecting a 2D grayscale image of the ultrathin glass; the distortionless phase deflection technology detection module is used for collecting a surface height map of the ultra-thin glass; the high-magnification microscope module is used for carrying out depth positioning on the interlayer defect judged by the central processing unit; the central processing unit is used for receiving the 2D gray level image and the surface height map, executing an image registration and defect classification algorithm and sending a depth positioning control instruction to the high-magnification microscope module, a multi-module cooperative work detection system is constructed, the detection efficiency and the detection precision are both considered, and data support is provided for production process improvement.

The invention provides an optical and acoustic image registration method for underwater structure crack detection, and the method comprises the steps: collecting a multi-mode image through the vision field and time synchronization, building a one-to-one correspondence relation according to the pose and time metadata, and improving the image quality through the denoising, distortion correction and other means; optical and acoustic multi-scale features are respectively extracted through a dual-channel network, and cross-modal semantic alignment is realized in combination with a shared weight and a channel attention mechanism; generating a high-resolution scale offset field by using local cross-correlation and micro-upsampling, and introducing manifold regularization constraint to ensure that a vector field is smooth and continuous; according to the method, the local scale field and the affine parameters are combined, non-rigid space mapping is achieved through thin-plate spline interpolation, feedback iterative optimization based on edge structure consistency is assisted, the registration precision of a key structure is enhanced, the automatic registration effect of the underwater multi-modal image is improved, and the method has high robustness and practical application value.

The invention relates to the technical field of industrial vision, in particular to an aviation part crack detection and repair method which comprises the following steps: acquiring an aviation part optical image at a reference time point and an aviation part optical image at a to-be-detected time point; according to the method, logarithmic polar coordinate transformation is carried out on different time point images, rotation, scaling and translation parameters are extracted, an image registration relation is established, the structural consistency of time sequence images in a local area is enhanced, a structural tensor is constructed for each pixel neighborhood, the change characteristics of the dual-time-phase tensor are compared, a structural change saliency map is generated, and the structural change saliency map is obtained. Sensitive capture of a tiny deformation area is achieved, the responsiveness to an initial crack is improved, then a crack propagation interval is further refined into a main crack path in a self-adaptive threshold segmentation and skeleton extraction mode, the tip acutance of the crack is calculated in combination with tip contour information of a geometric boundary of the crack, and the initial crack is obtained. And quantitative support is provided for the crack danger degree.

The invention relates to the technical field of bionic navigation orientation and positioning, in particular to a carrier course angle measuring method based on polarized light imaging in a non-Rayleigh atmosphere. Comprising the steps of real-time sky polarized light imaging, atmospheric polarization mode resolving based on Mueller matrix imaging, non-Rayleigh polarization information interference removal, polarization orientation resolving based on Rayleigh scattering, space projection conversion and sun position information resolving. Alignment information of the sun in a carrier coordinate system is obtained by utilizing a Rayleigh scattering model and actual sky polarization imaging, and then a carrier course angle is obtained according to the orientation of the sun in a local geographic coordinate system, so that an orientation function is realized; meanwhile, the error influence caused by image registration and time-sharing rotation measurement of the multi-view-field polarization detector and the interference of environmental factors on a single point are avoided, the method can adapt to sunny and cloudy weather, is also applicable to cloudy and other non-Rayleigh weather, and is wide in application range.

The invention discloses a real-time pose updating method based on feature prediction and point cloud registration, and belongs to the technical field of image registration, and the method comprises the steps: extracting a bone surface point cloud which at least covers a feature region and an adjacent region from a three-dimensional model of a bone tissue of a patient, the point cloud is used as a target point cloud for registration with a real-time ultrasonic image; an ultrasonic image collected by the ultrasonic patch in real time is obtained, and a two-dimensional ultrasonic image with a spatial position is obtained; predicting the two-dimensional ultrasonic image by using a pre-trained artificial intelligence model to obtain bone-soft tissue interface features, and extracting a plurality of points on a bone-soft tissue interface from the bone-soft tissue interface to form a source point cloud for registration; and registering the source point cloud to the target point cloud to realize real-time pose updating. According to the invention, the pose monitoring in the operation can be realized through the real-time image flow under the non-invasive condition.

The invention discloses an island shoreline erosion monitoring method and system based on image recognition, and the method comprises the steps: processing an optical image through meteorological data, obtaining a clear optical image, carrying out the multi-source feature fusion of the clear optical image and an SAR image, and obtaining a preliminary shoreline probability graph; carrying out fine extraction and image recognition on the initial shoreline probability graph according to the instantaneous tide level to obtain an island shoreline planar graph, carrying out image registration and constructing an island shoreline digital surface model, calculating a three-dimensional deviation field and quantifying an erosion index, calculating a driving factor, and determining a dynamic monitoring threshold according to the driving factor and a standard monitoring threshold. And carrying out island shoreline erosion early warning according to the erosion index and the dynamic monitoring threshold. The method not only can improve the efficiency and accuracy of island shoreline erosion monitoring, but also has good interpretability, and can be directly applied to an island shoreline erosion monitoring system.

The invention provides a pulmonary nodule display method and device, electronic equipment and a storage medium, and the method comprises the steps: segmenting a three-dimensional reconstruction image of the chest of a target patient to obtain a preoperative segmentation image, the three-dimensional reconstruction image being obtained based on preoperative CT data, and the preoperative segmentation image comprising the position information of a pulmonary nodule; segmenting the video image of the intraoperative lung tissue of the target patient collected by the thoracoscope in real time to obtain an intraoperative segmented image; feature matching is conducted on the preoperative segmented image and the intra-operative segmented image, the lens pose of the thoracoscope is determined, the preoperative segmented image is mapped into a two-dimensional reference image based on the lens pose, and the two-dimensional reference image comprises the position information of the pulmonary nodule; and carrying out image registration on the two-dimensional reference image and the intra-operative segmented image, and carrying out pulmonary nodule marking on the intra-operative segmented image to obtain a video image for displaying pulmonary nodules in real time. The position of the pulmonary nodule in the operation is accurately displayed in real time in a non-invasive mode, and the operation efficiency is improved.

The invention discloses a star map registration method based on a remote optical image, relates to the technical field of image processing, and solves the problems that the existing method is multi-oriented to a single telescope, most of the single telescope is aligned at a star point level, and the robustness is insufficient when parallax, distortion, optical difference and background pollution exist in remote multi-station imaging. The method comprises the following steps: acquiring a remote star map and extracting a star point centroid; angular distance calculation and triangle construction; registering the centroids of the star points in the whole image; performing similarity transformation and homography estimation; and carrying out sub-pixel refinement and full image registration. According to the invention, dual registration of star point centroids and pixel coordinates can be realized for star maps shot by a plurality of telescopes in different places, and meanwhile, a high-quality registration reference can be provided for subsequent three-dimensional information acquisition of a space target and identification and positioning of the space target.

The invention relates to the technical field of gas detection, and relates to an optimization method and system for SF6 gas detection in a multi-temperature environment, and the optimization method comprises the steps: obtaining dual-band infrared images of a scene to be detected at the same time, the dual-band infrared images comprising a detection band infrared image and a reference band infrared image; preprocessing the dual-band infrared image; performing registration processing on the preprocessed dual-band infrared image; performing differential operation on the registered dual-band infrared image to obtain a differential image; and determining a dynamic segmentation threshold according to the current environment temperature, and performing threshold segmentation on the differential image according to the dynamic segmentation threshold so as to identify the SF6 gas leakage area. According to the optimization method and the optimization system, the dynamic segmentation threshold value is determined according to the current environment temperature, the problem that the detection performance of a fixed threshold value is reduced due to background radiation changes at different environment temperatures can be solved, and stable and reliable detection sensitivity can be kept within a wide temperature range.

The invention provides a remote sensing image change detection method based on a CNN-Mama hybrid network, and relates to the technical field of image detection, and the method comprises the following steps: carrying out the preprocessing of a disclosed remote sensing image change detection data set, and constructing a dual-temporal high-resolution remote sensing image training set and a dual-temporal high-resolution remote sensing image test set; the preprocessing comprises image registration, cutting, normalization and data enhancement; constructing a change detection network model based on a CNN-Mama trunk, and performing training on the training set to obtain an optimal model; the network model comprises a twin encoder, a dynamic grouping attention module, a space adaptive fusion module, a multi-scale edge enhancement module and a decoder; and inputting a dual-temporal remote sensing image of a to-be-detected area into the optimal model, and outputting a binary change detection image with the same size as the input image through feature extraction, feature interaction, cross-layer fusion and step-by-step decoding. According to the method, the convolutional neural network and the state space modeling network are combined, and the local feature capture capability of the CNN and the long-range dependence modeling characteristic complementation of Mama are combined.

This disclosure describes an interactive augmented reality system for improving surgeon's view and context awareness during laparoscopic and video assisted surgeries. Instead of purely relying on computer vision algorithms for image registration between pre-operation (or intra-operation) images / models and later intra-operation scope images, the system can implement an interactive mechanism where surgeons may provide supervised information in initial calibration phase of the augmented reality function, thus achieving high accuracy in image registration. Besides the initialization phase before operation starts, interaction between surgeon and the system can also happens during the surgery. Specifically, patient tissue might move or deform during surgery, caused by for example cutting. The augmented reality system can re-calibrate during surgery when image registration accuracy deteriorates, by seeking additional supervised labeling from surgeons. The augmented reality system can improve surgeon's view during surgery, by utilizing surgeon's guidance sporadically to achieve high image registration accuracy.

The invention discloses a lake cyanobacterial bloom intelligent early warning method based on unmanned aerial vehicle remote sensing and image recognition, and the method comprises the steps: fusing an unmanned aerial vehicle multispectral image and a convolutional neural network, extracting the spectral features of cyanobacteria, and calculating the concentration distribution; image registration and a dynamic model are combined to track water bloom boundary change and drift trajectory, satellite remote sensing is used to verify precision and invert biomass density, early warning levels are divided according to the precision and the biomass density, decision information is generated, and intelligent monitoring and accurate early warning of cyanobacterial bloom are realized. The comprehensive technical effects of water bloom dynamic monitoring, accurate early warning and efficient management are achieved, and a scientific basis is provided for water environment treatment.

The invention provides a brain glioma CT-MRI (Computed Tomography-Magnetic Resonance Imaging) multi-modal fusion intelligent grading method and a brain glioma CT-MRI multi-modal fusion intelligent grading system, and relates to the field of medical image processing and intelligent diagnosis, and the method comprises the following steps: obtaining CT image and MRI image data of a brain glioma patient through a medical image acquisition device, carrying out standardized preprocessing on the CT image and the MRI image data, and generating a preprocessed image data set; and spatial alignment is carried out on the preprocessed image data set, CT-MRI registration image data are output, and a multi-scale registration method is adopted for spatial alignment. According to the CT-MRI multi-modal fusion intelligent grading method and system based on the brain glioma, by providing the CT-MRI multi-modal fusion intelligent grading method, the problem that in the prior art, image registration and feature fusion are not accurate is solved. A multi-scale registration method and a weighted combined feature extraction mode are adopted, accurate registration of CT and MRI images and efficient fusion of features are ensured, and therefore the accuracy and stability of a grading model are improved.

The invention discloses an icing monitoring method and system for a corner reflector deployed based on an unmanned aerial vehicle, and relates to the technical field of environment monitoring, and the method comprises the steps: deploying a corner reflector network, collecting radar cross-section characteristics and environment parameters to generate a data packet, and carrying out the icing monitoring based on the generated data packet; calculating a signal-to-clutter ratio, obtaining a standard deviation of target positioning and phase estimation by using a Cramer-Rao lower bound method, generating a dynamic weight by using the standard deviation, obtaining a corner reflector icing thickness estimation value through data fusion, and generating an optimized signal-to-clutter ratio based on the icing thickness estimation value; and carrying out cooperative judgment by utilizing the generated optimized signal-to-clutter ratio and combining the icing thickness and the attitude deviation, triggering combined adjustment based on a judgment result, and after the combined adjustment is triggered, obtaining the SAR image again and carrying out main and auxiliary image registration and interference pair optimization. According to the invention, the sensing precision and the monitoring robustness of the unmanned aerial vehicle on the state of the corner reflector in an icing environment are improved.

The invention belongs to the technical field of image processing, and provides a rock FIB-SEM sequence image multi-phase registration segmentation method and system, and the technical scheme of the method comprises the steps: outputting a registered image, a displacement vector field and a Jacobian thereof through affine correction and non-rigid registration based on a constructed detail-retaining non-rigid registration sub-network; segmenting the registered image based on the constructed deformation sensing type multi-phase segmentation sub-network to obtain a probability graph of a plurality of channels; performing joint training on the sub-networks by taking registration precision and segmentation accuracy as targets and combining the constructed loss function to obtain each sub-network after parameter optimization; and processing the target sequence image based on each sub-network after parameter optimization to obtain three-dimensional image volume data and a multi-phase semantic segmentation three-dimensional tag aligned with the three-dimensional image volume data. According to the method, high-precision image registration and semantic segmentation can be synchronously completed on the premise that high-frequency details and noise features with geological significance in an original image are reserved.

The invention provides a medical image registration method and equipment. The method is applied to the technical field of medical image processing. The method comprises the steps that a to-be-registered image pair used for medical image registration is acquired, the to-be-registered image pair comprises a floating image and a fixed image, the floating image is an image needing spatial transformation in the registration operation, and the fixed image is a reference standard correspondingly consistent with the spatial position and feature of the floating image in the registration operation; performing hierarchical feature extraction on the floating image and the fixed image through a registration model to obtain a first multi-scale feature pyramid and a second multi-scale feature pyramid, and performing correlation perception registration on multi-scale features in the first multi-scale feature pyramid and the second multi-scale feature pyramid to obtain deformation field data; and performing spatial transformation on the floating image based on the deformation field data to obtain a target registration image. According to the invention, efficient and accurate medical image registration is realized.

The invention relates to the field of medical image processing and artificial intelligence, in particular to a pulmonary nodule treatment effect AI evaluation system which comprises an image acquisition module, an image registration module, a feature representation module, a multi-scale analysis module, a trajectory analysis module, a response prediction module and a decision support module. According to the system, accurate alignment of CT images before and after treatment is realized through a 4D registration technology, manifold representation of a pulmonary nodule state is constructed based on a differential geometry theory, and nodule features are mapped into points on a high-dimensional manifold; extracting features of different time and space scales by adopting multi-scale space-time analysis, and constructing a manifold trajectory representing a treatment response process; geodesic prediction is realized by using a Riemann geometric framework, and long-term curative effect is predicted from early treatment response; the system not only evaluates the current treatment effect, but also can provide personalized treatment suggestions and optimal follow-up visit plans.

The invention belongs to the technical field of medical robots and minimally invasive surgery, and provides a robot control system and method for blue laser vaporization surgery of prostatic hyperplasia. Mapping the nuclear magnetic image volume data to a deformation field under an ultrasonic acquisition coordinate system, and deforming the preoperative nuclear magnetic image to an intra-operative ultrasonic space by using the deformation field to complete image registration; fusing the registered image with a stereoscopic vision system; the spatial depth of the surface of the target tissue is obtained from the endoscopic image so as to supplement navigation information; according to the utility model, the prostate deformation and probe posture change adaptive capacity in an operation is improved, the real-time visual closed-loop regulation and control capacity is improved, and the characteristics of small light spots, shallow heat diffusion, excellent hemostasis and the like of blue laser are combined, so that the vaporization and hemostasis precision in a tiny blood vessel dense area is ensured, and the problems of large tissue trauma and the like are avoided.

The invention belongs to the technical field of medical image processing, and particularly relates to a prostate MRI-TRUS deformable image registration method based on structure perception decoupling learning. The method comprises the following steps: designing an anatomical maintenance intensity disturbance module, simulating intensity and artifact differences between different modes on the premise of keeping an anatomical structure unchanged, and generating diversified appearance samples to improve the adaptability of the model to mode changes; a double-flow encoder structure is constructed, space attention latent consistency loss is introduced into a multi-layer feature space, the structure consistency is restrained from the feature level, and deep structure representation learning with the unchanged appearance is achieved; and an enhanced pyramid decoder is adopted to fuse multi-scale structural features layer by layer to predict a registration deformation field, so that high-precision cross-modal alignment is realized. According to the method, structural consistency constraint is carried out on the hidden space level, so that the influence of modal difference and artifacts can be effectively reduced, and the registration precision is improved.

Disclosed is a computer-implemented method which encompasses registering a tracked imaging device such as a microscope having a known viewing direction and an atlas to a patient space so that a transformation can be established between the atlas space and the reference system for defining positions in images of an anatomical structure of the patient. Labels are associated with certain constituents of the images and are input into a learning algorithm such as a machine learning algorithm, for example a convolutional neural network, together with the medical images and an anatomical vector and for example also the atlas to train the learning algorithm for automatic segmentation of patient images generated with the tracked imaging device. The trained learning algorithm then allows for efficient segmentation and / or labelling of patient images without having to register the patient images to the atlas each time, thereby saving on computational effort.

The invention relates to the technical field of medical images, in particular to a multi-modal medical image registration and fusion analysis method, which comprises the following steps of: eliminating image noise and artifacts based on a modal adaptive filtering strategy; constructing a pyramid type feature extraction network to realize multi-scale feature extraction, calculating feature matching degrees among different modal images, and dynamically adjusting matching weights by combining feature differences among modals; a focus area attention mask is constructed, targeted enhancement of registration image features is realized, and a hierarchical fusion strategy is adopted to evaluate the quality of a fused image; and constructing a multi-task deep learning model to complete focus automatic detection, segmentation and benign and malignant preliminary judgment on the fused image. According to the multi-modal medical image registration and fusion analysis method, a three-layer feature pyramid is constructed, a multi-feature fusion matching cost function is introduced, and cross-modal feature matching is optimized through an adaptive weight iteration nearest point ICP algorithm, so that the information richness, marginal definition and focus discrimination of a fused image reach the standard.

The invention relates to a scatter correction method, a cone beam computed tomography (CBCT) system and a computer storage medium. The method comprises the following steps: performing cone beam CT imaging on a to-be-measured object to obtain a first group of projection data of the to-be-measured object, and performing reconstruction according to the first group of projection data to obtain a first volume image containing scattering artifacts; performing fan-beam CT imaging on a plurality of different heights of the to-be-measured object to obtain a plurality of groups of fan-beam projection data of the to-be-measured object, performing reconstruction according to the plurality of groups of fan-beam projection data to obtain a plurality of fan-beam slice images, and using the plurality of fan-beam slice images as reference images without scattering artifacts; extracting a plurality of cone beam slice images from the first volume image; performing image registration on the plurality of fan-beam slice images and the plurality of cone-beam slice images; and training a scattering correction model based on deep learning by using the plurality of registered image pairs to obtain a trained scattering correction model.

The invention is suitable for the technical field of disaster monitoring and early warning, and provides a meteorological disaster monitoring and early warning method based on multi-source remote sensing data, and the method comprises the steps: obtaining the multi-source remote sensing image data of a target monitoring region; performing image registration and fusion processing on the multi-source remote sensing image data to obtain a registered multi-modal remote sensing image; identifying and extracting a potential meteorological disaster target area from the registered multi-modal remote sensing image based on a preset disaster target extraction model, and performing hierarchical locking identification; extracting disaster characteristic parameters of the potential meteorological disaster target area, and performing comprehensive assessment based on a preset disaster assessment model to generate a disaster risk level assessment result; meteorological disaster early warning information is generated according to the disaster risk level assessment result, and early warning is issued; the meteorological disaster monitoring accuracy is effectively improved, and the emergency response timeliness is greatly enhanced.

The invention relates to the technical field of automatic visual detection, and discloses a bottle body label quality detection-based model automatic generation method, which comprises the following steps of: acquiring a multi-angle image of a bottle body, carrying out denoising, enhancement and brightness equalization processing, extracting a curved surface label through target detection and a perspective transformation algorithm, and carrying out image processing on the curved surface label; further generating a distortionless complete label image through an image registration and fusion technology, and taking the distortionless complete label image as a training sample set of a label defect detection model constructed based on a double-branch deep learning network; and performing hot updating on the label defect detection model through a closed-loop feedback and continuous learning mechanism. According to the method, the technical problems of imaging deformation of the curved surface label, diverse defects, difficulty in detection and performance degradation after model deployment are effectively solved, and high-precision, high-robustness and sustainable-evolution automatic label quality detection is realized.

The invention discloses a crowd density detection method fusing optical flow and texture features, and relates to the technical field of computer vision, and the method comprises the following steps: collecting a real-time video stream of a camera, and carrying out graying, Gaussian filtering and perspective correction preprocessing; performing motion compensation by using image registration and offset transformation; modeling based on a Gaussian mixture model and extracting a foreground to generate a binary mask; analyzing a foreground coverage rate, an optical flow and texture features; pre-defining a multi-ROI and a density early warning standard; inputting the fusion features into a regression model and outputting initial density; dynamically calibrating and correcting the deviation; and generating a thermodynamic diagram superposition video to realize visualization. According to the invention, the optical flow and texture features are fused to improve density estimation precision, illumination resistance and dynamic background interference resistance; the dynamic calibration maintains long-term accuracy, and the dynamic ROI adapts to scene change; the thermodynamic diagram can quickly identify risks, is adaptive to multiple scenes, and meets real-time monitoring requirements.

The invention relates to the technical field of image registration, and provides a medical image registration method and system based on a diffusion model and a density guidance mechanism, and the method comprises the steps: obtaining a fixed image and a moving image as input, inputting the fixed image and the moving image into a trained registration network of an encoder-decoder structure, and carrying out the multi-scale feature extraction and fusion, preliminary registration features are obtained; generating a preliminary deformation field by a decoder, and generating a local density map based on the region registration difficulty; according to the density map, non-uniform down-sampling and up-sampling are carried out on the image, fusion features are extracted, and a final deformation field is generated; and transforming the moving image to a fixed image space by using the final deformation field to generate a registration image. According to the invention, by introducing the density guidance mechanism and the diffusion model, the problems of lack of uncertainty expression of deformation modeling and uneven distribution of computing resources in different areas in existing medical image registration are solved, and thus the rationality of registration and the precision of key areas are improved.

The invention relates to the technical field of image processing, in particular to an attention-based registration method and device for medical image registration, and the method comprises the steps: obtaining a fixed image and a moving image, and carrying out the preprocessing of the fixed image and the moving image, and splicing the images into a dual-channel input; extracting and fusing multi-scale features through a multi-scale feature stacking block in the lightweight feature extraction network; performing channel and space double attention enhancement on the fusion features; based on the enhanced features, a three-dimensional deformation field is generated through a decoder under unsupervised loss constraints; and transforming the moving image by using the deformation field to realize alignment with the fixed image. According to the method, a lightweight multi-scale feature stacking structure and a double attention mechanism are introduced, so that the high-precision image alignment capability is maintained, the model parameter quantity and the calculation complexity are remarkably reduced, the applicability and the reasoning efficiency on resource-constrained equipment are improved, and rapid, stable and high-quality medical image registration is realized.