143 results about "Rigid transformation" patented technology

The invention discloses a slope cutting house building deformation monitoring method based on unmanned aerial vehicle remote sensing, and the method comprises the steps: carrying out the periodic observation of an artificial slope through an unmanned aerial vehicle carrying a three-dimensional laser scanner, and obtaining slope three-dimensional point clouds in at least two different periods; performing denoising, down-sampling and coordinate normalization on the point cloud of each period to obtain a preprocessed data set; generating a global feature descriptor of each period by using a PointNet feature extraction unit; on the basis of a Lucas-Kanade iterative optimization algorithm, the descriptors of all periods are matched with the first period as the reference, and optimal rigid body transformation parameters are obtained; completing registration of the cross-period point clouds in a unified coordinate system according to the parameters; calculating displacement by adopting point-by-point difference and forming a differential deformation field; and comparing the displacement with a preset safety threshold, positioning an abnormal area and giving a risk level. According to the method, centimeter order registration and deformation identification of the cross-period point cloud under the unified stable reference are realized, and early warning and hidden danger assessment of a slope cutting and house building scene are supported.

The invention relates to a ship berthing position detection method and system based on shore end laser radars, and the method comprises the steps: deploying a plurality of laser radars at a shore end, and collecting original point clouds in parallel through an asynchronous thread model; constructing a rigid transformation matrix based on the radar installation attitude parameters, and further generating fusion point cloud data under a unified world coordinate system; then extracting 3D geometric edge points in a preset range of the front edge of the wharf as near-shore point cloud data, and further calculating the minimum ship-shore distance; based on the ship bow direction and a preset segmentation threshold value, the coastal point cloud data are segmented into ship bow area point cloud and ship stern area point cloud in the x-axis direction, the coordinate mean value of the ship bow area point cloud and the ship stern area point cloud is calculated and compared with an expected position in a wharf operation system, and berthing accuracy is judged; and the drift distance is calculated in real time, the drift distance, the attitude angle and the change rate of the attitude angle are compared with corresponding preset threshold values respectively for graded early warning and drift alarm, and high-precision ship berthing position dynamic detection is achieved.

The invention relates to the technical field of production process control, in particular to an MES system management and control method and system based on precision mold production and manufacturing, and the method comprises the following steps: obtaining a measurement point cloud and a theoretical model, calculating rigid body transformation matrix registration to generate a registration point cloud, searching a nearest projection point, measuring geometric deviation, and constructing a three-dimensional deviation vector; and calculating the actual residual material thickness, screening and generating margin insufficient risk points, analyzing a numerical control code to extract a coordinate instruction, retrieving a nearest risk point, calculating a compensation distance, and adjusting coordinates to generate a self-adaptive finish machining program. According to the method, the three-dimensional deviation vector is constructed to represent the entity state, the risk area is screened in combination with the machining allowance, the compensation distance is calculated according to the actual residual thickness, and the instruction is dynamically adjusted to generate the self-adaptive program, so that the path is automatically corrected according to the shape of the workpiece, and the overcut risk caused by insufficient allowance is effectively avoided; and accurate matching between the machining process and the workpiece state is ensured.

The invention discloses a multi-modal image registration processing method and system based on deep learning, and relates to the technical field of medical image processing, and the method comprises the steps: calculating a mutual information loss value after registration through a mutual information loss method, optimizing a CNN model and UNet model parameters in combination with the total loss generated by the fusion of global and local deformation fields, and obtaining a registered sCT image. A CNN model and a UNet model are combined to generate global and local deformation fields, overall rigid transformation and local nonlinear deformation are effectively captured, the spatial alignment precision of sCT and reference CT is improved, model parameters are optimized through mutual information loss and total loss, the intensity distribution consistency is ensured, feature weights and treatment plan parameters are automatically adjusted through registration quality feedback, and the accuracy of the treatment plan is improved. The coverage precision and efficiency of the radiotherapy plan are remarkably improved, automatic and high-precision image registration and treatment plan optimization are realized, and the reliability and practicability of clinical image processing are enhanced.

The invention discloses a visual alignment method and device for printing draft intelligent deflection and spraying vehicle cooperative control, and belongs to the technical field of machine visual localization, and the method specifically comprises the steps: firstly, arranging a calibration target in an ink-jet printing area, employing an image collection device to collect an image, and calculating a rigid transformation matrix of a visual coordinate system and a printer spraying vehicle coordinate system; establishing high-precision coordinate mapping; the control terminal reads the reference position and attitude information of the original draft in the spraying vehicle coordinate system from the production management system; the method comprises the following steps: after primary inkjet of a paperboard, shooting the whole paperboard by an image acquisition device, and extracting actual contour features and coordinates of a draft through image processing; the coordinates are mapped to a spraying vehicle coordinate system through a transformation matrix, and a rotation deviation angle and a translation deviation vector are calculated; analyzing to obtain angle and translation compensation values, packaging the angle and translation compensation values into a control instruction, and sending the control instruction to ink-jet printing control software through a real-time bus; finally, software drives a servo executing mechanism, the position and the posture of the spraying vehicle are adjusted according to a new target, and accurate alignment with the paperboard is achieved.

The invention discloses an intelligent detection method and system for a flexible circuit board based on machine vision, and relates to the field of circuit board detection.The method comprises the steps that firstly, through a global coarse registration step, rigid transformation such as translation, rotation and zooming of the whole to-be-detected FPC image is rapidly corrected, and preliminary alignment with a standard template is achieved; then, local fine registration is introduced for nonlinear deformation such as local stretching and wrinkles caused by the characteristics of the FPC material, so that the fine and non-uniform local dislocation is accurately sensed and compensated, and nearly pixel-level accurate matching between the to-be-detected image and the template is realized. On the basis of high-precision alignment, differential attention deficit segmentation is carried out, so that false differences caused by inaccurate deformation registration can be effectively eliminated, real defects can be accurately identified, the false alarm rate is remarkably reduced, and the reliability and practicability of a detection scheme are greatly improved.

The invention belongs to the technical field of industrial robot motion control, and discloses an industrial robot precise path generation system based on three-dimensional visual reconstruction, comprising an environment sensing device which is composed of a multi-baseline structured light projector and a multi-view camera array and obtains point cloud data through light interference suppression and spectrum correction; the three-dimensional reconstruction processing unit generates a three-dimensional model through rigid transformation and deep learning correction; the geometric feature analysis unit performs curvature analysis and boundary recognition to form region description; the trajectory planning device adopts a weighted B spline to generate a trajectory under curvature constraint and energy optimization; the dynamic correction unit is switched between vision feedback and force feedback according to an error threshold value; the control conversion processor combines inverse kinematics and dynamics compensation and predicts deviation to output a control instruction; and the execution driving mechanism utilizes a servo motor and closed loop feedback to execute actions, so that high-precision generation and stable execution of a track under a complex working condition are realized.

The invention discloses a surgical navigation point cloud registration system based on global optimization, and the system comprises the steps: reconstructing a target point cloud based on a preoperative medical image, collecting an intraoperative anatomical surface source point cloud, and converting the point cloud to a patient reference coordinate system; the method comprises the following steps of: selecting a registration point pair with anatomical representativeness to carry out initial coarse registration; on the basis of the coarse registration, executing fine registration, including establishing a point pair relationship, calculating weighting based on a distance residual error and a neighborhood structure, filtering out abnormal point pairs by adopting a bidirectional pruning mechanism based on statistical analysis, jointly estimating rotation and translation parameters through a robust optimization algorithm, and iteratively updating to convergence; and finally outputting a rigid transformation matrix for surgical navigation. According to the method, stable and accurate point cloud registration can be realized under the conditions of sparse point cloud, noise and low overlapping rate in an operation, reliable space mapping support is provided for operation navigation under image guidance, and the method has relatively high robustness and clinical adaptability.

The invention discloses an ultrahigh-voltage power transmission line tower gradient calculation method and device and a medium, and the method comprises the steps: fusing a high-definition camera image with a laser ranging eyepiece image, achieving the quick alignment of a heterogenous coordinate system through a rigid transformation matrix, and precisely capturing the depth information of the feature points of a tower target; a global-local detection model is constructed through cooperative work of a high-definition camera and a laser distance measuring sensor, and calculation complexity is reduced by adopting an SE-R3det network; the device and the medium can realize rapid and accurate tower inclination automatic measurement based on the method. The measurement efficiency is improved, laser ranging is adopted, feedback control is introduced, and automation of the whole process is achieved while the measurement precision is improved.

The invention discloses a point cloud anomaly detection method based on registration guide rotation invariant feature learning, which comprises the following steps: acquiring a normal point cloud, applying random rigid transformation to generate a point cloud template, and performing multi-scale voxel downsampling; a feature extraction model is built and trained, a loss function is optimized based on a block matching pair and a point matching pair, and an optimal feature extraction model is obtained through training; registering and aligning the normal point cloud with the template, integrating the local rotation invariant features and the registered coordinates to obtain the final features of the normal point cloud, and establishing a normal feature memory bank; during detection, the test point cloud and the template are registered and aligned, the final features of the test point cloud are obtained and then are compared with the features in the normal feature memory bank, and abnormal scores are calculated to judge abnormal points. According to the method, the problem of detection deviation caused by rotation change and insufficient feature representation capability in the prior art can be solved, the accuracy and robustness of point cloud anomaly detection in industrial quality inspection are remarkably improved, and the method is suitable for automatic defect detection of a high-precision three-dimensional scanning piece.

The invention provides a calibration method, a calibration device, a tracking type scanning system and a storage medium, and the method comprises the steps: scanning a reference object on the calibration device through a scanner, and obtaining first scanning data corresponding to the reference object in a scanning coordinate system; acquiring the pose of the marker on the scanner and the pose of the calibration device by using a tracker; calibrating equipment parameters of the scanner based on the true value corresponding to the reference object and the first scanning data; based on the pose of the marker and the pose of the calibration device, converting the first scanning data to a tracking coordinate system of a tracker; and based on the true value corresponding to the reference object and the first scanning data under the tracking coordinate system, completing rigid body transformation relation calibration between the local coordinate system corresponding to the marker and the scanning coordinate system. Through the method, calibration errors can be reduced to improve measurement precision.

The invention discloses an air-ground multi-source point cloud robust registration method based on structure entropy sampling, and the method comprises the steps: S1, reading a source point cloud and a target point cloud, and obtaining the space coordinates, normal vectors and curvatures of the source point cloud and the target point cloud; s2, performing standardized preprocessing on the source point cloud and the target point cloud; s3, performing structure entropy sampling on the source point cloud and the target point cloud after standardization processing, and constructing a sampling point set for registration; s4, adopting a point-to-point registration method with robustness to calculate an optimal rigid transformation matrix between the source point cloud and the target point cloud obtained through structure entropy sampling in the step S3, and realizing point cloud registration; s5, performing spatial restoration on a registration result in the step S4, and constructing a rigid transformation matrix under an original coordinate; and S6, outputting the transformation matrix and the registration point cloud after spatial restoration. The method has the advantages that the key points are optimized through structural entropy driving sampling, the anti-noise capability is high, the registration precision is high, and the registration requirement of small-error space-ground fusion data can be met.

The invention discloses a dual-stage CBCT and oral scanning tooth registration method and system, and relates to the field of tooth three-dimensional digital model registration, and the method comprises the following steps: extracting a tooth voxel structure in a CBCT image and a dental crown grid structure in an oral scanning image through a segmentation network; a coarse-to-fine depth registration network is introduced, geometric feature coding and feature matching are carried out on super-points obtained through multi-resolution down-sampling, local matching is carried out on a dense point set in a super-point neighborhood, a rigid transformation matrix is calculated in parallel, and an optimal registration result is screened out from the rigid transformation matrix; and dividing a registration object into a plurality of three-tooth groups containing three adjacent teeth, and performing local iteration alignment based on an initial registration result. The method still has higher robustness and accuracy under the complex conditions of large cross-modal difference, high input point cloud density and the like, is shorter in time consumption, and provides reliable data basis and automatic support for clinical scenes such as orthodontics and implantation.

The invention provides an equipment calibration method, a tracking type scanning system and a storage medium, the equipment calibration method is applied to the tracking type scanning system, the tracking type scanning system comprises a tracker and a scanner, and the method comprises the following steps: scanning a reference object by using the scanner to obtain scanning data corresponding to the reference object; the tracker is used for tracking a marker on the scanner, the pose of the marker is obtained, and the relative position between the marker and the scanner is fixed; converting scanning data under a scanning coordinate system of the scanner into a tracking coordinate system of the tracker based on the pose of the marker; and completing rigid body transformation relation calibration between the local coordinate system corresponding to the marker and the scanning coordinate system based on the true value corresponding to the reference object and the scanning data under the tracking coordinate system. According to the method, the equipment calibration efficiency can be improved.

The invention discloses an IOS image and CBCT image registration method based on particle swarm and single-tooth optimization, and belongs to the field of image data processing, and the method comprises the following steps: respectively obtaining tooth semantic segmentation results and point cloud data of an IOS image and a CBCT image through preprocessing and segmentation steps; introducing a particle swarm optimization algorithm, and performing iterative search and optimization on six-degree-of-freedom rigid transformation parameters between the IOS image data and the CBCT image data so as to complete global initial registration of the dental arch scale; and on the basis of a result after global registration, a registration object is divided into a plurality of local units based on a single tooth, and local iteration alignment is performed on each unit, so that tiny dislocation on the single tooth is eliminated, and the overall registration effect is optimized. According to the method, on the basis of processing cross-modal difference and realizing global registration, the registration precision of a single tooth level is further ensured through a local optimization mechanism, and reliable automatic registration support is provided for clinical scenes such as orthodontics and implantation.

The invention relates to the technical field of image processing, in particular to a positioning method for lamination of multi-layer flexible materials based on machine vision, which comprises the following steps: acquiring images of a first flexible material to be laminated and a second flexible material serving as a reference; contour geometry and surface texture features are extracted in parallel for the image, and a first feature map and a second feature map with the features as nodes and the spatial topological relation and gradient mutual information as weighted edges are constructed; calculating an initial nonlinear deformation field of the first flexible material; utilizing the initial nonlinear deformation field guide graph matching algorithm to obtain rough matching feature point pairs; and further solving to obtain a final non-rigid transformation model, and generating positioning parameters including non-uniform scaling and local distortion to guide fitting operation. According to the invention, the quality and efficiency of the flexible material laminating process can be improved.

The invention discloses a deep learning point cloud registration method for low-overlapping-rate point clouds. The method comprises the following steps: (1) carrying out multi-scale down-sampling on an input point cloud, calculating a weighted geometric centroid and a normal vector based on an adaptive neighborhood, and constructing a stable local reference point; (2) proposing a centroid geometric structure code, and embedding an attention mechanism by taking an inter-centroid Euclidean distance, a normal included angle and a neighborhood radius ratio as geometric priori to realize feature matching of structure constraint; (3) in a training stage, optimizing overlapping region positioning and point pair matching consistency by adopting a joint loss function formed by circular ring feature loss, overlapping perception loss and matchability loss; and (4) rigid body transformation is estimated by combining a coarse-to-fine matching strategy with a random sampling consistency algorithm, and global high-precision alignment is realized. According to the method, the registration precision and robustness under the conditions of low overlapping rate, noise and shielding are effectively improved, and the method is superior to an existing method in the aspects of a standard data set and real measurement data.

The invention provides a core image registration method based on multi-modal image segmentation and feature matching. The method is characterized in that the imaging difference between a rock core laser confocal image and a CT image is reduced through preprocessing color conversion and a normalization strategy, a marker is segmented by using deep learning, an algorithm target area is optimized in combination with morphological operation, feature points are detected under the guidance of an ROI area by using an ORB algorithm, and a target image is obtained. Matching of feature points is achieved by combining an FLANN feature point matching algorithm and a random K-d tree acceleration algorithm, image alignment is completed through a rigid transformation model, and registration precision is quantified through Dice. According to the method provided by the invention, the registration problem of the multi-modal core image caused by imaging difference is solved, the average Dice reaches 0.9210, and the registration efficiency and accuracy are remarkably improved.

The invention discloses a robotized autonomous assembly system and method for large aircraft components, and the system comprises a sensing module, a decision module and an execution module, the sensing module comprises a depth camera and a point cloud registration unit which is used for estimating the relative rigid transformation of a target point cloud of the current large aircraft component relative to a source point cloud of a known assembly pose; the decision-making module is used for converting relative rigidity into a tail end target pose under a base coordinate system of the mechanical arm based on a coordinate system chain type transmission principle and rigid body pose invariance constraint; the execution module is used for receiving the target pose of the mechanical arm and planning a collision-free, smooth and efficient motion track from the current pose to the target pose; the mechanical arm executes the assembling action according to the motion trail, meanwhile, the depth camera feeds back pose data in real time, and closed-loop control is formed till assembling is completed; according to the invention, the technical problems of insufficient sensing precision, poor system flexibility, limited measurement real-time performance and splitting of each link in the existing aircraft large component assembly technology can be effectively solved.

The invention relates to the technical field of medical image processing, in particular to a multi-time-point neural image comparative analysis method and system.The method comprises the following steps that a first three-dimensional neural image matrix and a second three-dimensional neural image matrix are obtained, rigid transformation matrix registration is calculated, and a third three-dimensional neural image matrix is generated; constructing a three-dimensional elastic grid model and adjusting node coordinates based on a mutual information criterion; generating a three-dimensional displacement vector field; constructing a Jacobian matrix and calculating determinant values; executing natural logarithmic transformation to generate a microscopic volume change rate map; and calculating a logarithm Jacobian mean value and determining a volume change state. According to the method, pose deviation is eliminated by calculating a rigid transformation matrix, a vector field reflecting fine displacement is constructed by utilizing a mutual information criterion, voxel-level quantitative analysis is realized in combination with a Jacobian matrix and logarithmic transformation, and minimal lesion recognition precision and diagnosis efficiency are improved in cooperation with a standard brain anatomy template.

This application belongs to the field of computer vision technology and discloses a cross-source point cloud registration method based on adversarial neural networks. The method includes: inputting a multi-frame LiDAR point cloud sequence and a UAV video sequence; segmenting each frame of the LiDAR point cloud into ground points and non-ground points and registering them to obtain a global LiDAR point cloud; performing motion recovery structure reconstruction on the UAV video sequence to generate a dense color point cloud; inputting the global LiDAR point cloud and the SfM dense color point cloud into an MMAlignNet network to construct a shared latent feature space; solving for the rigid transformation matrix from the LiDAR point cloud to the SfM dense color point cloud; and aligning the cross-modal point clouds according to the rigid transformation matrix to obtain a multimodal 3D reconstruction result. This application improves the registration accuracy, completeness, and robustness of large-scale outdoor environment 3D reconstruction by fusing UAV imagery and mobile LiDAR data, combined with cross-modal feature alignment and geometric constraint optimization methods.

The invention discloses a construction supervision method and system based on a squat silo, and relates to the technical field of construction digital monitoring, and the method comprises the steps: collecting geometric and image data, building a design reference model, generating a check point according to a fixed arc length, unifying actual measurement coordinates to design coordinates through image recognition and homonymy point rigid body transformation, and carrying out the calculation of the design coordinates; and on the basis of the knowledge graph, determining unique mapping of the actually measured coordinates and the design coordinates by process consistency, adjacent topology and geometric dissimilarity, and calculating a deviation to generate a report. According to the method, the comparison consistency is improved and the mismatching rate is reduced based on the unified aperture and the coordinate system, the robustness is enhanced through hierarchical semantic constraint and geometric discrimination, manual setting is reduced through threshold determination, the out-of-limit determination accuracy is improved, the local increment updating and versioning storage functions are achieved, and the efficiency and the traceability are better.

This invention discloses a point cloud registration method, apparatus, terminal, and computer-readable storage medium. The method includes: acquiring a source point cloud and a target point cloud to be registered; inputting the source point cloud and the target point cloud into a trained point cloud registration model, and sequentially processing them through an initialization module, a transformer-based overlap score module, a self-attention mechanism-based mismatch point removal module, and a singular value decomposition module in the trained point cloud registration model to obtain a final rigid transformation matrix between the source point cloud and the target point cloud; and using the final rigid transformation matrix to transform the source point cloud to obtain a final source point cloud, thereby completing the final registration between the source point cloud and the target point cloud. This invention reduces the impact of overlapping regions and mismatch points on registration by clearing non-overlapping regions and removing mismatch points from the point cloud to be registered in the point cloud registration model, effectively improving the accuracy of point cloud registration.

The application provides a three-dimensional point AI registration and tolerance analysis method and system for non-contact measurement, relates to the field of three-dimensional point AI registration and tolerance analysis, and comprises the following steps: synchronously collecting point cloud data through a plurality of sensors, and performing adaptive preprocessing to obtain a preprocessed point cloud pair; using local curvature and normal vectors to construct a multi-scale geometric descriptor, and performing multi-scale hierarchical feature extraction to obtain an enhanced point feature set; calculating a bidirectional matching probability matrix by using optimal transport theory, screening high-confidence point pairs through spatial compatibility constraints, obtaining a high-confidence matching point pair set and a matching score, performing transformation matrix estimation and quantitative uncertainty, and outputting an optimal rigid transformation and a covariance matrix; and performing adaptive optimization in layers, dynamically adjusting a search range, and verifying transformation parameters across scales to output a converged accurate registration result. The application is used to solve the defect of low registration accuracy at a low overlap rate in the prior art.

The invention relates to an automatic generation method of a 3D embossment photo frame capable of customizing embossment shapes and characters, and belongs to the technical field of computer vision, three-dimensional reconstruction and computer graphics. A high resolution image and a pixel level mask are generated to precisely define a relief core region. Secondly, Boolean operation fusion of the embossment and the photo frame is carried out, a 3D photo frame grid is generated through loading or programming, and an embossment main body grid and the photo frame grid are fused through Boolean union set operation; and finally, generating and finally fusing character grids, generating 3D character grids according to custom characters input by a user, positioning the 3D character grids to a designated area of a photo frame through a rigid body transformation matrix, executing Boolean union set operation again, and outputting a complete 3D model.

The application discloses a kind of component three-dimensional deformation field information acquisition method and device, belong to material manufacturing test technical field, the component three-dimensional deformation field information acquisition method includes: from different visual angle obtains the deformation process data of the component to be measured carrying speckle pattern, to extract each visual angle corresponding local deformation data, and it is spliced to obtain initial splicing result;Judgment whether there is overlapping part in the initial splicing result;If there is overlapping part then the initial splicing result is iteratively optimized, until the optimal rigid transformation matrix between each local deformation data is found, and the current optimized splicing result is used as fine splicing result;The continuity check is carried out to the fine splicing result;The fine splicing result that passes check is homogenized, to obtain the three-dimensional deformation field information of the component to be measured;The application can greatly improve the precision of three-dimensional deformation field information acquisition, reduce error.

The application discloses a non-rigid point cloud registration method and relates to the technical field of three-dimensional reconstruction. 2 The application further discloses a non-rigid point cloud registration method, which comprises the following steps: standardizing data points of a collected non-rigid point cloud and reference point cloud; calculating a local linear embedding weight matrix L of the data points; calculating a matrix M of the data points according to the matrix L; calculating a Gram matrix G of the data points; calculating a corresponding matrix P of the data points based on the reference point cloud; iteratively calculating a non-rigid transformation coefficient matrix W by using an L-M algorithm based on the matrix G, the matrix M and the matrix P; updating λ and σ in the non-rigid transformation coefficient matrix W according to the number of iterations; calculating the matrix P and the non-rigid transformation coefficient matrix W again; setting an iteration termination condition, that is, the number of iterations reaches a set value or a relative error of a target function value is less than a threshold value; and outputting a non-rigid transformation T based on the matrix G and the non-rigid transformation coefficient matrix W after the iteration is terminated.

The invention discloses a self-adaptive point cloud registration method based on mixed geometric constraint and statistical filtering, and belongs to the technical field of machine vision and three-dimensional point cloud processing. According to the method, an improved coarse-to-fine matching strategy is provided for solving the problems that a traditional distance signature algorithm is large in calculation redundancy, sensitive to noise and lack of scale adaptability. The method comprises the following specific steps: firstly, calculating normal vectors of a source point cloud and a target point cloud, carrying out rapid pre-screening by utilizing a normal included angle constraint, and removing candidate point pairs with inconsistent geometric features; secondly, a robust distance signature based on bilateral percentile truncation is constructed, and outlier noisy points in a distance spectrum are stripped through a statistical method; and finally, adopting a normalized dynamic matching threshold as a criterion to realize adaptive matching of point clouds with different densities, and solving a rigid transformation matrix through singular value decomposition (SVD). According to the method, the robustness and the calculation efficiency of point cloud registration in noise interference and partially overlapped scenes are remarkably improved.

The invention discloses a multi-modal point cloud tower inclination detection method based on Lie group structure self-attention. The method comprises the following steps: S1, carrying out multi-modal input data and space registration of a tower; s2, mapping the multi-modal features of the tower to the Lie group; s3, constructing a Lie group structure self-attention mechanism; s4, constructing an integral structure of the tower; and S5, carrying out structure segmentation and inclination angle output on the main pole of the pole tower. According to the method, tower point cloud and image features are uniformly mapped to a Lie group space, and explicit modeling and efficient multi-modal fusion of rigid transformation invariance features are realized by introducing Lie group structure position coding and a self-attention mechanism; the defects that an existing monitoring method is insufficient in geometric consistency utilization of tower multi-modal data, low in noise immunity and insufficient in spatial structure modeling are effectively overcome, the precision, generalization ability and complex environment adaptability of tower inclination detection are fundamentally improved, and the intelligent level of power distribution network tower health detection is greatly improved.

The application provides a medical image registration method and system based on a multi-dimensional loss function. The method comprises: preprocessing 3D-CT medical image data, including fixed images and floating images; constructing an initial model of a registration model using a transformer-based deep learning architecture, first performing coarse registration on the preprocessed 3D-CT medical image data to obtain coarse registration rigid transformation parameters; further performing fine registration on the initial model of the registration model based on the coarse registration rigid transformation parameters; and further optimizing the initial model of the registration model using a multi-dimensional adaptive loss function to obtain a final registration model capable of obtaining optimal results. The application comprehensively uses deep learning network technology, a coarse-to-fine two-stage registration structure based on a transformer deep learning network framework, and a multi-dimensional adaptive loss function, thereby improving the registration adaptation range and registration accuracy.