557 results about "Spatial transformation" patented technology

The invention provides an improved YOLOv11s safety helmet wearing detection model and an optimization method thereof, and relates to the technical field of computer vision target detection. According to the improved YOLOv11s safety helmet wearing detection model and the optimization method thereof, the improved YOLOv11s safety helmet wearing detection model comprises the following modules: a multi-modal fusion module, a space-time analysis module, a domain adaptation module, a topological optimization module and a dynamic architecture module; and the multi-modal fusion module is used for realizing feature decoupling by adopting channel separation convolution based on input RGB and near-infrared images, fusing visible light and thermal radiation features through a dynamic weight distribution algorithm, implementing affine transformation alignment on multi-scale features by utilizing a spatial transformation network, and generating a multi-modal feature graph. Through fusion of visible light and near infrared spectrum features and implementation of dynamic weight distribution, complementarity of target texture and thermal radiation features under a complex illumination condition is enhanced, and the problem of feature distortion of single-mode data in a strong backlight or low-illumination scene is solved.

The embodiment of the invention provides an end-to-end automatic driving control method and device based on multi-camera fusion, and multi-view target detection and tracking are realized through spatial transformation and coordinate mapping by combining front wide-angle camera information and left and right wide-angle camera information. A multi-view feature fusion network architecture is designed, the multi-view feature fusion network architecture comprises three sub-networks of feature extraction, dynamic weight distribution and feature fusion, and the fusion weight is dynamically adjusted based on image definition, detection confidence and view overlapping degree. A geometric consistency constraint between visual angles and a reconstruction loss function are introduced, a deep neural network model is constructed, abnormal conditions such as camera shielding are effectively handled, and an accurate control instruction is output. According to the method, the defects of the traditional technology in the aspects of multi-view information fusion, shielding processing and the like are overcome, and the sensing ability and the control reliability of the automatic driving system are remarkably improved.

The invention discloses an industrial surface defect detection method based on multi-scale feature fusion, and the method comprises the steps: collecting the multi-source data of a detected surface in real time through a multi-modal sensor array, and forming a structured data set through time-space synchronization and denoising; constructing an adaptive geometric correction model to realize spatial transformation and scale normalization of multi-scale features, and cooperatively realizing cross-modal alignment and preliminary fusion through texture and physical attribute branches of a double-flow decoding network; dynamically reweighting the fusion features based on a defect physical model, strengthening physical mechanism defect characterization and suppressing interference; combining optical flow compensation and three-dimensional convolution to extract spatio-temporal evolution characteristics, and forming dynamic defect characterization; and finally outputting defect type and severity evaluation through the classification model in combination with the process parameter library. Therefore, the adaptability of the method to a complex industrial environment is enhanced, and the detection stability can be maintained under different materials, illumination conditions and dynamic interference.

The invention discloses a scene topology understanding method and device, a storage medium and a program product, and relates to the field of computer systems based on a specific calculation model, and the method comprises the steps: inputting a multi-view environment image set into a backbone network, and generating bird's-eye view features corresponding to the environment image set; calculating a spatial transformation matrix of the aerial view features at the current moment and the aerial view features of the previous K frames, and performing space-time alignment on the obtained K + 1 frames of aerial view features to obtain multi-frame fusion features; inputting the multi-frame fusion features into a map prior model to obtain aerial view correction features; decoding the aerial view correction features based on a topological decoder, and generating a lane topological graph; comparing the lane topological graph with the annotation data, calculating an error loss function, and optimizing network parameters based on the error loss function; and generating an optimized topological graph, and determining a scene topology result based on the optimized topological graph. By implementing the method, the environment topology understanding capability in a complex scene can be improved, and the generation precision of the lane topological graph is optimized.

The invention relates to the field of image processing, and particularly discloses a welded pipe surface defect detection method based on robot visual inspection. The method comprises the steps that a robot carries a binocular camera and an annular LED light source and moves at a constant speed in the axial direction of a welded pipe to collect orthographic and inclined views, and a three-dimensional point cloud is constructed; and establishing a parameterized mapping function based on the point cloud, and converting the 3D coordinate into a 2D expansion surface coordinate. In the convolutional neural network, a first layer is inserted into a spatial transformation network to correct distortion of the expanded image, deformable convolution is adopted to extract edge, local deformation and specific defect response features, and standard convolution is combined to extract global features; and fusing multi-scale features and adding an attention mechanism to improve the weight of a defect region, and outputting a defect category and a bounding box offset after generating a candidate box. The method effectively solves the problems of stretching, deformation and defect distortion of welded pipe curved surface imaging, reduces the imaging difference of the same defect, and remarkably improves the defect positioning precision and recognition accuracy.

The invention provides a multi-view-oriented three-dimensional scene image reconstruction registration and optimization method and system, and relates to the technical field of image processing, and the method comprises the steps: obtaining multi-frame three-dimensional scene image data, extracting a multi-level feature set, constructing a cross-view-angle semantic association graph, building a feature corresponding relation, and calculating a multi-view-angle spatial transformation relation parameter. Performing coordinate system alignment on the image data to generate an initial three-dimensional reconstruction result, and performing optimization in combination with a multi-target joint optimization function and a dynamic adaptive weight regulation and control mechanism. According to the method, the precision and robustness of three-dimensional scene reconstruction are improved, and the problem of registration errors caused by large view angle difference in a complex scene is solved.

The invention relates to the technical field of surface defect recognition, and discloses an intelligent recognition method and system for surface defects of an autoclaved aerated concrete member. The method comprises the following steps: carrying out omnibearing shooting and spatial transformation matrix mapping on the surface of the autoclaved aerated concrete member to obtain a preprocessed image; performing texture feature extraction on the preprocessed image to obtain texture feature data; performing spatial feature analysis on the preprocessed image and the texture feature data to obtain spatial relation feature data; performing feature importance weighting and bidirectional cross fusion processing on the texture feature data and the spatial relation feature data to obtain comprehensive defect feature representation; and performing surface defect analysis on the autoclaved aerated concrete member based on the comprehensive defect feature representation, and outputting a surface quality comprehensive score and grade division result. According to the method, the problems of dead angles and image deformation existing in traditional single-view-angle detection are solved, and the completeness and accuracy of defect detection are ensured.

The invention discloses an image classification method for cross-domain transfer learning and related equipment, and relates to the technical field of image classification, and the method comprises the steps: obtaining a source domain image set and a target domain image set; extracting a first multi-level semantic feature of the source domain image set and a second multi-level semantic feature of the target domain image set based on a heterogeneous feature extraction network; generating a cross-domain migration weight matrix through a dynamic domain similarity measurement module based on the first multi-level semantic features and the second multi-level semantic features; based on the deformable feature pyramid, performing spatial transformation on the first multi-level semantic features to generate a migration feature map; based on the cross-domain migration weight matrix, channel recombination is carried out on the migration feature map, and target domain adaptive feature representation is constructed; and based on the target domain adaptive feature representation, outputting a classification result of the target domain image through a target domain classifier. According to the method, the feature matching precision and the classification robustness in a cross-domain scene are improved.

The invention provides an infrared and visible light image end-to-end registration method based on phase consistency enhancement, and belongs to the technical field of electric digital data processing. The invention discloses an infrared and visible light image end-to-end registration method based on phase consistency enhancement, which comprises the following steps of: firstly, preprocessing infrared and visible light source images in an image database, and dividing the infrared and visible light source images into a training set and a test set; and constructing a feature extractor based on a phase consistency attention mask, realizing cross-modal consistency feature extraction, and focusing an important space region by using phase information. A self-defined ResSobeNet network is adopted for parameter estimation, and Haar wavelet transform is used for replacing pooling operation, so that information loss in a traditional down-sampling method is avoided. A double-branch generator structure is designed, affine parameters and an optical flow field are generated respectively, meanwhile, constraint on rigid registration and non-rigid registration is achieved, and finally a registration result is obtained through resampling of a space conversion module. According to the method, the problems that cross-modal consistent features are difficult to extract in the infrared and visible light image registration process and explicit registration steps are complicated and difficult to generalize are solved, and the method has good universality and practicability.

The invention relates to a teacher behavior recognition method based on a cross-modal dynamic query network, and belongs to the field of artificial intelligence education application. The problem that in the prior art, adaptation to education scene dynamic interactivity, semantic complexity and environment interference is insufficient is solved. The method comprises the following steps: extracting video spatio-temporal characteristics through a time-space transformation network, and generating a semantic prototype by combining a comparison language image pre-training model; a multi-region attention module is adopted to realize hierarchical feature fusion, and semantic related visual clues are screened; generating a self-adaptive query vector driving classification decision by using a dynamic query decoding module; and outputting a behavior recognition result based on a cross attention mechanism. The method has the technical effects that the characterization distinction degree of complex teaching behaviors is remarkably improved, the robustness of a dynamic classroom environment is enhanced, the reliability of multi-class behavior decisions is optimized, deep education semantic understanding is realized, and the high-efficiency and low-consumption landing application requirements are met.

The invention discloses a tumor space-occupying brain network neural image alignment method based on multi-modal fusion, and belongs to the technical field of medical image processing and artificial intelligence crossing. The method comprises the following core steps of multi-modal image heterogeneous feature decoupling, tumor occupation deformation field modeling, functional network topological structure maintenance, cross-modal feature adversarial alignment, dynamic deformation constraint optimization and clinical interpretability verification, and construction of a three-dimensional non-rigid registration network based on a double attention mechanism. And differential homeomorphic mapping of a tumor focus area and normal brain tissue is realized through the cascaded spatial transformation module. Aiming at the problems of insufficient multi-modal feature alignment and brain network topology distortion in the prior art, the invention provides a function connection constrained cross-modal fusion strategy, a graph convolution network is adopted to encode resting state function connection features, and network node displacement caused by tumor occupation is dynamically corrected in combination with deformable convolution and a bidirectional feature competition mechanism; a space consistency loss function based on white matter fiber bundle tracing is designed, and through diffusion tensor imaging feature guide structure-function bimodal joint optimization, the problems of insufficient registration precision in a focus area and whole brain network connection distortion of a traditional method are solved. Experiments show that the registration precision of the method in glioma cases reaches 0.82 mm and is improved by 37% compared with that of a traditional method, and dissection-function consistency of functional network reconstruction around tumors is remarkably improved.

The invention discloses a deep learning-based multimodal image fusion method for soft tissue photoacoustic / ultrasound imaging. Steps: an ultrasound-photoacoustic imaging device acquires photoacoustic and ultrasound images of human soft tissue and performs size normalization processing; an input spatial transformation module converts the images to the YCbCr space; an input pre-convolution module modifies the number of data channels; an input multi-scale feature extraction module extracts salient features from the source images; an input filter prediction module derives multi-scale filters; and an input filter fusion and adaptive enhancement module combines the input source images to obtain the final fused result. The invention has superior fusion performance compared to several traditional fusion methods and deep learning-based fusion methods, and more importantly, it exhibits excellent real-time performance. Furthermore, various modes of photoacoustic / ultrasound fusion extension experiments have verified the effectiveness of the method proposed in the invention.

The invention relates to a light-weight low-light image enhancement method based on illumination iterative adjustment, and the method comprises the steps: carrying out the HVI color space transformation of a to-be-processed image, obtaining an HVI image, separating an illumination intensity component, obtaining an illumination intensity feature map, and carrying out the convolution processing of the HVI image, and obtaining a shallow feature map; feature fusion and brightness adjustment are carried out on the illumination intensity feature map and the shallow layer feature map to obtain a depth feature map, up-sampling is carried out step by step through two layers of decoders to recover the size, channel adjustment is carried out on each layer of decoder by adopting convolution of a specified size, and up-sampling is realized through bilinear interpolation; a Mama-based structure refining module is introduced into the tail end of each layer of decoder to improve the structure restoration capability, the number of channels of HVI features output by the last layer of decoder is adjusted to a specified value through a convolution of a specified size, residual connection is carried out on the HVI features and an initial HVI image, an enhanced HVI image is obtained, and a restored image is obtained through inverse HVI transformation. Adaptive enhancement is carried out under various low light conditions, multi-directional spatial context aggregation is realized, and texture representation is improved.

The invention discloses an adversarial sample generation method based on a partially perceptible patch, and the method enables a specific region of a target image to be visually close to the characteristics of a natural image through the optimization disturbance of the specific region of the target image, maximizes the interference to a target detection model, and improves the attack effectiveness while maintaining the concealment. The invention further provides a device of the adversarial sample generation method based on the partially perceptible patch. According to the method, color space transformation and random deformation are combined, the adaptability of the adversarial sample in different environments is further enhanced, and the method has higher robustness and universality in practical application.

The invention belongs to the technical field of medical image digital image processing, and particularly relates to a prostate TRUS-MRI registration method based on multi-scale weighting and an adaptive window, and the method comprises the steps: firstly collecting an ultrasonic image and a magnetic resonance image of a prostate, dividing the ultrasonic image and the magnetic resonance image into a training set, a verification set and a test set, and sketching a mask for a prostate gland in each image; then the input image is preprocessed; then, UNet is adopted as a feature extractor, a deformation field is output in combination with a spatial transformation network, deformation is applied through an STN pair, and a registration image is obtained; the method comprises the following steps: firstly, acquiring three parts of an ultrasonic image, then constructing a network loss function, finally, acquiring a total loss function according to the three parts of the network loss function, performing staged training, finally, outputting the total network loss function, processing a subsequent ultrasonic image and a magnetic resonance image, and outputting a registration image. Compared with the prior art, the method has the comprehensive advantages of cross-modal adaptability, detail optimization capability, calculation efficiency and generalization performance, so that the method is remarkably improved.

The invention belongs to the technical field of steel reinforcement framework three-dimensional reconstruction, and particularly discloses an automatic high-precision three-dimensional dense reconstruction method for a box girder steel reinforcement framework, which comprises the following steps of: dividing a scanning area into a plurality of sub-areas according to a depth image by moving an inspection robot along the steel reinforcement framework, and performing multi-angle local scanning in each sub-area; constructing a local point cloud model by combining camera poses during acquisition, further calculating a relative spatial transformation relationship by using an overlapping region between adjacent local models, constructing a pose map and implementing global optimization, and uniformly adjusting the poses of the local models in a world coordinate system; and finally, all corrected local point clouds are fused to generate a complete global three-dimensional model, so that collaborative reconstruction operation of domain-divided scanning, segmentation reconstruction and global optimization is realized, frame-by-frame accumulation of errors in traditional scanning-while-splicing reconstruction is effectively blocked, long-distance drift is remarkably inhibited, and the precision of three-dimensional reconstruction of the reinforcement cage is greatly improved.

The invention relates to the technical field of lock manufacturing machine tools, in particular to an error coupling compensation and positioning precision control method for a lock manufacturing machine tool. The method comprises the following steps: acquiring the structure and operation data of a lock making machine tool through a sensor, detecting the operation condition of the lock making machine tool and detecting the spatial transformation of each motion unit of the lock making machine tool, and generating the spatial transformation data of each motion unit; according to the spatial transformation data of each motion unit, error coupling positioning of a workpiece and a locking piece of the locking machine tool and operation coupling error analysis of each component of the locking machine tool are carried out, and operation coupling error data of each component are generated; setting precision compensation control parameters of the locking machine tool according to the operation coupling error data of each component, and generating precision compensation control parameter data; and real-time precision compensation and machining track correction of the locking machine tool are carried out on the operation error coupling data of all the components based on the precision compensation control parameter data. Error coupling positioning and precision compensation control of the lock manufacturing machine tool are achieved.

The invention belongs to the field of instrument identification, and particularly relates to a pressure gauge identification method and equipment in an industrial valve pressure testing process. The method comprises the following steps: extracting an instrument image from a collected video stream, and carrying out distortion correction and illumination normalization processing; feature extraction is carried out on the processed image, instrument types are identified based on a training model of multiple types of instruments, and an instrument image with angle deformation is corrected through a spatial transformation network; performing spatial sorting on the identified instrument images according to the coordinates of the central point of the detection frame, and generating a data array containing instrument information; and finally, performing enhancement processing on a single instrument image in the data array, positioning a dial plate area, performing perspective correction, determining geometric parameters of the dial plate, calculating an original reading based on the parameters, and outputting a final reading in combination with inclination compensation. Through the image processing and recognition steps, accurate recognition and reading of the pressure gauge in the industrial valve pressure testing process are achieved, and the recognition efficiency and accuracy are improved.

The invention discloses a photovoltaic module hyperspectral image registration splicing method, system and device. The method comprises the following steps: acquiring a hyperspectral image of a to-be-detected region and a high-resolution image of the same region; the acquired hyperspectral image and the high-resolution image are preprocessed; performing feature extraction and descriptor generation; preliminarily matching the feature points extracted from the hyperspectral image and the high-resolution image; performing accurate matching on the preliminarily matched feature points, calculating a spatial transformation model from the hyperspectral image to the high-resolution image, and performing registration on the hyperspectral image by using the transformation model obtained by calculation; and carrying out sequence splicing and fusion on the plurality of hyperspectral images and the registered image. According to the invention, the high-resolution image-assisted photovoltaic module hyperspectral image registration splicing reduces mismatching, reduces the number of times of repeated calculation and adjustment of an algorithm, reduces the consumption of calculation resources, and efficiently completes the registration splicing task under the condition of not increasing the hardware cost.

The invention discloses a multi-modal remote sensing image registration method based on domain self-adaption and an optimizer set algorithm, and relates to the field of image processing, and the method comprises the steps: respectively extracting the feature vectors of a reference image and an image to be registered, carrying out the feature matching of the feature vectors according to the class label containing conditions of the feature vectors, and obtaining a registration result; obtaining a feature matching vector based on a matching result; estimating transformation parameters corresponding to the affine transformation model by using a random sampling consistency algorithm, and optimizing the affine transformation model according to an estimation result and similarity measurement; and performing spatial transformation on a to-be-registered remote sensing image by using the optimized affine transformation model to realize spatial alignment between the reference image and the to-be-registered image. According to the method, firstly, the feature points of the multi-modal remote sensing image are extracted, the difference of feature vectors is reduced by using a domain adaptive method, the precision and robustness of feature matching are improved, transformation parameters are optimized through an optimizer set algorithm, and search is prevented from falling into local optimum.

The invention provides a friction stir welding seam intelligent correction method based on visual identification, and relates to the technical field of friction welding. Image information of a welding seam area of a to-be-welded workpiece is collected in real time, image feature points are recognized, and the spatial pose of the welding seam area is obtained through a spatial transformation model; based on the space position deviation of the actual space pose and the theoretical welding track, a dynamic deviation correction path is generated; and a multi-point iterative optimization algorithm of motion control parameters and a real-time feedback control method are further adopted, so that high-precision real-time adjustment and dynamic compensation of the welding seam track are realized. According to the method, the accuracy and the stability of friction welding seam deviation rectification are remarkably improved.

The invention discloses a method, device and equipment for generating a head model, a storage medium and a program product. The method is characterized by comprising the following steps of: receiving and verifying original three-dimensional T1 weighted image data and an original three-dimensional grid model; performing tissue segmentation based on the verified three-dimensional T1 weighted image data and the original three-dimensional T1 weighted image data to obtain a cerebral grey matter mask, a cerebral white matter mask, a scalp mask and a skull mask; performing topological structure repair on the brain grey matter mask and the brain white matter mask to obtain a brain mask; respectively converting the scalp mask, the skull mask and the brain mask into a scalp mesh model, a skull mesh model and a brain mesh model; calculating and applying a spatial transformation matrix from the verified three-dimensional grid model to the scalp grid model, and performing spatial registration and fusion on the scalp grid model, the skull grid model and the brain grid model to obtain a head model; the method has the advantages that the efficiency, precision and robustness of head model generation are effectively improved, and the method has good cross-platform deployment capability.

The invention provides an aircraft appearance abnormity identification method and system based on cross-category feature registration, and relates to the technical field of aircraft intelligent detection and maintenance, and the method comprises the steps: obtaining an aircraft appearance image, and carrying out the preprocessing of the image; inputting the preprocessed image into a cross-category feature registration network, extracting feature maps by using a DenseNet dense connection feature extraction module, fusing the feature maps of different dense layers, and inputting the fused feature maps into an STN-C spatial transformation and alignment module to obtain a transformed feature map; performing channel and space attention on the transformed feature map to obtain an attention weighted feature map, and inputting the attention weighted feature map into a twin network into which a negative cosine similarity loss function is introduced to obtain an embedded vector; calculating a mahalanobis distance between the embedded vector and normal distribution, generating a pixel-level abnormal thermodynamic diagram, and positioning an abnormal region; and adaptively adjusting the threshold of the abnormal score graph through dynamic threshold segmentation, and realizing segmentation of an abnormal region to obtain a final abnormal result.

A DBO-STN-Informer-based aquaculture water dissolved oxygen prediction method comprises the following steps: S1, adopting a dissolved oxygen sensor to collect dissolved oxygen data in aquaculture water online in real time, and dividing the data into a training set and a verification set according to a time sequence after preprocessing; s2, introducing a spatial transformation network STN on the basis of an Informer encoding mode to transform an input time sequence, constructing an STN-Informer prediction model, and performing spatial transformation and feature calibration on input time sequence data through the STN to help an encoder to understand a deep dissolved oxygen sequence; an attention mechanism parameter in the STN-Informer prediction model is optimized by adopting a dung beetle algorithm DBO; and finally, performing test scoring and optimization on the model by using the verification set. And S3, based on the trained and optimized model, outputting a dissolved oxygen concentration prediction value in a future time period. According to the method, the dissolved oxygen concentration in the future time period can be predicted according to the prior data, and the method has very high prediction accuracy, precision and efficiency.

The invention relates to a global measurement field construction method for a complex scene, and the method comprises the steps: fixedly arranging a plurality of target seats which are suitable for the placement of laser reflection target balls of a laser tracker in the scene, and enabling the target seats to serve as visual reference points; laying a laser tracker station; measuring three-dimensional coordinate values of the laser tracker stations and the visual reference point and laser interference ranging values of the laser tracker stations and the visual reference point, and obtaining measurement data of the visual reference point under different laser tracker stations; the method comprises the following steps: constructing a spatial transformation relationship among station coordinate systems of laser trackers, converting measurement data of the laser trackers to a global coordinate system, and constructing a global measurement field; and carrying out adjustment optimization on the obtained global measurement field. According to the method provided by the invention, high-precision measurement field construction oriented to large-size complex-structure parts is realized, the precision and robustness of coordinate calculation are improved, and the problem that traditional measurement field construction in a complex scene is easy to block is effectively solved.

The embodiment of the invention discloses a brain image registration method and device, electronic equipment, a storage medium and a program product. The method comprises the following steps: acquiring an individual brain image, a template brain image and a pre-trained registration model; inputting the individual brain image and the template brain image into a registration model, and registering the template brain image to the individual brain image through the registration model based on the following steps: carrying out affine registration associated with the individual brain image and the template brain image by using an affine registration network to obtain an affine transformation matrix, and carrying out a first spatial transformation layer to obtain a second spatial transformation layer; based on the affine transformation matrix, performing spatial transformation of the template brain image to obtain an intermediate brain image; and performing deformation registration on the intermediate brain image and the individual brain image by using a deformation registration network to obtain a deformation field, and performing spatial transformation of the intermediate brain image based on the deformation field by using a second spatial transformation layer to obtain a target brain image. The problem that the brain image registration process is tedious is solved.

The invention discloses a modal fusion method of camera-sonar equipment under the condition of unknown external parameters, and relates to the field of underwater multi-sensor data fusion. The method comprises the following steps: firstly, acquiring synchronous underwater optical images and sonar images, and converting sonar polar coordinate imaging data into a three-dimensional point cloud under a sonar coordinate system through a geometric model; then, under the condition that camera-sonar external parameters do not need to be calibrated in advance, a spatial transformation matrix between a camera coordinate system and a sonar coordinate system is automatically estimated and calibrated through the corresponding relation between images and target features in point clouds, and alignment fusion of two kinds of modal data is achieved. According to the method, optical and acoustic information can be effectively fused in a severe water quality environment, the sensing limitation of a single sensor is made up, the robustness and accuracy of underwater dam crack and other target detection are improved, and the method has the advantages of no need of manual calibration and high adaptability.

Various implementations disclosed herein include devices, systems, and methods that localize a device based on detecting planes in depth data acquired by the device. For example, an example process may include detecting first plane data in first sensor data acquired by a sensor at a first viewpoint location in a physical environment, detecting second plane data in second sensor data acquired by the sensor at a second viewpoint location in the physical environment, determining that the first plane data and the second plane data correspond to a same plane based on comparing the first plane data with the second plane data, and determining a spatial transformation between the first viewpoint location and the second viewpoint location based on the first plane data and the second plane data.

The invention provides a multi-scale implicit-explicit mixed medical image registration method and system. The method comprises the following steps: acquiring CT images of a moving image and a fixed image; processing the CT image by using the trained registration network model, performing down-sampling on a three-dimensional coordinate grid to different set proportions of an original resolution, and then predicting deformation of different scales by using a multi-layer perceptron network model to generate a first deformation field and a second deformation field; down-sampling the fixed image and the moving image to a third set proportion of the original resolution, predicting third scale deformation by using the CNN network model, and generating a third deformation field; and fusing the three deformation fields to obtain a final deformation field, and generating a final deformation image through a spatial transformation network. The method is simple in registration process, and has good practical application efficiency and expandability.

The invention provides a distortion image correction method based on optical imaging, and the method comprises the steps: training a distortion correction model through the historical data of a source imaging device, and obtaining the optical features and similarity weights of all regions of a source device and a target device; training an image correction model based on the historical data of the source device, and generating correction images and spatial transformation features of each region of the target device through the model in combination with the historical data and the similarity weight of the target device; and then calculating image quality evaluation indexes of each region according to the correction result, the transformation features and the optical features, and summarizing to obtain the distortion correction completion degree of the target equipment. According to the method, the distortion correction precision of the target equipment can be improved by effectively utilizing abundant data of the source equipment, and quantitative verification of the correction effect is realized.