31 results about "Projection reconstruction" patented technology

The invention relates to the technical field of image data processing, and discloses a few-sample scene-oriented industrial product surface defect image analysis method, which comprises the following steps of: obtaining surface gray level image data of a product to be analyzed, calculating a structure tensor matrix and generating an anisotropy degree graph; searching similar blocks in a preset search neighborhood, and constructing a local texture data matrix; performing singular value decomposition on the local texture data matrix to extract a main subspace; constructing a projection operator and utilizing the projection operator to carry out orthogonal projection reconstruction on the local image block to generate a reconstructed background image block; according to the method, through an orthogonal subspace projection mechanism, good product textures and defect signals are separated, random noise is removed, meanwhile, high-frequency structural features are completely reserved, and the method is high in robustness, high in robustness and high in robustness. And the defect detection precision of a complex texture surface in a few-sample scene is improved.

The invention relates to a remote sensing image super-resolution reconstruction method based on cross-scale Mama, and belongs to the technical field of remote sensing image super-resolution reconstruction. The method comprises the following steps: introducing a CSRSSG cross-scale residual state space group and a CARR channel attention back projection reconstruction module to improve an original MambaIR model, and obtaining a cross-scale Mamba super-resolution reconstruction model; extracting shallow layer features of the remote sensing image to be reconstructed based on a cross-scale Mama super-resolution reconstruction model; based on a cross-scale Mama super-resolution reconstruction model, deep layer features are extracted from the shallow layer features; and adding the shallow features and the deep features element by element to obtain residual connection features, and inputting the residual connection features and the remote sensing image to be reconstructed into a cross-scale Mamba super-resolution reconstruction model for reconstruction to obtain a super-resolution reconstructed remote sensing image. The method can effectively enhance the extraction and expression capability of the model on associated information among different scale features in the remote sensing image, and improves the super-resolution reconstruction quality of the remote sensing image.

The invention discloses a multi-modal data enhancement method and system based on feature space alignment. The method comprises the following steps: collecting modal original data streams of multiple unmanned aerial vehicles; reversely deducing an internal generation rule from a data final state, and constructing a sparse coding feature representation library through causal pilot signal extraction and singular trajectory analysis; constructing an adaptive weight map according to the feature representation library, and identifying a space alignment path through dynamic calibration processing to generate an alignment imbalance index; performing cross information entropy analysis to establish a collaborative enhancement chain, and identifying a collaborative enhancement mode based on the information coupling degree to generate a complementary enhancement vector; performing multi-dimensional projection reconstruction on the complementary enhancement vector, and determining feature mapping probability distribution through diffusion time inversion to obtain a multi-dimensional projection feature map; carrying out nonlinear enhancement processing to generate an enhancement expansion mode, and identifying an enhancement key node to construct an enhancement regulation and control sequence; and outputting hierarchical enhancement data based on the regulation and control sequence and the projection feature map, and realizing accurate space alignment and collaborative enhancement of multi-sensor data.

The invention discloses a method and a system for generating a long video with a controllable camera track based on a video diffusion model. Comprising a camera track and initial frame preparation stage, a point cloud construction and multi-view image generation stage, a scale factor alignment optimization stage, a camera motion prior injection and noise initialization stage, a diffusion inversion generation stage and a sliding window time consistency fusion stage. Three-dimensional camera track modeling, projection reconstruction and diffusion processes are combined, a video content generation process is explicitly guided to be aligned with a track path set by a user, and long video generation which is reasonable in structure, natural in vision and continuous in time is achieved.

The invention discloses a precipitation forecast correction method combining EOF projection and a U-Net network, and the method comprises the steps: extracting a main spatial mode of a precipitation abnormal field based on historical observation data, and carrying out the projection reconstruction of a mode forecast abnormal field on this basis, and obtaining a spatial structure with physical significance; an observation climate state is introduced to replace a mode climate state, so that systematic deviation of the mode is effectively eliminated; for residual terms which are not explained in the reconstruction process, U-Net is adopted for modeling and prediction so as to capture complex nonlinear error components in the mode; and finally, superposing a linear reconstruction result and a residual term of deep learning prediction to form a final rainfall forecast correction result. According to the method, a correction framework combining physical driving and data driving is constructed, and on the basis of keeping the physical interpretability of a traditional EOF method, a deep learning model is introduced to model a residual term, so that dual capture and correction of linear errors and nonlinear errors in mode forecasting are realized.

The invention discloses a cable safety monitoring and early warning system comprising the following modules: a disturbance injection module used for injecting a double-frequency disturbance electric signal; the temperature acquisition module is used for acquiring temperature change data of each monitoring point of the cable; the residual construction module is used for constructing a two-dimensional disturbance thermal residual spectrum; the thermal evolution analysis module is used for performing wavelet packet multi-scale decomposition on the disturbance thermal residual spectrum and generating a two-dimensional thermal evolution image through an improved back projection reconstruction algorithm; the first early warning calculation module is used for acquiring a thermal response function and calculating a time partial derivative to obtain a thermal evolution trend index as a first early warning factor; the vibration acquisition module is used for acquiring a reflection waveform data set; the second early warning calculation module is used for calculating an equal-amplitude decay time symmetry rate index as a second early warning factor; and the early warning generation module is used for generating a final early warning scheme according to a preset judgment rule. According to the invention, on the basis of dual-frequency disturbance excitation and an improved back projection reconstruction algorithm, intelligent early warning of cable monitoring is realized.

Disclosed in the present invention are a weighted analytic filtered back projection reconstruction method and system for asymmetric cone angle artifacts. The method comprises the following steps: dividing a reconstruction area into a plurality of weight regions on the basis of relative positions of a ray source ring and a detector ring; acquiring the projection data volume of voxel points in each weight area irradiated by X-rays; according to the projection data volume of the voxel points in each weight area irradiated by the X-rays, assigning a different initial weight to each weight area; performing smooth transition on the initial weight of each weight area by means of a transition weight to form a final weight assigned to each weight area; and according to different final weights of the weight regions, performing final weighted analytic reconstruction on projection data p (α, β, γ) to acquire a back projection image.

The present application relates to a kind of remote sensing image super-resolution reconstruction method based on cross-scale Mamba, belong to remote sensing image super-resolution reconstruction technical field.The method includes: introducing CSRSSG cross-scale residual state space group and CABR channel attention reverse projection reconstruction module are improved to original MambaIR model, obtain cross-scale Mamba super-resolution reconstruction model;Based on cross-scale Mamba super-resolution reconstruction model, extract the shallow feature of the remote sensing image to be reconstructed;Based on cross-scale Mamba super-resolution reconstruction model, extract deep feature from shallow feature;Element-by-element addition is carried out to shallow feature and deep feature, and residual connection feature is obtained, and residual connection feature and the remote sensing image to be reconstructed are input into cross-scale Mamba super-resolution reconstruction model and are reconstructed, and super-resolution reconstruction remote sensing image is obtained.The present application can effectively enhance the extraction and expression ability of model to the associated information between different scale features in remote sensing image, improve remote sensing image super-resolution reconstruction quality.

The invention relates to the technical field of multi-source servo intelligent calibration, and particularly discloses a servo control system autonomous intelligent calibration method based on multi-source information fusion, and the method comprises the steps: obtaining the equipment attribute information of all servo equipment and multi-source sensor data synchronized through a global time protocol in response to a system starting or calibration instruction; and according to the physical space distribution data and the transmission connection relation data in the equipment attribute information, constructing an equipment incidence matrix representing the coupling strength between the equipment. According to the method, various cross-device coupling conflict characteristics including transmission, electromagnetism and loads can be intelligently recognized, external interference noise is accurately stripped from sensor data through dynamic weight adjustment and a coupling projection reconstruction mechanism, it is ensured that a control system only carries out calibration for the real intrinsic error of the device, and the reliability of the system is improved. Therefore, false compensation and system instability caused by coupling interference are avoided.

The application discloses a kind of based on deep learning cascade enhancement multi-bandwidth photoacoustic imaging system and method, the system is by pulsed laser, two half annular array probes, multichannel data acquisition card and central computer composition, pulsed laser is used to excite the light-absorbing body in target area, generates photoacoustic signal;Two half annular array probes respectively have high-frequency probe with center frequency 5.5MHz and 2.5MHz low-frequency probe, high-frequency probe is used to obtain high-resolution image, low-frequency probe is used to detect larger light-absorbing body and low signal-to-noise ratio structural features;Multichannel data acquisition card is used to carry out digitization processing to photoacoustic signal received by two half annular array probes, and the data after processing is transmitted to central computer, reconstructs photoacoustic image based on filter back-projection reconstruction algorithm.The multi-bandwidth photoacoustic imaging system of the application shows very high image reconstruction capability, provides more accurate solution for biomedical imaging.

A sparse projection CBCT reconstruction method, device and equipment and readable storage medium, which comprises the following steps: obtaining X-ray projections of a target patient at multiple angles; obtaining a whole tissue mask prediction projection for reconstruction and a hard tissue mask prediction projection for reconstruction according to a preset tissue projection prediction model and the X-ray projections; supervising and training a preset neural field network initial model according to the X-ray projections, the whole tissue mask prediction projection for reconstruction and the hard tissue mask prediction projection for reconstruction to obtain a neural field network model corresponding to the target patient; obtaining whole tissue gray scale data of each coordinate point in a three-dimensional image to be reconstructed according to the neural field network model and coordinate information of each coordinate point in the three-dimensional image to be reconstructed, and forming the three-dimensional image to be reconstructed. Different tissues in CBCT are modeled by a multi-unit network, the interpretability and controllability of the neural field network model are improved, and the method has strong practical significance.

The application discloses a kind of multi-projection light field image synthesis method and system, it is related to image synthesis technical field, the method includes: obtaining the coordinates of projector array, the bottom of holographic screen and actual camera array;Determine multiple display restoration light;The display restoration light of each projector is carried out image synthesis, obtains multiple target images, and the holographic screen receives the projection of projector on target image and reconstructs three-dimensional light field;The display restoration light of a projector is carried out image synthesis process is: according to the column pixel of all display restoration light of projector corresponding target image determination column pixel synthesis process is: determining the ideal camera of display restoration light coordinates;According to the ideal camera coordinates and the coordinates of actual camera array determine the actual coordinates of two actual cameras;From according to display restoration light, active range and actual coordinates determine multiple actual imaging pixels.The application improves the accuracy of projection image synthesis and projection reconstruction light field display effect.

This application relates to the field of computer science and discloses an artificial intelligence task scheduling system and method for edge computing. The system includes modules for semantic manifold kernel extraction, heterogeneous resource potential field construction, gradient tensor projection reconstruction, and streaming gradient optimization scheduling. The method utilizes a potential energy function to map heterogeneous network resources into a scalar potential field and calculates negative gradient vectors; it generates a semantic manifold kernel mask based on the semantic entropy density of feature channels, and combines it with an environmental pressure tensor mapped by the negative gradient magnitude to project and reconstruct the model weights to generate a dynamic inference model; it synthesizes a scheduling vector using the negative gradient vector and Lyapunov drift correction to determine the next-hop target node. This invention achieves adaptive model compression and network load balancing while preserving key semantic features through potential field gradient collaborative optimization of the model structure and transmission path, solving the problem of balancing transmission efficiency and inference accuracy in edge computing scenarios.

The application discloses a kind of self-adapting direction and step of multi-energy spectrum CT iterative reconstruction method with scattering correction, comprising: step 1, setting multi-energy spectrum CT projection reconstruction model with scattering correction;Step 2, using iterative algorithm to solve out the increment of base material distribution projection value angle by angle, update the current base material image to new base material image;Wherein, through self-adapting direction and step to quickly obtain convergent base material projection image.The scheme provided by the application is used to remove scattering, and the result of base material decomposition is more accurate, and the reconstruction result artifact is less, which lays a foundation for quantitative analysis.

The present application relates to a kind of step edge high-precision three-dimensional reconstruction method, using the parallel single-pixel imaging principle based on projection reconstruction, in combination with edge detection algorithm and adaptive bidirectional projection strategy solve the light ray aliasing problem of step edge, to realize the high-precision acquisition of step edge three-dimensional stereo matching point.First, the sequence of phase-shifted fringe pattern of bidirectional periodic continuation is generated, is projected to measurement scene by projector in turn, and the coded image of scene is obtained by camera synchronous acquisition.Then, the lower frequency image in fine positioning fringe image is extracted to calculate absolute phase map, carries out edge extraction, generates edge mask image.Secondly, the one-dimensional projection function calculation of bidirectional is carried out to the edge pixel in mask image, further calculate the intersection of projection line to obtain multiple intersection points of bidirectional projection line as candidate matching point, and two direct matching points from upper and lower surfaces of step edge are screened out by epipolar distance constraint.Finally, using calibration parameter and matching point, three-dimensional reconstruction point cloud of step edge is obtained by solving stereo matching algorithm.The application effectively solves the light ray aliasing problem of step edge caused by camera visual angle obstruction in triangulation camera-projector system, reconstructs the upper and lower surfaces of step edge completely, and improves the three-dimensional measurement precision of step edge.Based on the optimized bidirectional projection reconstruction parallel single-pixel imaging, while ensuring the measurement accuracy, the measurement efficiency is improved, which is conducive to the application and popularization of the method in actual industrial measurement field.

The invention discloses a remote sensing image semantic segmentation method and system guided by linear positioning attention, and the method comprises the following steps: carrying out the direction-sensitive feature projection reconstruction of an input feature map through a linear positioning attention mechanism, and adjusting the direction distribution of a feature vector through a nonlinear power positioning function, the low-rank limitation is broken through in combination with depth separable convolution, and direction enhancement features are generated; a deformable edge attention mechanism is used, a geometric self-adaptive edge attention graph is generated based on a deformable multilayer perceptron, edge features are enhanced through serial secondary attention modulation, and edge enhancement features are obtained; a symmetric coding and decoding network is constructed, an encoder uses ResNet18 to extract multi-scale features, a decoder uses a multistage linear positioning ViT block to perform step-by-step up-sampling and fuse with jump connection features, shallow jump connection introduces edge enhancement features, and finally a pixel-level semantic segmentation map is output through a segmentation head.

The invention belongs to the technical field of computer vision and pattern recognition, and relates to a face recognition method and device based on deep semantic perception, and the method comprises the steps: obtaining a face image training set, and constructing a visual feature matrix of the face image training set; performing singular value decomposition on the matrix to obtain a first-order standard projection matrix; introducing a learnable matrix to construct a second-order projection matrix, constructing a two-stage projection reconstruction matrix, and introducing a pre-trained face representation network to construct a loss function; iteratively solving an optimal learnable matrix according to the loss function, and obtaining an optimal second-order projection matrix according to the matrix; obtaining a to-be-recognized face image, and recognizing the to-be-recognized face image according to the matrix and the face image training set to obtain a face recognition result; according to the method, the sample is projected into the high-level semantic space of the network for subspace learning, so that the discrimination stability and interpretability under complex disturbance are improved.

The invention provides a single-view projection fast fluorescence tomography reconstruction system based on a double-path condition guide diffusion model, and the system comprises a view collection module which generates an original data set composed of a fluorescence three-dimensional distribution image and a single-view fluorescence projection measurement image, and collects single-view fluorescence projection data in a real experiment; the model construction module is used for defining a condition model framework, a diffusion model framework and a dual-path condition guiding mechanism to obtain an initial single-view projection reconstruction model; and the model optimization module is used for carrying out training updating and hyper-parameter setting and optimization on the initial reconstruction model by utilizing the original data set to obtain a single-view projection reconstruction model, and then verifying and evaluating the single-view projection reconstruction model through the test set. According to the method, the aggravated ill-conditioned problem caused by extremely sparse single-view data and overlapped depth information is solved, rapid dynamic living body three-dimensional imaging of target monitoring target molecules can be realized, and high-quality reconstruction is realized.

The invention relates to the technical field of computer vision and optical character recognition, and discloses a clustering-based character component matching method and system. The character component matching method based on clustering comprises the following steps: performing optical character recognition processing on a to-be-processed image to obtain a plurality of characters; the method comprises the following steps: firstly, performing spatial clustering processing on all characters to obtain at least one initial space group, and then performing angle clustering processing on the characters in each initial space group to obtain at least one character subgroup; performing projection reconstruction on the character subgroup, and determining the arrangement sequence of all characters; sorting all the character subgroups according to the sorting sequence of all the characters to obtain an ordered character string sequence; according to the method, the discrete OCR recognition result is reconstructed into an ordered character string sequence by combining double clustering with a dynamic projection reconstruction technology, so that the character sequence reconstruction accuracy and the automation degree are effectively improved.

The invention provides an underground cable three-dimensional imaging method based on terahertz data and X-ray data, and the method comprises the steps: solving dielectric constant distribution and material density distribution according to the terahertz data and the X-ray data, building the incidence relation between the dielectric constant distribution and the material density distribution, and determining projection reconstruction initial parameters; a fitness function is constructed according to the projection data and the projection reconstruction initial parameters, dielectric constant distribution and material density distribution are reconstructed through a quantum genetic optimization algorithm based on the fitness function and the incidence relation, and an initial reconstruction result is obtained; performing cross iterative optimization on the dielectric constant distribution and the material density distribution by using a regularization item constructed by an initial reconstruction result until convergence; and based on the converged dielectric constant distribution and material density distribution, after an X-ray reconstruction image and a terahertz reconstruction image are generated, data fusion and feature optimization are carried out, and a three-dimensional image of the underground cable is obtained. Therefore, a high-precision three-dimensional image of the underground cable is generated by fusing the optimized bimodal image data.

The invention relates to a multi-camera 360-degree panoramic image splicing method based on depth information guidance. The method comprises the following steps: calibrating internal and external parameters of a multi-camera system and constructing a panoramic coordinate system; the images are subjected to distortion removal and then paired according to an adjacent sequence, feature points are extracted and matched, the real depth under a camera coordinate system is calculated through binocular intersection, and a sparse real depth map is obtained; meanwhile, obtaining an affine invariant depth map by using a monocular depth estimation network, carrying out scale alignment on the sparse real depth map and the corresponding affine invariant depth map to obtain a global scale factor and a depth offset, and recovering a dense real depth map; performing depth projection on the three-dimensional point clouds reconstructed by back projection of the plurality of dense depth maps on a spherical surface to obtain a panoramic depth map; and establishing a pixel mapping relation between the panoramic image and each camera image by means of panoramic depth image information, constructing a pixel value mapping table, and generating a 360-degree panoramic image. According to the method, the artifact problems of image dislocation, ghosting and the like caused by parallax can be effectively relieved.

The invention discloses a deep learning-based X-ray image lossless counting method and device for cartons, and the method comprises the following steps: obtaining gray images of different energy levels, generating a pseudo-color image, optimizing the exposure time and dosage according to the material and thickness of a carton body, and carrying out the projection collection of four angles at the same time; the collected images are preprocessed; inputting the preprocessed image into a density perception U-Net model, extracting features through an edge guide branch and a material perception branch, generating an attention map and strengthening key features, and outputting an article density map and a bounding box set by a detector; processing the density map based on a density peak detection algorithm to obtain a predicted object number, and performing error correction through multi-angle projection reconstruction and density map fusion; and the counting result and the defect alarm are visually displayed. According to the invention, articles in the box body can be counted without opening the box and unsealing, and the problem of article counting in a complex scene is effectively solved.

Image data is read out from a left part (LL) of a left field of view (FOV) of a left image sensor (L) and a right part (RR) of a right FOV of a right image sensor (R). The left part of the left FOV extends horizontally towards a right side of a gaze point (X) till only a first predefined angle (N1) from the gaze point. The right part of the right FOV extends horizontally towards a left side of a gaze point (X) till only a second predefined angle (N2) from the gaze point. The image data is processed to construct a left part of a left image and a right part of a right image. A right part of the left image and a left part of the right image are reconstructed using image reprojection. The left image and the right image are generated by combining respective parts.

A method for calibrated X-ray imaging is described, comprising an X-ray imaging system (10) with a variable-pose radiation source (1) (P1) and an X-ray detector (2) with a variable-pose position (P2) that can be controlled independently of the variable-pose position (P1) of the radiation source (1). In this method, an initial calibration (KAL0) is first performed by projecting an X-ray image of an X-ray phantom (PH) onto the X-ray detector (2) for a plurality of different positions (P1(n), P2(n)) of the radiation source (1) and the X-ray detector (2), thereby determining a plurality of initial projections (C(n, R0, t0)).The X-ray imaging system (10) is then used to acquire X-ray projection data (PD) of an object under investigation (OB). During the acquisition of the X-ray projection data (PD), optical image data (OBD1) representing relative poses (k(i)) of the X-ray detector (2) to the X-ray source (1) are determined for a plurality of different poses (P1(n), P2(n)) of the X-ray source (1) and the X-ray detector (2). Based on the relative poses (k(i)) of the X-ray detector (2) to the X-ray source (1) determined from the optical image data (OBD1) and based on the initial calibration (KAL0), corrected projections (C(n, R1, t1)) are determined. Finally, a reconstruction of X-ray image data (BD) is determined based on the acquired X-ray projection data (PD) and the corrected projections (C(n, R1, t1)). A calibration and control device (50) is also described. Furthermore, an X-ray imaging system (10) is described.

The application discloses a kind of binary Fourier single-pixel imaging method, system, device and medium, method includes: based on error diffusion template, by m kind of pixel scanning order to Fourier base pattern is error diffusion, obtains m group binary Fourier base pattern;M group binary Fourier base pattern is projected respectively to processing, and corresponding m two-dimensional target image is reconstructed to obtain;Select n two-dimensional target image to carry out fusion processing, and obtain target image.The application obtains multiple groups of binary Fourier base pattern by different scanning order, reconstructs by the projection of binary Fourier base pattern, finally, the reconstructed two-dimensional target image is fused, effectively reduce the quantization noise and environmental noise of image, while the high-frequency information of image is preferably retained, so that the imaging quality is greatly improved, while realizing the function of good noise reduction, without losing image detail part due to excessive smoothing, can be widely applied in optical imaging technology field.

The application discloses a zooming static CT system and an image reconstruction method, wherein the system comprises a cylindrical detector array arranged in an arc and an X-ray source array arranged in a preset shape; the X-ray source array and the detector array remain static during scanning; projection data is acquired at different angles by sequentially emitting beams of different X-ray sources; the distance from the X-ray source to the detector is variable; a data acquisition system is used for controlling the detector to acquire data; a control system controls the emission of the X-ray source and the acquisition of data by the detector; and an image processing system processes acquired image projection data by using an image reconstruction algorithm. The image reconstruction algorithm in the embodiment of the application adopts a weighted filter back-projection reconstruction algorithm, which can optimize the system geometry, improve the efficiency and accuracy of the reconstructed image, and improve the applicability and reliability of the system.

The application relates to the technical field of image data processing, and discloses an industrial product surface defect image analysis method for a few-sample scene, which comprises the following steps: acquiring surface gray image data of a product to be analyzed, calculating a structure tensor matrix and generating an anisotropy degree graph; searching for similar graph blocks in a preset search neighborhood, and constructing a local texture data matrix; performing singular value decomposition on the local texture data matrix to extract a main subspace; constructing a projection operator and performing orthogonal projection reconstruction on the local graph blocks by using the projection operator to generate reconstructed background graph blocks; splicing the reconstructed background graph blocks into a full-width reconstructed background image, calculating a gray difference graph, and performing adaptive threshold segmentation to extract defect features; and through the orthogonal subspace projection mechanism, the good product texture and the defect signal are separated, the random noise is removed, the high-frequency structural features are completely reserved, and the defect detection precision of a complex texture surface in a few-sample scene is improved.

This invention provides a method and apparatus for measuring the volume of a material pile in a construction scenario, relating to the fields of computer vision and building engineering technology. The invention acquires binocular image data and generates left and right view pairs through epipolar correction; then, it uses a cross-cost aggregation algorithm to estimate disparity in the corrected binocular image data to generate a disparity map; it performs back-projection reconstruction on the disparity map, extracts color information and confidence values, and generates a binocular point cloud image, then segments and extracts the scaled point cloud; next, it inputs the binocular point cloud image into an AtlasNet network architecture to generate a scale-free point cloud; it aligns and restores the scale of the scale-free point cloud, and finally fuses the restored point cloud with the scaled point cloud to calculate the fused height value and the volume of the material pile. This invention generates a complete 3D model with realistic physical scale by jointly registering and fusing scale-free and scaled point clouds, achieving high-precision material pile volume calculation and significantly improving measurement accuracy and efficiency.

The present invention provides a generalized segmentation method and system in the field of medical images, which belongs to the technical field of medical image processing. After preprocessing multi-center medical image data, the method constructs a network architecture based on U-Net and ResNet for feature extraction and segmentation prediction. The image features are converted into frequency domain representations through the frequency domain decomposition module, and the low-frequency amplitude and phase information are separated. The style representation space is constructed using the frequency domain style projection reconstruction module, and the unknown domain style is projected into the source domain style library and the feature distribution is perturbed and expanded. Through the multi-level contrast decoupling module, the decoupling of domain-independent and domain-dependent features is achieved by combining prototype-like contrast learning and pixel-level semantic contrast learning. Finally, the segmentation loss and contrast learning loss optimization model are combined to generate the final segmentation result. The present invention adopts the above method to solve the problem of domain offset of multi-center data of medical images through frequency domain style alignment and contrast decoupling.

The application discloses a brain perfusion parameter prediction blood vessel recanalization grade method and system, and belongs to the technical field of medical image processing. The method comprises the following steps: acquiring a preoperative CTP parameter map and a CTA maximum density projection reconstruction map of an acute large blood vessel occlusive stroke patient and a dynamic scanning sequence of CTP scanning; extracting functional perfusion features from a CTP channel and morphological structure features from a CTA channel through a multi-modal feature fusion network; extracting a collateral compensation vector based on a CTP dynamic sequence through a pre-constructed collateral circulation embedding module; modeling six discrete grades into an ordered category by using an mTICI hierarchical ordered regression prediction head of a cumulative link model; and modulating a prediction output by the collateral compensation vector through an attention gate mechanism, and outputting an mTICI grade prediction probability distribution. The application realizes preoperative prospective prediction of an mTICI grade after mechanical thrombectomy.