165 results about "Spatial reconstruction" patented technology

The invention provides a magnetic resonance image reconstruction method and device, a model training method and device, equipment and a medium. The method comprises the following steps: inputting acquired to-be-reconstructed under-sampling k-space data into a pre-constructed image reconstruction model, and carrying out continuous k-space reconstruction processing for multiple times to obtain a magnetic resonance reconstruction image; wherein the k-space reconstruction processing in the image reconstruction model comprises the steps of performing frequency domain restoration on input k-space data based on a Transformer encoder introduced with symmetric weight, performing image domain signal restoration on the k-space data after frequency domain restoration, and performing data consistency processing on the k-space data after image domain signal restoration. According to the method, the image reconstruction model is constructed in an end-to-end expansion training optimization mode, so that the frequency domain and image domain information of the undersampled data can be fully utilized, the accuracy of image reconstruction is effectively improved, the computing resource overhead and the time cost can be effectively reduced, and the high efficiency of the reconstruction process is ensured.

The invention relates to the field of grain storage management, and discloses an intelligent granary grain pile dew point early warning method and system, and the method comprises the steps: carrying out the continuous collection of the temperature and humidity, airflow velocity and pile density of each grain pile in a granary, and generating a humidity and heat perception graph reflecting the spatial layering characteristics; performing time sliding and spatial reconstruction analysis on the damp-heat perception graph, and constructing a damp-heat evolution mapping model; based on the damp and hot evolution mapping model, calling a multi-dimensional risk identification engine to perform aggregation judgment on the micro-region with the dew point close to a critical value, and outputting a potential condensation risk set; performing joint correction on the potential condensation risk set, real-time airflow distribution, grain pile accumulation form and ventilation response delay characteristics, and performing dynamic weight regression on different grain varieties and seasonal factors to generate dynamic stability indexes reflecting risk migration and diffusion trends; and adaptively adjusting the sensing sampling density and the early warning judgment threshold according to the dynamic stability index. The method has the advantage of improving the early warning precision of storage environment management.

The invention relates to the technical field of unmanned aerial vehicle obstacle avoidance, and discloses an unmanned aerial vehicle autonomous obstacle avoidance method based on deep learning and binocular vision. According to the method, original data streams of a left view and a right view are acquired through a binocular vision acquisition terminal, and stereoscopic vision feature representation is extracted through parallel convolutional coding branches of a deep neural network model; inputting into a three-dimensional space reconstruction module to generate a dense depth map and an obstacle initial position coordinate set; the dynamic obstacle analysis engine calculates a dynamic threat evaluation index by combining real-time flight attitude parameters of the unmanned aerial vehicle, and the space-time trajectory prediction model calculates future moving path probability distribution of the obstacle according to the dynamic threat evaluation index; and fusing the distribution with preset navigation path planning data to generate a three-dimensional obstacle avoidance course correction vector, converting the three-dimensional obstacle avoidance course correction vector into a flight control instruction set, and downloading the flight control instruction set to an execution module. According to the method, the reliability and adaptability of autonomous obstacle avoidance of the unmanned aerial vehicle in a complex dynamic environment are improved, and a powerful guarantee is provided for safe navigation of the unmanned aerial vehicle.

The invention discloses an infrared unmanned aerial vehicle target detection method based on multi-scale self-enhancement cross-layer fusion, and aims to solve the problem of difficulty in small target detection of an infrared unmanned aerial vehicle under a complex background. According to the method, a novel target detection network is constructed, and a lossless down-sampling module and a cascade asymmetric convolution module based on spatial reconstruction are designed in a backbone network of the novel target detection network, so that multi-scale context features are extracted while information loss is reduced. The neck network adopts an enhanced pyramid structure, deep semantics and shallow details are integrated through a cross-layer feature fusion module, and a multi-scale adaptive channel attention module is introduced to dynamically re-calibrate fusion features so as to focus key information. The network adopts four detection heads to output multi-scale prediction, and uses an EIoU loss function to optimize small target positioning precision. The method can effectively improve the feature extraction and detection capability of'weak, small and dark 'infrared unmanned aerial vehicle targets, and has higher robustness and accuracy in a complex scene.

The invention relates to the technical field of bridge health monitoring, and discloses a method for monitoring the bearing capacity state of a bridge latticed column structure. The method comprises the following steps: constructing an original monitoring field according to a sensor time sequence signal, and generating a state evolution sequence representing the overall dynamic evolution of the bridge latticed column through multi-dimensional space reconstruction, so as to capture the nonlinear characteristics of the structure behavior and provide a basis for anomaly detection. And positioning a potential abnormal region based on the deviation degree from the reference model, and extracting a time-varying characteristic spectrum of the potential abnormal region to drive the digital twin model to perform simulation so as to obtain a simulation response spectrum. The simulation response spectrum and the actual measurement state evolution sequence are subjected to space-time fusion, a mixed state field fusing real data and physical mechanism inference is generated, a monitoring blind area is made up, and the data reliability is improved. And by performing multi-scale decomposition on the mixed state field, the specific state mode of the bearing capacity of the bridge latticed column is identified, and more accurate and more complete evaluation of the bearing capacity abnormity is realized.

The invention relates to the technical field of ultrasonic image processing systems, and provides an ultrasonic image processing system based on three-dimensional reconstruction, which comprises an ultrasonic data acquisition terminal, an image processing terminal, a three-dimensional reconstruction terminal, a quality evaluation terminal and a doctor interaction terminal, the ultrasonic data acquisition terminal is used for acquiring ultrasonic original radio frequency data and a B mode image and synchronously acquiring probe position information of the ultrasonic probe; the image processing terminal is used for performing image denoising, edge enhancement and tissue feature labeling according to the ultrasonic original radio frequency data and the B mode image, and extracting key section information for subsequent reconstruction; the three-dimensional reconstruction terminal is used for performing spatial reconstruction based on the key section data and the probe position information to form a three-dimensional ultrasonic image body; the quality evaluation terminal is used for scoring the three-dimensional reconstruction effect of the three-dimensional ultrasonic image body and generating three-dimensional reconstruction quality evaluation information; and the doctor interaction terminal is used for a doctor to check the three-dimensional ultrasonic image body and the three-dimensional reconstruction quality evaluation information in real time and generate and display optimization suggestion information. The method has the effect of improving the accuracy of three-dimensional reconstruction quality evaluation.

The invention relates to the technical field of meteorology and hydrology, in particular to a rainfall space reconstruction method, system and device based on terrain and meteorological multi-factor fusion driving and a medium, and the method comprises the steps: obtaining space terrain factors and meteorological observation data of a target area, and employing a BP neural network to progressively interpolate missing measurement values of meteorological driving variables in a layered manner; setting a missing detection elimination rule based on a wet season and a dry season to process precipitation data; constructing a 14-dimensional high-dimensional input feature system fusing a space terrain factor, a time sequence factor and a meteorological driving factor; nonlinear models such as an XGBoost model, a BP neural network model or an LSTM model are used for training, and finally the monthly scale precipitation space reconstruction of the grid is achieved. According to the method, the problems that a traditional interpolation method is poor in adaptability in a complex terrain area and insufficient in multi-factor driving relation description are effectively solved, and the precision and the physical consistency of precipitation space distribution are remarkably improved.

The invention relates to the technical field of remote sensing land prediction, in particular to a farmland non-grain spatial distribution prediction method and system, and the method comprises the steps: abstracting a farmland plot into a graph node, defining an edge weight as a weighted combination of an inter-plot Euclidean distance, crop type similarity and irrigation system connectivity, constructing a three-layer dynamic graph structure comprising land parcels, neighborhoods and administrative units; taking the three-layer dynamic graph structure as input, and performing node feature extraction by adopting a graph attention long-short-term memory network to obtain a non-grain ratio preliminary predicted value; non-stationary state space-variational Kalman filtering is adopted to correct the preliminary prediction value of the non-grain ratio, and a plot-level prediction value is obtained; and performing spatial reconstruction on the plot-level predicted value by using policy-sensitive Kriging interpolation. According to the invention, spatial continuity and time serialization cultivated land non-grain trend prediction is realized.

The invention discloses an EEG-to-three-dimensional fMRI volume generation method based on multidirectional time-frequency convolution attention and VM-UNet, and the method comprises the steps: S1, carrying out the preprocessing of an inputted EEG signal, constructing an EEG input sample in a three-dimensional tensor form, and carrying out the normalization processing of EEG and fMRI data; s2, mapping an EEG input sample to a high-dimensional feature space by using an EEG spectrogram projection module to obtain an initial embedding representation; s3, processing the initial embedded representation through a multi-directional time-frequency convolution attention encoder to obtain a multi-level fusion global feature representation, and adjusting the multi-level fusion global feature representation into a feature map with a target size; s4, the adjusted feature map is input into a Vision-MambaU-Net decoder, multi-scale decoding and spatial reconstruction are carried out, and a three-dimensional fMRI volume is generated; and S5, performing end-to-end training on the model by adopting a joint loss function to realize the generation of the volume from the EEG to the three-dimensional fMRI. According to the method, the fine anatomical features of cortex wrinkles and deep grey structures are accurately reserved, and the video memory consumption and the reasoning time are greatly reduced.

The invention provides a hub-oriented dynamic flexible adaptive space reconstruction method and system, and the method comprises the steps: constructing a hub space plane graph of a to-be-constructed hub space in a preset coordinate system, and determining boundary coordinates, obstacle coordinates, entrance coordinates and exit coordinates of the hub space plane graph; determining a first route by adopting a first algorithm based on the boundary coordinates, the obstacle coordinates, the entrance coordinates and the exit coordinates of the hub space plane graph, and determining a first path based on the first route; if the hub space is a preset key area, updating the first route by adopting the second algorithm to obtain a second route, and determining a second path based on the second route; and determining a path boundary in the first path or the second path, arranging the spatial reconstruction units along the path boundary, determining coordinates of the spatial reconstruction units, and matching the coordinates with the corresponding spatial reconstruction units.

The invention provides an atmospheric environment particulate matter pollution detection method and device, and the method comprises the steps: carrying out the grading analysis of the scattered light intensity at each environment detection point in an environment pollution detection region, and obtaining the particle size distribution characteristics of air particulate matters at each environment detection point in a plurality of particle size intervals; determining a spatial distribution field of the particulate matter concentration in the air based on the particle size distribution characteristics; determining a dynamic correction factor of diffusion and aggregation of particulate matters in the air according to an atmospheric physical state, and further performing dynamic correction on the spatial distribution field based on the dynamic correction factor; performing spatial reconstruction on the spatial distribution field obtained by dynamic correction on a regular grid, and determining the distribution pattern of the air particulate matter concentration in the environmental pollution detection area according to a reconstruction result; and generating a particulate matter pollution index in the environmental pollution detection area according to the distribution pattern. By adopting the scheme of the invention, the reliability of atmospheric environment particulate matter pollution detection can be improved based on particulate matter concentration space reconstruction of multi-point dynamic correction.

The invention relates to the technical field of image generation, in particular to a medical beauty 3D model construction method and system based on an image large model, and the method comprises the following steps: obtaining rasterized image data based on a 3D full-head point cloud model through space coordinate projection and pixel mapping, judging connectivity and a missing region, screening region features, and generating texture guidance parameters; and screening textures according to partition features, determining a multi-region fitting index, and outputting a continuous boundary fusion region. According to the invention, through carrying out multi-angle spatial reconstruction and fine pixel mapping on the three-dimensional point cloud data, texture coverage is not limited to a single view angle, spatial feature partitioning and target attribute identification, so that accurate supplementation of the texture of the missing region is realized; the locally generated texture content and the original input region are highly fused in the aspects of color structure and detail representation, and region mutation is effectively eliminated through spatial adaptation and boundary adaptive processing, smooth connection of multi-part texture splicing presentation, color transition and gradient fusion.

The invention discloses an equipment surface defect detection method and system based on three-dimensional space reconstruction. The method comprises the following steps: acquiring multi-view spatial data of standard equipment and equipment to be detected, and respectively constructing three-dimensional models containing geometric structures and surface features; registering the three-dimensional point cloud data of the standard equipment and the to-be-detected equipment into the same world coordinate system, and automatically generating two-dimensional rendering pictures of the standard equipment and the to-be-detected equipment with the same visual angle and pose in the same coordinate system; and further performing surface defect detection on the two-dimensional rendering pictures corresponding to the two-dimensional rendering pictures, and visually presenting a defect detection result at a corresponding position of the three-dimensional model. According to the invention, artificial leak detection and error detection are effectively avoided, and the efficiency and accuracy of equipment surface defect detection are remarkably improved.

The invention belongs to the technical field of computer vision and image processing, and particularly relates to an efficient single image super-resolution system based on perception loss guidance. The system comprises a feature extraction module, a denoising module and an up-sampling module. The feature extraction module adopts a multi-branch structure and supports structure re-parameterization so as to improve the reasoning efficiency; the up-sampling module realizes spatial reconstruction based on sub-pixel rearrangement. A small denoising module formed by depth separable convolution is introduced before up-sampling, and artifacts and noise are effectively suppressed at a feature level. Meanwhile, a feature extractor formed by a randomly initialized lightweight convolutional network is introduced and used for calculating the perception loss, so that the structure reduction capability is enhanced, and the training cost is also remarkably reduced; according to the invention, each module is activated and optimized in stages, so that the training stability and the final performance are improved; and excellent performance is realized under a lightweight design, and the method is suitable for various resource-limited scenes.

The invention discloses an underground pipeline detection method and system based on image recognition. The method comprises the following steps: obtaining a preprocessed RGB image sequence and registered laser radar point cloud data; generating a target detection frame set with semantic tags; segmenting the obtained initial two-dimensional semantics into a mask set; generating a two-dimensional semantic segmentation mask set after depth calibration and a semantic point cloud cluster set after depth calibration; obtaining a two-dimensional semantic segmentation mask set after cylinder verification and a semantic point cloud cluster set after cylinder verification; outputting a two-way binding semantic point cloud cluster set; and spatial reconstruction processing is executed based on the two-way bound semantic point cloud cluster set, and underground pipeline spatial recognition detection is completed. According to the method, three-dimensional geometric consistency is introduced as a verification condition on an algorithm structure, so that a segmentation result simultaneously meets semantic rationality and spatial physical authenticity.

The invention discloses a salinized frozen soil roadbed settlement prediction method and system, and relates to the technical field of roadbed settlement prediction, and the method comprises the steps: obtaining multi-source monitoring data of a target salinized frozen soil roadbed, the multi-source monitoring data comprises temperature, moisture content, conductivity, stress and ground surface settlement time sequence observation values distributed along the depth, and a ground surface remote sensing image; establishing a physical mechanism model reflecting the heat-water-salt-force coupling effect of the salted frozen soil based on the multi-source monitoring data; integrating the physical mechanism model as a constraint condition into a neural network training process, and constructing a settlement response prediction sub-model with physical consistency in combination with multi-source monitoring data; performing spatial reconstruction on the multi-source monitoring data to form a two-dimensional profile image containing a temperature field, a humidity field and a salt field, and inputting the two-dimensional profile image into a convolutional neural network to extract spatial structure features; and constructing graph structure data by utilizing spatial structure characteristics, modeling a mutual influence relationship among different salinized areas through a graph neural network, and outputting settlement tendency distribution.

The application discloses a kind of multiple immunofluorescence detection method, equipment and medium based on space reconstruction, it is related to histopathological analysis technical field, including, preparation tissue sample and join fluorescent reference microbead to obtain continuous section;To the tissue sample cycle pretreatment, and utilize fluorescent reference microbead to carry out intensity normalization and bleaching time shift correction, obtain correction image sequence;Spectral unmixing is carried out under three-dimensional voxel coordinate system and introduces space regular constraint, obtain voxel level fluorescence intensity vector;Combining nuclear dye signal and fluorescent reference microbead, using rigid and non-rigid registration and supplemented with three-dimensional deconvolution, reconstruct high-resolution voxel stack and segment cell voxel set;The adjacent relationship and colocalization index of different phenotype cells are calculated, and detection result is generated.The application realizes high-resolution tissue and cell reconstruction under three-dimensional space, improves spatial continuity and accuracy, and improves the stability and accuracy of voxel level fluorescence signal quantification.

The invention relates to the technical field of data processing, in particular to an anti-vibration multi-group camera motion capture system and a data correction method thereof, and the method comprises the following steps: collecting image data of a falling body target through a plurality of infrared motion capture cameras on a plurality of groups of mutually independent metal frame supports; based on spatial reconstruction consistency, performing multi-view cross validation processing on the image data acquired by the plurality of groups of cameras, and identifying and eliminating abnormal parallax data caused by camera vibration; performing trajectory filtering processing on the data subjected to cross validation processing, and performing smoothing processing on trajectory data in combination with Kalman filtering and adaptive weighted median filtering; dividing the filtered trajectory data into a plurality of overlapped time windows according to time; accurate trajectory tracking can be realized in a high dynamic impact scene, and the authenticity and reliability of captured data are improved.

The invention belongs to the technical field of image processing, and particularly discloses a face forgery detection method and system based on mask and frequency diffusion reconstruction, and the method comprises the steps: applying spatial masking to a central region of an input face image, reconstructing the masking region through a diffusion model of a mask condition, and generating a mask reconstruction image; carrying out discrete cosine transform on an input face image, decomposing the input face image into a low-frequency component and a high-frequency component, and reconstructing the low-frequency component and the high-frequency component into a frequency domain enhanced image through inverse DCT; inputting the mask reconstruction image and the frequency domain enhancement image into a detector, and learning a spatial reconstruction error and a frequency domain texture fidelity feature through a contrast learning training detector; and inputting a to-be-detected image into the trained detector to carry out face counterfeiting detection, and outputting to obtain a detection result. The method solves the problem that the existing deep forgery detection method faces severe challenges in practical application due to the limitation that the existing deep forgery detection method is special, precise and single in type but is difficult to generalize.

The invention discloses a case site reconstruction and restoration method and system based on a multi-modal fusion reasoning technology. The method comprises the steps that case site information is collected; performing on-site three-dimensional space reconstruction and material evidence photographing extraction by adopting a mode of structured light and a binocular camera; related multi-modal data semantic features of the case are extracted; constructing a multi-modal knowledge graph; the method comprises the following steps: describing a field material evidence by adopting a multi-modal heterogeneous graph, and reasoning the multi-modal heterogeneous graph by adopting cross-modal graph convolution; key coefficients in the knowledge graph are deduced by adopting a dynamic deduction algorithm based on a recurrent neural network, and a basis is provided for constructing the heterogeneous knowledge graph of the case scene; performing visual display on the case logic; a method of combining an inference chain and a VR technology is adopted, and a logical relationship of case occurrence is visually displayed.

The invention discloses a GPS trajectory tracking and dynamic path analysis method for an automobile data recorder. The method comprises the following steps: S1, collecting and preprocessing vehicle position and timestamp data; s2, constructing an initial trajectory point sequence, and generating a trajectory data set; s3, constructing a multi-section trajectory model, and realizing path dynamic tracking and spatial reconstruction; s4, mapping a current vehicle driving path, and identifying an offset point and an abnormal behavior according to a track fragment matching strategy; s5, extracting vehicle track offset characteristics, and calculating an angle offset accumulated value to generate a deviation score in combination with a control point change trend; s6, triggering a rule engine to adjust a path based on the score, and sending an optimization result to the driving assistance terminal; s7, establishing a multi-segment track model and a video timestamp index, and realizing track and video synchronous playback; and S8, uploading an analysis result to a terminal and a background to realize path monitoring and behavior feedback. The method improves the track precision, the recognition accuracy and the path optimization efficiency, and is suitable for intelligent driving dynamic decision making.

The invention provides an unmarked mandibular movement dynamic acquisition method, and relates to the field of dynamic reconstruction of a jaw position relationship. According to the method, feature extraction and posture calculation are performed completely based on the video image and the tooth three-dimensional model, intervention of a mark point or a mechanical device is not needed, the testee can complete collection in a natural state, an artificial interference source is eliminated, and the physiological measurement accuracy and the user experience are remarkably improved. A multi-view-angle high-speed camera system is combined with a dental crown point topological structure for modeling, multi-view-angle space reconstruction and time sequence compensation are utilized, stable tracking and high-precision reconstruction are still kept under the conditions of shielding and motion blur, and the risk of collection failure is effectively avoided. According to the method, based on dentition overall modeling and posture calculation, a complete mandibular movement track curve can be output, complex movements such as chewing, deviation, occlusion and mandibular lateral rotation can be accurately expressed, and high-quality original data is provided for clinical occlusion evaluation generation.

The invention relates to the technical field of defect detection imaging, and discloses a high-precision three-dimensional imaging method and system based on continuous magnetic flux leakage data. Collecting surface magnetic flux leakage signals of a measured target to form an initial signal set, and extracting feature vectors by using a convolutional neural network model; separating three-dimensional space data, and obtaining initial three-dimensional position coordinates of the defect through an iterative mapping algorithm; mining spatial geometric features, determining potential geometric boundaries, and fusing data to construct a refined position model; extracting a detection subset, correcting a shape error, constructing a space reconstruction grid, filling holes, and obtaining a complete three-dimensional shape representation; and through precision calibration and signal matching verification, a high-precision three-dimensional image is output. According to the method, the defect positioning and shape reconstruction precision can be effectively improved, the imaging integrity and accuracy are guaranteed, the problems of fuzzy positioning and shape distortion of traditional magnetic flux leakage imaging are solved, and the defect detection requirements of a complex detected target are met.

The invention provides a sedimentary strata three-dimensional geology construction method and system integrating carbon footprint evaluation and intelligent block modeling, and relates to the field of energy and geology. According to the method, carbon footprint evaluation and sedimentary stratum three-dimensional geological construction are organically fused, so that full-process integration of data acquisition, dynamic partitioning, spatial reconstruction and low-carbon scheduling is realized, and the bottlenecks that data and carbon emission information are separated and model precision and energy consumption control are difficult to balance in the traditional technology are broken through; the real-time performance of fine expression of geological structures and environmental risk assessment is greatly improved, and more quantitative and intelligent technical support is provided for carbon capture site selection and green exploration decision.

The invention relates to the technical field of image analysis, in particular to a health management system for the real-time safety state of a foundation pit supporting structure, and the system comprises a three-dimensional scene point cloud construction module which synchronously collects left and right camera images and preset camera parameters, calculates the parallax value of each pixel, and transmits the parallax value to a display module; and according to the parallax value of each pixel, the pixel coordinate and the internal and external parameters of the camera. According to the method, pixel-level parallax extraction and camera parameters are combined, the three-dimensional space coordinates of the underground diaphragm wall are calculated, spatial reconstruction of high-density point locations on the surface of the structure is achieved, a reference form capable of being tracked quantitatively is constructed, and corresponding point iterative registration and normal offset analysis are introduced, so that the accuracy of the method is improved. A real deformation track of a wall body can be extracted from tiny deformation, a deviation value set which is highly coupled with a structure state is constructed, a pixel point extraction strategy aiming at anchor rod wall construction characteristics is adopted at a node level, and real space coordinates of constructed nodes are locked.

This application relates to a robotic method for precise ore sampling in complex mining environments, specifically in the field of ore feature recognition. The method includes: acquiring LiDAR point cloud data and industrial camera image data; after voxelizing the point cloud data, obtaining LiDAR BEV features through an improved SPConv sparse pyramid convolution operator feature extraction network, spatial reconstruction unit, and channel reconstruction unit; extracting semantic features from the image data to generate an image BEV feature map; fusing the two data elements-wise in the BEV space; extracting multi-scale features from the fused BEV feature map using a dual-branch Decoder network; introducing a KA attention mechanism to weight the fused features; and calculating a sampling priority score based on the weighted feature map to determine the optimal sampling point and plan the path. This method can improve ore recognition accuracy in complex mining environments and shorten the decision-making time for the entire process of sampling location determination and path planning.

A machine-oriented video coding (VCM) decoding apparatus is proposed in one example of the disclosure. The VCM decoding apparatus includes an internal decoder that generates a feature map by decoding a bitstream and parses information about the feature map by decoding the bitstream, and an image reconstructor that reconstructs at least one skipped channel in the decoded feature map based on the information about the feature map, wherein the information about the feature map can include at least one of the following: group-wise coding information including whether each channel of the feature map is skipped, a quantization compensation parameter, feature map truncation information, temporal reconstruction information, spatial reconstruction information, and packing information.

The embodiments of the present application relate to the field of image processing technology, and in particular to a multi-source heterogeneous image segmentation method based on self-supervised learning, comprising: collecting multi-source heterogeneous sample images and superimposing them in the channel dimension to obtain multi-channel sample input data; constructing a multi-source heterogeneous image segmentation model composed of a spatial branch network and a channel branch network, the spatial branch network and the channel branch network respectively process the multi-channel sample input data to obtain spatial reconstruction features and channel reconstruction features to construct a mask loss function; taking visible light sample images and radar sample images as positive sample pairs, and performing projection transformation and prediction transformation respectively to obtain visible light contrast features and radar contrast features to construct a contrast loss function; constructing a total loss function based on the mask loss function and the contrast loss function, iteratively training the model until convergence, and obtaining a trained model to apply to segmentation tasks to achieve high-precision segmentation of multi-source heterogeneous images.

The present invention discloses a hyperspectral and multispectral image fusion method based on adaptive multi-scale features. First, a hyperspectral image with low spatial resolution and high spectral resolution and a multispectral image with high spatial resolution and low spectral resolution are used as the input of the network, and efficient feature expansion is achieved through upsampling by a channel interaction upsampling module. Then, they are fused into a super-multispectral image with the same size as the reference image. Next, diverse features are captured and spatial information is reconstructed via a multi-scale adaptive spatial reconstruction module. The optimized feature map enters a spectral self-attention enhancement module for spectral attention enhancement, calculates the correlation between feature maps to generate an attention map, and reconstructs spectral information. This avoids problems such as loss of spatial details, insufficient cross-domain generalization ability, and spectral distortion caused by low spectral resolution during the image fusion process.

The invention discloses an air endoscope image processing device and method, equipment and a medium, and relates to the field of medical image recognition, and the device comprises an image acquisition module, a preprocessing module and a tumor recognition module. The image acquisition module is used for acquiring a to-be-processed gas cavity mirror image; the preprocessing module sequentially performs enhancement and segmentation processing on the to-be-processed gas cavity mirror image; the tumor recognition module adopts a tumor recognition model to classify the preprocessed images; the tumor identification model comprises an encoder, a spatial reconstruction model, a student model and a discriminator; the encoder carries out multiple times of down-sampling on the preprocessed image; the spatial reconstruction model carries out feature space reconstruction on the coded image; the student model adopts a convolution attention mechanism to carry out multiple times of up-sampling on the spatial reconstruction image; and the discriminator compares the difference between the final reconstructed image and the preprocessed image so as to determine the category of the to-be-processed gas cavity mirror image. According to the invention, the processing precision and speed of the gas cavity mirror image are improved.