3598 results about "Low resolution" patented technology

The invention discloses a stainless steel surface defect detection method based on machine vision, and relates to the field of material defect detection.The stainless steel surface defect detection method comprises the steps that stainless steel surface images are collected, the collected multispectral surface images are subjected to image fusion, and a multispectral composite image is generated; performing detail enhancement on the low-resolution image through a super-resolution reconstruction algorithm to obtain a high-resolution image set; positioning a defect area position from the high-resolution image set through a defect detection algorithm, and obtaining a defect position data set; extracting a historical image defect feature set and a real-time image defect feature set from the defect position data set, and constructing a defect classification model according to the historical image defect feature set; inputting the real-time image defect feature set into a defect classification model, and performing defect classification and evaluation to obtain a defect classification data set; and generating and distributing a defect detection report. By combining the multispectral imaging algorithm, the super-resolution reconstruction algorithm and the defect classification model, the automation, precision and efficiency of stainless steel surface defect detection are improved.

The invention relates to the technical field of tunnel engineering, in particular to a tunnel surrounding rock grading method and system, comprising intelligent sensing and data acquisition, multi-source data fusion and modeling, hybrid model dynamic grading, real-time decision and support optimization, online learning and dynamic feedback, and risk early warning and emergency response. Compared with the prior art that a geological data acquisition mode combining manual drilling coring and low-resolution geophysical prospecting is adopted, efficiency is low, subjective errors are large, and a complex geological structure is difficult to cover, unmanned aerial vehicle LiDAR scanning, intelligent rock core image analysis and a high-density IoT sensor network work cooperatively, and the working efficiency is greatly improved. Real-time dynamic acquisition of full-section geological information is achieved, manual intervention errors are eliminated in combination with a multi-source data fusion algorithm, the automation level and three-dimensional space representation precision of data acquisition are remarkably improved, and a high-resolution holographic data base is provided for surrounding rock classification.

The invention discloses a super-resolution image enhancement system and method based on a variational mode decomposition algorithm. The system comprises an adaptive decomposition module, an enhancement processing module, a fusion module and an optimization module. The adaptive decomposition module receives low-resolution image signals, generates modal component signals containing different frequency band characteristics, and outputs modal quantity parameter signals according to image frequency domain energy distribution. The enhancement processing module comprises a high-frequency enhancement unit and a low-frequency reconstruction unit, and generates a high-frequency enhancement signal and a low-frequency reconstruction signal. And the fusion module receives the modal quantity parameter signal, the high-frequency enhanced signal and the low-frequency reconstructed signal, and performs spatial adaptive weighted fusion on the high-frequency signal and the low-frequency signal through a dynamic weight coefficient to generate an initial high-resolution signal. And the optimization module carries out adaptive nonlinear filtering processing on the initial high-resolution signal. The super-resolution image enhancement system based on the variational mode decomposition algorithm can solve the problem that the prior art is difficult to adapt to a complex image structure.

The invention discloses an image super-resolution reconstruction method based on deep learning, and belongs to the technical field of image signal processing. Comprising the following steps: S1, realizing multi-dimensional feature representation of a low-resolution input image by using a deep learning network; s2, realizing accurate alignment and lossless mapping of cross-scale features, and effectively solving the problems of feature dimension inconsistency and information distortion; s3, explicitly learning an increment mapping relation between low-resolution input and a reconstructed high-resolution image, and relieving the gradient disappearance problem of a deep learning network structure; s4, enhancing the naturalness and detail fidelity of the reconstructed image through adversarial training; s5, integrating pixel-level reconstruction loss, perception loss and structural similarity loss, and effectively balancing objective indexes and subjective visual quality of the reconstructed image; and S6, realizing joint optimization of deep learning network parameters through a unified gradient back propagation mechanism, and finally outputting a high-quality super-resolution reconstructed image.

The invention provides a remote sensing strip mine area detection method based on a double-branch structure and a feature fusion mechanism, and the method comprises the steps: obtaining multi-source high-resolution remote sensing image data of a strip mine mining area, carrying out the preprocessing of atmospheric correction, radiation calibration, color correction, image registration, cutting operation and the like, and obtaining time sequence remote sensing image data; a deep learning algorithm is adopted to construct a strip mine area detection model based on a double-branch structure and a feature fusion mechanism, training is carried out through the time sequence remote sensing image data, a remote sensing image detection model is obtained, the double-branch structure comprises a feature extraction branch and a feature generation branch, and the feature extraction branch comprises a feature extraction branch and a feature fusion branch; the feature fusion mechanism comprises a cross attention fusion module and a feature adaptive fusion module; inputting to-be-detected remote sensing image data into the remote sensing image detection model to obtain a detection result of the strip mine mining area. According to the method, the recognition precision of small target details and mining area boundaries of low-resolution images is improved, and the calculation efficiency and the detection precision are both considered.

The invention belongs to the technical field of digital restoration of computer vision and cultural heritage, and particularly relates to a traditional picture restoration method fusing low-resolution prior and efficient visual selection, which comprises the following steps: constructing a multi-source image data set, taking images in the multi-source image data set as high-resolution images, preprocessing the high-resolution images to obtain low-resolution images, and carrying out high-resolution priori and high-efficiency visual selection on the low-resolution images. The high-resolution image and the low-resolution image are respectively masked to generate simulated damage mask images, and the simulated damage mask images comprise a regular damage mask image and an irregular damage mask image; taking the multi-source image data set and the preprocessed multi-source image data set as training data, and training a multi-source image model; the dual-stage repair network comprises a coarse repair network and a fine repair network; according to the method, the problems of structural semantic loss, high priori information dependency and insufficient global and local coordination when an existing image restoration method is used for processing a complex scene and a large-range missing region are solved.

A high-fidelity three-dimensional reconstruction method for a specular reflection plane comprises the steps of decomposing pixel colors into diffuse reflection and specular reflection components based on a 3D Gaussian sphere, introducing a dynamic reflection ratio parameter and a spherical harmonic function coefficient to respectively represent two reflection characteristics, and simulating light multi-reflection behaviors through weight fusion of cumulative projection and a reflection ratio map. Secondly, in combination with monocular inverse depth calibration, depth smoothing constraint of color gradient weighting and edge mutual exclusion loss, geometric consistency is enhanced, and artifacts are suppressed; a progressive multi-resolution training strategy is further adopted, low resolution is gradually optimized to complete resolution, reflection parameters are activated in stages, 3D Gaussian sphere overgrowth and floating artifacts are inhibited, and efficiency and precision are balanced. According to the method, the reconstruction fidelity of the specular reflection scene is remarkably improved while the real-time rendering advantage of the 3DGS is reserved, and the method is suitable for the high-precision modeling fields of virtual reality, augmented reality and the like.

The invention discloses a medical image super-resolution reconstruction method based on multi-level attention guidance, and the method comprises the following steps: S10, constructing a deep learning network model based on a generative adversarial network architecture, which comprises a generator and a discriminator; the generator is based on an improved U-Net architecture, a hierarchical attention module and a dual-path feature processing module are configured in an encoder and a decoder of the generator, the hierarchical attention module adopts different attention strategies according to network levels to consider structure and texture, and the dual-path feature processing module separates and processes low-frequency and high-frequency information; the generator further comprises a multi-level feature fusion module for integrating the multi-scale features of the decoder, and an attention guide up-sampling module for final enhancement and dimension raising. The discriminator adopts a spectrum normalization U-Net architecture and uses multi-scale features for matching; s20, training the network model by adopting a composite loss function comprising pixels, adversarial, perception and total variation loss; and S30, inputting the low-resolution image into the trained model, and outputting a high-resolution image. According to the method, through deep fusion of multi-level attention and multi-scale feature processing, the image restoration quality can be remarkably improved, the texture detail definition can be enhanced, the anatomical structure accuracy can be ensured, and the noise robustness can be improved.

The invention discloses a medical image segmentation method based on high-resolution modal guidance and cross-modal boundary perception, and the method comprises the steps: carrying out the data preprocessing and enhancement of multi-contrast magnetic resonance imaging data, obtaining a boundary mask through a Canny operator and a Dilatation operation, constructing a multi-modal low-resolution data set, and carrying out the recognition of the multi-modal low-resolution data set; meanwhile, a high-resolution T2f modal data set is reserved, and the data set is divided into a training set, a verification set and a test set; a segmentation model is constructed, and the segmentation model comprises a high-resolution mode-guided double-encoder architecture module, a cross-level attention collaboration mechanism module, and a segmentation branch and boundary prediction branch decoder module; designing a training strategy of joint optimization of boundary contour detection and region segmentation, training the segmentation model by using a training set, and storing optimal model parameters on a verification set; and carrying out model performance verification in the test set, and segmenting a to-be-tested medical image by using the verified segmentation model.

The invention relates to the technical field of topographic mapping, in particular to a DEM super-resolution reconstruction method and system combined with a depth map, and the method comprises the steps: constructing a multi-modal data set which comprises an optical image, a depth map, a low-resolution DEM image and a high-resolution DEM image which are paired; designing a multi-modal fusion super-resolution network; designing a pseudo Siamese feature extraction module to extract low-level features and high-level features of the optical image, the depth map and the low-resolution DEM image; fusing the low-layer feature map generated by the pseudo Siamese feature extraction module in a shallow layer through a multi-scale feature fusion module, and fusing the deep-layer feature map in a deep layer; outputting a high-resolution DEM image through a DEM reconstruction module; and training the multi-modal fusion super-resolution network by using the multi-modal data set and the collaborative loss function. According to the method, the depth map generated by the optical image is used as priori knowledge to guide the DEM super-resolution reconstruction task based on the basic depth large model, a complex fine adjustment process is not needed, and the DEM reconstruction precision and efficiency are improved.

The invention provides an image restoration and super-resolution reconstruction system and method based on deep learning, and belongs to the technical field of digital image processing. The invention aims to solve the problems of high calculation complexity and resource consumption, limitation of long sequence processing, high training difficulty and texture scene deficiency when a multi-scale residual network based on a Transform architecture is used for image resolution conversion. The reconstruction system comprises: an image preprocessing module performing window division and video memory optimization on an input low-resolution image; the multi-layer fusion network dynamically adjusts the characteristics of the low-resolution image, captures channel information in different scenes, performs interactive fusion, performs comparison supervision, establishes an information communication channel, dynamically adjusts and optimizes parameters through negative feedback, and obtains a super-resolution image. And the loss function module maximizes the similarity of the super-resolution image and the high-resolution image in the segmentation feature space to obtain a final super-resolution image.

The invention discloses a semantic segmentation method for a low-resolution road scene, and aims to solve the problems of difficulty in small target recognition, fuzzy details, texture information loss and the like existing in a low-resolution image in the conventional semantic segmentation technology. The method comprises the following steps: (1) collecting a low-resolution road scene image and a corresponding semantic tag; (2) constructing a semantic segmentation model consisting of an edge guidance module (BGM), a double-domain feature decomposer (DDFD), a domain alignment attention fusion module (DAAFM) and a double-layer attention context aggregation module (HACAM); (3) designing a joint loss function to carry out multi-scale supervision on semantic regions, edges and middle features; (4) carrying out model training by utilizing the road scene image; and (5) outputting a semantic segmentation result map and an edge prediction map. The boundary perception capability is enhanced by introducing learnable pixel difference convolution, the extraction precision of a small target and a global structure is improved by combining frequency domain and spatial domain feature alignment, and context semantic relationship expression is optimized by fusing a channel and a spatial attention mechanism. The method effectively improves the semantic segmentation precision and boundary restoration capability of the model in a low-resolution complex road environment, and is suitable for intelligent analysis tasks of road images in scenes of automatic driving, intelligent traffic, severe weather and the like.

The invention discloses a plateau mountain road disaster identification method and system based on multi-source remote sensing image restoration and super-resolution reconstruction. The method comprises the following steps: acquiring and preprocessing a multi-source remote sensing image of a plateau mountain region, and extracting landform measurement parameters based on a digital elevation model; a super-resolution reconstruction network fusing deformable convolution and Transform is constructed, and a low-resolution image is reconstructed by using constraint training of a composite loss function containing geomorphic measurement parameters; performing feature extraction and adaptive weighted fusion on the preprocessed image and the reconstructed high-resolution image; based on the fused image, utilizing a multi-task deep learning model to identify landslide, debris flow and roadbed subsidence disasters along the highway; and carrying out morphological optimization and boundary refinement under GIS constraint on an identification result, and outputting a disaster thematic map. According to the invention, the precision and reliability of road disaster identification in a complex terrain environment are effectively improved.

The invention relates to the technical field of forest fire recognition and early warning, in particular to a forest fire early warning method and system based on a super-resolution neural operator, and the method specifically comprises the following steps: extracting an initial feature map # imgabs1 # of a to-be-recognized low-resolution forest fire image # imgabs0 #; constructing a super-resolution feature capture network, three HAFM hierarchical attention fusion modules, an MSMO multi-scale mapping operator module, a super-resolution feature fusion module and a super-resolution image reconstruction module, inputting the initial feature map # imgabs2 # into the super-resolution feature capture network, and finally outputting a super-resolution image # imgabs3 #; and constructing a forest fire recognition module, inputting the super-resolution image # imgabs4 # into a ResNet-18 neural network model to obtain a recognition feature # imgabs5 #, finally performing full connection layer processing on the recognition feature # imgabs6 # to obtain a recognition result of the forest fire image, and judging whether to perform early warning or not according to the recognition result. According to the invention, the problem that a low-resolution fire image cannot be accurately identified and timely early warned in existing forest fire identification and detection can be solved.

The invention discloses a feature integration method for interactive convolution and dynamic focusing of infrared images. The feature integration method comprises the steps that infrared image pairs with different resolutions in various real scenes are obtained through an infrared camera; performing degradation preprocessing on a part of original high-resolution images to obtain low-quality high-resolution images to form a mixed low-resolution data set, and dividing the processed data set into a training set and a test set; constructing a double-layer feature extraction module for feature modeling; training a network by using the processed training set, and optimizing a loss function; and inputting a low-resolution infrared image into the trained network, and outputting a high-resolution reconstruction result. According to the method, local and global features are fused, so that the super-resolution reconstruction quality of the infrared image in complex scenes such as low contrast and fuzzy edges is remarkably improved, and meanwhile, relatively high calculation efficiency is kept.

The invention provides a processing method and system for reconstructing an LDI sub-pixel image through sparse compression and application, and the method comprises the steps: calculating a low-resolution to-be-exposed image corresponding to each phase structure in advance through a computer image compression algorithm; the low-resolution to-be-exposed images obtained through calculation are loaded to the digital micro-reflector according to the time sequence, and the display moment of each pattern is matched with the rotation position of the corresponding phase structure; coupling the pattern of the digital micro-mirror and the phase encoding mask in a frequency domain through a 4-f optical system, and reconstructing a high-resolution exposure pattern at a sub-pixel level by utilizing a diffraction effect; projecting a high-resolution light field of the reconstructed high-resolution exposure pattern to the surface of the photoresist, and forming a target circuit pattern by accumulating exposure dose; and the micro-nano structure with sub-pixel precision is obtained after development. The system comprises an image compression module, a pattern matching module and a pattern forming module. According to the invention, the manufacturing cost of the laser direct imaging equipment is reduced under the condition of the same precision.

The invention belongs to the technical field of remote sensing super-resolution images, and particularly relates to a remote sensing image super-resolution system and method based on an adaptive Mamba-attention network. Comprising a feature extraction module used for carrying out shallow feature extraction on an input low-resolution image to obtain shallow features; the multiple cascaded adaptive state space blocks are used for processing the shallow layer features to obtain reconstruction features; and the reconstruction module maps the reconstruction features to a target resolution space through sub-pixel rearrangement operation to obtain a high-resolution remote sensing image. High-frequency details and a low-frequency structure are cooperatively processed in a feature space by using the remote sensing frequency sensing modulation module, and high-resolution output is generated by combining sub-pixel rearrangement up-sampling, so that high-quality reconstruction of a complex remote sensing scene is realized.

The invention discloses a target detection method, system and equipment based on knowledge distillation and a medium, belongs to the technical field of computer vision, machine learning and artificial intelligence, and aims to solve the technical problem of how to overcome the limitation of the existing knowledge distillation technology in smart city target detection application. According to the technical scheme, the method comprises the steps of hierarchical attention fusion: in combination with local attention and global attention mechanisms, capturing detail information of a target through a high-resolution feature map, extracting overall features of the target through a low-resolution feature map, and carrying out hierarchical attention fusion; a final attention feature map is generated through a self-adaptive fusion strategy, and the focusing capability of a student model on target features is enhanced; carrying out weighted fusion on the fusion strategy in combination with a channel attention mechanism, and further enhancing the expression ability through a lightweight convolutional network, thereby improving the expression ability of a student model for target features; improving a distillation loss function; optimizing a training strategy; and network training.

The invention discloses a metal structural part surface damage identification method based on machine vision, and belongs to the field of machine vision, and the method comprises the steps: obtaining reference image data with known damage features, carrying out the preprocessing, analyzing the change trend of a system detection state, and judging whether there is a deviation correction demand or not. And if the deviation exists, carrying out geometric correction processing on the lens distortion error to obtain a corrected reference image. Further separating the real change of the damage from the system deviation, and combining low-resolution and high-resolution detection to obtain the distribution data of the suspected damage area and the specific characteristic parameter data of the damage. According to the method, quantitative data of damage levels are obtained through automatic classification, detection differences among multiple devices are calibrated, visual presentation information of damage positions and levels is generated, and finally camera parameters and algorithm thresholds for subsequent detection are optimized and adjusted, so that high-precision damage detection and evaluation are realized.

The present application relates to the field of computer vision. Provided are a visual encoding method and apparatus, and a visual encoding model training method and apparatus, which are used for using the same visual encoding model to encode images of different resolutions, and are applied to encoding scenarios for images of more sizes. The visual encoding method comprises: first, acquiring an input image, wherein the input image may be a high-resolution image and may also be a low-resolution image; and then inputting the input image into a visual encoding model, so as to output visual encoding data, wherein the visual encoding model is used for dividing the input image into a plurality of image blocks according to positional embedding, extracting features from each image block, and outputting visual encoding data on the basis of the features of each image block and corresponding positional encoding, the positional embedding is obtained by means of adjusting initial positional embedding on the basis of the difference between the input image and a preset resolution, and the positional embedding may specifically comprise a matrix corresponding to the division of the input image

Approaches presented herein provide for visual localization by matching features of a query image with features obtained from representation of a three-dimensional (3D) environment. A model such as a neural radiance field (NeRF) can be trained to represent the 3D environment. When a query image is received, query features can be extracted at two different resolutions. A lower resolution set of query features can be compared against NeRF descriptor features for a set of training images, to narrow the search space by finding a set of coarse matches. Higher resolution query features can then be compared against sampled features of these coarse matches, to identify 2D-3D correspondences that can be used to calculate camera pose information for the query image.

Systems and methods for upsampling low-resolution content within a high-resolution image include obtaining a composite image and a mask. The composite image includes a high-resolution region and a low-resolution region. An upsampling network identifies the low-resolution region of the composite image based on the mask and generates an upsampled composite image based on the composite image and the mask. The upsampled composite image comprises higher frequency details in the low-resolution region than the composite image.

The invention relates to the technical field of remote sensing hyperspectral images, in particular to a remote sensing hyperspectral image super-resolution reconstruction method based on a hybrid neural network, and the method comprises the steps: firstly carrying out the preprocessing of a remote sensing hyperspectral image data set, and obtaining a training set; constructing a hybrid neural network with a spectrum hypergraph branch, a space hypergraph branch, a semantic hypergraph branch, a sensitive wave band extraction branch and a recovery and reconstruction branch; training the hybrid neural network by using a training set to obtain a hybrid network model; and finally, inputting a to-be-reconstructed low-resolution remote sensing hyperspectral image into the hybrid network model to obtain a corresponding reconstructed remote sensing hyperspectral image. According to the method, space, spectrum and semantic feature information are fully combined, and a main body structure and detail change characteristics in an image are captured; the sensitive wave band extraction branch is provided to enhance cross guidance and fusion among multi-domain features, and the effect of super-resolution reconstruction is fully improved.

The invention relates to the technical field of metal defect detection, and discloses an aluminum alloy round aluminum rod surface defect image super-resolution method. The method comprises the following steps: acquiring a low-resolution original image sequence of surface defects of the aluminum alloy round aluminum rod, and synchronously acquiring gray value distribution at different illumination angles through a multi-channel optical sensor; constructing a dynamic degradation model according to pixel displacement of adjacent frames in the original image sequence, extracting cross-scale defect features in the original image sequence, and taking output parameters of the dynamic degradation model as spatial constraint conditions of a feature extraction network; and a high-resolution defect image is generated through the multi-stage residual error reconstruction network, the high-resolution image output by the reconstruction network is fed back to the dynamic degradation model, and the frequency domain response coefficient of the spatial fuzzy kernel function is updated to form closed-loop optimization. The identification degree of defect features is improved, and a reliable image data basis is provided for accurate detection of the surface defects of the aluminum alloy round aluminum rod.

The invention provides a multi-band radar data space downscaling method based on data fusion. The method comprises the following steps: step 1, acquiring S-band and X-band observation data from a weather radar; step 2, carrying out data preprocessing on the obtained observation data; step 3, constructing a spatial downscaling model; 4, performing training verification and testing on the spatial downscaling model through the observation data after data preprocessing; step 5, according to the trained spatial downscaling model, performing spatial downscaling of the S-band observation data; and step 6, judging whether the generated data is close to a true value or not. By means of the deep learning neural network, the model can extract and fuse the characteristics of the S-band observation data and the X-band observation data, the original low-resolution S-band data is converted into high-resolution data, the requirement for high resolution of radar data in practical application is met, and dependence on hardware facilities is reduced.

The invention relates to a small focus segmentation and feature learning method based on Transform and CNN fusion. The method comprises the following steps: preprocessing an input three-dimensional image to obtain original input features, and inputting the original input features into a segmentation network model of an encoder-bottleneck-decoder structure; after passing through an encoder based on channel-Transform fusion, obtaining fusion features through a bottleneck used for expanding a receptive field; wherein the encoder comprises a channel convolution module for local feature capture and a local attention mechanism module for global modeling; the preprocessed input features pass through a channel convolution module and a local attention mechanism module in sequence, and then pass through a bottleneck containing cavity convolution to carry out feature fusion; and inputting the fused features into a decoder based on a jump connection mechanism to gradually restore the low-resolution features to the same size as the input image, and generating a final segmentation result or classification result. According to the method, efficient feature extraction and fine prediction of three-dimensional images such as medical images can be realized.

The present invention relates to the technical field of image restoration, and in particular to a remote-sensing hyperspectral image super-resolution reconstruction method based on a hypergraph neural network, comprising: S1, acquiring a hyperspectral image, and preprocessing the hyperspectral image to obtain a training set and a verification set; S2, constructing a hypergraph neural network; S3, using training images in the training set to construct a three-layer hypergraph, and on the basis of the three-layer hypergraph, constructing hypergraph operators corresponding to the training images; S4, using the training images and the hypergraph operators corresponding to the training images to train the hypergraph neural network, and using a loss function to iterate network parameters of the hypergraph neural network, to obtain a trained hypergraph neural network; and S5, inputting, to the trained hypergraph neural network, low-resolution hyperspectral images to be reconstructed for reconstruction to obtain a high-resolution hyperspectral image. The present invention has an excellent reconstruction result.

The present invention relates to the field of defect detection technology, and in particular to an OLED screen defect detection method based on optical-magnetic multimodal fusion, comprising the following steps: S1: synchronous excitation and data acquisition based on pulsed current drive; S2: cross-modal spatial calibration and noise suppression; S3: polarization-magnetic field coupling feature extraction; S4: dynamic correlation decision-making and closed-loop feedback, including realizing defect classification determination through correlation analysis of multimodal data, and verifying the authenticity of defects through current overload testing. This method effectively solves the problems of ambient light interference, missed detection, high cost, low resolution, etc. existing in the prior art, has significant beneficial effects, can improve detection accuracy, efficiency and reliability, meet the needs of large-scale production lines, and can efficiently locate and reduce errors and redundant steps in the detection process.

The invention provides a CT (Computed Tomography) image super-resolution reconstruction method based on a generative adversarial network. The method comprises the following steps of: 1, collecting high-resolution image data and corresponding low-resolution image data; 2, establishing a CycleGAN model comprising a first generator, a second generator, a first discriminator and a second discriminator; a third step of training a CycleGAN model by using the collected high-resolution image data and the corresponding low-resolution image data so as to optimize parameters of the CycleGAN model; and a fourth step of performing image conversion on the to-be-enhanced medical image which is input into the trained CycleGAN model.

The invention discloses a non-contact heart rate detection method, system and device based on visual Transform and multi-scale feature aggregation and a medium. The method comprises the steps that a face visible light video is obtained, the unified video frame rate is re-sampled through the frame rate, face key point positioning and region division are carried out on each frame of image, and a face visible light video sequence is obtained; performing time migration operation on the video sequence to obtain a difference frame sequence, performing channel fusion on the video sequence and a difference frame, and performing down-sampling on a spatial dimension to obtain a low-resolution feature tensor; constructing a non-contact heart rate extraction model, inputting the low-resolution feature tensor into the non-contact heart rate extraction model, and outputting a heart rate value of the target; the non-contact heart rate extraction model comprises a feature enhancement module, a multi-scale mask feature aggregation module, a Transform time sequence modeling module and an rPPG predictor, the system, the device and the medium are used for achieving the non-contact heart rate detection method based on visual Transform and multi-scale feature aggregation, and the precision of rPPG signal extraction is improved.