101results about How to "Keep details" patented technology

The invention discloses a generative adversarial network sonar image denoising method and device fused with comparative learning, and the method comprises the steps: constructing a generative network, inputting a noisy image, and carrying out the image denoising processing through a fusion attention mechanism module and a comparative learning module; constructing a discrimination network, inputting a noise-free image without noise and a de-noised image generated by the generation network, and carrying out image discrimination; s3, training an adversarial network composed of the generative network and the discriminant network, and storing final parameters after training is finished; and S4, inputting the noise image into the trained generative network to obtain a denoised image. The fusion attention mechanism module and the contrast learning module can improve the ability of the generative network to understand and process the input image, enhance the ability of the network to capture and retain the image details, and accurately restore the main body features and detail information of the noise image.

The invention belongs to the field of CT image processing, and particularly relates to a CT image reconstruction method based on a three-dimensional diffusion model and a related device.The method comprises the steps that a medical CT scanner is adopted to obtain three-dimensional CT voxel data, and the three-dimensional CT voxel data is preprocessed; extracting a medical object surface point cloud from the preprocessed three-dimensional CT voxel data by adopting a voxel-level reconstruction algorithm, and generating three-dimensional sparse point cloud data; processing the three-dimensional sparse point cloud data by adopting a 3D Gaussian splashing technology, and converting the three-dimensional sparse point cloud data into three-dimensional Gaussian distribution data by using a pre-trained three-dimensional Gaussian kernel; gradually adding Gaussian noise to the three-dimensional Gaussian distribution data by adopting a three-dimensional diffusion model, and gradually removing the Gaussian noise to generate reconstructed three-dimensional Gaussian distribution data; and projecting the reconstructed three-dimensional Gaussian distribution data to a two-dimensional plane by adopting a rendering technology to generate a two-dimensional CT reconstructed image. According to the method, the advantages of the 3D Gaussian splashing technology and the three-dimensional diffusion model are combined, efficient, real-time and denoising medical image reconstruction can be achieved under the low-dose CT imaging condition, and meanwhile the problem of white floater artifacts possibly occurring in the reconstruction process is effectively solved.

The invention discloses a glaucoma distinguishing detection method and system based on multi-modal fusion. The method relates to the technical field of glaucoma distinguishing detection, and comprises the following steps: collecting glaucoma distinguishing detection data; analyzing glaucoma distinguishing detection data; and comprehensively analyzing, optimizing and adjusting the glaucoma distinguishing detection data quality accuracy. The glaucoma distinguishing detection data is collected and processed, the glaucoma distinguishing detection data is analyzed, the glaucoma distinguishing detection data integrity evaluation index and the glaucoma distinguishing detection data image quality evaluation index are obtained, and the glaucoma distinguishing detection data quality accuracy method is comprehensively analyzed, optimized and adjusted. The accuracy of the glaucoma distinguishing and detecting method and system based on multi-modal fusion is improved, and the problem that the accuracy of the glaucoma distinguishing and detecting method and system based on multi-modal fusion is insufficient in the prior art is solved.

The invention relates to the technical field of images, and particularly provides a video-level spectral imaging system and imaging method based on compressed sensing, and the video-level spectral imaging system based on compressed sensing is used for shooting a target to obtain an original multi-spectral image. Each pixel in the original multispectral image comprises spectral information of one wave band, the spectral information corresponds to the spatial position of a spectral channel in the light splitting element, and the original multispectral image is subjected to convolution operation to obtain a preliminary pseudo panchromatic image; performing sharpening operation on each pixel of the preliminary pseudo panchromatic image to obtain a pseudo panchromatic image; obtaining a demosaicing residual error based on the original multispectral image and the pseudo panchromatic image; and finally, restoring the pseudo panchromatic image by using the demosaiced residual error to obtain a demosaiced single-spectrum image of each wave band. The spectrum channels are arranged on the target surface of the light splitting element in a pseudo-random mode, and all spectrum images are accurately and efficiently restored.

The invention relates to a power transmission line fire smoke identification method based on improved YOLOv9. The method comprises the following steps: constructing a fire smoke image data set; constructing a power transmission line fire smoke detection model; inputting the training set into the power transmission line fire smoke detection model for training, and storing the optimal power transmission line fire smoke detection model obtained through training; and inputting a to-be-identified power transmission line fire smoke picture into the optimal power transmission line fire smoke detection model to obtain a power transmission line fire smoke picture detection result. According to the method, depth separable convolution is introduced, so that the calculation amount can be reduced, parameters are reduced, and the network detection speed is increased; a parallel weighted mixed domain attention module is adopted to enhance extraction of target features in the fire smoke image; the up-sampling content perception feature recombination network structure CARAFE has a larger receptive field, the perception of important content is enhanced, and the lightweight level of the model can be kept due to the fact that the operation amount is small.

The invention provides a feature weight adaptive super-pixel segmentation method based on a sonar image. The method comprises the following steps: Step 1, converting an original sonar image into a grayscale image; step 2, denoising processing is carried out on the image by using guide filtering; step 3, extracting specific features of the sonar image; step 4, carrying out superpixel segmentation preprocessing of automatic adjustment of feature weights; step 5, carrying out maximum flow-minimum cut superpixel aggregation; step 6, carrying out aggregation on the super pixels through a preset threshold value; and Step 7, separating the target from the background to obtain a segmentation result. According to the method, a segmentation technology suitable for the target object area of the acoustic image can be realized, and a better image foundation is laid for subsequent target object disease type judgment and volume and area measurement through analysis of the segmented area, so that segmentation is more accurate, the type of the target object is easier to judge, and the size of the disease is more accurately measured.

The invention provides a bridge crack identification method and system, and the method comprises the steps: carrying out the preprocessing of image data of a target bridge, so as to obtain a processed image, and carrying out the foreground segmentation of the processed image, so as to obtain a segmented image; performing first feature extraction on the segmented image to obtain a first feature image; performing second feature extraction on the segmented image to obtain a second feature image; performing fusion processing on the first feature image and the second feature image to obtain a fused image; and inputting the bridge training image into a preset crack recognition model for training, and inputting the fused image into the trained preset crack recognition model for crack recognition to obtain a crack recognition result, so that the foreground segmentation accuracy can be improved, meanwhile, the interference of noise on the image can be effectively avoided, and the image recognition accuracy is improved. The degree of distortion is small, and detail information in the image can be effectively reserved, so that the accuracy of subsequent model crack recognition is improved.

The invention belongs to the technical field of electronic information and communication, and discloses a self-adaptive signal transmission path optimization system of a chip, which comprises a data acquisition module used for acquiring monitoring video data, chip internal parameter data and chip transmission environment data; the data processing module is used for preprocessing the acquired monitoring video data, chip internal parameter data and chip transmission environment data to obtain a video feature data set, a chip internal parameter feature data set and a chip transmission environment feature data set; analyzing the influence of the chip transmission environment feature data set on the video feature data set and the chip internal parameter feature data set to obtain a transmission influence feature data set; the transmission error analysis module is used for training and obtaining a transmission error prediction model according to the transmission influence feature data set, and obtaining errors generated in the chip internal transmission process of the monitoring video signals based on the transmission error prediction model; the reliability of signal transmission is improved, and the chip resource allocation is more efficient.

The present invention belongs to the field of image defogging technology, and proposes an adaptive image defogging enhancement method and system in a dusty and foggy environment in a mine. First, an image processing model including an image grading preprocessing module, an adaptive processing module and an optimization module is established; the image processed by the image grading preprocessing module is input into the adaptive processing module; the image processed by the adaptive processing module is further optimized by the image optimization module; and the optimized image is output. In practical application, the present invention can accurately, efficiently and stably defog images in a dusty and foggy environment in a mine, significantly improving the image quality and the ability to identify safety hazards in coal mines.

The invention discloses an anti-shake method for an image transmission video of an unmanned aerial vehicle. The anti-shake method comprises the following steps: step 1, data acquisition; step 2, multi-frame optical flow analysis; step 3, feature extraction and jitter judgment; step 4, generating a jitter-free frame by the generative adversarial network; and step 5, post-treatment and synthesis. And step 6, performing real-time feedback and optimization. Through combination of multi-frame optical flow analysis and deep learning feature extraction, different types of jitters can be accurately identified. The accurate jitter judgment not only improves the reliability of detection, but also provides an important basis for subsequent processing, so that the video quality can be effectively guaranteed in practical application no matter in a complex environment or in a rapid movement condition, and the scheme can remarkably improve the video quality on the premise of keeping details and definition. The visual effect of the image is further enhanced through subsequent color correction and sharpening processing, so that the finally output video is smooth and has higher watching value.

The invention discloses a high dynamic range image tone mapping method and device, a medium and a program product, and relates to the field of image processing. The method comprises the following steps: calculating a brightness logarithm and a large-scale texture layer logarithm of a high dynamic range image to enhance image details and generate a global detail enhanced image; the brightness distribution is optimized by counting the brightness histogram, cutting the abnormal brightness value and compensating the abnormal brightness value; determining the average brightness and the standard deviation of the low dynamic range image based on the optimized histogram, obtaining a normalized brightness grey-scale map, and performing local detail enhancement; and restoring the brightness of the enhanced image to an original HDR range, mapping a color channel, and merging color spaces to obtain a final LDR image. According to the method, the dynamic range of the high-dynamic-range image can be compressed, and meanwhile, the detail features and brightness distribution reasonability of the image are effectively kept, so that the subjective visual quality and user experience of the low-dynamic-range image are remarkably improved.

The invention discloses a small unmanned aerial vehicle intrusion detection method based on improved YOLOv8, and relates to the technical field of unmanned aerial vehicle intrusion detection. The invention provides an improved YOLOv8 network model, an SPD-Conv layer and a DP-MobileNetv3 layer are introduced to a Backbone module of the YOLOv8 network model, the efficient feature extraction capability is maintained, the parameter quantity and calculation complexity of the model are reduced, the size of the YOLOv8 model is further reduced, redundant information in the feature extraction process is eliminated, details of a small target of an unmanned aerial vehicle are reserved, and the accuracy of feature extraction is improved. And the accuracy and detection speed of the model are improved.

The invention discloses a weak supervision defogging method based on uncertainty driving, and the method employs a weak supervision defogging network of a U-Net architecture, and the weak supervision defogging network comprises a regional dynamic attention module RDAM, a lower sampling layer, a multi-scale feature fusion module MSFFM, and an upper sampling layer. The RDAM is used for processing an input image or the output of a previous layer of RDAM, constructing a dynamic convolutional layer and obtaining a regional dynamic attention feature map through the dynamic convolutional layer; the down-sampling layer is used for down-sampling the output of the RDAM and reducing the spatial resolution; the MSFFM is used for processing input features through a window multi-head self-attention mechanism so as to capture a long-distance dependency relationship between pixels and performing multi-scale feature extraction and fusion; the upper sampling layer is used for recovering the feature map output by the MSFFM to the resolution ratio of the feature map output by the MSFFM of the upper layer and performing feature splicing with the feature map output by the MSFFM of the upper layer; and enabling the up-sampled feature map to sequentially pass through a 3 * 3 convolutional layer, a ReLU activation function and a 1 * 1 convolutional layer, and finally outputting a defogged image.

The invention provides a high-signal-to-noise-ratio long-wave infrared multispectral imaging system and an imaging method thereof, which are used for detecting and identifying infrared weak and small targets under a dense fog condition. A dark channel prior method is adopted to carry out defogging processing on each acquired spectrum image so as to eliminate imaging quality degradation caused by atmospheric scattering and absorption; a block matching technology is used for carrying out block matching on a defogged image, similar blocks in a multispectral image are extracted, a collaborative filtering technology is combined for processing the similar blocks, noise is removed, the contrast ratio of the image is enhanced, and the signal-to-noise ratio is doubled. According to the multispectral imaging system, the imaging quality can be effectively improved by utilizing spectral feature differences of different objects and combining an image processing technology, and the multispectral imaging system has high anti-interference performance, is beneficial to foggy day target detection and recognition and has wide application prospects.

The invention discloses a physical intelligent magnetic resonance diffusion weighted image denoising and quantifying method, and relates to diffusion magnetic resonance weighted imaging. Comprising the following steps: 1) acquiring diffusion weighted magnetic resonance data acquired by using a multi-diffusion-direction planar echo imaging sequence; 2) designing a physical intelligent magnetic resonance diffusion weighted image denoising and quantification deep learning network model and a loss function; 3) using the training data set obtained in the step 1) to train a physical intelligent magnetic resonance diffusion weighted image denoising and quantification deep learning network model; and 4) inputting diffusion weighted data to be denoised into the trained network to generate a final denoising and quantification result. Through combined application of physical regularization loss and low-b-value data, the problem of low signal-to-noise ratio in magnetic resonance diffusion weighted imaging is effectively solved, image denoising is realized, and an accurate quantitative result is further provided. And the image quality and the ADC quantitative accuracy can be maintained and even improved while the acquisition times are reduced.

The invention discloses an infrared image non-uniformity correction method based on notch filtering and guide filtering, which comprises the following steps of: inputting an infrared focal plane array image, performing discrete Fourier transform on the input image to obtain an amplitude spectrum and a phase spectrum, predicting a central frequency spectrum area of the amplitude spectrum, estimating a notch filter compensation parameter, and correcting the infrared image non-uniformity. Further constructing a notch filter; enabling the amplitude spectrum to pass through the constructed notch filter to obtain a new amplitude spectrum after filtering, and obtaining a filtered background image layer through inverse discrete Fourier transform by utilizing the new amplitude spectrum after filtering and the original phase spectrum; carrying out subtraction on the original input infrared focal plane array image and the filtered background image layer to obtain a noise image, carrying out weighted improvement on a regularization coefficient by utilizing anisotropic guided filtering, and extracting a part of detail image layers from the noise image; and superposing all the detail layers, and outputting a correction result. The method can effectively overcome the problem that image details are lost in the non-uniform correction process.

The invention relates to the technical field of artificial intelligence, in particular to an intelligent monitoring method and device and a storage medium. The intelligent monitoring method comprises the steps of driving a video acquisition module to perform video acquisition, obtaining at least one video frame, and performing scene recognition based on the at least one video frame to obtain a current application scene; when the current application scene is the privacy scene, driving an infrared sensor to perform thermal imaging acquisition to obtain at least one thermal imaging frame, and closing a video acquisition module; improving a YOLOv7-tiny model to obtain a thermal imaging model, replacing an original spatial pyramid pooling module with a spatial pyramid pooling module based on a channel attention mechanism, and adding a convolution module based on a Meta-ACON activation function; and inputting the at least one thermal imaging frame into a thermal imaging model for scene recognition to obtain the current actor state. According to the invention, only the current actor can be monitored and identified in the privacy scene, and the privacy information of the user is effectively protected.

The present invention relates to the field of drone control technology, and specifically to a drone swarm collaborative control system and method in a complex environment based on image recognition. The system preprocesses collected real-time environment image data of the drone swarm to generate real-time environment image feature data of the drone swarm, builds real-time environment three-dimensional model data of the drone swarm based on the real-time environment image feature data, and matches the data with the collected preset flight route data of the drone swarm. According to the matching result, preset flight route adjustment data of the drone swarm is generated. If the preset flight route adjustment data of the drone swarm needs to be adjusted, the flight route of the drone swarm is planned by using a constructed drone swarm flight route planning model to generate drone swarm flight route planning data. Finally, drone swarm collaborative control data is built, and drone swarm flight route adjustment operations are performed to achieve collaborative control of the drone swarm and scientific adjustment of the flight route.

The invention relates to a monitoring ball signal extension system and method based on a multi-mode networking module, and belongs to the technical field of power transmission and transformation construction safety monitoring. The system comprises a signal detection module, a networking mode selection module, a signal processing module, a signal transmission module, a monitoring system integration module and an external signal enhancement device. The signal detection module monitors signal intensity and interference conditions; the networking mode selection module automatically selects a networking mode in sequence; the signal processing module optimizes the video signal; the transmission module transmits signals; the monitoring system integration module interacts with the monitoring platform; the external signal enhancement device comprises an antenna gainer and a signal repeater. The system can realize automatic networking, adapts to a complex environment, ensures stable transmission of signals, improves imaging quality, enhances monitoring flexibility, optimizes data processing, is suitable for safety monitoring of a power transmission and transformation construction site, and ensures smooth engineering.

The invention discloses an ice condition video image recognition method based on TS-MaskFormer. A rapid ACE defogging algorithm is adopted to enhance a night image, and an ice condition image detection and identification data set covering different stages of freezing, starting of ice melting, floating of floating ice and the like is constructed based on a semi-supervised labeling method of SAM; a TS-MaskFormer ice condition image recognition model is provided, a MaskFormer model and a fusion module of two different backbone networks are integrated, and accurate ice condition image detection is achieved by training, adjusting and optimizing model parameters. And identifying an ice condition image based on the optimized model, and calculating an ice surface coverage rate through an identification result. According to the method, more detail information can be reserved in the aspect of ice condition image recognition, and higher ice surface recognition precision and higher calculation speed are achieved.

The invention relates to the technical field of unmanned aerial vehicle control, in particular to an unmanned aerial vehicle group cooperative control system and method in a complex environment based on image recognition, and the system comprises the steps: carrying out the data preprocessing of the collected real-time environment image data of an unmanned aerial vehicle group, and generating the real-time environment image feature data of the unmanned aerial vehicle group; and constructing unmanned aerial vehicle group real-time environment three-dimensional model data based on the unmanned aerial vehicle group real-time environment image feature data, matching the unmanned aerial vehicle group real-time environment three-dimensional model data with the collected unmanned aerial vehicle group preset flight route data, generating unmanned aerial vehicle group preset flight route adjustment data according to a matching result, and if the unmanned aerial vehicle group preset flight route adjustment data needs to be adjusted, determining that the unmanned aerial vehicle group needs to be adjusted. And if yes, performing flight route planning on the unmanned aerial vehicle group through the constructed unmanned aerial vehicle group flight route planning model, generating unmanned aerial vehicle group flight route planning data, finally constructing unmanned aerial vehicle group cooperative control data, and executing unmanned aerial vehicle group flight route adjustment operation, thereby realizing cooperative control of the unmanned aerial vehicle group and scientific adjustment of the flight route.

The invention relates to an image denoising method, a model training method and device, electronic equipment and a computer program product, relates to the technical field of image processing, and can be applied to a scene for denoising an image containing noise. The method comprises the steps of obtaining a to-be-processed image, and performing decomposition transformation processing on the to-be-processed image to obtain a plurality of frequency sub-graphs; performing feature extraction processing based on the plurality of frequency sub-images to obtain intermediate image features; performing up-sampling processing on the intermediate image features to obtain recovered image features, the up-sampling processing being an inverse process of decomposition transformation processing; and generating a de-noised image according to the restored image features and the to-be-processed image. According to the method, video memory occupation in the image processing process is effectively reduced, the calculation efficiency of the model is improved, and meanwhile, detail and texture information of the image is reserved; moreover, the image denoising effect is optimized, and the information loss in the denoising process is reduced.

The invention discloses a Chinese character recognition method based on overall semantic feature extraction, and the method employs an Encoder-Decoder architecture, and comprises the steps: carrying out the preprocessing of an input image; performing feature extraction and down-sampling on the preprocessed input image through a convolutional neural network to obtain local feature maps of different depth levels of the input image; feature fusion: extracting feature vectors of different levels in the local feature map by using Transform Blocks of different depths, and splicing the local feature map and the feature vectors of different levels on a channel dimension to form spliced feature vectors; and decoding: carrying out autoregressive decoding on the spliced feature vectors by using Transform Decoder, and generating an identification result. The method has high efficiency, accuracy and adaptability in a Chinese character recognition task, meets diversified recognition requirements, and has very high practical value and popularization potential.

The present application relates to the technical field of fault location of power systems, and specifically to a primary and secondary fusion ring network cabinet with intelligent fault location function, which includes: a ring network cabinet traveling wave signal acquisition module, which collects traveling wave signals when a transmission line fails; a ring network cabinet traveling wave signal analysis module, which divides the traveling wave signal into equal parts; according to the fluctuation degree of the signal amplitude in the signal segment, combined with the energy spectrum of the signal segment, the mutation coefficient is obtained; according to the frequency domain characteristics and frequency domain shape characteristics of the energy spectrum, the low-frequency concentration of energy is obtained; the filter window size of each sampling point in the signal segment is obtained; the ring network cabinet traveling wave fault location module filters the traveling wave signal, locates the fault, and issues instructions to the primary and secondary fusion switch. The present application aims to effectively remove the pulse noise in the traveling wave signal, retain the detailed information of the fault traveling wave signal, and enhance the accuracy of fault location.

The invention discloses an X-ray and NeRF algorithm-based luneberg lens internal structure analysis method, and relates to the technical field of three-dimensional imaging of optical lenses, and the method comprises the following implementation steps: S1, configuring an RGB-X-ray combined imaging system, and carrying out data acquisition, and S2, carrying out the calibration of external parameters of a camera through a calibration plate: carrying out the imaging through a specially-made calibration plate, and carrying out the calibration of the external parameters of the camera through the calibration plate. The method comprises the following steps: acquiring a plurality of frames of RGB (Red, Green, Blue) images to obtain external parameters of an RGB camera under a calibration plate coordinate system, acquiring a plurality of frames of X-ray images to obtain an external parameter matrix of an X-ray camera about the calibration plate coordinate system, performing implicit representation on a continuous field in a lens by utilizing NeRF, mapping three-dimensional coordinates to differentiable density and radiation values, and calculating the calibration plate coordinate system. According to the method, high-fidelity reconstruction from multi-angle X-ray projection to a three-dimensional structure is achieved, detail information in the lens can be effectively reserved, meanwhile, the continuity and symmetry of the structure are guaranteed, and compared with a traditional voxelization or CT reconstruction method, the problems that noise artifacts, fractures and unstable numerical values are likely to be generated in the traditional voxelization method are solved.

The invention discloses a geometric constraint-driven complex roadway body point cloud dynamic extraction method. The method comprises the following steps of S1, collecting three-dimensional point cloud of a roadway by adopting three-dimensional laser scanning equipment; s2, constructing a coordinate system according to the longitudinal central axis of the roadway, and realizing coordinate transformation operation of the three-dimensional point cloud; s3, a virtual cloth grid is constructed on the periphery of the three-dimensional point cloud, and each cloth particle in the virtual cloth grid dynamically shrinks from outside to inside under the combined action of radial attraction pointing to the X axis of the coordinate system and internal elastic force maintaining the shape of the virtual cloth grid; through multiple iterations, the virtual material distribution grids can be attached to the surface of the inner wall of the roadway; calculating the shortest distance from each point in the three-dimensional point cloud to the surface of the virtual distribution grid, and performing classification operation on roadway body points and non-body points in the three-dimensional point cloud; and S4, carrying out refined processing on the roadway body points in the three-dimensional point cloud obtained after classification, and finally outputting a roadway body model.

The present invention discloses a method for detecting the intrusion of a small unmanned aerial vehicle based on an improved YOLOv8, and relates to the technical field of unmanned aerial vehicle intrusion detection. The present invention proposes an improved YOLOv8 network model, introduces an SPD-Conv layer and a DP-MobileNetv3 layer into the Backbone module of the YOLOv8 network model, reduces the number of parameters and the computational complexity of the model while maintaining efficient feature extraction capabilities, further reduces the size of the YOLOv8 model, eliminates redundant information in the feature extraction process, retains the details of the small target of the unmanned aerial vehicle, and improves the accuracy and detection speed of the model.

The invention discloses a head and neck tumor image segmentation method based on combined analysis of CT and PET, and the method comprises the steps: obtaining a head and neck tumor data set, and carrying out the preprocessing of the head and neck tumor data set; constructing a parallel common attention mechanism module; an improved UNet3D network model is constructed; optimizing the improved UNet3D network model by adopting a combined loss function; and inputting the preprocessed head and neck tumor data set into the optimized and improved UNet3D network model, and outputting a segmentation task result of the bimodal image. According to the method, a parallel common attention mechanism and a multi-scale feature fusion technology are introduced, multi-modal image information is efficiently combined, features of the tumor under different scales are captured, accurate segmentation is carried out on the features, the precision and efficiency of head and neck tumor segmentation are remarkably improved, the limitation that a single-modal image is difficult to comprehensively reflect the tumor features is overcome, and the accuracy and efficiency of head and neck tumor segmentation are improved. And the problem of scale diversity of the tumor in the image is solved.

The present invention discloses a method for removing artifacts from sparse angular CT. The method is based on perceptual loss and feature fusion attention residual network. It includes: preprocessing the input sparse CT image with artifacts and the original CT image; extracting features from the preprocessed sparse CT image and the original CT image to obtain a feature map; training the obtained feature map with an improved feature fusion attention residual network with perceptual loss as the loss function, using the original CT image as a label, so that the sparse CT image can remove artifacts but keep its structure unchanged; finally, the sparse CT image is used as input, and the artifacts are removed using the trained network model, and the peak signal-to-noise ratio and structural similarity are calculated. The method uses perceptual loss as the loss function, which can enhance detail information. And adding an attention mechanism to the feature fusion residual network to improve the performance of the network, and finally achieving a fast and effective artifact removal effect.

This invention discloses a two-branch image dehazing method based on edge constraints, comprising the following steps: Step 1: Constructing an image dehazing dataset; Step 2: Obtaining edge maps of the images using the Canny edge detection operator, and constructing a training sample using the edge maps, corresponding fog maps, and clear maps; Step 3: Randomly segmenting the images in the training samples as input to the network, and using data augmentation methods to expand the sample size; Step 4: Constructing a two-branch dehazing network based on edge constraints, including a dehazing subnetwork and an edge subnetwork; Step 5: Initializing network parameters and constructing a target loss function to optimize the parameters of the two-branch dehazing network; Step 6: Using the trained model to dehaze the test fog map to obtain a clear, fog-free image. This invention can utilize edge constraints to recover clear, fog-free images, solving the problem of lost spatial details and semantic inconsistencies in the repaired images during image dehazing tasks.