92results about How to "Good details" patented technology

A method for segmentation of paint cracks on an ICGA image based on a conditional generative adversarial network includes: (1), collecting an original ICGA image, extracting a complete fundus oculography image, labeling it with gold standard, normalizing fundus oculography image and gold standard, splicing it into a group of images as sample data, distributing the sample into a training set and atest set according to proportion; (2) based on the principle of conditional generative countermeasure network, constructing the network of generators and discriminators; (3) inputting that data of thetraining set into the network for adversarial train, defining a loss function, and generating a paint crack image correspond to the original picture by the training generator; (4) in the testing phase, inputting the data of the test set, and getting the corresponding paint crack segmentation result diagram through the trained generator G. The segmentation method provided by the invention can be used for solving the problems that the sample size of the ICGA image is small and the acquisition of the contrast image is difficult, and has the characteristics of high accuracy of the segmentation result.

The present invention provides an image processing method based on a mobile terminal and a mobile terminal. The mobile terminal comprises a first camera and a second camera, the first camera is used for shooting color images, and the second camera is used for shooting black and white images. The method comprises: obtaining the first color original image and the black and white original image of the same shooting object captured by the first camera and the second camera at the same time; performing synthetic processing of the first color original image and the black and white original image, and generating a first middle image; and performing image processing of the first middle image, and generating an object image. The image processing method based on a mobile terminal and the mobile terminal are able to allow the first middle image to keep the data bit wide, are larger in data volume, higher in image quality, smaller and less in noise and better in detail presentation, and may obtain an object image with higher quality through image processing by employing the first middle image with larger data volume.

The invention discloses a quick magnetic resonance imaging method based on a recursive residual U-type network. The method comprises the three steps of data processing, model training and image reconstruction. By adopting the method, a recursive residual module is added to the U-type network, the problems of gradient blast and gradient vanishing caused by back propagation of the network are solved, new network parameters are not added while the layer number of the network is increased, the quality of a reconstructed image is obviously improved, and more image details can be restored.

The invention relates to tooling fabric with a cooling function and an uvio-resistant function. A method for manufacturing the tooling fabric with the cooling function and the uvio-resistant function comprises a first step of using cooling master batch as raw materials and manufacturing the cooling master batch to slicing sheets, a second step of drying the slice sheets, a third step of carrying out spinning on the slicing sheets through a spinning box to manufacture nylon base cooling fibers, a fourth step of adopting a siro spinning technology to spin cotton fibers and the nylon base cooling fibers into blended yarn, and a fifth step of adopting technological measures of 'early opening, later weft insertion and on-machine tension slightly larger' to carry out weaving. A formula of the cooling master batch comprises PA6, bowlder powder, zirconium dioxide powder, silicon dioxide powder, stearic acid dispersants, polystyrene dispersants, low molecular wax dispersants and titanate coupling agents. The tooling fabric has functions of fast heat absorption and fast heat dissipation, and an everlasting cooling feature is guaranteed. Meanwhile, the tooling fabric further has a notable uvio-resistant feature.

The embodiments of the invention disclose an image shooting method and device, an electronic device and a storage medium and relate to the technical field of electronic devices. The image shooting method is applied to the electronic device. The electronic device is provided with a rotatable camera. The method comprises the following steps: when a shooting request is received, obtaining a first image shot by the camera; determining a plurality of shooting main bodies in the first image based on the first image; controlling the camera to rotate and obtain a plurality of second images, wherein the preset position of each second image in a plurality of the second images focuses one shooting main body in a plurality of the shooting main bodies respectively; the shooting main bodies focused by each second image are different; and a plurality of the second images are synthesized to obtain a target image. According to the image shooting method and device, the electronic device and the storagemedium provided by the embodiments of the invention, a plurality of images, which respectively focus different shooting main bodies, are shot in different angles and synthesized through the camera arranged on the electronic device, so as to obtain an image with finer details.

A computer-implemented method for processing data includes receiving a collection of traces corresponding to signals received over time at multiple locations due to reflection of seismic waves from subsurface structures. A measure of correlation among the traces as is computed a function of a set of wavefront parameters, which determine respective moveout corrections to be applied in aligning the traces. A matrix having at least three dimensions is generated, wherein the elements of the matrix include the computed measure of the correlation. Using the matrix, values of the wavefront parameters are identified automatically or interactively along the time axis or along selected horizons to maximize the measure of the correlation, and a seismic image of the subsurface structures is generated by aligning and integrating the traces using the moveout corrections that are determined by the identified values of the wavefront parameters.

The invention discloses a multi-depth feature fusion image saliency target detection method and system. The method comprises the steps of obtaining to-be-detected image information in a set scene; inputting the image information into a trained multi-depth feature fusion neural network model; wherein the multi-depth feature fusion neural network model adopts convolution to perform feature extraction in a coding stage, restores information of an input image in combination with an up-sampling method of convolution and bilinear interpolation in a decoding stage, and outputs a feature map with significance information; learning feature maps of different levels by adopting a multi-level network, and fusing the feature maps of different levels; and outputting a final saliency target detection result. According to the invention, a multi-depth feature fusion neural network is used to carry out saliency target detection on an image in a scene, the detection precision is guaranteed, and the speedof a subsequent processing process is accelerated; a contour detection branch is added, and the boundary details of the to-be-detected target by using contour features are refined.

The invention belongs to the technical field of automatic detection, and particularly relates to a conveying belt longitudinal tearing detection device based on a double visual sensor. The conveying belt longitudinal tearing detection device comprises a shell body, an infrared light emitting module, a visible light emitting module, a power supply module, an infrared light detection module, a visible light detection module, an acquired data preprocessing module, a storage module and a data processing module are fixedly arranged in the shell body, the shell body is fixedly arranged on a connecting support below an upper conveying belt through a fixed base, linear light beams emitted through the infrared light emitting module and the visible light emitting module are received through the infrared light detection module and the visible light detection module respectively after being incident on the upper conveying belt and then sent to the acquired data preprocessing module, the acquired data preprocessing module is used for preprocessing acquired image data and then sending the preprocessed image data to the data processing module, and the data processing module is used for analyzingthe acquired data and judging whether the conveying belt is torn or not. According to the conveying belt longitudinal tearing detection device, the reliability of longitudinal tearing detection of theconveying belt is improved.

The invention discloses an image fusion method. An RGB image and an MONO image that are taken via dual cameras in a night scene mode are subjected to image fusion operation; the RGB image and the MONO image are input, the RGB image is converted into a YUV format, brightness information of the RGB image in the YUV format is separated from the same and put in an image Y1, color information is separated from the RGB image and put in an image UV1, UV1 is subjected to wave filtering operation via a wave filtering algorithm, color noise in UV1 is removed, a new image UV2 is obtained, the MONO image is subjected to de-noising operation via a de-noising algorithm, an image Y2 is obtained, Y1 is subjected to edge extraction operation via an edge detection algorithm, a new image Y3 is obtained, an image Y4 is obtained after Y2 is superposed on Y3, and a result image is obtained after Y4 and UV2 are fused. After color noise and white noise in the fused image are removed, brightness information of MONO and RGB is used, and good detail information can be obtained while the fused image is high in noise level.

The invention provides a target detection method based on an attention mechanism deep learning network. The method is characterized by comprising: extracting a feature map of an image to be detected through a target detection model containing an attention mechanism module, detecting the position and the category of a target from the feature map, and the attention mechanism module comprising at least one attention module M1 used for generating an attention weight matrix with the same size according to the feature map and acting on the feature map; at least one attention receptive field module M2 used for carrying out feature extraction on the feature map; and at least one attention feature fusion module M3 used for fusing the features of different levels of the network. According to the target detection method, high detection speed is ensured on the basis of high detection accuracy, and meanwhile, the model is simple in structure and small in calculated amount.

The embodiment of the invention provides a target detection method. The method comprises the steps that based on a preset feature extractor, conducting multi-layer convolution operation on an obtainedimage to be detected, target feature maps corresponding to at least two layers of target convolution operations are obtained, and the at least two layers of target convolution operations at least comprise the last layer of convolution operation; based on a preset region extraction network model and a target feature map corresponding to the last layer of convolution operation, determining region position information of a suspected target region of the to-be-detected image; and based on a preset classification model, the obtained target feature map and the determined area position information of the suspected target area, obtaining target information from the to-be-detected image, the target information comprising the target type and the target position information of the target object in the to-be-detected image. By applying the embodiment of the invention, more accurate detection of the target object can be realized.

The invention provides a dark channel prior and edge information-based image defogging method, and belongs to the technical field of image defogging methods. The method mainly comprises the following steps of: (1) estimating a value of atmospheric light by adoption of a block-based technology; (2) carrying out edge detection and expansive operation on a fog-degraded image, and extracting edge information of an original foggy image; (3) self-adaptively selecting a window size according to an obtained edge information graph so as to obtain a rough transmissivity graph; (4) refining the transmissivity graph by adoption of a rapid gradient domain steerable filter method, and importing a tolerance coefficient and a correction factor to correct the refined transmissivity graph; and (5) substituting the value of the atmospheric light and the corrected transmissivity graph by utilizing a fog-degraded model, so as to obtain a fog-free image. According to the method provided by the invention, when the transmissivity graph is solved, the edge information of the image is combined to self-adaptively select the window size, so that the halo phenomenon at the edge area can be effectively solved.

The invention discloses a smog elimination method of a remote sensing image based on content and characteristics and a multi-scale model. The smog elimination method is completed through the followingsteps of obtaining auxiliary information of the image; preprocessing the type of the input remote sensing image; judging the scene content of the image coverage area, setting the reasonable number ofscale parameters and assigning the corresponding value, and using the Retinex theory to process the image; using the Retinex algorithm to compensate the color of the image; converting the RGB image to HSI model image, and processing the I component by histogram equalization; after processing, converting the RGB image to RGB image; converting the image according to the auxiliary information. The method of the invention not only can effectively eliminate the haze phenomenon in the remote sensing image, achieve the purpose of enhancing the visual effect of the remote sensing image and improvingthe quality of the remote sensing image, but also has good retention ability to the detail information in the image.

The invention provides a color image fusion system and method based on ternary number wavelet transform, and relates to the technical field of image processing. The system comprises an image input interface circuit for inputting an image to be fused; an image fusion chip for ternary number wavelet transform and fusion processing of the image to be fused to generate a fusion image; a first storage chip for storing fusion data generated when the image fusion chip processes the image to be fused; a second storage chip for storing auxiliary coefficients generated when the image fusion chip carries out ternary number wavelet transform for the image to be fused; and a display chip for displaying the generated fusion image and fusion data. The image fusion can be performed by constructing a novel technical framework, and combining a ternary number wavelet transform function using coarse-scale and fine-scale coefficients, the color distortion of the fusion image can be minimized, and the quality of the fusion image is improved.

The invention discloses a single-frame image super-resolution reconstruction method based on cascade regression base learning. The method comprises the following steps: taking a super-resolution reconstruction technology of a single-frame low-resolution image as a research object, learning a multi-layer over-complete sub-dictionary for representing an image structure, constructing a mapping relation between a low-resolution image and a high-resolution image, and learning an optimized regression base and a corresponding coding coefficient; and then, complete super-resolution reconstruction is realized for the low-resolution image set, and the reconstructed image is used as a low-resolution image of the next layer for feature extraction. The invention discloses a single-frame image super-resolution reconstruction method. learning by utilizing a meta-dictionary learning method to obtain a low-resolution dictionary; a weighted linear regression method is used for carrying out multilayer regression base learning on a reconstructed high-resolution training set image and an original high-resolution image in a cascading mode so as to approach a complex nonlinear mapping relation between alow-resolution image and a high-resolution image, and instance regression super-resolution reconstruction with high processing speed, small memory occupation and high reconstruction quality is achieved.

The invention relates to a pedestrian detection data expansion method based on a generative adversarial network, and the method comprises the steps: S1, building a three-layer cascaded generative adversarial neural network model, and setting a target function of model training, wherein each layer of generative adversarial neural network adopts a Bicycle GAN structure, the generator adopts a residual Unet structure, and the input of the next layer of network is the pedestrian instance mask picture and the output of the previous layer of network; s2, preprocessing the training data; s3, traininga three-layer cascade generative adversarial neural network model by adopting the preprocessed data; and S4, completing expansion of pedestrian detection data through the three-layer cascade generative adversarial neural network model. Pedestrians generated by adopting the scheme of the invention are fused with the background more naturally, and details of the generated pedestrians are finer by improving the Unet structure of the generator; the multi-scale pedestrian picture is generated based on the cascade structure, so that the quality of the large-size and high-resolution pedestrian picture is improved; diversified pedestrians can be generated, and the data expansion efficiency is improved.

The invention discloses a nighttime traffic monitoring enhancement method based on gradient domain fusion, and relates to a digital image processing method. Because there are vehicle lamps, street lamps, building lamps and other active light sources at night, a large amount of halo will be generated till the objects nearby the lamps cannot be seen. In particular, the strong halo generated by the lamps of vehicles on a highway severely affects the visual effect of traffic sign boards, and causes the severe degeneration of the quality of the traffic sign boards. The method enables image fusion and image halo removal to be combined, achieves the processing of the image in the gradient domain, and enables the traffic signs in a final result image to be easy to recognize through the interframe complementation information of a video.

The invention discloses an RGB-D saliency object detection method based on a cascade improved network, and belongs to the technical field of image processing. Most existing RGB-D models directly aggregate the features from the CNN networks of different levels, and the noise and interference information contained in low-level features are easily introduced. The invention creatively provides a cascade improved structure, a saliency map generated by the features of a high-level part is used as a mask to improve the features of a low-level part, and then a final saliency map is generated by aggregating the improved low-level features. In addition, in order to eliminate the interference information of the depth map, the invention provides a depth enhancement module for preprocessing before thedepth features and the RGB features are mixed. According to the method, four evaluation indexes are used for carrying out experiments on seven data sets, and the results show that the method surpassesall current most advanced RGB-D saliency object detection methods.

The invention discloses an MAP-MRF super-resolution image reconstruction method based on attitude information constraint. The method comprises the following steps: S1, carrying out modeling and calculating; s2, extracting image feature points; s3, establishing an MAP-MRF model of the image sequence; and S4, super-resolution reconstruction: carrying out fuzzy kernel estimation by using an iterativereweighted least square method, and solving an MRF optimal solution by using a belief propagation algorithm to complete super-resolution reconstruction. Compared with a traditional super-resolution image reconstruction method, the method has the advantages that attitude information constraint is added, and non-redundant space-time information is provided outside the image; the MAP-MRF model better conforms to an actual image sequence imaging model, errors caused by mismatching of a prior model can be effectively avoided, a reconstructed high-resolution image is clearer, detail information ismore prominent, noise amplification can be effectively inhibited, and the singularity problem of an observation matrix can be effectively solved.

The invention relates to the field of computer vision, and aims to remove moire and output a corresponding clean picture. The invention discloses a screen shot image moire removing method based on a convolutional neural network AMNet, and the method comprises the following steps: analyzing moire features, and building a moire model in a screen shot image; building a data set; designing a network framework; designing a network structure; setting the learning rate of the network and the weight of the loss function of each part, training the convolutional neural network by using a deep learning framework Tensorflow until the loss converges, and generating a training model; and inputting the test picture with the moire pattern into a network to obtain a corresponding clean picture without themoire pattern. The method is mainly applied to occasions for removing the screen moire patterns.

The invention belongs to the technical field of intelligent cooking, and discloses an intelligent cooking control system and method with a thermal image. The intelligent cooking control system with the thermal image comprises a thermal image collection module, a temperature detection module, a cooking parameter setting module, a main control module, an image processing module, a timing module, a heating module, a stirring module and a display module. Through the cooking parameter setting module, the problem of cumbersome operation caused by the fact that a user manually conducts adjusting stepby step through a button to obtain cooking parameters can be avoided, and the setting efficiency and intelligence of the cooking parameters are improved; and meanwhile, the threshold of a food boundary is calculated through the image processing module, outline extraction of food is achieved through threshold segmentation and a segmentation threshold window extraction method for color components,green components and blue components, and the precise outline of the food is obtained.

The invention discloses an image generation method and device. The method comprises the steps of acquiring a first image of a shooting target and a first ambient light angle used for indicating the relative position relation between an irradiation light source and the shooting target when the first image is shot; obtaining a first 3D model generated by fusing the depth information of the shootingtarget, the plurality of second images and the first lighting information; wherein the plurality of second images are a plurality of two-dimensional images shot from a plurality of angles of the shooting target; wherein the first lighting information comprises a first illumination angle which is the same as the first ambient light angle and first light intensity of which the corresponding brightness is greater than or equal to a preset brightness threshold; and according to the first image and the first 3D model, performing fusion to generate a third image. By adopting the method, when a userperforms selfie in a dark environment, the 3D model containing the first lighting information can be fused with the actually shot three-dimensional image to obtain the selfie image which does not generate stereoscopic impression and skin detail loss, so that the user experience is better.

A material dispensing unit (1; 10) for a three dimensional printing apparatus (100; 200) has a nozzle (3) for depositing particulate material (5; 220a, 220b; 224; 226) on a build surface (7), where the nozzle defines a through passage (9) for the material. The through passage has an inlet end (11) for receiving the material and an outlet end (15) for dispensing the material. A valve (21) is provided at least one of at, within or in fluid communication with the through passage for controlling flow of the material via the through passage, the valve being operable between open and closed positions. Flow of said material into the through passage is blocked when in the closed position and, when in the open position, flow of the material into the through passage is allowed. A method of forming a three dimensional object (31) is also disclosed, as is a three dimensional printing apparatus (200) comprising one or more dispensing units (la; b) for dispensing particulate material (220a, 220b; 224; 226), an enclosure (137) for containing the material dispensed by the one or more dispensing units and one or more heating elements (210, 212) for heating the material contained in the enclosure to a first predetermined temperature.

The invention discloses a leather processing agent. The leather processing agent comprises, by weight, 25-35% of sodium polyacrylate, 1-4% of an emulsifier, 1-4% of sodium tripolyphosphate, 1-4% of asilver loaded product, 1-4% of nano titanium dioxide, 5-15% of modified rapeseed oil, 15-25% of an amphoteric acrylic acid retanning agent, and 25-35% of water. The leather processing agent is used inthe fields of textile industry, leather or furry product industry, is capable of providing leather with antibacterial performance, mildew resistance, tightness, uniform dyeing degree, high dye uptake, high solvent resistance, and high grease resistance, so that using requirements can be satisfied.

The invention provides a grid generation method and device and a storage medium. The method comprises the following steps: firstly, acquiring a point cloud and a corresponding relation between the point cloud and an original image; then, based on the point cloud, constructing a tetrahedron corresponding to the point cloud; taking the tetrahedron as a vertex and the overlapped surfaces between the adjacent tetrahedrons as directed edges to construct a directed graph; according to the corresponding relationship between the point cloud and the original image, determining the weight of a directed edge contained in the directed graph; and according to the weight of the directed edge, generating a target grid corresponding to the point cloud through a graph cut algorithm. In the application, the weighted value of the directed edge is determined according to the corresponding relationship between the point cloud and the original image, and compared with a weight set as a constant, the target grid corresponding to the point cloud generated according to the weight is more accurate, so that when the target grid obtained through the method is used for reconstructing a large-scale scene, the higher the precision of a reconstructed three-dimensional model is, the better the detail effect is.

The invention provides a pollen image detection method based on feature fusion, and the method comprises the steps: inputting a preprocessed pollen image into a convolutional neural network, and obtaining shallow features; generating a spatial attention weighted feature map and a spatial attention weight matrix through a spatial attention module based on the shallow features; generating a deep feature map from the spatial attention weighted feature map through convolution and down-sampling, and generating a channel attention weighted feature map from the deep feature map through a channel attention module; inputting the space attention weight matrix and the channel attention weighted feature map into a cross-connection attention mechanism to obtain a feature map after feature fusion; and inputting the feature image after feature fusion into a prediction module to obtain a detection result of pollen information in the pollen image. The pollen detail information in the superficial layer features is weighted and fused to the deep layer features through the cross-connection attention mechanism, feature fusion is performed after the deep layer features are optimized, and more pollen details of the pollen image can be recovered.

The invention discloses an image restoration method based on gradient sparsity and non-local similarity information, comprising a first stage of image non-blind restoration based on gradient sparsityand a second stage of image quality improvement based on non-local autoregressive model. The non-blind image restoration based on gradient sparsity in the first stage includes the following core steps: current gradient image auxiliary variable calculation; Current restored image update; Lagrange multiplier calculations. The invention is based on the sparsity of the image transformation domain, sothat the edge structure in the image can be recovered, and on this basis, the quality of the image restoration is further improved by utilizing the non-local similarity information in the image.

The invention discloses a fresh jujube wormhole detection method based on a hyperspectral image convolutional neural network, and the method employs a convolutional neural network model which can be used for the detection of fresh jujube wormholes for detection, and the construction method of the convolutional neural network model comprises the following steps: S1, collecting the data of a hyperspectral image of a sample; S2, extracting an optimal characteristic wavelength; S3, performing data preprocessing; S4, training a convolutional neural network model by using the training sample set; and S5, performing classification verification on the data set by using the model. According to the fresh jujube wormhole detection method based on the hyperspectral image convolutional neural network, the detection of the fresh jujube wormhole can be realized without manual intervention by processing the hyperspectral image data under the selected characteristic wavelength, the detection precision of the network model on the fresh jujube wormhole is improved. The method solves the problem of misjudgment caused by interference of color blocks, spots, fruit stems and the like on the surfaces of the fresh jujubes in the existing computer vision detection process of the fresh jujubes, and has the characteristics of simplicity, feasibility and high recognition efficiency.

The invention discloses an image processing device and an image processing method for an LED curved surface body, and belongs to the image processing technology field. The device comprises a three-dimensional software simulation module, an area coding graph module, a curved screen module, a simulation camera module and an image rendering module. The method is based on the 3Ds Max camera principleand the image processing technology, the image is unfolded and deformed to adapt to LED screens of different curved surface shapes, a normal video image is converted into an image matched with an LEDcurved surface body, the playing requirement of the curved surface screen is met, and the correct visual feeling is formed. The method is simple to operate, convenient to simulate, wide in application, high in image quality and definition and perfect in detail expression.

The invention relates to a non-uniform fog image defogging algorithm based on a transmission attention mechanism. The non-uniform fog image defogging algorithm comprises the following steps: S1, performing sparse smooth dilated convolution feature extraction; S2, performing non-uniform haze feature processing based on a transmission attention mechanism; S3, obtaining a loss function. According to the non-uniform fog image defogging algorithm provided by the invention, an end-to-end mapping relation between a degraded image and a clear image is directly constructed under the guidance of an attention mechanism for a real non-uniform haze image, a good defogging effect is achieved for a non-uniform foggy image and a synthesized foggy image, and an obtained restored image has relatively good detail information; the color is more natural.