522results about How to "Rich in details" patented technology

The invention relates to an adaptive low-illumination visible image and infrared image fusion method, and belongs to the field of digital image processing. The method comprises the steps: image preprocessing, NSCT transformation, frequency domain coefficient fusion and an NSCT inversion. The method mainly achieves the multi-scale decomposition of a bright image and an infrared visible image, extracts high-frequency components and low-frequency components, and selects different fusion rules to carry out the frequency domain fusion according to the characteristics of the high-frequency components and low-frequency components. The NSCT inversion part is mainly used for carrying out the multi-scale inversion of the obtained fusion high-frequency and low-frequency components, obtaining the gray scale image after fusion, then carrying out the weighting of the gray scale image after fusion and an original color visible image, and obtaining a color fusion image. The method can effectively keep more detail information of the original image, can improve the contrast and resolution of the fusion image, and can be widely used in the fields of intelligent traffic, video monitoring, medical diagnosis, target detection and national defense security.

The invention relates to a fault diagnosis method, in particular to a rolling bearing fault diagnosis method based on the vibration detection. The method comprises the following steps of firstly decomposing the rolling bearing data collected by an acceleration sensor into three layers of wavelet packets, solving the energy of a third layer of wavelet packet coefficient rebuilding signals, selecting a frequency band with centralized energy to rebuild approximate evaluation of an original signal according to the variation of energy values of each frequency bands of the third layer; and utilizing a cepstrum to further analyze the rebuilt signal, and comparing the rebuilt signal with a theoretically-computed fault characteristic frequency and an edge frequency characteristic. By combining multiple resolutions of the wavelet packet and the cepstrum, the periodic component on a power spectrum, a separated-side frequency-band signal and the characteristics which are slightly subjected to the transmission route can be well detected. Meanwhile, the method is strong in manipulability and practicability.

The invention provides a train fault diagnosis system and method based on a vehicle and cloud, and relates to the technical field of high-speed rail large data fault diagnosis. The system comprises a vehicle fault diagnosis subsystem and a cloud fault diagnosis subsystem. The vehicle fault diagnosis subsystem is combined with the cloud fault diagnosis subsystem. Cloud service is used to monitor the running process of a train. Cloud data are continuously updated, and a fault diagnosis and prediction model is continuously optimized and updated, and can visualize a diagnosis process. According to the invention, the details of the fault diagnosis and prediction model are more and more perfect; the precision is improved continuously; the performance is beyond the control of a traditional single real-time fault diagnosis module; the requirements of high reliability and high precision of the train fault diagnosis subsystem in the running process are satisfied; and the safety of the train in the running process is improved.

The invention relates to the digital image processing field and particularly relates to a white balance and dark primary color adaptive histogram underwater image enhancement method. The method comprises steps that step 1, color correction for an image is carried out through utilizing a dynamic threshold white balance algorithm; step 2, a dark channel map of the image after color correction is solved through utilizing a dark channel model; step 3, a weight factor of block images is calculated; step 4, underwater image processing is carried out through utilizing a CLAHE method; step 5,the weight factor and gray mapping relationship tables are utilized to acquire a final gray mapping relationship table of a block through fusion calculation; and step 6, a bilinear interpolation algorithm is utilized to calculate a gray mapping value corresponding to each pixel in block images one by one, enhanced block images are acquired, interpolation calculation for pixels connected between the adjacent block images is carried out, and images of underwater images after enhancement are acquired. The method has advantages of simple calculation and strong timeliness.

The invention discloses an efficient super-resolution method based on a deep back projection network. The method comprises the following steps: (1) obtaining a total training set and a test set; (2) preprocessing the total training set to complete data enhancement; (3) scaling the images in the total training set at different scales; and (4) image super-resolution reconstruction is realized basedon the convolutional neural network, and the convolutional neural network totally comprises 27 convolutional layers and specifically comprises three parts of feature extraction, error back projectionand image reconstruction. The combination of the group convolution and the 1 * 1 convolution is used for replacing the traditional convolution to redesign the iteration submodule, and the strategy caneffectively reduce the model parameter quantity and improve the model efficiency; Each iteration sub-module comprises an error feedback mechanism, so that error correction can be carried out in time;in addition, a channel weighting module is introduced, so that the model efficiency can be further improved.

A hyperspectral data dimensionality reduction method based on tensor distance patch alignment belongs to a hyperspectral remote sensing image processing method. The hyperspectral data dimensionality reduction method aims at the tensor characteristic of the hyperspectral data. Firstly, the hyperspectral data is converted into a tensor form through a window area and maintains space information of every pixel; secondly, the tensor distance is introduced to construct a high-quality tensor distance neighbor graph containing determination information; thirdly, a globally optimal spectrum-space information is acquired according to a patch alignment framework expanded to tensor space; fourthly, solutions of tensor sub-space are obtained by using a iteration optimization method of the alternating least square algorithm; and lastly, categories of samples are discriminated on the basis of the tensor nearest neighbor method. The hyperspectral data dimensionality reduction method has the advantages that relatively high overall classification accuracy and the Kappa coefficient through effective utilization of the space area characteristic and the spectrum characteristic of the hyperspectral data, and the acquired classification effect picture is very clear and smooth with rich details; the dimensionality reduction framework can process 2-order data, 3-order data and data in higher orders.

The invention relates to a virtual camera and real camera switching system and method. The system comprises a real camera, wherein the real camera is connected with a chroma key device, a synthesis rendering server is connected with a virtual camera controller, a virtual camera parameter generating device connected with the real camera and the virtual camera controller is arranged inside the synthesis rendering server, a virtual camera and the virtual camera parameter generating device are connected with a synthesis device, the synthesis device is connected with the chroma key device, and a foreground generating device, a background generating device and an output device which are connected with the synthesis device are further arranged inside the synthesis rendering server. According to the method, virtual foregrounds, virtual backgrounds and virtual persons are fabricated in field and are fast synthesized, the switching of a field real scene and a virtual scene is realized through the operation of the virtual camera controller to the scene, the limitation of the real camera when the real camera shoots a virtual object is compensated, and the program expressive force of a virtual studio is greatly improved.

The invention relates to an image and video amplification method and a relevant image processing device. The method comprises a preprocessing module and a composite amplification module, wherein the preprocessing module is used for executing high-pass filtering processing to an input image to extract the high-frequency part of the input image and is used for executing image decomposition processing to the input image to decompose the input image into smooth areas and marginal areas; and the composite amplification module is used for conducting amplification processing to the original input image and the smooth areas through simple interpolation operation and is used for conducting amplification processing to the marginal areas and the high-frequency part through both complex interpolationoperation and simple interpolation operation. By adopting the method, the amplification results of the original input image, the smooth areas, the marginal areas and the high-frequency part can be fused, i.e. an output image with features of saw tooth resistance, clear-cut margin, rich detail, high contrast and the like can be output according to a preset amplification scale under the circumstance that the operation workload is small, the complexity is low and the speed is high.

The invention provides an automatic retinal blood vessel segmentation method for clinical diagnosis of glaucoma. Through the method, the influence of bright regions such as optic disc and exudate is eliminated by fusing five models depending on different image processing technologies, namely a matched filter, a neural network, multi-scale line detection, scale space analysis and morphology. At thesame time, massive data is not needed to establish a retinal blood vessel segmentation model, so the method greatly reduces the amount and complexity of data to be processed, is easy to realize, andcan effectively improve the efficiency of retinal blood vessel segmentation. The method also utilizes the region growth method and the gradient information to carry out iterative growth on the background and the blood vessel region on the basis of a multimodal fusion result, the segmentation results exhibit better continuity and smoothness, more retinal vascular details and more complete retinal vascular network can be kept, and thus the method effectively assists ophthalmologists in diagnosing diseases and lightens the burden of ophthalmologists.

The invention discloses a semantic segmentation method based on an improved full convolutional neural network. The semantic segmentation method comprises the steps of: acquiring training image data; inputting the training image data into a porous full convolutional neural network, and obtaining a size-reduced feature map through a standard convolution pooling layer; extracting denser features while maintaining the feature map size through a porous convolutional layer; predicting the feature map pixel-by-pixel to obtain a segmentation result; using the stochastic gradient descent method SGD totrain parameters in the porous convolutional neural network in the training; acquiring image data that needs to be subjected to semantic segmentation, and inputting the image data the trained porous convolutional neural network, and obtaining a corresponding semantic segmentation result. The invention can improve the problem that the feature map of the final upsampling recovery in the full convolution network loses sensitivity to the details of the image, and effectively expands the receptive field of a filter without increasing the number of parameters and the amount of calculation.

The invention relates to an image restoration system capable of improving imaging quality greatly. The image restoration system comprises an image data storage module, an edge-method based edge spread function calculating module, a linear spread function calculating module, a normalized MTF(modulation transmission function) calculating module, an MTFC (modulation transfer function compensation) digital filter system coefficient calculation and storage module, an image data boundary processing module, a row image digital filtering module, a row filter image noise suppression module, a row filter image storage module, a line image digital filtering module, a line filter image noise suppression module and a restored image storage module. By the image restoration system capable of improving imaging quality greatly, real-time on-track processing of remote sensing images is realized, image definition is improved and noises are suppressed effectively.

The invention discloses an image super-resolution reconstruction method based on double-dictionary learning, which mainly solves the problem that detailed information cannot be effectively supplemented in the prior art when super-resolution reconstruction is performed on a low-resolution image. A realization process comprises the following steps of: firstly, inputting a low-resolution image XL to be processed, constructing five pairs of high-resolution dictionaries and low-resolution dictionaries (Dh1, Dl1), (Dh2, Dl2),..., (Dh5, Dl5), and reconstructing five high-resolution estimation images under the five pairs of dictionaries; constructing one pair of high-frequency dictionary and low-resolution dictionary Df={Dhf, Dlf} by virtue of the high-frequency information and low-frequency information of the input low-resolution image, and reconstructing five pairs of high-resolution estimation images with different neighbor parameters; and finally, performing low-rank decomposition on the ten pairs of reconstructed high-resolution estimation images, and solving a mean value of a low-rank matrix obtained from the decomposition to obtain a final reconstructed high-resolution image XH. The method provided by the invention can be used for obtaining the high-resolution image with clear edges and rich details when being used for performing the super-resolution reconstruction on the low-resolution image and is suitable for super-resolution reconstruction on various natural images.

The invention discloses a high-resolution remote sensing image weak target detection method based on deep learning. For a remote sensing image with low resolution, a small target size and fuzzy quality, the method comprises the following steps: firstly, improving the resolution of an image by adopting a WGAN-based super-resolution reconstruction method; inputting the image with the enhanced quality into a target detection framework; carrying out deep feature extraction on the image by using a residual network; fusing the extracted low-level features with the extracted high-level features; it is ensured that the fused multi-layer feature map has rich detail information and also contains high-level semantic information; and carrying out region-of-interest coarse extraction on the feature mapby using the fused multi-layer features and the region suggestion network, mapping the extracted region to the same dimension by using a region-of-interest alignment method, and carrying out subsequent target accurate classification and position refinement to obtain a final target detection result. According to the method, the weak and small target detection precision and recall rate under the conditions of low remote sensing image resolution and complex background are effectively improved.

The invention relates to a medical image fusion method and an image detection method based on fusion medical image learning, which relate to an image detection technology based on fusion medical imagelearning. The invention solves the technical problems that the medical image is polluted by noise, the signal-to-noise ratio is low, the gray level difference between different tissues is small, theapplication of the medical image is affected, and the single mode image cannot provide richer information of the pathological tissue from different angles, and the like. The method comprises the stepsof: reading the two kinds of modal images and preprocessing the two kinds of modal images respectively to obtain denoised images; using improved shearing wave transform for multi-scale image segmentation. According to the fusion rules, two kinds of modal images are fused to get fused images. All the fused images are combined into a fused image data set. The improved YOLO v2 depth learning algorithm is used to train the dataset, and the training network is generated. The method performs detection with a trained network. Different modalities of medical images are fused together to provide moreinformation from different perspectives.

The invention discloses a control method of a monitoring ball machine. The method comprises the following steps of (1) vertically and horizontally dividing a to-be-monitored space into a plurality of small partitions, and setting a shooting focal distance for each small partition; (2) setting corresponding preset positions for a central point position and an edge point position of each small partition, and storing horizontal position information, vertical position information and shooting focal distance information of the ball machine; (3) reading a video frame, and performing target detection on the video frame; (4) according to direction information of a detected target in a monitoring scene, mapping the target to the corresponding preset position; (5) calling the preset position by the ball machine to acquire a monitored image. The defects that the automation degree of control is not high, the real-time property and the flexibility are not enough, and human manual interference is required in a ball machine of the traditional video monitoring system are overcome; the control method is convenient in operation, high in automation degree of control and good in instantaneity, and is particularly good in capture effect on the monitored image of a quickly moving target.

The invention discloses an improved two-dimensional maximum entropy division night vision image fusion target detection algorithm which comprises the steps that a two-dimensional histogram is improved, that is the two-dimensional histogram is established according to the maximum gray scale of a gray scale-weighted area, a weight is selected, the maximum entropy is calculated by using the histogram, and infrared and low light images are divided. Compared with the traditional maximum entropy division algorithm, an effect in the aspect of target detection is obvious, and background suppression and target extraction functions are provided; and then a phase of the multi-dimensional characteristic and the validation of operation are verified, and the divided infrared and low light images are subjected to characteristic level fusion target detection. The detection algorithm has a better effect and better applicability in the aspects of target detection under a complicated background and multi-target detection.

The invention discloses a method for reconstructing real-time three-dimensional face data with multiple images. The method includes the following steps: constructing shooting systems of eight cameras; performing multi-camera calibration; performing image matching, to be more specific, calculating global matching points of an ith group of images, and calculating accurate matching points of the ith group of images; calculating three-dimensional face coordinates via the matching points and a projection matrix of the cameras, to be more specific, imaging an object point on the third image of the ith group, and taking an intersection of three projection lines as the object point so as to complete reconstructing of the real-time three-dimensional face data. The face data obtained in the method are global, dense and real-time and can be used directly, and the method can be widely applied to the fields such as the animation industry, computer 3D (three-dimensional) games, human-computer interaction, identity recognition, medical virtual surgery, videophones, face recognition, face expression, age simulation, film and TV advertising production and computer cognition.

The invention relates to a high-dynamic-range image dual-screen display method based on color space switching. The method is characterized by comprising the following steps: S1, calculating a high-dynamic-range image and converting into an RGB (Red Green Blue) space image; S2, converting the RGB space image in the step S1 into an HSL (High Speed Logic) space image, and finishing the separation of brightness and tone; S3, performing separate self-adaptive logarithm mapping on the brightness L of the HSL space image in the step S2, finishing compression of the brightness, and enhancing the saturation S; S4, converting the HSL space image in the step S3 into an RGB space image; S5, performing front-back panel image segmentation on the RGB space image in the step S4; S6, performing dual-screen segmented display of front and back panel images generated in the step S5 on an LCD-FED (Liquid Crystal Display-Field Emission Display) dual-screen display screen. By adopting the high-dynamic-range image dual-screen display method, the problem of color shift is solved, and the detail expression capability is enhanced greatly.

The invention discloses a self-adaptive low-light level image intensification method for reducing color cast, relates to low-light level image intensification methods, and aims to solve the problems that the image color cast is intensified when a conventional low-light level image intensification method is used, and a relatively bright area of an image is over-inhibited or over-intensified when being not well processed. The self-adaptive low-light level image intensification method comprises the following steps: firstly, converting a low-light level image into a RGB (Red, Green, Blue) color space, performing inverted S-shaped conversion, performing inversion, calculating minimum values of different pixel points of reversed images at three RGB channels so as to obtain initial dark channel images, and performing median filtering so as to obtain atmosphere light intensity estimation values; converting the inversion images into an HSV color space, and calculating self-adaptive intensification parameters by taking average gray level values of a V channel as average brightness; calculating transmissivity images according to atmosphere imaging equations, modifying so as to obtain transmissivity smooth images, with the atmosphere imaging equations, performing demisting operation on the three RGB channels of the inversion images, performing inversion, and performing S-shaped conversion, thereby obtaining finally intensified images. The self-adaptive low-light level image intensification method is applicable to intensification processing on images.

The invention relates to a multi-sensor fusion low-illumination video image enhancement method and belongs to the video image processing field. According to the method, matching is carried out according to characteristic similarities between videos from different sources; registration is performed on images from different sources by adopting a multi-scale SIFT algorithm; an accurate transformation matrix can be obtained based on the combination of the multi-scale SIFT algorithm and a RANSAC algorithm; the transformation matrix is utilized to perform interpolation on each frame in infrared video images and visible light video images, and therefore, images with different resolutions can be transformed into images with the same resolution, and problems in the registration of images with different resolutions can be solved; and fast fusion of the images is realized by using an alpha-based weighting algorithm, the fusion time of the images satisfies the real-time requirement of the videos, and therefore, real-time display of the videos can be realized. With the multi-sensor fusion low-illumination video image enhancement method adopted, the definition of the videos can be improved, and information contained by the clear videos is also rich and colorful, and follow-up processing can be facilitated.