54results about How to "Guaranteed reconstruction accuracy" patented technology

The invention discloses a regional depth edge detection and binocular stereo matching-based three-dimensional reconstruction method, which is implemented by the following steps: (1) shooting a calibration plate image with a mark point at two proper angles by using two black and white cameras; (2) keeping the shooting angles constant and shooting two images of a shooting target object at the same time by using the same camera; (3) performing the epipolar line rectification of the two images of the target objects according to the nominal data of the camera; (4) searching the neighbor regions of each pixel of the two rectified images for a closed region depth edge and building a supporting window; (5) in the built window, computing a normalized cross-correlation coefficient of supported pixels and acquiring the matching price of a central pixel; (6) acquiring a parallax by using a confidence transmission optimization method having an acceleration updating system; (7) estimating an accurate parallax by a subpixel; and (8) computing the three-dimensional coordinates of an actual object point according to the matching relationship between the nominal data of the camera and the pixel and consequently reconstructing the three-dimensional point cloud of the object and reducing the three-dimensional information of a target.

An improved self-adaptive sparse sampling fault classification method belongs to the technical field of fault diagnosis. A traditional sparse classification method is improved. Firstly, a wavelet module maximum value and a kurtosis method are used for carrying out feature enhancement processing on signals, and on the premise that signal sparsity is guaranteed, a unit matrix is adopted to replace aredundant dictionary. Secondly dimension reduction is carried out on data by adopting a Gaussian random measurement matrix, thereby reducing redundant information in the signal, and reserving effective and small amount of data. Then, a sparse coefficient is solved by adopting a sparsity adaptive matching pursuit (SAMP) algorithm, and the compressed signal is reconstructed; and finally, a cross correlation coefficient is adopted as a judgment basis of the category of the fault, so that an improved adaptive sparse sampling fault classification method is provided. Experimental verification proves that redundant information in signals is effectively reduced, the influence of time shift deviation on fault type judgment is avoided, meanwhile, the operation complexity is reduced, and the calculation speed and the reconstruction precision are improved.

The invention discloses a rapid compression perception reconstruction method facing to a wearable device, and belongs to the signal processing field. The method comprises the steps of: performing compression measurement to an original electrocardiogram signal x through a constructed m*n dimension random sparse binary observation matrix phi, so as to get an observation vector y, the length of which is m; and based on the observation matrix phi and the observation vector y, reconstructing the original signal by the compression perception reconstruction method learned from the alternating direction multiplier method and the block sparse Bayes, so as to get an estimation vector of the original signal, which is as shown in the description. Compared with the conventional compression perception reconstruction method learned from the block sparse Bayes, the method provided by the invention is advantaged in reaching quicker reconstruction speed on the premise of ensuring the reconstruction precision.

The invention relates to a sparse unmixing method based on optical calculation. The method comprises the following steps: S1, an optical filtering method is adopted to acquire a transform domain sparse coefficient matrix for hyperspectral signals; S2, a deterministic measurement matrix is designed to realize perception measurement on the hyperspectral signals; S3, parallel multiplication calculation is carried out on the sparse coefficient matrix and the deterministic measurement matrix through a hardware FPGA for acquiring endmember signals; and S4, the sparse structural information of the hyperspectral signals and a structured reconstruction algorithm are used for reconstructing an abundance matrix. The optical filtering technology and an electrical system are adopted to realize multiscale geometric analysis, the sparse unmixing efficiency is greatly improved on the premise of ensuring the reconstruction precision, and quick and effective sparse decomposition can be realized.

The invention belongs to the field of distributed compressed sensing reconstruction technology, and particularly relates to a variable step size distributed compressed sensing reconstruction method based on a recurrent neural network. The method comprises the following steps: acquiring structural information of a vector to be reconstructed of each channel by using the recurrent neural network, obtaining each non-zero conditional probability of the vector in each channel, then estimating an optimal atom of the current iteration, and then determining a value of a non-zero term of each channel by solving a least square problem, and completing the reconstruction of signals. According to the method, non-combined sparse multi-channel signals can be reconstructed, and meanwhile, the computational complexity of the coding end cannot be increased.

The invention discloses a teleoperation method for an unstructured environment based on 5G + AR. The method comprises the following steps: firstly, carrying out rapid three-dimensional reconstructionon an unstructured environment where a robot is located, and mapping a reconstructed scene into a virtual reality space; and then, capturing a gesture action of a user, establishing a virtual model ofthe hand of the user, and mapping the gesture action into the virtual reality space, thereby realizing interaction between the user and the virtual object; finally, introducing an auxiliary technology to help the user finish teleoperation work more accurately and quickly.

The invention discloses a reconstruction method and device for target electromagnetic-scattering-characteristic data, and relates to the technical field of radar imaging. The method includes the stepsthat a three-dimensional echo signal model of a target is built, and the three-dimensional echo signal model is converted into a state-space equation; according to scattering echo data of the target,a generalized Hankel matrix is built, and the state-space equation is solved based on the Hankel matrix to determine a parameter estimating value of the target scattering center; according to the parameter estimating value of the target scattering center, electromagnetic-scattering-characteristic data of the target is reconstructed. Through the steps, the reconstruction accuracy of the target electromagnetic-scattering-characteristic data is guaranteed, and efficient compression and real-time loading of the full-space target electromagnetic-scattering-characteristic data can also be achieved.

The invention relates to an eddy current array detection apparatus based on compressed sensing and a detection method thereof, and an eddy current array probe. The eddy current array detection apparatus comprises an excitation signal generator, a power amplifier, the eddy current array probe, a signal acquisition module, a FPGA controller and a PC host computer. The eddy current array probe is composed of eddy current probe units, and each eddy current probe unit is composed of a probe shell, a tip, a bottom cover, mounting nuts, excitation coils, coil formers, an iron core, a TMR magnetic sensor and a printed circuit board. According to the invention, in virtue of the characteristic of sparsity of array signals, the precision of signal reconstruction is guaranteed and signal sampling frequency is greatly reduced, so requirements on hardware modules like sampling circuits are lowered, the amount of sample data is reduced, hardware burden in data acquisition, transmission and storage is mitigated, and the service life of equipment is prolonged; and for portable instruments powered by limited energy and equipped with array probes, energy consumption for sampling and calculating can be greatly saved, and the service life of the portable instruments can be substantially prolonged.

The invention discloses a circular arc front-and-back edge of an aviation blade reconstruction method based on error control, and is used for solving the technical problem that anti-noise performance of the existing blade section characteristic parameters extraction method based on the measured data is poor. The circular arc front-and-back edge of the aviation blade reconstruction method based on the error control is composed of selecting region points covering the front-and-back edge of the aviation blade, arranging the region points, and matching the arranged region points with a cubic spline so as to gain a curve L0. The method carries out discretion and encryption on the matched spline curve L0 in an arc equal-length mode, respectively marks the head discrete point as P0i and the end discrete point as P1i, utilizes the least square method to carry out a circular matching for the discrete points between P0i and P1i containing P0i and P1i, calculating the average error epsilon i from the discrete points to an ith matching circle, and setting the iteration control error as epsilon until the epsilon i is less than or equal to the epsilon, the iteration ends. Based on the error control based on the matching of the least square method for fitting, utilizing the iterative matching method based on the error control, the circular arc front-and-back edge of the aviation blade reconstruction method reconstructs blade circular arc front-and-back edge, avoids that the points in the selected area involve the matching, wherein the points are not on the circular arc, and improves the anti-noise performance.

The invention relates to a hyperspectral sparse unmixing device and belongs to the technical field of hyperspectral unmixing. The device comprises a first box body, an air inlet fan, an air inlet cylinder, a filtering grid plate, a noise reduction assembly, a sponge, a damping cotton and a filtering assembly, wherein one end of the air inlet cylinder is inserted and fixed in the side wall of the first box body, the air inlet fan is arranged between the upper inner wall and the lower inner wall of the air inlet cylinder, the filtering grid plate is vertically arranged between the upper inner wall and the lower inner wall of the air inlet cylinder and is positioned at one end port of the air inlet cylinder. According to the invention, the noise reduction assembly filters redundant waves of the signal in the low-pass filter, and real-time additive noise such as cloud layer, atmosphere and the like in the acquisition process is removed, filtered and quantized, so that the quality of the hyperspectral image is improved, and the extraction precision of the end element generating sparse demixing is improved; the multi-scale geometric analysis is achieved by adopting the optical filteringtechnology through the filtering assembly, the sparse demixing efficiency is greatly improved on the premise of ensuring the reconstruction precision, and the rapid and effective sparse decompositionis achieved.

The invention discloses a monocular 3D reconstruction method with depth prediction. The monocular 3D reconstruction method comprises the following steps: A, obtaining a depth map and rough pose estimation of an RGB image by using a monocular depth estimation network; b, calculating camera pose estimation in combination with an ICP algorithm and a PnP algorithm, and executing loopback detection on a local level and a global level to ensure the consistency of a reconstruction model; and C, converting the depth map into a global model, and then inserting the random fern code of the current frame into a database. According to the invention, the defects in the prior art can be overcome, and large-scale high-quality three-dimensional reconstruction is realized.

An eddy current testing method of the irregular wear-out defect of a thimble tube by introducing multimedia units comprises steps as follows: firstly, a degenerate magnetic vector potential method discrete control equation is obtained; secondly, the multimedia units containing a substrate and the defect simultaneously is introduced into a unit attribute of the irregular wear-out defect of the thimble tube, the conductivity and the magnetic conductivity attribute in the gauss point position are judged according to the area where the gauss integral point position in the multimedia units is located, and the process of the degenerate magnetic vector potential method discrete control equation is corrected; then, an eddy current detection signal is obtained according to an experiment system; finally, the actual size of the irregular wear-out defect of the thimble tube is obtained according to a conjugate gradient algorithm. After the multimedia units are introduced to the eddy current testing method of the irregular wear-out defect of the thimble tube, the size of the units can be reasonably increased in the premise that the reconstruction accuracy of the eddy current nondestructive detection of the irregular wear-out defect of the thimble tube is guaranteed, so that the total number of the units is notably reduced, the calculation efficiency is effectively improved, and the method has very high practical value.

The invention provides an image processing method, system and device and a storage medium, which are used for image sparse reconstruction, image denoising, compressed sensing image reconstruction or image restoration. The image processing method comprises the following steps: establishing a general linear optimization inverse problem under the norm regular constraint of a sparse signal; on the basis of a standard or learned iterative soft threshold shrinkage algorithm, establishing a differentiable deep network model based on convex combination to solve the problem; introducing a deep neural network of any structure in the solving step to accelerate the solving step, and reducing the number of iterations required for convergence of the algorithm. According to the method, a traditional iterative optimization algorithm and a deep neural network of any structure are combined, the image reconstruction performance is improved, rapid convergence is ensured, and the current demand of image sparse reconstruction is met.

The invention relates to a feedback method based on ELM superposition CSI in an FDD large-scale MIMO system, which belongs to the technical field of wireless communication. The feedback method comprises the steps that a user side reads downlink channel state information and an uplink user sequence, carries out the superposition of the downlink channel state information and the uplink user sequence, obtains a superposition sequence, and then achieves transmission; and a base station end uses an ELM interference elimination network to recover downlink channel state information and the uplink user sequence in the obtained receiving sequence. Compared with non-superposition coding CSI feedback, the method has the advantages that uplink bandwidth resource occupation is completely avoided. Compared with superposition of CSI feedback based on deep learning, the method has the advantages that under the condition of keeping equivalent network recovery performance, the number of parameters is reduced, the memory occupation space is reduced, and the network training time is shortened.

The invention discloses a sheet metal part rectangular contour three-dimensional measurement method and system based on feature fitting, and belongs to the field of three-dimensional measurement. According to the method, preliminary contour extraction is realized based on a sheet metal part image, a rectangular contour is detected, and approximate quadrilateral fitting is carried out; sub-pixel contour extraction is carried out, corresponding point matching is realized, and a three-dimensional point cloud is reconstructed; and constructing a three-dimensional rectangular projection fitting model based on iterative rotation, and calculating three-dimensional rectangular parameters. Compared with a traditional method of fitting rectangles after three-dimensional point cloud reconstruction based on structured light, the method can reduce the problem of reconstruction errors of contour edge parts caused by illumination variation, image noise and the like, improves contour point cloud reconstruction precision, realizes automatic recognition of contours and accurate parameter acquisition, and improves the accuracy of contour point cloud reconstruction. And the accuracy of three-dimensional measurement of the profile of the sheet metal part is

The invention belongs to the field of machine vision research, and particularly relates to a three-dimensional reconstruction method based on GPU parallel acceleration. The method mainly comprises thefollowing steps: initializing an algorithm; loading the image sequence modulated by the structured light, and copying the image sequence in the computer memory to a common memory of the GPU equipmentside in parallel; the GPU is used for carrying out background segmentation on images shot by the left camera and the right camera in parallel, three-dimensional reconstruction is only carried out onsegmented foreground targets, redundant calculation is reduced, and the calculation efficiency is improved; solving an image sequence phase principal value by using GPU parallel acceleration and carrying out phase unwrapping to obtain an absolute phase value; distortion correction is carried out on the phase unwrapped image; stereoscopic matching is carried out in a serial and parallel combined mode, that is, serial processing is carried out on local areas of an image, parallel processing is carried out between the areas, and stereoscopic matching is carried out; disparity map preprocessing; and calculating a three-dimensional point cloud according to the disparity map. According to the method, the parallel computing capability of the GPU is fully utilized, the overall computing speed of the algorithm is increased, and the application scene of three-dimensional reconstruction is expanded.

The invention relates to sensors, and discloses a multifunctional senor sample selection method based on kernel subtractive clustering. The multifunctional senor sample selection method includes the following steps: step 1, sampling is carried out on a sensor which needs to be subjected to signal reconstruction, and sample data are obtained; step 2, the sample data obtained in the step 1 is subjected to normalization processing; step 3, the kernel subtractive clustering is used for mapping the sample data, which are already subjected to the normalization processing, to a high-dimensional space to enable the sample data to be linearly separable, and the sample data are classified; and step 4, a clustering center of each type of the sample data is used as a sample point to be selected. According to the multifunctional senor sample selection method, on the basis that the reconstruction accuracy is guaranteed, the number of the sample data which are needed for the signal reconstruction is reduced. The fact that each senor is subjected to high-density sampling is not needed, and therefore workloads which are needed for the sampling are greatly reduced.