96 results about "Projection algorithms" patented technology

The purpose of the invention aims to provide an object and indoor small scene restoring and modeling method based on RGB-D camera data. In summary, the method comprises: in the RGB-D depth data (point cloud) registering and modeling, integrating the restraining conditions for point-to-face and point-to-projection together so that they are applied to the accurate registering of the sequence depth data (point cloud) obtained by the RGB-D camera; and finally obtaining a point cloud model (.ply format) for an object or a small scene wherein the model is able to be used for object measurement and further CAD modeling. The registering method of the invention considers the speed advantage of the point-to-projection algorithm and integrates the precise advantage of the point-to-tangent plane algorithm. This overcomes the problems with the slow speed and low scene precision in a traditional point cloud registering method, and is capable of finding out the corresponding point of a point on the source point cloud to a target point cloud quickly and accurately so as to realize the cloud splicing of multiple view points.

A method and apparatus for generating a two-dimensional output image of a slab multi-planar reformatting (MPR) volume data set from a selectable view point and view direction is described. The method comprises factorising a view transform matrix into an intermediate transform matrix and a warp transform matrix. The intermediate transform matrix is such that voxels in the volume data set map to pixels in an intermediate image in which neighbouring voxels in a plane of the volume data set which is most parallel to tie plane of the output image are separated by an integer number of pixel along each of tie axes of the intermediate image. The intermediate transform matrix is applied to the data set to provide an intermediate data set. A projection of the intermediate data set along the view direction is performed according to a desired projection algorithm to provide the intermediate image. The wrap transform matrix is then applied to the intermediate image to provide the output image.

The invention provides a GPS (global position system) terminal-based map matching method for vehicle position tracking, which belongs to the field of map matching. According to the method, the problem of inaccurate vehicle positioning caused by GPS signal deviation can be solved, a vehicle is matched with the right position on the current running road, the distance between the vehicle and the current road destination node is calculated, when a GPS signal is invalid, the position of the vehicle can be reckoned, and the time that the vehicle reaches the node within the critical zone of the road destination node is calculated. Considering that the GPS terminal for operating a map matching program has limited resources, the method synthetically utilizes a weight-based map matching thought and a curve fitting-based map matching thought which have good real time and combines road-network topology information, and different projection algorithms can be selected according to the current position of the vehicle. By using the method, vehicle position information with accurate real time is provided, and according to the method, vehicle monitoring, real-time traffic information collection, vehicle navigation, etc can be realized.

The invention provides an inertia / astronomy / satellite high-precision integrated navigation system and a navigation method thereof. The inertia / astronomy / satellite high-precision integrated navigation system comprises a sensor module, an integrated navigation resolving module and a real-time display module, wherein the sensor module is sequentially connected in series with the integrated navigation resolving module and the real-time display module and comprises an inertia sensor, a star sensor and a satellite receiver; the integrated navigation resolving module comprises a four-step Runge-Kutta strapdown navigation resolving module, an astronomy attitude determination resolving module, an unequal-interval Kalman filtering module and a position and speed compensation algorithm module; and the real-time display module comprises a three-dimensional real-time display module. The navigation method realizes the real-time acquisition of sensor data, utilizes the inertia sensor data to carry out strapdown navigation resolving based on a four-step Runge-Kutta algorithm and can carry out the integrated navigation resolving on the inertia sensor data, star sensor data and satellite receiver data and carry out the real-time three-dimensional display on an integrated navigation result by utilizing a projection algorithm.

The invention discloses a three-dimensional reconstruction method for a concrete CT (computed tomography) image, comprising the following steps of: firstly, performing tomography on a concrete test sample by a CT machine; secondly, performing three-dimensional reconstruction on the concrete CT image by using a ray-casting algorithm, so as to obtain real images of aggregates, mortar and holes; and finally, performing three-dimensional stereo reconstruction on the concrete CT image by using a visualization tool VTK (visualization tool kit). The three-dimensional reconstruction method for the concrete CT (computed tomography) image, disclosed by the invention, has the advantages of being strong in real-time performance, high in operation efficiency, and capable of clearly and reliably distinguishing the aggregates, the mortar and the holes. A real meso-structure model of concrete can be obtained through a three-dimensional reconstruction result, the three-dimensional reconstruction result can be subjected to a test comparison with a concrete mesoscopic numerical simulation result, and a numerical analysis model is corrected.

Provided is a multiple-sparse-representation face recognition method for solving the small sample size problem. In the method, two modes are adopted to solve the small sample size problem during face recognition, one mode is that given original training samples produce 'virtual samples' so as to increase the number of the training samples, and the other mode is that three nonlinear feature extraction methods, namely a kernel principle component analysis method, a kernel discriminant analysis method and a kernel locality preserving projection algorithm method are adopted to extract features of the samples on the basis that the virtual samples are produced. Therefore, three feature modes are obtained, sparse-representation models are established for each feature mode. Three sparse-representation models are established for each sample, and finally classification is performed according to representation results. By means of the multiple-sparse-representation face recognition method, virtual faces are produced through mirror symmetry, and then norm L1 based multiple-sparse-representation models are established and classified. Compared with other classification methods, the multiple-sparse-representation face recognition method is good in robustness and classification effect and is especially suitable for a lot of classification occasions with high data dimensionality and few training samples.

The invention discloses a process monitoring method which is based on non-Gaussian component extraction and support vector description. The method comprises the following steps: read-in of training data and data to be diagnosed, data preprocessing, establishment of a principal component analysis model, particle swarm optimization algorithm, non-Gaussian projection calculation, support vector data description, residual analysis, principal component estimation, fault detection and the model updating. By the method, the non-Gaussian components can be automatically extracted from operating data of an industrial process, thus avoiding the disadvantage that the conventional statistical process monitoring method assumes that data is subject to normal distribution, and the non-Gaussian projection algorithm based on the particle swarm optimization algorithm ensures the maximization of the non-Gaussian properties of the extracted independent components, and avoids the problem that the independent component analysis method is easy to be involved in the locally optimal solution. Compared with the conventional statistical process monitoring method, the method can find abnormity in time, effectively reduce the rate of false alarm, and obtain better monitoring effect.

The invention discloses a forward-looking scanning radar imaging method. The method comprise the steps of initializing imaging system parameters; performing range pulse compression on echo data; subjecting a radar right forward-looking area to super-resolution imaging processing; performing back projection imaging processing on a large-azimuth area; and splicing imaging results of the two areas. According to the forward-looking scanning radar imaging method, an ill-posed problem regularization method is utilized, and accordingly, the azimuth resolution is improved, and a back projection algorithm is combined to improve the azimuth resolution of the large-azimuth area. The two imaging results are spliced into a complete scene image, so that the forward-looking scanning radar imaging scene range is expanded, and not only the defect that synthetic aperture radars cannot perform forward-looking imaging is overcome, but also the problem that azimuth imaging scene ranges of prior forward-looking scanning radars are small is solved.

A multiple-input-multiple-output radar waveform design method belongs to the radar communication technical field, and aims to provide a design method with lower related sidelobe and frequency spectrum inhibition depth, high efficiency, less consumption, high robustness, and excellent time frequency anti-interference performance; the method comprises the following steps: pre-evaluating an autocorrelation sidelobe inhibition fuzzy region according to a relative position between a strong scatterer and a to be measured object in a radar scene, thus forming a corresponding object function; analyzing MIMO radar waveform orthogonality constraint so as to form the object function satisfying the orthogonality constraint; pre-evaluating a frequency domain interference fuzzy frequency band zone according to scene prior information, thus forming the corresponding object function; forming a constant modulus phase coding waveform constrained condition; forming a loose alternative projection algorithm framework; solving a waveform design according to the loose alternative projection algorithm framework, thus providing three waveform optimization output modes. The loose alternative projection constant modulus waveform coding design enables the MIMO radar to have batter detection performance.

The invention discloses a river water level monitoring method based on monocular camera, which is an information processing method belonging to the video monitoring field, that is using light reflection characteristic difference between the river plane surface and the bank plane surface to determine water level height with horizontal pixel projection algorithm (HPIP) for video collected images. The method includes the steps: (1) using affine transformation for executing geometric correction to image; (2) executing histogram equalization process for video image greyscale; (3) determining water level height with horizontal pixel projection algorithm; (4) real time and robust tracking water-level fluctuation. The invention has the advantages of easy realization, strong robustness, high reliability, high real time property or the like, and can provide real time dependable information for hydrology monitoring.

The invention discloses a kind of method of measurement the solid particle three-dimensional density field in the gas and solid two phase flow, velocity field and equipment. The particle flux transmitting tube is laid at the front of high rate stereoscopic vision system which is consisted by the double look-out angle lens stereoscope and the high rate camera and the jet stream particle is refracted through the double look-out angle lens stereoscope mirror face and the single high rate camera target area can achieve the particle flux graphic with two different angles at the same time and after the graphic is input the computer a program module which is consisted by the pattern recognition algorithm procedure and three-inversion projection algorithm procedure treats the particle diagram and then get the transient three-dimensional density field. The program module treats series microsecond class time sequence changeable particle graphic and gets the particle three-dimensional movement trace and the velocity field. It is stabilize by using the single camera system and overcomes the problem of multiple camera system's constructer is complex and the synchronism is bad.

The invention relates to a monocular real-time three-dimensional reconstruction method based on loop testing, and belongs to the technical field of three-dimensional reconstruction. The method comprises: carrying out pairwise matching in an image sequence of a specified scene on the basis of an image feature point matching theory to obtain image matching point pairs; solving an essential matrix, and then utilizing a singular-value decomposition theory to acquire an initial pose; utilizing the initial pose or a previous-frame pose to obtain an estimated pose through a pose tracking model; judging whether a current frame is a key frame; then utilizing a random fern algorithm to calculate similarity of the current frame and the key frame, and if the similarity reaches a threshold value, it isconsidered that a loop is formed; utilizing a pose of the key frame to optimize the current pose if the loop is formed; utilizing the above-obtained pose to obtain a point cloud, and fusing the sameinto a TSDF global-model; and adopting a light ray projection algorithm to visualize a surface. According to the method, accuracy of the acquired pose is enabled to be high, the cumulative-error problem in three-dimensional-reconstruction processes is eliminated, and a real-time reconstruction result has higher accuracy.

The invention discloses a signal modulation type identification method and a signal modulation type identification system. The method comprises the following steps: preprocessing a signal to be identified, extracting a predetermined number of feature parameters, and using a feature vector formed by the predetermined number of feature parameters to represent the signal to be identified; using an optimal projection matrix to extract the features of the signal to be identified, and projecting the signal to be identified into a low-dimensional feature subspace, wherein the optimal projection matrix is obtained through a locality preserving projection algorithm; and calculating the Euclidean distance between the signal to be identified in the low-dimensional feature subspace and a training signal of which the signal modulation type is known, and determining the signal modulation type of the signal to be identified based on the nearest neighbor algorithm of Euclidean distance. According to the technical scheme, the signal to be identified is projected into the low-dimensional feature subspace by using the optimal projection matrix, which reduces the amount of calculation; and the optimal projection matrix obtained through the locality preserving projection algorithm reduces the deviation of the signal in the process of projection, and has better robustness and higher identification rate.

The invention discloses a video processing-based body height measurement method. According to the method, a monocular camera is taken as a realization platform. Calibration is carried out on an image including reference lines to obtain projection parameters of three reference lines; then a video including a to-be-measured human body is processed. The method comprises the following specific steps: (1) accurately identifying a human body in a video frame; (2) segmenting out a foreground human body by an image processing algorithm; and (3) extracting pixel coordinates of human body feature points (a overhead point and a foot point), and finally inputting data of the projection parameters, the camera height, the feature point pixel coordinates and the like into a core projection algorithm, so as to obtain body height information. According to the method, only an ordinary camera is needed in hardware; the others are achieved by software; the method is simple, feasible and high in reliability; the body height measurement efficiency is greatly improved.

The invention discloses a method for extracting cross sections of radian tunnels on the basis of land-based LiDAR (light detection and ranging), and belongs to the technical field of civil engineering. The method includes intercepting point cloud data of the same cross section in different periods for monitoring deformation of tunnels and analyzing the deformation; establishing corresponding relations of points between deformation point cloud and reference point cloud, namely non-deformation point cloud, and computing deformation of the integral cross sections by the aid of minimum-distance projection algorithms; applying bidirectional projection processes, RANSAC (random sample consensus) algorithms and least square processes to extracting radian central axes pertinently for the radian tunnels; verifying the precision of ultimately extracted deformation of the cross sections and measurement results of total-station instruments. The method has the advantages that the quantity of the point cloud can be increased if the tunnels are long, rotation is only useful for to-be-extracted cross sections and the adjacent point cloud and is equivalently useless for the point cloud at other locations, rotation is carried out whenever each cross section is extracted for the radian tunnels, the concept of point buffer areas is proposed, only the point cloud adjacent to the extracted cross sections rotates, and accordingly the tunnel deformation computation efficiency can be greatly improved.

The invention discloses a tracing accumulation method of coherent detection based on dynamic programming and a back image projection algorithm, comprising the following steps of: obtaining a predication track of a target through the dynamic programming; and then carrying out energy accumulation through the back image projection algorithm to improve an output signal to noise ratio, so as to realize the detection on a low observable target. Compared with a non-coherent TBD (Track Before Detect) algorithm, the method uses phase information of an echo signal when inter-frame accumulation is carried out, and coherent accumulation is carried out on a multi-frame echo to improve an output signal to noise ratio and further improve the detection efficiency; a distance migration compensation is directly carried out on a time domain, so that approximate unfolding of a target distance does not need to be carried out, and Doppler migration of a target is not considered; and therefore, an accumulation error is reduced, the detection possibility and the detection precision are improved, and the false alarm probability is reduced.

The invention relates to the field of computer vision and particularly relates to a license plate recognition method based on mixed feature and gray projection. For the specific problem of license plate recognition, the invention aims to improve the accuracy and reliability of a system, and a real-time requirement can be satisfied. The method comprises the steps of (1) license plate positioning based on mixed characteristics, (2) license plate character segmentation, and the license plate character segmentation based on an improved gray projection algorithm, and (3) character recognition and output after the license plate character segmentation. The characteristics of a combined direction gradient histogram and a kernel principal component analysis method are put forward, the advantages of the direction gradient histograms of a binary image, a grayscale image and a 16-value image are integrated, and the structural features of a Chinese character can be extracted. The method provided by the invention can be embedded into an FPGA to realize actually and is applied to a camera or video camera with a real-time image output function and a license plate positioning and recognition function.

The invention discloses an automatic convergence imaging method based on ground penetrating radar echo data. The method comprises the following steps: S1) carrying out filtering on an echo image through a mean de-background algorithm to remove ground clutters; S2) removing echo image noise through a median filtering algorithm; S3) obtaining the number of radars required by pixel points through threshold analysis; and S4) carrying out convergence imaging on the echo image through a back-projection algorithm based on cosine distance weighting. The method can quickly and automatically carry out convergence imaging on the ground penetrating radar echo image, and can still converge echo signals quickly and effectively under the complex urban environment by utilizing the back-projection algorithm based on cosine distance and threshold analysis image segmentation technology, thereby meeting practical and application requirements.

An objective assessment method for a stereoscopic image quality combined with manifold characteristics and binocular characteristics trains a matrix after dimensionality reduction and whitening obtained from natural scene plane images through an orthogonal locality preserving projection algorithm, for obtaining a best mapping matrix. Image blocks, not important for visual perception, are removed. After finishing selecting the image blocks, through the best mapping matrix, manifold characteristic vectors of the image blocks are extracted, and a structural distortion of a distorted image is measured according to a manifold characteristic similarity. Considering influences of an image luminance variation on human eyes, a luminance distortion of the distorted image is calculated according to a mean value of the image blocks. After obtaining the manifold similarity and the luminance similarity, quality values of the left and right viewpoint images are processed with linear weighting to obtain a quality value of the distorted stereoscopic image.

The invention relates to a projection algorithm for generating curved surface non-interference five-axis processing locus, belongs to the milling processing technology field and aims to guarantee that a final knife contact point is on an original workpiece curved surface, improve processing precision and guarantee relatively good processing surface quality. According to the algorithm, a knife contact point of a cutter projected on a triangular sheet model is selected as an initial point; the shortest distance from the initial point to a workpiece curved surface is calculated, and a corresponding point on the workpiece curved surface is further calculated; a tangent plane on the workpiece curved surface passing the corresponding point is established; the cutter is projected to the tangent plane to acquire projection points on the tangent plane; an adjusting search step algorithm is executed for an oscillation convergence problem; the projection point on the tangent plane satisfying an error is taken as the knife contact point. The algorithm is advantaged in that based on a Newton-Raphson algorithm, a continuous iteration method based on the tangent plane is employed, that the knife contact point is on the original workpiece curved surface is guaranteed, and calculation precision of the cutter contact point is improved.

The invention provides an inversion method and apparatus for water quality parameter concentration based on an optimal combination of multiple bands. The method comprises the following steps: acquiring the measured water quality parameter concentration at a sampling site and the hyperspectral reflectance data of a water body at the sampling point; screening the bands of the hyperspectral reflectance data by using a continuous projection algorithm so as to obtain a first band set; constructing a first partial least square inversion model according to the measured concentration and the corresponding hyperspectral reflectance data of the first band set; calculating the contribution value of the corresponding hyperspectral reflectance data of the first band set in the first partial least square inversion model; screening the first band set of the hyperspectral reflectance data according to the contribution value so as to obtain a corresponding second band set; constructing a second partial least square inversion model by using a partial least square algorithm and according to the measured water quality parameter concentration and the corresponding hyperspectral reflectance data of the second band set; and inputting the corresponding hyperspectral reflectance data of the second band set into the second partial least square inversion model for concentration inversion so as to obtain the inversion value of water quality parameter concentration.

The invention discloses a single-grain fixed-point recognition technology by aiming at different categories of wheat seeds. The technology comprises the following steps that: obtaining the hyperspectral image data of a wheat seed to be identified, and after a black and white correction method is adopted to correct the hyperspectral image, carrying out principal component transformation; selectinga whole grain of wheat seed as a region of interest to extract the average spectrum as the spectral data of a grain sample; adopting a S-G smoothing method to carry out smoothing processing on the spectral data; using a Kennard-Stone method to divide the sample spectral data into a training set and a prediction set at a ratio of 3:2, and adopting a successive projection algorithm to select an optimal waveband for identifying the wheat seeds of different hybrid varieties; constructing a support vector machine model; and inputting an optimal waveband into the established SVM (Support Vector Machine) model, and showing a prediction result by different colors to finish hybrid seed fixed-point recognition. By use of the technology, the online detection of the wheat grains of different hybrid varieties is realized, and single-grain fixed-point recognition is realized.

The invention provides an apple interior and exterior quality fast nondestructive testing method based on the hyperspectral imaging technology. Apple spectrum information is collected through a hyperspectral imaging system, features of an apple spectrum curve are analyzed, and spectrum data is analyzed; then, single-band images are determined and subjected to secondary threshold segmentation, corresponding areas are extracted, and spectrum curves of the corresponding areas are extracted; a secondary continuous projection algorithm is disclosed according to apple exterior features like diseases and pests to extract a feature wavelength set SP1; a multi-parameter continuous projection algorithm is disclosed according to apple interior features like sweetness and hardness to extract a feature wavelength set SP2; finally, the feature wavelength set SP1 of apple interior quality and the feature wavelength set SP2 of apple exterior quality are fused, and apple interior and exterior quality fast nondestructive detection is achieved through a BP neural network. The method is easy and convenient to operate, high in detection speed and high in detection result precision, and apples are not damaged.

The invention discloses a model-free self-adaptive control method for an attitude of an assembled spacecraft. The method comprises the following steps: S1) establishing a full-format dynamic equivalent linear data model for an attitude movement of a general assembled spacecraft; S2) designing a parameter self-adaptive estimating equation and estimating a pseudo gradient for the pseudo gradient in the full-format dynamic equivalent linear data model; and S3) establishing an error criterion function according to the dynamic equivalent linear data model, and designing a weighting one-step forward controller through a minimizing error criterion function. The dynamic equivalent linear model is established for a general discrete time nonlinear system, and a modified projection algorithm is adopted for performing online parameter estimation for the unknown variable in the dynamic equivalent linear model, so that the attitude determination accuracy of a flexible spacecraft movement simulator is increased.

The invention provides an adaptive iterative learning control method based on non-strict repetition which solves the problem of convergence of system trajectory tracking error occurring when multiplenon-strict repetition problems exist in a controlled system. A projection algorithm is used for design of unknown parameter vector estimators, convergence conditions are given and rapid convergence oftrajectory tracking error is realized. Based on the state-space reconstruction technology, the non-strict repetition rule is embedded into the adaptive iterative learning control method, so that thejitter range of tracking error is effectively reduced and the problem of poor anti-interference capability of the conventional projection algorithm is solved.

The invention relates to a nondestructive identification method of rice with industrial wax. The nondestructive identification method comprises the steps of shooting a hyperspectral image of the rice with a visible-near-infrared hyperspectral imaging system to obtain three-dimensional information of the rice, correcting with a black and white board to convert spectral response strength of a rice sample under each wavelength into reflectivity, preprocessing reflectivity data with multiplicative scatter correction (MSC), extracting eight characteristic wavelengths by a successive projections algorithm (SPA), then shooting unknown rice to obtain a hyperspectral image of the unknown rice, and establishing a mathematical regression model according to a partial least squares (PLS) method to identify whether the unknown rice is coated with the industrial wax, wherein two-dimensional information is spatial information, and one-dimensional information is spectral information. The method is quick and convenient, and simple to implement, and a feasible method is provided for quick nondestructive identification of the rice with the industrial wax on a market.

The invention belongs to the technical field of remote sensing image processing, and relates to a green tide information extraction method. The method comprises: obtaining a sea area satellite remote sensing image with a conventional method, screening images of green tide regions, and performing preprocessing operations of geometric correction and image mosaicking; calculating and determining a green tide range by adopting a normalized difference vegetation index; cutting a region of interest in any shape to generate an irregular image file containing the green tide regions; extracting green tide information by adopting a split Bregman fast projection algorithm of a variational level set two-phase image segmentation based Chen-Wees model; and finally, quantitatively calculating the green tide regions by adopting a quantitative formula. The image segmentation based variational model accurately extracts and quantifies the green tide information on the satellite remote sensing image, so that a conventional manual threshold method can be completely replaced and operational application can be realized; and method is scientific and reasonable in principle, low in human factor quantity, high in calculation speed, accurate and stable in result, high in operability, friendly in application environment, high in practicality and easy to popularize.

The invention provides a method for detecting canopy moisture content of leaf vegetable crops. The method comprises the following steps: filtering spectrum, performing gradient extraction on characteristic wavelength by adopting a backward interval partial least square algorithm, a genetic algorithm and a continuous projection algorithm; performing wavelet denoising on a front view image and a top view image, extracting canopy projection area, canopy perimeter and plant height as growth characteristics of the leaf vegetable crops, extracting color component mean value of each of RGB space and HSI space as the color characteristic, extracting entropy, secondary moment, contrast ratio and homogeneity in six color co-occurrence matrixes constructed by the RGB space and the HSI space; and reducing dimension of the characteristic wavelength and the image characteristic by using kernel principle component analysis, establishing a detection model of the canopy moisture content of the leaf vegetable crops by using an extreme learning machine algorithm. Through the adoption of a plurality of chemometrics algorithms and image processing methods, the internal and external information of the canopy of the leaf vegetable crops are fully acquired, and the nondestructive detection of the canopy moisture content of the leaf vegetable crops is realized.

The invention discloses a bumpy platform SAR rapid time domain imaging method based on a mixed coordinate system. The method comprises the following steps of carrying out distance direction matching filtering processing on a baseband echo signal to obtain a distance-compressed signal; dividing a subaperture to construct a mixed coordinate system imaging grid; carrying out backward projection on each subaperture data on the mixed coordinate system imaging grid; and carrying out a rapid backward projection or rapid decomposition backward projection algorithm in the mixed coordinate system on a sub-aperture image in the mixed coordinate system to obtain an SAR image with a full spatial resolution. In the invention, a polar coordinate subaperture imaging grid in a rapid time domain algorithm is replaced with a mixed coordinate grid, a two-dimensional interpolation in the time domain imaging algorithm is optimized, execution efficiency of the time domain algorithm is improved, and a processing speed and imaging quality of the time domain algorithm are increased and improved.

The invention provides a method and system for reconstructing a three-dimensional image of an electronic speculum. The method for reconstructing the three-dimensional image of the electronic speculum comprises the following steps: a central processing unit (CPU) divides an object in an image to be subjected to iteration reconstruction into slices, and constructs a frame that the CPU cooperatively calculates with a many integrated core (MIC) processor; the CPU schedules all equipment in the frame to obtain reconstruction image values and measurement projection image values of respective current slices in a multi-threading parallel mode, and calculates final reconstruction image values of the current slices through a re-projection algorithm and a back projection algorithm according to the reconstruction image values and the measurement projection image values in sequence till calculating the final reconstruction image values of all slices. According to the method and the system for implementing the electronic speculum three-dimensional image reconstruction, through the cooperative calculation of the CPU and the MIC, the execution of an iteration reconstruction method is greatly accelerated, so that image reconstruction tasks can be rapidly accomplished.