48 results about "Iterative strategy" patented technology

The invention discloses a BP decoding method and device for a polarization code, and belongs to the technical field of channel coding. A BP decoding algorithm based on an early termination iteration strategy is used for decoding the polarization code, and a judgment condition of early termination of the BP decoding algorithm based on the early termination iteration strategy is that a symbol of the log-likelihood ratio of the leftmost end of a factor graph does not change in continuous two iteration processes. The invention also discloses a BP decoding device for the polarization code. Through the convergence characteristic of the symbol of the log-likelihood ratio, the method and device carry out the judgment of iteration termination of BP coding, can remarkably reduce the number of iteration times under the condition of causing no loss of decoding performance, and are more remarkable in effects under the conditions of medium and high signal-to-noise ratios. The method is simple and easy, is low in calculating complexity, and is simple in hardware implementation.

The invention provides a Split Bregman weight iteration image blind restoration method based on a non-convex higher-order total variation model, and belongs to the technical field of image processing. The method is characterized in that firstly, a non-convex higher-order total variation regularization blind restoration cost function is obtained by introducing image border sparse prior information meeting a hyper-Laplacian model and by combining a high-order filter bank capable of generating piecewise linear solutions; secondly, a weight iteration strategy is provided, a minimization problem of the non-convex higher-order total variation regularization blind restoration cost function is converted into a minimization problem of an approximate convexity cost function with the updated weight; thirdly, the minimization problem of the approximate convexity cost function with the updated weight is converted into a new constraint solving problem through an operator split technology, and the constraint solving problem is converted into a split cost function through the method of adding a penalty term; fourthly, the split cost function is solved through a Split Bregman iteration solving frame. According to the Split Bregman weight iteration image blind restoration method based on the non-convex higher-order total variation model, an image can be restored effectively and rapidly, the shortage that a staircase effect is generated in a traditional total variation regularization blind restoration method is overcome, and meanwhile a better restoration effect on manually degraded images and actually degraded images is achieved.

The invention provides a method for improving BP (belief propagation) decoding by use of a polarisation code based on early termination of the iterative strategy. According to the method, the convergence condition of the likelihood ratio of the information bit of the polarisation code is used as the criterion for stopping iteration in the BP decoding algorithm, the iteration number is reduced, and the purpose of reducing the complexity of the BP decoding algorithm can be achieved; simulation results show that the strategy can greatly reduce the decoding iterations, when the maximum iteration number is 60 and the signal-to-noise ratio is 3.5dB, the average iteration number is reduced by 80% compared with the original BP decoding algorithm, and the method is better than the existing early termination method of the iterative strategy.

The invention provides a sparse singular value decomposition scanning radar forward-looking imaging method, which is used for the radar imaging technology field. A radar forward-looking echo signal is analyzed and scanned, and an azimuth echo model of a target scattering coefficient and antenna measuring matrix operation is constructed, and then an azimuth high resolution problem is converted into a matrix inversion problem. A target function is constructed by adopting the sparse singular value decomposition method provided by the invention. The solution of the echo azimuth signal is realized by adopting an iterative strategy. A limitation of conventional TSVD methods of contradiction between resolution improvement and noise suppression is broken through, and sensitivity of an imaging result to noises is reduced, and therefore the azimuth resolution improvement is realized.

The invention discloses a mesh surface curve design method based on distance constraint. The designed curve passes through a given interpolation point and is smooth (in the sense of discreteness) andstrictly located on the mesh surface. The method transforms the complex manifold constraints into distance constraints and describes them as optimization problems together with smooth constraints andinterpolation constraints. The local surface is approximated by the tangent plane, and the distance constraint is relaxed to the distance from the point to the tangent plane. Because the points on thecurve used to calculate the distance are interdependent with their corresponding tangent points, a global-local iterative strategy is adopted and Gauss- Newton idea method is used to control its convergence behavior: in the whole stage, it is relaxed into a convex optimization problem by distance approximation to solve the iterative step size; In the local phase, a robust and efficient projectionmethod is used to map the optimized curves to the surface to update the tangent points. Finally, all the relaxed polygons are mapped to the mesh surface by using the cutting plane method. Compared with the existing methods, this method has many advantages in efficiency, robustness and application range.

The invention relates to a power distribution network optimization scheduling method and system considering dynamic reconstruction of a network frame. Firstly, carrying out active power distribution network element time sequence modeling is carried out; then, establishing an active double-layer optimization scheduling model considering network dynamic reconstruction with the economy of power distribution network dispatching and the stability of fast voltage as objectives; and performing decimal encoding on branches in each basic loop, setting an iterative condition of a network frame population, adding a radial constraint condition of a power distribution network frame into an iterative strategy, and solving the active double-layer optimization scheduling model. The method can effectivelyreduce the operation cost of the power distribution network, smooth the voltage level of the power distribution system and improve the system voltage stability.

The invention belongs to the technical field of radar electromagnetic simulation, and discloses a rough surface and multi-target composite scattering simulation method based on iterative physical optics. The method comprises the following steps: inputting a rough surface power spectral density function and a roughness parameter, obtaining a rough surface geometric contour by the Monte Carlo method; using the simulation software FEKO to geometrically model the targets; adding the geometric models of the targets to the rough surface model to generate a composite model; using the physical opticsmethod to calculate the surface induced electromagnetic current directly scattered by the rough surface and each target; calculating the surface-induced electromagnetic current between the rough surface and the targets and between the targets according to the iterative strategy and Huygens principle; obtaining a far-field total scattering field of the composite model by the Huygens principle; andobtaining a two-station radar scattering coefficient of the composite model based on the total scattering field and the incident field of the composite model. The rough surface and multi-target composite scattering simulation method based on iterative physical optics has the advantages such as low memory requirement, high simulation efficiency and strong versatility of the simulation method.

The invention discloses a Raman spectrum image demixing method based on nonnegative matrix approximation. Non-pure pixels do not exist in a Raman spectrum image, unknown limitation on ANC, ASC and end elements exists in Raman spectrum demixing, various regularization factors such as sparsity are added in the nonnegative matrix approximation under the condition that the Raman spectrum image meets requirements of an LMM (labor market model), so that mixed pixel components and corresponding abundance thereof can be obtained effectively. Compared with the existing other method, the Raman spectrum image demixing method based on the nonnegative matrix approximation has the following advantages that the number of end elements (substances) is not required to serve as a priori knowledge; an iteration strategy is adopted, and a gain result is not relevant to the number of the end elements; and a graph relation and 1 / 2 bound norm are used, so that the noise immunity and the stability of an algorithm are improved.

The invention discloses an intelligent real-time prediction method for a high-speed large-range maneuvering target track. The method comprises the following steps of firstly, proposing a learning sample establishment method; constructing a target motion law learning and training mechanism based on an improved BP neural network; and finally, through a single-step prediction and rolling prediction method, realizing the intelligent, rapid and accurate prediction of the high-speed large-range maneuvering trajectory of the aerospace moving target. According to the invention, only the history of theaerospace moving target and the position data at the current moment need to be known, the motion model of the target is not needed, meanwhile, by designing a momentum factor and adopting a variable step size iterative strategy, the convergence speed of the traditional BP neural network is increased, and oscillation during the convergence process is reduced, and the precision of trajectory prediction is greatly improved. The method can be directly applied to the trajectory prediction problems of various high-speed and high-maneuverability targets, has the higher applicability, and provides thetheoretical basis and the technical reserves for the subsequent tasks, such as monitoring, tracking and intercepting the hypersonic aircrafts, such as X-37B, etc., and the like.

Multi-scale deep reinforcement machine learning for N-dimensional segmentation in medical imaging. Multi-scale deep reinforcement learning generates a multi-scale deep reinforcement model (22) for multi-dimensional (e.g., 3D) segmentation of an object. In this context, segmentation is formulated as learning an image-driven policy (38) for shape evolution (40) that converges to the object boundary.The segmentation is treated as a reinforcement learning problem, and scale-space theory is used to enable robust and efficient multi-scale shape estimation. By learning an iterative strategy to findthe segmentation, the learning challenges of end-to-end regression systems may be addressed.

The invention discloses an analysis method for a structure response interval comprising interval parameters based on a hyper-volume iterative strategy. According to the method, a structure response variable is taken as a target function, the maximum / minimum value of the response is solved through an optimization algorithm, and therefore an upper / lower boundary of the response is determined. The adopted optimization algorithm is a global optimization algorithm based on the hyper-volume iterative strategy, a strategy of Simpson integral is adopted to self-adaptively arrange integral points within a range of uncertain parameters, the corresponding maximum / minimum value of the target function in the integral points is founded, then an integral point corresponding to the maximum / minimum value of the target function is taken as a starting point, the global maximum / minimum value is obtained on the basis of a Newton method, and namely the upper / lower boundary of the uncertain parameters which is propagated in a nonlinear system. According to the analysis method for the structure response interval comprising interval parameters based on the hyper-volume iterative strategy, under an uncertain input condition, the interval response propagation boundary of a structure can be determined accurately.

The invention relates to an improved particle swarm-based offshore static target search scheme generation method. The method comprises the steps of establishing a calculation model of an offshore search task success rate; mapping the problem space to an algorithm space, and initializing a population; executing the first stage: performing iterative search in the solution space by utilizing the population through an iterative strategy, and analyzing the form of the solution space; executing a second stage: updating the population iteration strategy according to the solution spatial form, and performing iterative search to obtain a result after the iterative search is updated; and mapping the result after the update iteration search back to the problem space to obtain a search scheme and thetask success rate of the scheme. Historical data obtained by the population in the search process can be effectively utilized, the form of the solution space is analyzed, the search strategy is updated, and finally a high-quality search scheme is generated.

The invention discloses a video stabilization method based on a iterative strategy of a recurrent neural network. The method comprises the following steps: capturing paired video data by using a jitter video acquisition and stabilization processing hardware device; preprocessing an acquired video stream sample; designing and constructing an end-to-end full convolutional deep neural network based on an intra-frame and inter-frame iterative strategy of the recurrent neural network; inputting the preprocessed training data into a recurrent neural network, and guiding a training process of networkparameters by using linear weighting of four losses to obtain a trained model; and inputting the low-quality jitter test video into the trained neural network to obtain a stable version of the targetvideo. According to the invention, the historical motion state information is transmitted to each current video frame through the iterative strategy of the recurrent neural network in the aspect of time sequence, and the perception capability of the network for the jitter sequence frame information is enhanced, so that a stable picture can be predicted more accurately.

The invention relates to a satellite positioning technology, and discloses a satellite integrity monitoring method and system. According to the satellite observation value and the ephemeris obtained by a global navigation satellite system receiver, the integrity system of the satellite is calculated and monitored by adopting a weight selection iterative strategy and a robust estimation method. Thesatellite integrity calculation method has the advantages that when the number of the fault satellites is more than one, monitoring can be carried out, the fault satellites are eliminated. Meanwhile,the calculation efficiency and the calculation accuracy of satellite integrity calculation are effectively improved through the weight selection iterative strategy and the robust estimation method introduced by the satellite integrity calculation method.

The invention discloses an adaptive selection method for an adaptive region in real-time target matching and locating, and belongs to the technical field of computer image identification. The method comprises the steps of performing preprocessing and threshold segmentation on a real-time image; building a potential function by utilizing a spatial position relationship between a landmark and a target; by utilizing a local region stability index and a local region intensity index, performing first layer screening under the constraint of a spatial position potential function; performing non-maximum suppression on obtained candidate matching regions, and removing the regions with the relatively high overlapping degree; then performing correlation surface feature index screening and sorting; and finally selecting an optimal adaptive region through a potential function iterative strategy. Under the condition of insufficient matchable region quantity in target matching identifying and locating processes, the adaptive region which is abundant in information amount, is stable and reliable and is free of a repeat mode is selected in real time adaptively.

The invention relates to a detection method and system applied to Flash intelligent analysis and detection, an intelligent terminal and a computer readable storage medium. The detection method comprises the steps: S1, acquiring a Column total set, a Page total set and a Block total set, and presetting a bad Column total set, a bad Page total set, an error threshold value and an initial bad Block template; S2, alternately acquiring bad Page elements and bad Column elements in sequence from the Block total set based on an error threshold value, and alternately updating a bad Page total set and a bad Column total set in sequence; S3, on the basis of all the bad Column total sets corresponding to different error thresholds and all the bad Page total sets corresponding to different error thresholds, updating the error thresholds according to a bad template iteration strategy, obtaining a final Column total set from all the bad Column total sets, and obtaining a final Page total set from all the bad Page total sets; S4, obtaining a final bad Block template. The method has the advantages that the influence of the bad Page on follow-up operation of selecting bad Column elements is reduced, generation of false bad Column is reduced, and then the effective capacity of Nand Flash after analysis and detection is improved.

The invention discloses a process early warning method and system for analysis and estimation of a full measurement point coupling structure. The invention provides a full-measurement-point synchronous estimation and monitoring model based on analysis of a coupling structure between measurement points and measurement point estimation errors, and aims to solve the problems that an original monitoring method based on working condition estimation is incomplete in monitoring and insufficient in modeling of a coupling relation between the measurement points. On the basis that full measurement points are regarded as a full measurement point diagram, a multi-kernel graph convolutional layer is provided and applied to a full measurement point synchronous estimation and early warning model, and synchronous working condition estimation and monitoring of full sensor measurement points are achieved by explicitly modeling the coupling relation between the measurement points. According to the method, a self-iteration strategy based on feature approximation is designed for the provided model, so that the problem that estimation values of part of measuring points are abnormal due to strong coupling between the measuring points when a system is abnormal in a traditional method is solved, and the method has important significance on accurate monitoring of the industrial process.

The invention discloses an input vector-oriented RTL-level circuit reliability calculation method. Based on the principle that the modules can be calculated, an integrity linked list corresponding toeach module is constructed for each module through a depth-first search algorithm, so that an input source of each module can be obtained from an upper-layer sub integrity linked list of the module. The smooth calculation of each module is guaranteed. Based on a recursion principle, the reliability of the circuit module is analyzed by means of an SCA method; according to the constructed integritylinked list, an iterative strategy is utilized to realize reliability calculation of the RTL-level circuit; for the extracted modules, a recursive algorithm is used for detection, when it is found that all composition units of some sub-modules are basic gates, an SCA method is called to carry out calculation to guarantee precision, and when a calculation result is fed back to the top-layer linkedlist, the next module is called to carry out iterative calculation until the end of the top-layer linked list is reached. The method is suitable for the calculation of a super-large-scale integrated circuit and the use of a parallel strategy, and is high in calculation precision.

The invention provides an edge gateway, and an edge gateway dynamic policy service implementation method, device and system. The edge gateway comprises a policy module, a digital signature module, a pull module and a drive registration module. A script vm is embedded in the strategy module, and when a new strategy is updated, the strategy module is directly executed; the digital signature module is used for recording the md5 signature of the service subfile; the pull module is used for sending md5 signatures of all files to a strategy verification service by the edge gateway when an updated strategy exists, calculating sub-files needing to be updated, and pushing the strategy needing to be updated; and the drive registration module is used for interacting with the device based on the updated new strategy and the edge gateway. According to the method, the updating speed is high, the error-tolerant rate of the weak network environment is high, full-amount updating is not needed through dynamic strategy issuing, the updating size is small, and only necessary parts are updated; fast iteration is supported; the strategy verification service is simple, control and data separation can be achieved only through a control end, and the problem of compatibility with multiple versions is not needed.

The invention provides a conjugate gradient large-scale MIMO detection method based on deep learning for signal detection for a large-scale MIMO system in which the number of transmitting antennas is greater than that of receiving antennas. According to the method, proper trainable parameters are selected, a CG iterative algorithm with the trainable parameters is expanded into a deep neural network, and optimal parameters of each layer are found through network training. And through an initial value iteration strategy based on characteristic value estimation, the convergence of the algorithm is improved. According to the method, rapid convergence during parameter estimation can be ensured. In addition, the number of trainable parameters is only related to the number of network layers and is not related to the number of antennas. The characteristics ensure the rapid and stable training process of the method and the reasonable expandability of a large-scale system. According to the invention, a lower bit error rate is achieved with lower complexity.

The invention discloses a sectional calibration method for a successive approximation analog-to-digital converter. The successive approximation analog-to-digital converter for realizing the method comprises two N-bit SARADCs which are independent from each other and have deviation and two groups of digital calibration modules. Redundancy is introduced into two independent N-bit SARADC capacitor array structures, and back-end processing is carried out segment by segment through a digital method by using the redundancy, so that the ADC precision is improved. The digital calibration module realizes an LMS algorithm through a digital circuit so as to achieve a calibration purpose. The calibration algorithm is continuously carried out in the background, and normal work of the foreground is not affected. Compared with the existing calibration design, the segmented calibration algorithm uses two groups of different iteration steps, the configuration of two different iteration strategies is more flexible to adjust, and the calibration precision and speed can be better considered.

The invention discloses a method applicable to screen capture of various non-rectangular sections, and the method comprises the following steps: constructing a deep confidence multi-shape screenshot model based on a least square support vector machine principle; performing optimization training on the multi-shape screenshot model by using an optimal iterative strategy, and importing the trained multi-shape screenshot model into a screenshot mode; selecting the screenshot mode at a position to be subjected to screenshot, and triggering the multi-shape screenshot model to run a touch point received on a detection screen; when the number of the touch points is greater than a set threshold value, determining a screenshot range and a screenshot shape according to the number of the touch pointsand the positions of the touch points on the screen so as to carry out screenshot. According to the method, the number of the touch points can be accurately obtained under various scenes so as to carry out operation of screenshot of various shapes, the real-time performance and applicability of screenshot can be guaranteed while complex scene conditions are met, and the wide applicability and userexperience of screenshot are improved.

The invention discloses a semi-supervised multi-label learning method based on SPUM data. The method comprises the following steps of: considering single positive example unlabeled multi-label data (SPUM) data in which only one label in a plurality of labels of data is definitely labeled as positive and the other labels are in an unlabeled state, and solving a globally optimal solution through information transmission when the unlabeled data and the SPUM data are dispersed in different data nodes. According to the method, a real-time label prediction loss function based on positive class number constraint is adopted, so that the algorithm can learn a large amount of information by constraining the positive class number and alternately optimizing label prediction and model output in a scene that only a single label of partial data is labeled as a positive example, and the cost of a labeling task is greatly reduced. In a distributed network in which resources such as communication bandwidth are limited, an iterative strategy based on event triggering is adopted, so that the gradient information amount is allowed to be transmitted after reaching a certain degree, and the network burden is greatly reduced while the performance is kept.

The invention discloses a UAV path planning method based on second-order cone optimization, including: establishing a two-dimensional UAV motion model and transforming it; Description form; Convexize and relax the non-convex constraints existing in the problem, so that it can satisfy the second-order cone optimization framework, and introduce the L1 penalty function in the no-fly zone constraint to ensure the solvability of the problem; based on the above-mentioned various The cost function is constrained to design, so that the transformed problem is equivalent to the original problem; the problem is discretized, and a successive iteration strategy is given to solve it under the framework of second-order cone optimization. The present invention adopts the path planning method based on second-order cone optimization, which not only ensures that the designed path is the global optimum, but also directly solves the state quantity and control input of the UAV, avoiding additional design of control laws, and, in When applied, it also has the advantages of high calculation efficiency and easy expansion of the number and types of constraints.

The invention provides a Split Bregman weight iteration image blind restoration method based on a non-convex higher-order total variation model, and belongs to the technical field of image processing. The method is characterized in that firstly, a non-convex higher-order total variation regularization blind restoration cost function is obtained by introducing image border sparse prior information meeting a hyper-Laplacian model and by combining a high-order filter bank capable of generating piecewise linear solutions; secondly, a weight iteration strategy is provided, a minimization problem of the non-convex higher-order total variation regularization blind restoration cost function is converted into a minimization problem of an approximate convexity cost function with the updated weight; thirdly, the minimization problem of the approximate convexity cost function with the updated weight is converted into a new constraint solving problem through an operator split technology, and the constraint solving problem is converted into a split cost function through the method of adding a penalty term; fourthly, the split cost function is solved through a Split Bregman iteration solving frame. According to the Split Bregman weight iteration image blind restoration method based on the non-convex higher-order total variation model, an image can be restored effectively and rapidly, the shortage that a staircase effect is generated in a traditional total variation regularization blind restoration method is overcome, and meanwhile a better restoration effect on manually degraded images and actually degraded images is achieved.

This patent discloses a multi-temporal point cloud automatic registration method based on shape-invariant features: aiming at the point cloud registration characteristics of multi-temporal changing scenes, the point cloud rough registration is completed through shape-invariant point extraction and matching, and then the multi-temporal point cloud is searched The shape-invariant region in the phase point cloud estimates the rotation and translation parameters between multi-temporal point clouds, and completes the point cloud fine registration. Specifically, for point cloud coarse registration, firstly determine the four-point pairs with approximately the same name in the multi-temporal point cloud, and then calculate the feature descriptors of the points in the neighborhood centered on the four-point pairs, and determine the multi-temporal point cloud through feature matching and spatial geometric constraints Matching point sets with the same name between them, based on which the rotation and translation parameters are estimated, and the rough registration of the point cloud is completed; in the fine registration stage, this patent adopts an iterative strategy to extract the shape-invariant region in the multi-temporal point cloud, and further optimizes the rough registration accordingly. The initial rotation and translation parameters obtained in the quasi-phase, and finally the optimized parameters are used for the rigid transformation of the whole point cloud, and the automatic registration of the multi-temporal point cloud is completed.

The invention discloses a video stabilization method based on a cyclic neural network iterative strategy. The method includes: capturing paired video data with a shaking video collection and stabilization processing hardware device; preprocessing the collected video stream samples; designing and constructing End-to-end fully convolutional deep neural network based on the intra-frame and inter-frame iteration strategy of the cyclic neural network; input the preprocessed training data into the cyclic neural network, and use the linear weighting of four losses to guide the training process of the network parameters, Get the trained model; input the low-quality jitter test video into the trained neural network to get a stable version of the target video. The present invention transmits historical motion state information for each current video frame in time series through the iterative strategy of the cyclic neural network, which enhances the network's ability to perceive the shaking sequence frame information, thereby more accurately predicting stable pictures.

Multiscale deep reinforcement machine learning for N-dimensional segmentation in medical imaging. Multiscale deep reinforcement learning produces multiscale deep reinforcement models for multidimensional (e.g., 3D) segmentation of objects (22). In this context, segmentation is formulated as learning an image-driven policy (38) for shape evolution (40) that converges to object boundaries. This segmentation is treated as a reinforcement learning problem, and scale-space theory is used to achieve robust and efficient multi-scale shape estimation. The learning challenge of end-to-end regression systems can be addressed by learning an iterative strategy to find segmentations.

The invention discloses an image feature migration method based on self-organizing graph constraint non-negative matrix factorization. The problem that image non-negative feature migration is not seenacross fields is mainly solved. The method comprises the following steps: (1) respectively selecting a plurality of image samples from an auxiliary domain Ds and a target domain Dt to form a trainingsample set; (2) initializing a basis matrix A and a feature matrix S of an auxiliary field and a target field; (3) calculating a basis matrix graph GA and a feature matrix graph GS; (4) setting the number of iterations T, and optimizing by using an iterative strategy to obtain a final basis matrix A and a final feature matrix S; and (5) calculating the label of the test image through the basis matrix A. Characteristic self-organizing graph constraints are adopted, characteristic regression parameters are introduced, the characteristic migration robustness of an unseen image is enhanced, and the method can be used for solving the problem of image non-negative characteristic migration.