69 results about "Variational model" patented technology

A method of optimizing MOSFET device production which includes defining key independent parameters, formulating those key independent parameters into a canonical variational form, calculating theoretical extracted parameters using at least one of key independent parameters in canonical variational form, physics-based analytical models, or corner models. The method also includes calculating simulated characteristics of a device using the key independent parameters and extracting target data parameters based on at least one of measured data and predicted data, comparing the simulated characteristics to the target data parameters, and modifying the theoretical extracted parameters or key independent parameters in canonical form as a result of the comparison. Then, calculating and outputting the simulated characteristics based on the modified theoretical extracted parameters and the modified key independent parameters in canonical form.

Embodiments of the invention disclose a foreground target extraction method and apparatus, a device and a storage medium. The method includes the steps of obtaining a non-foreground target image and aforeground target image obtained by shooting the same region under the same environment; initializing an objective function, constructing a total variational model based on the non-foreground targetimage, the foreground target image and the objective function; updating the objective function iteratively, solving the total variational model and obtaining an optimal objective function; and performing threshold filtering on the optimal objective function according to preset filtering rules, and determining a foreground target corresponding to the foreground target image according to the filtering result and the foreground target image. The technical scheme of the embodiments of the invention overcomes the defects that the existing method can perform detection only at a specified location and cannot adapt to the shape change in the foreground target, and the extraction process is vulnerable to the influence of uneven lighting and environmental changes, and achieves complete extraction ofthe foreground target.

The invention discloses a method for restoring a motion blurred image based on a total variational model and a neutral network, which mainly solves the problem existing in the prior method that the image cannot be accurately restored. The implementation process comprises the following steps of: (1) constructing a Toeplitz matrix; (2) working out gradients in horizontal and vertical directions; (3) initializing the neutral network; (4) working out a neuron output; (5) working out the output of the neutral network; (6) working out a first variation delta E1 of a network energy function; (7) if the neurons are completely updated, jumping to the step (4); otherwise, jumping to a step (8); (8) if the set iterations are reached, outputting a restoration result, otherwise jumping to a step (9); (9) working out a restoration error; (10) if the restoration error is smaller than the set error, outputting the restoration result, otherwise jumping to a step (11); (11) working out the current input bias matrix; (12) working out a second variation delta E2 of the network energy function; and (13) if the summation of the delta E1 and the delta E2 is less than 0, jumping to a step (2); if the summation of the delta E1 and the delta E2 is more than or equal to 0, jumping to the step (3); and if the delta E1 is equal to 0, outputting the restoration result. The method can obtain relatively accurate restored images, and be applied to the restoration of motion blurred images.

The invention discloses a light and face identifying method based on light estimation. The method includes the following steps: firstly, in accordance with an image imaging model, estimating a preliminary light from an original face image, then in accordance with the preliminary light, computing a light shielding correction light and light fusion factor, constructing a light fusion model, realizing light shielding light correction, then using a total variational model to conduct smoothing filtering on the light after the light shielding correction to acquire the final light of the original face image, finally using a light invariant extraction model to acquire the light invariant of the original image. According to the invention, the method can better process the problem of shadow edge, extracts the light invariant which contains rich and detailed face characteristics and is closer to original face intrinsic, can meet imaging theories of images, and is applied to the complex lighting environment by means of slow changing characteristics of light.

The invention discloses a fault diagnosis method under the action of a central frequency convergence trend. The method comprises the following steps: (1) collecting a dynamic signal x (t) of rotary mechanical equipment; (2) setting initial decomposition parameters of the variational model; (3) decomposing the dynamic signal x (t) by using a variational model with set initial decomposition parameters, and traversing the signal analysis frequency band under the guidance of a central frequency convergence trend to iteratively decompose the dynamic signal x (t) to obtain an optimized mode {m1... Mn... MN} and a corresponding central frequency {omega1... Omegan... OmegaN}; (4) searching a fault correlation mode mI, guiding parameter optimization by the central frequency omega I of the fault correlation mode mI, and extracting an optimal target component containing fault information; and (5) performing envelope analysis on the optimal target component, and diagnosing the rotating mechanicalequipment according to an envelope spectrum. According to the fault diagnosis method provided by the invention, intelligent decomposition of the original dynamic signal of the diagnosis target equipment is realized by adopting a decomposition mode guided by a central frequency convergence trend, the acquired equipment dynamic signal can be adaptively analyzed, and the difficulty of performing mechanical fault diagnosis by a technician by using a variational mode decomposition method is reduced.

The invention discloses an infrared image enhancement method based on transmission image fusion. The method comprises: carrying out under-segmentation and super-segmentation on an image under different scales through a clustering algorithm, and generating two clustering segmentation results; for the clustering segmentation result, generating two transmission images by using image pure pixel priori; constructing a fusion weight map for each transmission map, constructing an image pyramid structure model for each transmission map and the fusion weight map thereof, and performing layer-by-layer fusion to obtain a fused transmission map; based on the variational model, carrying out edge-preserving smoothing processing on the fused transmission map to obtain an optimized transmission map; and obtaining an enhanced infrared image by using the optimized transmission image. According to the method, a large number of originally hidden scene details in the infrared image can be recovered, and negative effects cannot be introduced.

The invention discloses a renewable energy short-term prediction method based on data mining and variational mode decomposition. The renewable energy short-term prediction method comprises the following steps: 1), dividing a clustering process into data clusters: dividing the data clusters into some different groups in a given data set through a clustering algorithm, and extracting historical power generation data similar to environmental features into the same class; 2) processing historical data by variational modal decomposition: determining the bandwidth and the center frequency of each modal component by continuously iterating and searching the optimal solution of a variational model, and performing adaptive subdivision on the signal in a frequency domain; and 3) utilizing an extremelearning machine to predict future data. The technical scheme of the embodiment of the invention is combined with the K-means clustering technology, utilizes the variational mode decomposition and theextreme learning machine to predict the power generation amount of renewable energy, and has better accuracy compared with a traditional prediction method.

The invention discloses a hyper-spectral image super-resolution method based on hyper-parameter fidelity and depth prior joint learning. The method comprises the following steps: establishing a hyper-spectral and multi-spectral image fusion variational model based on depth prior regularization of a hyper-parameter fidelity model; optimizing a hyperspectral multispectral image fusion variation model; carrying out tensor representation on the model optimization iteration process; performing network expansion on the iterative process of variational model optimization, and executing the iterative process of optimization; and training the network by using the L1 norm as a loss function. The method has the capability of representing the hyperspectral image degradation model and the data prior in the network at the same time, and has excellent performance when being applied to hyperspectral multispectral image fusion.