33 results about "Natural gradient" patented technology

The invention discloses a glass cover plate surface color gradient printing process and relates to the technical field of cover plate surface processing. The printing process comprises the steps of (1) manufacturing a five-layer screen printing plate; (2) conducting screen printing through A-color printing ink; (3) conducting screen printing through B-color printing ink; (4) conducting screen printing through white semi-transparent ink; (5) conducting screen printing through white ink; and (6) conducting screen printing through black ink, wherein the color A and the color B are different colors. According to the glass cover plate surface color gradient printing process, the structure of the screen printing plate is designed, and multiple layers of ink in colors is printed on the surface ofa cover plate through the screen printing plate; through mutual overlapping and permeation of the multiple layers of the ink, the natural gradient effect of the colors in a printing region is achieved; the gradient colors are uniform, natural, clear and not dull, and the appearance visual effect is good. Through the glass cover plate surface color gradient printing process, there is no need to purchase expensive equipment, the cost is low, the preparation process is simple, and the process meets the requirements for industrial production.

A natural gradient, single well, tracer dilution non-continuous mixing apparatus and method for measuring flow of liquids through porous media are described. The method has been applied to concentrations of Light Nonaqueous Phase Liquid (LNAPL) in monitoring wells to measure the rate of LNAPL flow through the wells and through the adjacent formation. Similar measurements were made for concentrations of water in other formations. A LNAPL-soluble fluorescing tracer was added to LNAPL in the wells, and a water-soluble fluorescing tracer was added to water in the wells. The tracer is initially uniformly-mixed into the LNAPL or water in the well, and the tracer concentration is measured using a fiber optic cable and a spectrometer. The LNAPL or water with dissolved tracer is then allowed to flow from the well without any mixing. At a later time, the LNAPL or water and tracer in the well are mixed to a uniform tracer concentration, and the tracer concentration is remeasured. Using the initial tracer concentration, the subsequently mixed tracer concentration, the elapsed time, and the well diameter, a LNAPL flow rate is calculated.

The invention discloses an underground cascade reservoir and a method for constructing the same, wherein, the underground cascade reservoir comprises at least two grooves, impermeable geomembranes paved at the bottoms of the grooves; a cutting undercurrent filter pipe arranged along the top of the impermeable geomembranes of the grooves, control valve inlet wells arranged at the downstream ends of the cutting undercurrent filter pipes; an impervious wall arranged at the downstream ends of the grooves, wherein, the control valve inlet wells are orderly connected by conduits being connected orderly. The construction method comprises the following steps that: (1) the grooves are dug; (2) impermeable parts are prepared; (3) the cutting undercurrent filters are paved; (4) the control valve inlet wells are arranged; (5) the conduits are paved. The invention has the advantages that: the underground cascade reservoir is constructed under the condition of the natural gradient of the natural hypsography of the river, and is positioned in the seasonal mountains and rivers and utilizes the natural gradient of the mountains and rivers to collect the natural water resources maximumly.

A method and apparatus is disclosed for performing blind source separation using frequency-domain normalized multichannel blind deconvolution. The multichannel mixed signals are formed as frames of N samples, which consist of r consecutive blocks of M samples. The frames of mixed signals are separated using separating filters in the frequency domain in an overlap-save manner using a discrete Fourier transform(DFT). The separated signals are then converted back into the time domain using the inverse DFT to be applied to a nonlinear function. The cross-power spectra between separated signals and nonlinear-transformed signals are computed and are normalized by the power spectra of separated signals and the power spectra of nonlinear-transformed signals to have flat spectra. The invention then applies the time domain constraint to preserve the first L cross-correlations. These alias-free normalized cross-power spectra are further constrained by nonholonomic constraints. The invention then computes natural gradient by convolving alias-free normalized cross-power spectra with separating filters. After the length of separating filters is constrained to L, separating filters are updated using the natural gradient and normalized to have unit norm. The terminating conditions are checked to see if separating filters converged.

The invention provides a speech signal blind separating method based on a variable step size natural gradient algorithm. (1) A speech signal collecting device is used for collecting mixed speech singles of speeches of multiple speakers, and the number of microphones is larger than or equal to that of the speakers; (2) the collected mixed speech signals are preprocessed, and then mixed speech signals which have zero mean, irrelevant signal sources and the high signal to noise ratio are obtained, wherein the preprocessing comprises the steps of adopting an LMS digital filter, removing the mean and removing related whitening processing; (3) the estimation of the speech of each speech source is restored and obtained from the mixed human speeches by using the variable step size natural gradient algorithm for regulating step size based on gradient factors. The speech signal blind separating method based on the variable step size natural gradient algorithm is capable of separating real mixed speech signals, high in separating speed and accurate and stable in separating effect.

The invention relates to an irrigation method and system for a desertified river bank wind prevention and sand fixation forest belt. The system comprises a water taking facility which is arranged in the upstream river channel of a river and formed by multiple water taking permeable pipes, the water taking facility is connected with at least one water collecting facility arranged on the lower portion of the water taking facility through pipelines, the water collecting facility is connected with irrigation facilities arranged on the lower portions of the water collecting facilities, and the irrigation facilities are arranged on the outer side of the wind prevention and sand fixation forest belt. Accordingly, the mode of infiltration water taking under a river bed is adopted, blocking of silt and sundries is avoided, and excellent clean water is improved for irrigation. Meanwhile, the water taking facility, the water collecting facilities and the irrigation facilities are arranged on the river from the upper portion to the lower portion according to the homeopathy of the terrain, gravity-flowing irrigation is formed by means of the natural gradient, irrigation is conducted by completely replying on the natural gravity without using any manual power facility, and the operation cost is greatly lowered. Due to the fact that no mechanical power device is arranged, loss in daily operation is very low, regular maintenance is not needed in general, and the operation cost is lowered.

The present invention provides a natural gradient ramp anti-fake method and a product. The method comprises the steps of: printing codes onto a printing stock, printing B-color ink liquid to A-color ink without drying to allow the A-color and B-color ink to be intersected, fused, diffused and mixed so as to form a random pattern with the gradient ramp trending to C color; and collecting personality characteristic information on the random pattern to combine and store the personality characteristic information and the codes into a cloud database to take as anti-fake files for checking true andfalse by customers. The natural gradient ramp anti-fake method and the product are zero in device investment, simple in manufacturing process and low in production cost, cannot be copied and are easyto identify.

The invention relates to a mixed signal adaptive fast separation method based on proportional-derivative control, including the steps: firstly, the initial separation signal is calculated; the initialvalue of the step length is determined, and the first transformation matrix and the separation signal are calculated by using a natural gradient blind source separation algorithm; then, the next variable step size related to the separated output state is calculated by the transformation of the difference of the kurtosis cumulants of the separated signals between the separated iterative calculations, and a new separation matrix is calculated from the next variable step size; it is judged whether the new separation matrix meets the control precision requirement; and if so, the separation signalis calculated by using the separation matrix which meets the control precision requirement. As the initial step length is within a certain value range, and the maximum convergence speed is close to the uniform order of magnitude under the same stable error. In particular, the technical parameters of the transformation are related physical quantities output from the separation process, and the composition of the separation algorithm has self-adaptability and practical use value.

The invention discloses a neural network noisy vocal print recognition method based on residual error and batch normalization, which comprises the following steps: preparing a data set in a noisy environment, and respectively extracting corresponding Mel-frequency cepstral coefficients from voice signals; constructing a ResTDNN network structure: after the construction is completed, training the network by using a natural gradient random descent method, after the training is completed, capturing vocal print information of a speaker by using voice information about 10 seconds through x-vector extracted by dimension reduction, and adopting a cross entropy loss function; and performing dimension reduction by using linear discriminant analysis, performing length normalization on an x-vector after the dimension reduction, training linear probability analysis, respectively calculating likelihood functions of two voices from different spaces, and evaluating a recognition system by adopting equal error rate and minimum detection cost. According to the method, the robustness of the time delay neural network in the noise environment can be improved, the neural network degeneration and gradient disappearance phenomenon requirements are relieved, and the vocal print recognition rate is improved.

The invention provides a semi-open type reclamation sludge mudflat and a mudflat building method thereof. The mudflat structurally comprises a built seawall, a newly-built siltation-promotion submerged dyke, a hydraulic reclamation block and a hydraulic reclamation pipeline, wherein the built seawall is connected with the newly-built siltation-promotion submerged dyke, the hydraulic reclamation block is arranged between the built seawall and the newly-built siltation-promotion submerged dyke, and the hydraulic reclamation pipeline penetrates through one side of the built seawall to be connected with the hydraulic reclamation block. The newly-built siltation-promotion submerged dyke comprises a siltation-promotion submerged dyke body, a levee crown, a levee slope and a water-permeable structure. Compared with a traditional technology, under the conditions that a closure zone has not been closed, one side or two sides are opened, and water inside the closure zone mutually exchanges withwater outside the closure zone, on the dependence of an existing shoreline or dyke, the submerged dyke for reducing water power is properly built so as to achieve semi-open type hydraulic reclamationor naked blowing, and therefore, the aim of increasing mudflat resources is achieved. Meanwhile, through the natural gradient of dredger fill, the good ecological mudflat with high, middle and low beaches is formed. The technology is simple, the mudflat building speed is high, and the cost is economical; dredged spoil matter can be utilized, and the mudflat meets ecological green building requirements.

The invention provides a voice signal blind separation method based on a variable step size natural gradient algorithm. (1) collect the mixed voice signal of multiple voices with the voice signal acquisition device, and the number of microphones is more than or equal to the number of speakers; (2) preprocess the mixed voice signal collected, including: adopting LMS digital filter, removing the mean value Whitening processing with decorrelation, and then get a mixed speech signal with zero mean value and high signal-to-noise ratio that is irrelevant to each source; (3) Using the variable step size natural gradient algorithm based on the gradient factor to adjust the step size to recover from the mixed human voice Estimation of the sound of each sound source. The invention not only can separate real mixed voice signals, but also has fast separation speed, accurate and stable separation effect.

The invention belongs to the field of SAR image speckle noise reduction processing, and particularly relates to a polarized SAR image speckle noise reduction method based on the natural gradient of a PID. The polarized SAR image speckle noise reduction method based on the natural gradient of the PID includes the steps that n polarized SAR images collected at the same moment are converted to be a one-dimensional vector image; the one-dimensional vector image is preprocessed; iteration solution is carried out on the preprocessed image signal through a natural gradient algorithm of the PID to obtain a separation matrix, and one-dimensional signal output of each image is obtained by multiplying the image signal by a whitened signal; the images are reconstructed; a one-dimensional signal with the maximum equivalent number of looks is found out; an image with the maximum equivalent number of looks is the SAR image with reduced noise, the one-dimensional signal with the maximum equivalent number of looks is converted into a two-dimensional image, and the two-dimensional image is converted to be a gray level image with N*M pixel dots according to a two-dimension to one-dimension inverse conversion method. The method with the PID applied adjusts the separation step length of the natural gradient algorithm, so that the processing result image is clearer than that obtained according to a traditional method, and speediness is achieved.

The invention discloses an MIMO-OFDM (Multiple Input Multiple Output-Orthogonal Frequency Division Multiplexing) signal blind separation method for increasing the natural gradient algorithm convergence speed. The method comprises the following steps: firstly, making improvements on a Flexible ICA algorithm under the constraint of ICA preprocessing in combination with an ICA principle to obtain an improved iterative algorithm; secondly, giving a method for adaptively feeding back and adjusting an iteration step size based on a separation error in combination with a BP neural network feedback thought with the separation error; and lastly, giving a comprehensive ending criterion of algorithm iteration from the aspects of the convergence of a separation matrix, iteration times and the separation error. Compared with the prior art, the method has the advantages that the requirements on the convergence speed and the separation accuracy are met on the basis of combining the MIMO-OFDM system features, and an effective separation effect is achieved.

A computer-implemented method for training a normalizing flow. The normalizing flow predicts a first density value based on a first input image. The first density value characterizes a likelihood of the first input image to occur. The first density value is predicted based on an intermediate output of a first convolutional layer of the normalizing flow. The intermediate output is determined based on a plurality of weights of the first convolutional layer. The method for training includes: determining a second input image; determining an output, wherein the output is determined by providing the second input image to the normalizing flow and providing an output of the normalizing flow as output; determining a second density value based on the output tensor and on the plurality of weights; determining a natural gradient of the plurality of weights with respect to the second density value; adapting the weights according to the natural gradient.

The embodiment of the invention discloses a method and device for identifying overlapped digital signals. Two-order, four-order and six-order cumulants of standard signals in different modes are calculated in advance; characteristic values of the standard signals in various modes are calculated and stored according to the cumulant result; and the identification process comprises the following steps of calculating a separate matrix by using a natural gradient ICA algorithm according to a received overlapped signal matrix; calculating the overlapped signal matrix by using the separate matrix to obtain a final estimated value matrix of source signals, wherein each component, which is not zero, in the final estimated value matrix of the source signals is a separated source signal; calculating the two-order, four-order and six-order cumulants of each component, which is not zero, and calculating the characteristic value of each component, which is not zero; and determining the modulation manner of each separated source signal according to the characteristic value of each component, which is not zero, and the characteristic values of the standard signals in various modes. By means of the embodiment of the invention, the overlapped signals can be separated; the identification performance of the overlapped signals is improved; and the overlapped signals are demodulated easily.

The invention discloses a device for using a stable isothermal layer to testing structure dynamic properties. The isothermal layer exists in a temperature box. After the starting of a heat source, the temperature in the box forms a natural gradient along the vertical direction due to a gas heat-expansion and cold-contraction characteristic and the natural adjustment of gravity. The temperature at the same horizontal height is identical and varies little with time. According to the invention, a temperature field of the stable isothermal layer is used, the uniform temperature of a test-piece is ensured and the dynamic property test of the test-piece is completed. According to the invention, the isothermal layer is formed through the natural convection of gas, the temperature control difficulty is lowered, the reliability is high, the power dissipation is low, the space is saved; the rapid switching of a temperature point can be realized through moving the test-piece into different isothermal layers; the device is simple, and the disassembly and assembly are convenient; in addition, micro-adjustment is performed on the temperature gradient by means of thermal insulation, heat reflection and heat dissipation processing of the box, and the temperature control is flexible.