49 results about "Fixed point algorithm" patented technology

The invention provides a CNN-based low-precision training and 8-bit integer quantization reasoning method. The method mainly comprises the steps of carryin gout low-precision model training; Performing model training by using a 16-bit floating point type low-precision fixed point algorithm to obtain a model for target detection; Quantifying the weight; Proposing an 8-bit integer quantization scheme, and quantizing the weight parameters of the convolutional neural network from 16-bit floating point type to 8-bit integer according to layers; carrying out 8-bit integer quantitative reasoning; quantizing the activation value into 8-bit integer data, i.e., each layer of the CNN accepts an int8 type quantization input and generates an int8 quantization output. According to the invention, a 16-bit floating point type low-precision fixed point algorithm is used to train a model to obtain a weight; Compared with a 32-bit floating point type algorithm, the method has the advantages that the 8-bit integer quantization reasoning is directly carried out on the weight obtained by training the model, the reasoning process of the convolutional layer is optimized, and the precision loss caused by the low-bit fixed point quantization reasoning is effectively reduced.

Approaches for managing and maintaining a state of charge of an energy storage device by adjusting (biasing) responses to electrical grid operator commands to perform ancillary services are disclosed. In embodiments, methods and systems regulate a set point regulation in an energy system. In an embodiment, a method determines when the set point needs to be changed, calculates a new set point, and moves the output of the system from an old set point to the new set point at a defined ramp rate. The method then incorporates, as part of a set point algorithm, the capability to restore the energy storage device to a desirable state of charge (SOC). Embodiments implement Dynamic Bias, SOC and Signal Bias Range Maintaining, Operational Limits, and Fixed Signal Bias algorithms and perform Intelligent Algorithm Selection to manage and maintain the SOC of an energy storage device by biasing responses to grid operator commands.

The invention discloses a high-efficiency stable multifunctional single-stage photovoltaic single-phase grid-connected control method, and belongs to the technical field of photovoltaic power generation control. The method is characterized in that: the method adopts a high-speed digital signal processor and applies a C language dynamic fixed-point algorithm to form a high-performance control algorithm of a whole grid-connected control system; the method adopts a single-stage grid-connected control structure and a quick smooth maximum power point tracking method to directly transmit the maximum energy sent by a photovoltaic array to a local urban power grid through first-stage DC / AC transformation; the harmonic and idle work produced by a local load can be compensated by adopting an FBD method, so the method is favorable for improving the electrical energy quality of the local power grid; and a proportional resonant converter is used to implement precise closed-loop control for the grid-connected current so that the photovoltaic grid-connected current and the commercial single-phase voltage are totally synchronous while implementing maximum power point tracking of the photovoltaic array. The power factor is high, the system efficiency is high, the operation of the whole system is safe and reliable, and the aberration rate of the grid-connected current harmonic is lower than 3 percent.

The invention provides an L-curve-based hyperspectral unmixing method for estimating a regularized parameter automatically. An expected target is reached by using the method in hyperspectral unmixing. The method comprises the following steps: (1) reading hyperspectral remote sensing image data V with a computer; (2) determining the number of end members in the hyperspectral image by a virtual dimensionality (CD) method; (3) initializing an end member matrix W and an abundance matrix H by a vertex component analysis (VCA) algorithm, and performing projection operations on the W and the H; and (4) determining a regularized item, determining an iteration expression of the regulated parameter alpha, the end member matrix W and the abundance matrix H, and iterating the alpha, the W and the H until simultaneous constringency of a target function and the regularized parameter. The algorithm is characterized by overcoming the unreasonable disadvantages caused by selecting the regularized parameter manually in an NMF in the prior art, utilizing a fixed point algorithm to estimate an optimal regularized parameter, and dynamically weighing a fitting degree function and the regularized item automatically so as to achieve a better unmixing effect.

The method allows, by means of an iterative inversion algorithm, to predict the spatial distribution of the lithologic composition of sediments deposited in a sedimentary basin during a geologic time interval, and the temporal evolution of the depositional profile throughout filling of the basin, while respecting exactly the thicknesses of the sedimentary sequences measured otherwise. The input data consist of the thickness maps of the sedimentary layer studied, data relative to the location and to the composition of the sediment supply at the boundaries of the sedimentary basin studied, and physical parameters characterizing transport of the sediments during the period considered. These data are determined by interpretation of seismic surveys, by measurements and observations. This set of data is applied to an iterative inversion loop initialized by the accommodation provided by a stationary multilithologic diffusive model. This loop then works as a fixed-point algorithm correcting the accommodations by means of a preconditioning of the residue on the sequence thicknesses obtained by an instationary model. This preconditioning is obtained by applying the tangent application of the stationary multilithologic model to the residue on the thicknesses. At the output, we obtain the accommodation at the end of the sequence considered, as well as the bathymetry and surface composition solutions at any time of the sequences of the instationary model calibrated to the layer thicknesses. Application: location of hydrocarbon reservoirs notably.

The invention provides a high-efficiency stable photovoltaic single-phase grid connected control method with double MPPT functions and belongs to the technical field of photovoltaic generating control. The method is characterized in that: a high-speed digital signal processor and a C-language dynamic fixed point algorithm are utilized for forming a high-performance high-reliability control algorithm for an overall photovoltaic single-phase grid connected control system; by using a two-stage type grid connected control structure, a maximum power point tacking strategy of double MPPT rapid smoothing, and a grid connected current prediction dead-beat control algorithm, multiple strings of photovoltaic arrays are enabled to be at optimization operation states and the maximum energy emitted by the photovoltaic arrays is transformed through a stable bipolar type structure for outputting to a local electrical network; and at the time when the multiple strings of photovoltaic arrays are provided with double MPPT maximum power point tacking and grid connected current prediction dead-beat control, photovoltaic grid connected current is enabled to be completely consistent with a local single-phase voltage phase. The advantages are as follows: the power factor is 1, and the grid connected current harmonic wave distortion rate is lower than 3%, thus the system operation is safe and reliable.

A mechanism for reconstructing a signal (e.g., an image) based on a vector s, which includes measurements of the signal. The measurements have been acquired using at least a portion of a measurement vector set represented by a matrix H. Each of the measurements corresponds to a respective row of the matrix H. (For example, each of the measurements may correspond to an inner product between the signal and a respective row of the matrix product HD, wherein D is a generalized permutation matrix.) A total-variation primal-dual hybrid gradient (TV-PDHG) algorithm is executed based on data including the matrix H and the vector s, to determine an estimate for the signal. The TV-PDHG algorithm is implemented in fixed-point arithmetic.

The invention provides a floating point calculation method based on a FPGA. The method comprises the steps of processing a group of data which is identical in index and different in mantissa and used as a data block, expressing the data in calculation by adopting a fixed point format, adjusting the data precision by virtue of left shifting, right shifting the data to avoid the overflow error in the fixed point calculation, and dividing result data by a preset gain to obtain correct data after the calculation is ended. By adopting the floating point calculation method, the contradiction between a fixed point algorithm and a floating point algorithm is solved, the floating point calculation efficiency is improved, and the cost is reduced.

The invention provides an operation method for FFT. The method comprises the following steps: performing block floating point operation, moving data blocks leftwards or rightwards to regulate the signal power of FFT input during operation, performing data regulation according to an internal calculation result of each stage after the data operation of each stage by virtue of a special block floating point operation module, storing a set of data sharing a shift factor with an independent data field on hardware, and dividing result data by a preset gain to obtain correct data after operation, wherein the shift factors of the data blocks in block floating points are determined by maximum values of all the data in the whole data blocks. According to the floating point operation method, conflicts between a fixed-point algorithm and a floating point algorithm are solved, floating point operation efficiency is improved, and cost is lowered.

This invention discloses one output key message mixture feedback chaotic flow code decoding method, which multiple overlap one section number parameter section linear chaotic mapping signals to generate Key flow. After merging with chaotic odd bit to generate key code flow, couple bit and message merge as back feedback to keys generator to overcome the limit accuracy effect through sequence. The invention has the following technique effects : a, good security saving without stealer to get enough chaotic track information and to output key message mixture key codes system to resist each attack method; b, simple type to realize system by set formula; c, good flexibility to adjust code speed.

A system including an integrated circuit chip also includes a microcontroller in the chip and an algorithm for execution by the microcontroller. The algorithm includes addition, subtraction, and multiplication operators (e.g. 25,15,20) and shift-left and shift-right operators (e.g., 48,21) configured for solving particular equations (Eqns. 1-4). Input numbers are within particular ranges to allow the shift operators to shift binary bits so each number so it fits within a register of a particular width. An IR sensor (4) may convert IR radiation (3) to produce a voltage (Vobj) representing the temperature (Tobj) of an IR emitting object (2). The algorithm (100) operates in conjunction with the microcontroller (7) to convert the voltage (Vobj) into a value representing the temperature (Tobj) of the remote object (2) without keeping track of decimal points and resolution of the numbers.

The invention discloses a DWT-SVD-ICA-based digital audio watermarking algorithm. The algorithm is performed according to the following steps: the first step, preprocessing is performed on an original audio signal, and a watermark is embedded in a selected appropriate segment; the second step, when the watermark is embedded, three-level wavelet conversion is performed on the original audio signal, singular value decomposition is performed on an approximate component of the wavelet conversion, and then watermark embedment is completed by utilizing a random mixing method; and the third step, when the watermark is extracted, wavelet conversion and singular value decomposition are performed on the audio signal embedded with the watermark, and then watermark blind detection is achieved by utilizing a negentropy maximization fixed point algorithm. Compared with a traditional ICA-based audio watermarking algorithm, the DWT-SVD-ICA-based digital audio watermarking algorithm is advantageous in that watermark robustness is improved to some extent, that is to say, the watermark robustness is high; and imperceptibility of the watermark can be guaranteed, and blind extraction of the watermark can also be guaranteed.

The invention discloses a three-dimensional face feature point positioning method based on a noise reduction self-coding network. The method comprises the following implementation steps: extracting anose tip point coordinate from a messy face point cloud; extracting a human face region through the position of the nose tip point and preprocessing the human face region; and training a face model, carrying out manual region segmentation on the face by the model, and dividing the to-be-detected face into a plurality of regions according to the segmented face and a rigid matching algorithm; performing shielding detection on each region, and converting the shielding degree into coefficient representation; training a noise reduction auto-encoder for each divided shielding area, wherein each noise reduction auto-encoder outputs a feature point positioning result; fusing the positioning results of the plurality of feature points through the shielding coefficient to obtain a final result, and finishing the whole fixed point algorithm.

The invention relates to a brain function connectivity detection system based on self-adaptive priori information guidance and a brain function connectivity detection method utilizing the system. The method comprises the steps that S1, blind source signal separation is performed separately on functional magnetic resonance data of all single-subjects in group-subjects collected by the same mask through an independent component analysis method, so independent functional components corresponding to all the single-subjects are obtained; S2, adaptive prior information used for guiding functional magnetic resonance data analysis on the group-subject and single-subject levels is extracted from the functional components corresponding to all the single-subjects; S3, by utilizing the adaptive prior information, based on a multi-objective optimization framework, in combination with a weight summing algorithm and a fast fixed-point algorithm, blind source signal separation is performed on the functional magnetic resonance data on the group-subject level, group functional components reflecting all subject commonalities in the group are obtained, so that brain function connectivity detection is completed. The brain function connectivity detection system can position a brain function connectivity area more accurately.

The invention discloses a fixed-point acceleration method for data. The method comprises the following steps: acquiring test data and first model parameters; inputting the test data and the first model parameters into a fixed-point algorithm model needing fixed-point parameter adjustment to obtain a first characteristic value; Converting the data format of the first characteristic value into a floating point format from a fixed point format, and obtaining a second characteristic value represented by the floating point format; and performing fixed-point operation on the second characteristic value by using a GPU acceleration mechanism to obtain output data. The invention further provides a fixed-point acceleration device, electronic equipment and a computer storage medium. According to theinvention, the calculation of the fixed-point algorithm model can be accelerated, and the fixed-point iteration period is accelerated.

The invention belongs to the technical field of automatic knitting machines, in particular to an electromagnetically needle selecting method of a computerized flat knitting machine based on a Q15 fixed-point algorithm. The method is characterized by comprising the following steps: (1) reading needle selecting information; (2) storing information of needle pressing states of blades on a needle selector; (3) reading selected needles in a nose direction; (4) utilizing a coder to feed back pulse signals and adopting the Q15 fixed-point algorithm to carry out real-time data processing so as to determine positions of the blades of the needle selector; (5) calculating state values of the blades of the needle selector, i.e. determining the state values of the positions of the blades of the needleselector by judging a state value of the machine head direction, i.e. controlling 0.1 values of the blades; (6) comparing the states of the blades; (7) driving the blades with different state values to act; and (8) judging whether a needle selecting process is ended or not. On one hand, the needle selecting method improves the speed and the precision of needle selection, and on the other hand, solves the problem of reliability of the needle selection, thereby not only saving manufacturing cost, but also effectively overcoming technical contradiction of the speed and the reliability of the needle selection.

The invention provides an operation method of an FFT (Fast Fourier Transformation) processor. The method comprises the following steps: setting multiple stages of butterfly operation in an FPGA (Field Programmable Gate Array) program, wherein all the stages share a block floating point shifting factor; performing judgment through data according to the judgment state of the previous block floating point factor before each stage of operation to decide a shifting choice during the output of a data memory; controlling a gain which is output finally through the shifting sum of all the stages. The invention provides a floating point operation method, the contradiction between a fixed point algorithm and a floating point algorithm is solved, the floating point operation efficiency is improved, and the cost is reduced.

In a computational device, a method is provided for determining attributes for results of arithmetic operations, including a result for a series of arithmetic operations, with fixed-point numbers, in an environment for fixed-point algorithm development. An indication is received of a plurality of possible word lengths to be used for storing a plurality of digits representing a result of arithmetic operation with fixed-point numbers. An indication is received of how a placement of a radix point is to be determined in the plurality of digits representing the results of the arithmetic operation with fixed-point numbers. When calculating a selected fixed-point arithmetic operation, one of the word lengths in the plurality of possible word lengths is employed for storing the plurality of digits representing a result of the selected fixed-point arithmetic operation. A placement of a radix point in the plurality of digits representing the result of the selected fixed-point arithmetic operation is based on the received indication of how the radix point is to be determined. A growth rate for a number of digits in a result of a series of arithmetic calculations is less than N, and preferably proportional to logb (N), where N is equal to the number of arithmetic operations performed, and b is the base of operands in the arithmetic operations.

The invention discloses a design method of a neural network activation function. The method includes designing a neural network framework, selecting a saturation activation function as the neural network activation function; According to the application scenario of the neural network, selecting the initial fixed-point bit width of the data. Determining The initial fractional bit width and the initial integer bit width according to the precision requirement and the data characteristics of the neural network. Performing Binary conversion of the fractional part and the integer part; Taking the converted fixed-point format data as input, training the neural network and recording the training results. Record the training test results; Repeating steps 2 to 6 until the overall fixed-point width,decimal-bit width and integer-bit width meet the requirements are found as the final neural network fixed-point architecture. Determining saturation value of activation function of neural network; Implementing forward propagation function; Implement back propagation function. The invention saves power consumption and cost, and solves the problem of fixed-point algorithm overflow.

A system including an integrated circuit chip also includes a microcontroller in the chip and an algorithm for execution by the microcontroller. The algorithm includes addition, subtraction, and multiplication operators (e.g. 25,15,20) and shift-left and shift-right operators (e.g., 48,21) configured for solving particular equations (Eqns. 1-4). Input numbers are within particular ranges to allow the shift operators to shift binary bits so each number so it fits within a register of a particular width. An IR sensor (4) may convert IR radiation (3) to produce a voltage (Vobj) representing the temperature (Tobj) of an IR emitting object (2). The algorithm (100) operates in conjunction with the microcontroller (7) to convert the voltage (Vobj) into a value representing the temperature (Tobj) of the remote object (2) without keeping track of decimal points and resolution of the numbers.

Aiming at solving the defects of large error and high calculation and hardware implementation complexity of QAM (quadrature amplitude modulation) demodulation method in prior art, the invention proposes a fixed-point soft-decision demodulation method using QAM technology. The fixed-point soft-decision demodulation method for using QAM technology of the invention carries out soft decision on part defined bits of a received symbol with the mean of the absolute values of real part and imaginary part of the received symbol and then carries fixed-point soft decision on other bits with the hard-decision information of the part defined bits, thus simplifying the calculation and guaranteeing the modulation performance. In addition, the method is easy to be implemented with chip fixed-point algorithm.

In a computational device, a method is provided for determining attributes for results of arithmetic operations with fixed-point numbers, in an environment for fixed-point algorithm development. An indication is received of a plurality of possible word lengths to be used for storing a plurality of digits representing a result of arithmetic operation with fixed-point numbers. An indication is received of how a placement of a radix point is to be determined in the plurality of digits representing the results of the arithmetic operation with fixed-point numbers. When calculating a selected fixed-point arithmetic operation, one of the word lengths in the plurality of possible word lengths is employed for storing the plurality of digits representing a result of the selected fixed-point arithmetic operation. A placement of a radix point in the plurality of digits representing the result of the selected fixed-point arithmetic operation is based on the received indication of how the radix point is to be determined.

A technique to parameterize the quantization scheme of the attribute coding of point cloud compression algorithms is described herein. Based on fixed-point arithmetic, the algorithm calculates the quantization step size (QS) in fixed-point notation, given a user-input quantization parameter (QP).

The invention discloses an optimized drawing method of a power grid planning wiring diagram. The partial regulation of stations and circuit positions is realized in a mode that distances among primitives are amplified with a same proportion, the regulation mode is simple, the viewing effect of the wiring diagram after the regulation is good, and coordinates are defined accurately and completely; and through coordinates of the wiring diagrams in a province diagram and a city diagram, the coordinate mapping is realized through using a fixed-point algorithm, and the power grid wiring diagram is prevented from being drawn repeatedly on the province diagram and the city diagram by drawing people.

A system, method, and apparatus for calculating non-linear functions with finite order polynomials are presented herein. Use of finite order polynomials allow calculation of the non-linear functions using fixed point arithmetic operations resulting in significant cost savings.

The invention belongs to the signal processing technology field, and especially relates to a high-efficiency fast fixed point processing method suitable for communication signal blind separation. Aimed to the defect that a present non-circle fast fixed point algorithm (nc-FastICA) cannot selects an optimal non-linear function adaptively, an optimal non-linear function is selected adaptively through analysis of characteristics of different communication signals, and a fast fixed point processing method used for communication signal blind separation is provided. The separation efficiency is high, performances are stable, the convergence speed is fast, the provided is is raised by about 8-14dB compared with a traditional nc-FastICA algorithm and an ACMN algorithm when the separation communication signals are separated, and the provided method can be widely used for communication signal blind separation.

The invention discloses a high-efficiency stable multifunctional single-stage photovoltaic single-phase grid-connected control method, and belongs to the technical field of photovoltaic power generation control. The method is characterized in that: the method adopts a high-speed digital signal processor and applies a C language dynamic fixed-point algorithm to form a high-performance control algorithm of a whole grid-connected control system; the method adopts a single-stage grid-connected control structure and a quick smooth maximum power point tracking method to directly transmit the maximum energy sent by a photovoltaic array to a local urban power grid through first-stage DC / AC transformation; the harmonic and idle work produced by a local load can be compensated by adopting an FBD method, so the method is favorable for improving the electrical energy quality of the local power grid; and a proportional resonant converter is used to implement precise closed-loop control for the grid-connected current so that the photovoltaic grid-connected current and the commercial single-phase voltage are totally synchronous while implementing maximum power point tracking of the photovoltaic array. The power factor is high, the system efficiency is high, the operation of the whole system is safe and reliable, and the aberration rate of the grid-connected current harmonic is lower than 3 percent.

Computer vision systems and methods for end-to end training of neural networks are provided. The system generates a fixed point algorithm for dual-decomposition of a maximum-a-posteriori inference problem and trains the convolutional neural network and a conditional random field with the fixed point algorithm and a plurality of images of a dataset to learn to perform semantic image segmentation. The system can segment an attribute of an image of the dataset by the trained neural network and the conditional random field.