335results about How to "Fast operation" patented technology

The invention discloses a log-based APT attack detection and tracing method, and belongs to the technical field of attack detection. In order to solve the problems of high efficiency, fine granularity and accuracy of APT attack detection, the method comprises the following steps: acquiring original system log data, and then preprocessing the original system log data to obtain a data set; and constructing a traceability graph based on the four operation types. The four operation types comprise file operation, process operation, authority operation and network operation; performing node anomaly detection by using a BERT model and a VAE model to obtain a threat node set; and based on the obtained threat node set and the traceability graph, obtaining an attack path comprising all threat nodes by using a Steiner tree algorithm. The method is suitable for learning under the condition that a large number of labeled samples and priori knowledge do not exist, the problem that a traditional algorithm cannot detect zero-day attacks can be solved, and a simple APT attack path traceability graph can be extracted from a large number of system logs.

The invention provides a PCBA surface defect detection method and system based on image recognition, and the method comprises the steps: collecting and constructing a multi-class PCBA overall defect image data set, and carrying out the preprocessing; constructing a pre-training neural network model, and classifying the preprocessed PCBA overall defect image data set; constructing a local target identification method based on a YOLOX network according to an overall defect classification result; and carrying out local defect detection on the PCBA surface based on a local target identification method. According to the method, the overall classification and local detection strategies are comprehensively applied to reduce the burden of the model, and the detection precision and speed are improved. And an attention mechanism is introduced to improve the performance and generalization ability of the model. Data enhancement, consistency processing, classification labeling and the like are carried out on an image data set, and the defect detection effect of the model is improved. And multiple optimization strategies such as structure improvement, loss function improvement and pruning lightweight are adopted for the network model, so that the image target detection performance is effectively improved.

The invention discloses a real-time automatic focusing method based on interference fringe phase extraction, and the method specifically comprises the steps: importing an initial interference image, setting a focusing state position as an initial phase shift value, and carrying out the zero setting of the initial phase shift value; when the detection system runs, keeping interference image video stream input, and calculating a phase shift value and a phase value in real time through an improved iterative algorithm; and the calculation result is converted into a defocusing value, and the PZT controller is controlled to complete focusing. According to the automatic focusing method provided by the invention, the defocusing amount is obtained according to the phase movement of the interference fringes, a high-precision displacement measurement method is introduced into the automatic focusing technology, and compared with a traditional microscopic focusing method, the focusing precision is improved in a breakthrough manner, and the method has very high innovativeness; based on an interference image method, phase information closest to reality is extracted through iterative operation, real-time processing is continuously carried out through video streams of interference images, the operation speed is high, and real-time focusing can be completed in the scanning process of a detection system.

The invention relates to the technical field of power grid interaction, in particular to a two-way energy control method and intelligent equipment, and aims to solve the problem of how to perform charging and discharging control on two-way energy equipment according to power grid load information and peak and valley electricity prices on the basis of an optimal cost principle. The method comprises the steps that the initial electric quantity of an energy storage module is acquired; acquiring historical electrical load information of a set time period; and predicting a power control curve of a set time period according to the initial electric quantity, the historical electric load information and the expected electric quantity based on a cost optimal principle. On the basis of the predicted power control curve, charging and discharging control is performed on the energy storage module through the bidirectional energy equipment, so that the stability of the power grid can be enhanced, the power supply condition of the power grid can be optimized, the electric charge expenditure of the user can be reduced, the user experience can be improved, income can be brought to the user, and the return on investment of the user can be improved. And by adopting a dynamic programming algorithm, the operation speed is improved, and engineering realization is more convenient.

The invention discloses a machine vision-based hollow fresh soybean detection method and a sorting device. The method comprises the following steps: acquiring a fresh soybean RGB image with transmission information and surface information; extracting a single fresh soybean image through an edge detection algorithm, and constructing a fresh soybean data set with center coordinates; marking the fresh soybean data set, and dividing the fresh soybean data set into a training set, a verification set and a test set; constructing an improved YOLOv8 classification model by using the labeled fresh soybean data set; the improved YOLOv8 classification model is used for conducting hollow fresh soybean judgment on the fresh soybeans on the conveying line, and the hollow fresh soybeans are separated out. The sorting device comprises a material conveying mechanism, an illumination mechanism, an image acquisition unit, a central controller, a sorting mechanism and the like, an improved YOLOv8 classification model is embedded in the central controller, light-weight and high-accuracy classification of fresh soybeans is realized, a multi-target detection task is converted into a single-target classification task, characteristics of the fresh soybeans are maximized, and the sorting efficiency of the fresh soybeans is improved. And the classification accuracy is improved.

The invention provides a two-stage training method for a lightweight Chinese-English bilingual large language model. The method comprises the following steps: step 1, collecting and preprocessing Chinese-English training data; the preprocessing process comprises special mark clearing, sensitive data clearing, data duplicate removal, advertisement removal and formatting processing; 2, performing first-stage high-resource language training, namely performing model training by using English; 3, second-stage bilingual mixed training is carried out, and the Chinese data and the English data are mixed for second-stage training by adopting language migration training; wherein the step 3 comprises the following steps: step 31, performing word list expansion; and step 32, language migration training. The bilingual processing capability can be improved.

The invention provides an ionosphere TEC prediction model integrating double mechanisms of self-attention and self-correlation, and belongs to the technical field of ionosphere delay prediction, and the model comprises a trend prediction module, a correlation analysis module and a sequence fusion module. Wherein the trend prediction module predicts the overall change trends of an ionized layer TEC sequence and space environment parameters by means of an autocorrelation operation mechanism; the correlation analysis module analyzes the mutual relation among the ionized layer TEC sequence, the space environment parameters and the longitude and latitude positions by using a self-attention mechanism; and the sequence fusion module performs further integration and optimization on the overall trend of the ionized layer TEC sequence by using the correlation information obtained by the correlation analysis module, and finally obtains an ionized layer TEC prediction result fusing the space environment and the latitude and longitude position information. According to the method, the spatial environment data and the longitude and latitude position information are fully utilized, and the accuracy of model prediction is greatly improved.

The invention provides a sleeve length matching optimization method and system suitable for a mandrel type hanger and a terminal, and the method comprises the steps: converting a length matching problem of a mandrel type hanger sleeve into a combined optimization problem based on a simulated annealing algorithm, carrying out the random search in a solution space, and carrying out the step-by-step optimization to find an optimal length matching closest to a target, thereby achieving the optimization of the length matching of the mandrel type hanger sleeve. Through random jump and local optimization among different solutions, a local optimal solution can be effectively avoided, a global optimal solution is finally converged, the calculation process is simple, the speed is high, the error is small, the stability is good, the cost is low, and on-site popularization and application are facilitated.

The invention belongs to the technical field of wood defect detection, and particularly relates to a wood surface defect detection method based on improved YOLOv81, electronic equipment and a storage medium. The method comprises the following steps of: improving the Backbone of the traditional YOLOv81 by using Faster Net, and reducing the parameter quantity and the calculation quantity by using the design of partial convolution PConv in the Faster Net; an enhanced weighted bidirectional feature pyramid network EFPN is provided, information exchange between different feature layers is enhanced through an auxiliary branch, and an expansion re-parameter module DRM is designed to fully capture features on different scales. The improved YOLOv81 is more suitable for detecting the multi-scale defects on the surface of the wood, the detection precision is improved, the parameter quantity and the calculated quantity are smaller, and an efficient and accurate solution is provided for detecting the defects on the surface of the wood.

The invention provides a 10T SRAM (Static Random Access Memory) storage unit applied to a multi-bit storage and calculation integrated multiplication circuit, which comprises a first phase inverter, a second phase inverter, a first NMOS (N-channel Metal Oxide Semiconductor) tube, a second NMOS tube, a third NMOS tube, a fourth NMOS tube, a fifth NMOS tube and a first PMOS (P-channel Metal Oxide Semiconductor) tube, the output end of the first phase inverter and the input end of the second phase inverter are connected with the drain electrode of the second NMOS tube; the grid electrode of the first NMOS tube is controlled by a word line, and the source electrode of the first NMOS tube is connected with the first write-in end; the grid electrode of the second NMOS tube is controlled by the word line, and the source electrode of the second NMOS tube is connected with the second write-in end; the grid electrode of the third NMOS tube is connected with the first storage node, the drain electrode of the third NMOS tube is connected with the source electrode of the fourth NMOS tube, and the source electrode of the third NMOS tube is grounded; the drain electrode of the fourth NMOS tube is connected with the first reading end, and the grid electrode of the fourth NMOS tube is connected with a first control signal. According to the invention, the in-memory calculation of the multi-bit multiplication circuit is realized.

The invention discloses a Zynq-based convolutional neural network hardware acceleration system, which is realized based on a Zynq heterogeneous chip and comprises a PL end module, a PS end module and a storage module. The PL end module comprises a convolution acceleration module, a full connection layer acceleration module and a Softmax activation module and is used for completing convolution calculation, full connection operation and activation function processing; the PS end module comprises an input data processing module and a post-processing module and is responsible for loading and preprocessing input data and sorting and screening detection results. The storage module is composed of an on-chip memory and an off-chip memory, wherein the on-chip memory stores an intermediate processing result of input data, a convolution kernel and intermediate feature map data; and the off-chip memory stores the original image data, the weight parameter of the full connection layer and the final detection result. All the modules are connected through a 64-bit AXI bus, the PS end module is responsible for coordinating data flow, and the PL end module executes efficient parallel computing. The system makes full use of the high-performance characteristic of the Zynq heterogeneous architecture, is suitable for an efficient reasoning acceleration scene of the convolutional neural network, and particularly has a wide application prospect in embedded AI equipment.

The invention relates to a GPU-based high-IO perceptibility long-sequence text injection method, medium and equipment, and the method comprises the steps: obtaining and coding a long-sequence text input, processing a coded matrix vector through a multi-head attention mechanism, and realizing text injection, the matrix vector comprises a query Q, a key K and a value V. The method is characterized in that the value V is equal to the query Q, the key K and the value V. In the multi-head attention mechanism, the similarity is calculated by using an L2 distance, the matrix vector is segmented into blocks to realize thread block parallel processing during forward propagation, and the similarity-based result matrix and the gradients of the query Q, the key K and the value V are recalculated by using a recalculation strategy during back propagation. Compared with the prior art, the method has the advantages of reducing the overhead of attention calculation memory access, improving the operation efficiency and the like.

The invention relates to the field of sparse matrix calculation, and discloses a sparse calculation unit, method, device and medium under a GPGPU architecture, the sparse calculation unit comprises a sparse matrix operation unit arranged in a streaming multiprocessor, and the sparse matrix operation unit is connected with a transmitting module and a write-back module; the sparse matrix operation unit comprises a selector, an algorithm module, an accumulator and a plurality of second buffer modules; the sparse matrix operation unit is used for calculating the sparse matrix; a calculation algorithm is stored in the algorithm module and used for calculating matrix elements, and the calculation algorithm comprises an improved SPGEMM algorithm and an addition algorithm; and the memory access unit scheduling module is connected with the plurality of streaming multiprocessors and is used for coordinating and scheduling the memory access units on the plurality of streaming multiprocessors. And the sparse matrix calculation unit is added, so that the processing capability of the GPGPU on the sparse matrix is improved, and the operation speed of the GPGPU on the sparse matrix calculation task is improved.

The invention relates to the technical field of digital model construction of network construction type control energy storage power equipment, in particular to an energy storage electromechanical transient model construction method and system for analyzing the characteristics of a synchronous control equipment grid-connected system, and the method comprises the steps: applying double droop control to an energy storage converter system, analyzing a regulation and control principle and grid-connected equivalent external characteristics, constructing a grid-connected equivalent circuit of the double-droop control energy storage converter system; designing the structure of the energy storage converter system; according to the designed energy storage converter system structure, a double-droop control part electromechanical transient model, an energy storage body electromechanical transient model and a grid-connected interface electromechanical transient model are constructed, and a complete energy storage converter control system electromechanical transient model capable of being embedded into a PSASP main interface is formed. And the output power change of the converter is reflected at the access node, so that the functions of characteristic analysis, load flow calculation and the like of a regional power grid system accessed to the energy storage converter can be realized, and a technical scheme is provided for guiding the planning design of accessing the energy storage converter to the regional power grid.

The invention relates to the crossing field of artificial intelligence and bird ecology, and particularly discloses a large-scale bird sound recognition method based on a deep learning technology, and the method comprises the steps: collecting a bird sound audio through a project region recording device, training a bird sound event detection model through a Bird audio detection challenge 2018 data set, carrying out the class prediction through field recording data, and carrying out the recognition of a bird sound event. Effective bird sound and background noise are separated; constructing a bird sound recognition data set in combination with a bird species catalog of the Chinese bird watching recording center and an audio file of Xen-Canto, training a bird sound recognition model by using background noise and the bird sound recognition data set, and performing verification; and finally, inputting the effective bird sound data into the model to carry out label prediction so as to identify the bird sound. The invention aims to carry out wide bird voice recognition on a project area under construction through a deep learning technology, so that the data processing cost is saved, and the recognition accuracy in a complex scene is improved.

The invention discloses a resistive random access memory manufacturing method, a resistive random access memory chip manufactured by the resistive random access memory manufacturing method comprises a first resistive random access memory unit and a second resistive random access memory unit, a resistive random layer of the first resistive random access memory unit comprises a high-voltage material layer, and a resistive random layer of the second resistive random access memory unit is a high-dielectric constant material layer. The first resistive random access memory unit has the characteristic of long storage time due to the fact that the resistive random layer comprises the high-voltage material layer, and the second resistive random access memory unit has the characteristics of fast response and high operation speed due to the fact that the resistive random layer is the high-dielectric-constant material layer. According to the manufacturing method of the resistive random access memory, the two resistive random access memory units can be subjected to process integration through the innovative process, and the manufactured resistive random access memory chip can meet the requirements of two performances.

The invention provides an instance segmentation detection method and application based on improved YOLOv8s. The method comprises the following steps: collecting a to-be-segmented data set, preprocessing the to-be-segmented data set, constructing an improved YOLOv8s model, and training the preprocessed data set; the method is applied to target detection of images and / or videos, such as segmentation detection of second-phase particles of nickel-based superalloy. According to the improved model, an SPDConv layer and a small target detection layer are added, and an SIoU loss function is adopted, so that the small target detection performance of the model is greatly improved, and the omission ratio is reduced; second-phase particle dispersion calculation is updated and embedded into the second-phase particle identification process, batch detection is achieved, and indexes such as the average area fraction, the average number, the average size and the dispersion of the second-phase particles can be directly calculated; the model segmentation performance is good, the operation speed is high, the target detection process is opened, and the method can also be applied to other processes, intelligent recognition of macroscopic objects and online recognition and calculation of videos.

The invention relates to the technical field of wind power plant active power control, and discloses an active power control method and system for a wind power plant, and the method comprises the steps: obtaining the historical data of a fan, and constructing a training set according to a wake flow influence result of a fan state; based on an AdaBoost algorithm, constructing a strong classifier in combination with the training set; the method comprises the following steps: determining fans influenced by wake flow by using a strong classifier, determining upstream fans generating wake flow by using a wake flow center line deflection angle, and defining all wind turbine generators which are not influenced by wake flow and generate wake flow which does not influence downstream fans as a controllable fan group; and solving the wind turbine generator groups with the minimum number of execution actions in the wind power plant by utilizing a dynamic programming algorithm, and executing active power control operation. According to the method, the space-time coupling influence caused by the wake effect in the active power control process of the wind power plant can be eliminated, the environmental disturbance generated by the execution of the power control action of the fan to the wind power plant is reduced to the minimum, and the real-time accurate control of the active power of the wind power plant can be realized.

The invention discloses a large-data-volume dense signal processing method based on a domestic multi-core DSP (Digital Signal Processor). The large-data-volume dense signal processing method comprises the following steps: step 1, constructing a data processing external storage space; 2, designing a data transmission program; 3, designing a DSP kernel data processing program; 4, designing a data transmission processing state monitoring module; and 5, outputting a data processing result by designing an output data processing result module. The method aims at improving the real-time performance and universality of dense signal processing of a large data volume and reducing the engineering implementation difficulty. According to the method, the OpenMP parallel programming program interface is combined to carry out parallel processing on the data so as to improve the program running efficiency.

The invention discloses an instruction set system and a CPU (central processing unit) system based on an RISC (Reduced Instruction Set Computer) architecture, which are applied to a system of the RISC architecture. Instruction set coding is in a 32-bit equal-length instruction set coding format and comprises an operation code field with variable length, a first register index field, a second register index field, an address offset field, a pointer field, a return field, a storage field, a bit number field, an immediate number field, a process field, a time counting field and a mode field. The operation code fields with different bit lengths are combined with at least one field in the fields, or all the operation code fields form different instruction coding format types, and the operation code fields are used for executing preset operation instructions. Compared with the traditional instruction set operation, the method can be realized by needing fewer instructions, so that the instruction execution times are reduced, the operation speed is improved, complex operation can be realized by only needing one instruction, the occupied storage space is reduced, the programming difficulty is reduced, and the program performance is improved.

The invention discloses a lithium ion battery life prediction method based on a life prediction combination model, and the method specifically comprises the following steps: S1, carrying out the fitting analysis according to the data of a battery capacity recession experiment, and relates to the technical field of lithium ion battery life detection. According to the lithium ion battery life prediction method based on the life prediction combination model, the prediction precision of the battery life is improved, the life prediction model is established on the basis of a capacity attenuation curve fitting formula and an empirical model, and model parameters are updated on the basis of the improved empirical model; the established prediction model not only can accurately reflect the trend of battery capacity decline, but also can effectively combine the actual use state of the battery, so that the high-precision prediction of the residual service life of the lithium ion battery in different health states can be realized, and the prediction accuracy of the lithium ion battery in different health states is improved. And the development of the battery echelon use industry is facilitated.

The invention provides an electric smelting magnesium oxide production process safety monitoring method based on supervision matrix regression, and relates to the technical field of fault monitoring and diagnose.The method comprises the steps that firstly, a fixed camera is used for collecting furnace body image data of an electric smelting magnesium oxide furnace, and after graying and normalization processing, a furnace body image matrix is obtained; secondly, establishing and training an objective function model of the furnace body image matrix in a low-dimensional space through a public information extraction method, and solving a unique mapping matrix of each piece of sample data in the low-dimensional space; then, establishing a regression model of the mapping matrix of each sample and the corresponding class label vector by adopting a supervised matrix regression method; then, determining a hyper-spherical boundary and an angle boundary according to the category label of the sample; finally, furnace body image data are collected and processed in real time, the category of a new sample is judged according to the hyper-sphere boundary and the angle boundary, and then whether a fault occurs at the current moment or not is judged. According to the invention, the accuracy of industrial fault monitoring classification can be improved.

The invention discloses a method for determining control parameters of an ammonia-coal mixed combustion system and a related device and system, and relates to the field of ammonia-coal mixed combustion systems.The method comprises the steps that the actual value of the ammonia input flow, the actual value of the air input flow, the actual value of the nitrogen oxide content and the actual value of the ammonia remaining amount are collected; utilizing the mechanism model to obtain a predicted value of the nitrogen oxide content and a predicted value of the residual ammonia gas; determining the range of the nitrogen oxide content and the range of the residual amount of ammonia gas; training the neural network model by using the range of the nitrogen oxide content and the range of the residual ammonia gas to obtain an optimal excess air coefficient determined by the neural network model; obtaining a target ammonia gas input flow and a target air input flow corresponding to the optimal excess air coefficient; and determining the target ammonia gas input flow and the target air input flow as control parameters of the ammonia-coal mixed combustion system. The control parameters of the ammonia-coal mixed combustion system can be quickly and accurately determined, and emission of nitric oxide and ammonia gas is reduced.

The invention discloses a model training method, a product recommendation method, equipment, a medium and a program product, and relates to the technical field of big data, and the model training method comprises the steps: obtaining initial feature information of a training user group; performing feature extraction on the initial feature information according to the correlation between each feature information in the initial feature information and the target product acceptance degree to obtain training feature information of the training user group; and training the to-be-trained model according to the training feature information and the target hyper-parameter group to obtain a recommendation group determination model. Therefore, on one hand, the correlation between the feature information and the target product acceptance degree is fully considered, and the group determination accuracy of the recommendation group determination model is improved. And on the other hand, only the target product is recommended to the user group, so that the user with relatively low acceptance on the target product is prevented from receiving the recommendation of the target product, and the marketing effect is improved. And on the other hand, the time cost in the marketing process is reduced.

The invention relates to the technical field of semiconductors, in particular to a multiplexing-based optical computing chip and a design method thereof.The multiplexing-based optical computing chip comprises a computing area formed by crossing a first waveguide and a second waveguide, and the computing area comprises a branch waveguide connected with the first waveguide and the second waveguide, a beam splitting part, an intensity modulation part and a first signal conversion part are sequentially arranged on the branch waveguide, the beam splitting part is used for splitting optical signals in a specific proportion, the intensity modulation part is used for modulating the signal intensity of the optical signals into a specific value, and the first signal conversion part is used for setting or maintaining the optical signals in a specific mode; the specific mode means that the optical signal has a specific polarization direction and / or a specific order; the second waveguide is provided with a calculation unit for calculating the sum of the optical powers of the plurality of optical signals. According to the multiplexing scheme provided by the invention, the data processing amount of the optical computing chip can be improved.

The invention belongs to the technical field of 3D modeling based on image processing, and particularly relates to a 3D modeling method based on animation design and a control system.The animation modeling subsystem completes automatic three-dimensional modeling by obtaining a front image of a single character to obtain a character three-dimensional model; the animation action driving subsystem firstly outputs final motion parameters according to the input motion data, and then adjusts a role three-dimensional model; finally, after the output motion parameters and the role three-dimensional model are obtained, frame-by-frame motion parameter adaptation is carried out, layer-by-layer driving skeleton operation is carried out, and preliminary role motion roaming is generated; and the rendering display module is used for rendering and displaying the initial role motion roaming to obtain final role motion roaming. According to the invention, rapid role three-dimensional modeling can be carried out according to a single front image, the motion parameters are obtained according to the motion data, and a plurality of role three-dimensional models and a plurality of motion parameters are randomly matched to directly and rapidly obtain three-dimensional animation roaming of a plurality of roles in different motion states.

The invention belongs to the field of satellite imaging algorithms, and particularly relates to a method for predicting an earth observation imaging window of an imaging satellite, which is used for imaging an imaging target and comprises the following steps of: 1, roughly searching; step 2, performing imaging estimation; and step 3, performing imaging correction. Compared with the prior art, the imaging satellite task planning method solves the problems that in the prior art, imaging satellite task planning takes little consideration to constraint conditions, calculation errors are large, algorithms are complex, and the imaging effect is poor, and perturbation and errors caused by the factors such as the earth oblateness, the orbit eccentricity rate and the atmospheric resistance are considered in the scheme. And a simple and rapid iterative algorithm is further applied to solve a satellite imaging window, so that the efficiency and the accuracy of an autonomous task are improved.

The invention belongs to the technical field of biological detection, and provides an electrical impedance detection system based on an FPGA (Field Programmable Gate Array), which comprises an FPGA main control board, an impedance measurement chip, a peripheral circuit of the impedance measurement chip, an RS232 serial port transmission module, a PC (Personal Computer) end upper computer and a power supply module, the power supply module supplies power to the FPGA main control board, the impedance measurement chip, the peripheral circuit of the impedance measurement chip and the PC end upper computer, the FPGA main control board is arranged between the PC end upper computer and the impedance measurement chip and the peripheral circuit of the impedance measurement chip, the impedance measurement chip is connected with the FPGA main control board through an I2C bus, and the FPGA main control board communicates with the PC end upper computer through the RS232 serial port transmission module. The electrical impedance detection system is higher in detection precision, better in stability and low in power consumption and cost, and has extremely high production and application value.

The invention discloses a laser radar hardware parameter design method based on a laser radar equation simulation model, and belongs to the technical field of laser radar remote sensing detection. Based on a laser radar equation simulation model, laser radar parameters such as different laser wavelengths, energy, pulse widths, divergence angles, field angles, repetition frequencies, telescope calibers and laser inclination angles of visible light bands and chlorophyll concentrations, absorption coefficients, scattering coefficients, scattering phase functions and wind speeds of water bodies can be analyzed; the influence of water body biological optical parameters such as zenith angle and environmental parameters on a multi-platform laser radar receiving signal; and finally, by changing different parameters, the laser radar receiving signal strength is analyzed, and the system parameter design is optimized. The method is high in operation speed and high in calculation efficiency, and input parameters can be conveniently adjusted according to a laser radar system and the environment; the method can be used for simulating signals of shipborne, airborne, satellite-borne and other multi-platform ocean laser radars.

The invention relates to the field of floating-point multiplication, and particularly discloses a floating-point multiplier based on an FPGA, a calculation method and equipment, and the floating-point multiplier comprises a symbol calculation module which is used for determining the symbol of a target output floating-point number through an exclusive-OR gate and the symbol bits of a first input floating-point number and a second input floating-point number; the index addition module is used for adding the indexes of the first input floating-point number and the second input floating-point number and subtracting the deviation value of the corresponding format to obtain the index output of the target output floating-point number; the binary multiplier module is used for carrying out multiplication operation by using a Karatsuba algorithm and a Urdhva-Tiryagbhyam algorithm based on the mantissa bit widths of the first input floating-point number and the second input floating-point number so as to obtain a mantissa product of the target output floating-point number; and the result normalization module is used for performing normalization operation based on the mantissa product. The calculation delay can be reduced, and the percentage increase of the hardware area is also reduced.