48results about How to "Fewer calculation parameters" patented technology

The invention discloses a bottom material sonar image classification method based on a convolution neural network, belonging to the technical field of image classification. The method comprises obtaining the sonar image of seafloor bottom, preprocessing the image with denoising and enhancement, extracting the edge shape based on a canny algorithm, and generating a gray scale-basic element co-occurrence matrix, constructing a convolution neural network classifier structure and a sample set, training a neural network, obtaining classification model and realizing bottom material sonar image classification. The present invention is directed to the graphical characteristics of bottom material sonar images of seafloor, the disadvantage of using single method is solved, and the learning strategyof the convolution neural network classifier is used to learn and train different types of seabed sediment, and finally the classification model with classification function is obtained, and the purpose of fast and accurate classification of seabed sediment sonar images is achieved.

The present invention relates to a modeling method of broadband radio frequency power amplifier. In the method, a memory-free non-linear nominal model is connected with a memory non-linear identification model in parallel so as to realize the modeling of the power amplifier PA. The modeling method makes full use of the prior knowledge of the PA, and adopts the mixed modeling idea which combines the traditional memory-free non-linear nominal model and the memory non-linear identification model, realizes the modeling of the broadband radio frequency PA, and greatly improves the performance of the whole PA model.

The invention discloses an adaptive optimization method and system for power generation scheduling of a cascade hydropower system. The method comprises the steps of determining a constraint conditionset of each hydropower station; Setting calculation parameters and generating an initial population; Calculating the individual fitness and updating the historical position of the individual and the global optimal position of the population; Updating the position of each individual by adopting a standard sine-cosine algorithm; Adopting a population position center variation strategy to increase individual diversity, introducing a neighborhood search strategy to improve population global optimization capability, and utilizing a simplex dynamic search strategy to enhance algorithm convergence speed; And continuously updating the position of each individual through multi-round iteration, and taking the population global optimal position obtained in the last round as an optimal scheduling process. The Yangtze River basin engineering application result shows that the method has the advantages of being high in calculation precision, high in optimizing speed, high in searching capacity, highin robustness and the like, the overall power generation benefit of the cascade hydropower system can be remarkably improved, and the method has important engineering application value.

A fracture-vug type oil reservoir dynamic reserve calculation method provided by the present invention relates to a method for calculation aiming at the oil reservoir reserves of the elastic drive stages, and aims to realize the effects that the needed calculation parameters are less, the parameter values are all measured at an oil recovery site and are objective and accurate, and the calculation results are more credible. The fracture-vug type oil reservoir dynamic reserve calculation method of the present invention comprises the following steps of S01 measuring the bottom hole crude oil density rho 1 for the first time by a manometry method; S02 exploiting an oil well for a period of time, and recording the cumulative oil production as Np; S03, measuring the bottom hole crude oil density rho 2 for the second time by the manometry method; S04 substituting the measured rho 1, rho 2 and the Np into a formula to obtain the fracture-vug type oil reservoir dynamic reserves.

The invention provides a drug combination recommendation method based on a time attention mechanism and a graph convolutional network. According to the invention, reasonable drugs can be recommended for treatment of critical patients in a complex medical environment, and clinicians can be helped to treat patients. Diagnosis and treatment in electronic health records are coded in a unified coding format, time sequence information in diagnosis and treatment is stored, codes are converted into vectors, and the time attention mechanism composed of two layers of recurrent neural networks is used for capturing the time sequence information. The method aims at medicines of prescriptions issued by doctors in the electronic health records and medicines having adverse reactions with known medicines,graph network structure data are converted to describe the relation between different medicine combinations, and medical medication knowledge in a medicine graph network is learned by utilizing the graph convolutional network. Compared with the prior art, the simplified graph convolutional network reduces the calculation parameters of the neural network model and reduces the training and learningtime under the condition of maintaining the prediction accuracy unchanged.

The invention discloses a vehicle load measurement method which comprises the following steps: controlling a vehicle to run at a constant speed on a preset paving material, and acquiring running parameters of the vehicle running at a constant speed on the preset paving material; determining the own weight of the vehicle and the dynamic friction coefficient between the vehicle and the preset pavingmaterial according to the vehicle model information of the vehicle; and calculating the load of the vehicle according to the own weight of the vehicle, the dynamic friction coefficient and the running parameters. According to the invention, after the vehicle is controlled to run at a constant speed on the preset paving material, running parameters of the vehicle running at a constant speed on a preset paving material are obtained, and the own weight of the vehicle and the dynamic friction coefficient between the vehicle and the preset paving material can be determined according to the vehiclemodel information of the vehicle, so that the weight of the vehicle can be calculated according to the own weight of the vehicle, the dynamic friction coefficient and the running parameters, the measurement operation is simple, and the measurement cost is low, the calculation mode is simple, the calculation error is small, and therefore the load of the vehicle can be simply, rapidly and accurately measured.

The invention provides a human body inertial parameter recognition method. According to the method, a human body is simplified to a five-rigid-body model, the momentum theorem and the theorem of moment of momentum are used for obtaining a human body five-rigid-body kinetic equation, a motion capture system is used for obtaining the motion parameters like displacement, speed and accelerated speed of the tested human body, a pressure sensing measuring device is used for obtaining contact force of feet and the ground, and finally obtained data is substituted into the kinetic equation, and the interactive least square method is used for calculating the human body inertial parameters which need to be measured. The human body inertial parameters can be used for kinetic analysis of human body gait, and have important application in the field of physical exercise science and medical rehabilitation and other fields.

The invention discloses a metering control method based on three quantities of a water supply system. The metering control method is to firstly use the flow rate of the water supply system to collect data, collect the required basic data, and use the collected data to calculate the amount of pipe network leakage, Calculation of apparent loss and production-sales difference water volume, measurement analysis and control of pipe network leakage, apparent loss and production-sales difference water volume. It has the following advantages: Simplify the complicated quantitative control of leakage loss to the measurement and analysis of the three quantities of pipe network leakage, apparent loss, and production-sales difference in the water supply system only through the flow rate, and realize the control of pipe network leakage, Early warning of apparent loss to reduce production-sales gap.

The invention provides a method and system for determining an electric load adjustment interval of a low-vacuum heat supply unit, and the method comprises the steps: obtaining a main steam flow correction coefficient according to a preset main steam flow and a main steam flow under a reference working condition, and obtaining an enthalpy drop correction coefficient and a regenerative steam extraction coefficient according to the iteration of steam exhaust parameter data and acquiring the total heat supply amount of the current working condition; when the total heat supply amount of the currentworking condition is larger than the preset heat supply requirement, acquiring the electric load according to the obtained main steam flow correction coefficient and steam enthalpy drop correction coefficient; calculating electric loads under different preset main steam flows and preset steam extraction flows, and taking a range between the maximum electric load and the minimum electric load of the unit as an electric load adjustment interval of the low-vacuum heat supply unit; according to the invention, the correction coefficient of the key parameter is introduced to calculate the electrical load, and the calculation efficiency can be greatly improved on the premise of ensuring the accuracy.

The invention discloses an image deblurring method based on depth features. The invention relates to the field of low-quality image enhancement, solves the problem of blurred image restoration by using a standard non-blind deconvolution deblurring method, and comprises the following steps: 1, carrying out graying and decoloration algorithm preprocessing on an original blurred image; 2, enlarging a receptive field of an input image by adopting a VGG16 network, and extracting fuzzy features; 3, selecting a local image block to obtain an initial blurring kernel, and reconstructing a clear image by adopting a standard non-blind deconvolution deblurring method; and 4, constructing a double-loss function, namely optimizing the restored clear image by using a smoothness loss function and a color loss function.

The invention relates to a knowledge graph visualization method for electric power scientific and technological achievements. Compared with the prior art, the defect that semantic analysis is difficult to carry out due to the fact that the electric power scientific and technological achievements are numerous in data size is overcome. The method comprises the following steps: obtaining electric power scientific and technological achievement data; setting a structural representation model based on the activation vector; setting a text representation model based on the CNN; carrying out knowledgerepresentation based on joint training; and obtaining a visual result of the knowledge graph. According to the method, the structural knowledge representation method based on the activation vector and the text information knowledge representation method based on the CNN are fused, calculation parameters can be effectively reduced, model operation is efficient, features in the text structure and the text information are effectively obtained, and the representation result is more accurate.

The invention relates to a method for determining the uplift amount after the bottom pile foundation of the saline soil cutting foundation bed is strengthened, which can scientifically and reasonablydetermine the uplift amount of the bottom pile foundation of the saline soil cutting foundation bed after consolidation, and can meet the needs of practical engineering. The method comprises the following steps: determining the neutral point depth l0 of the pile foundation in the saline soil, the atmospheric cooling influence depth l below the top surface of the cutting foundation bed, and calculating and determining the calculated thickness h0 of the pile foundation uplift deformation at the bottom of the cutting foundation bed; determining the temperature Ti 1 and other parameters of the i-layer soil before cooling in winter; carrying out salt dilatancy test, compression test, water content test, soil particle specific gravity test and weight test on saline soil samples to determine thesalt dilatancy index Csi of the i-th layer soil; determining the uplift amount se after the bottom pile foundation of the saline soil cutting foundation bed is strengthened by the following formula (shown in the description).

The invention discloses a night infrared pedestrian detection method and system based on improved YOLOv3, and the method comprises the steps: collecting a plurality of infrared pedestrian detection data sets through a plurality of infrared cameras, carrying out the pixel value contrast enhancement of the data sets, increasing the pedestrian pixel values, and reducing the background pixel values; a night infrared pedestrian detection network model YOLOv3-SAB is improved and constructed based on YOLOv3, a stem down-sampling module and asymmetric convolution are introduced to improve the network feature extraction capability and feature expression capability, calculation parameters in a bottleneck residual error reduction model are introduced, and the pedestrian detection speed of the model is improved; a specific prior aiming frame is generated through clustering by using a mean value clustering algorithm, and the model target positioning precision is improved; cIoU is used as a YOLOv3-SAB network bounding box regression loss function, so that model convergence is accelerated, and the accuracy of a prediction box is improved; a YOLOv3-SAB network is trained to generate a night infrared pedestrian detection model; and performing real-time infrared pedestrian detection at night by using the night infrared pedestrian detection model. According to the invention, the detection precision and the detection speed of pedestrians at night are effectively improved.

The invention provides a state space controller, a control method and a gradient power amplifier. The state space controller comprises an integrating part which is used for integrating control deviation which is acquired based on the difference between gradient coil current and reference current, and a delay compensator which is used for generating a digital control amount used for generating a pulse-width modulation driving signal according to the control deviation integrated by the integrating part. The delay compensator comprises a feedback loop which is used for multiplying the digital control amount by a predetermined multiple after a delay of N calculation periods and subtracting from the digital control amount, wherein the N is 1 or 2. According to the invention, a simple and fast state space control method is realized.

The invention provides an underwater fish body posture dynamic identification method and device. The method comprises the following steps: determining a to-be-identified fish body image; and performing target detection on the spliced image based on the fish body posture recognition model to obtain a target detection frame image, performing down-sampling on the target detection frame image, generating a feature pyramid based on the down-sampled target detection frame image, and determining a fish body posture based on features in the feature pyramid. The method can determine the posture of the fish body in the moving state with high precision.

The invention discloses a method for determining the upheaval quantity after reinforcing a low embankment basement salinized soil pile foundation to determine the upheaval quantity of low embankment basement salinized soil after reinforcing the pile foundation in a scientific and reasonable manner and to be adapted to the actual engineering demand. The method comprises the following steps: determining the neutral point depth 10 of the pile foundation in salinized soil and the atmosphere cooling influence depth 1 below the top surface of the low embankment, and calculating to determine the lowembankment pile foundation upheaval deformation computation thickness h0; determining the temperature Ti1 of the ith-layer soil and other parameters before temperature reduction in winter; carrying out a salt expansion test, a water content test, a specific gravity test of soil particle and a weight test on the soil sample of salinized soil and determining the salt expansion index Csi of the ith-layer soil and other parameters; and determining the upheaval quantity se after reinforcing the low embankment basement salinized soil pile foundation through the formula (refer to the Specification).

The invention discloses a soil deformation prediction method, system and device and a readable storage medium. The method comprises the steps: obtaining geometric design data of a target tunnel which passes through a large-thickness collapsible loess stratum, a soil internal friction angle of the large-thickness collapsible loess stratum, and a collapsibility coefficient which passes through the stratum; according to the geometric design data of the target tunnel, the friction angle in the soil body and the collapsibility coefficient of the penetrating stratum, the width of a settling tank at any vertical depth from the tunnel vault to the earth surface and the stratum loss rate after the target tunnel is excavated are calculated; according to geometric design data of the target tunnel, the stratum loss rate after excavation of the target tunnel and the width of a settling tank at the vertical depth in the large-thickness collapsible loess stratum, vertical displacement and horizontal displacement at any point between the earth surface of the large-thickness collapsible loess stratum and the arch crown of the target tunnel are calculated; a soil deformation prediction result is obtained; the method has the characteristics of clear meaning, simple formula, few calculation parameters and better accuracy.

The invention discloses a vehicle detection method in a video stream, comprising the following steps: acquiring a video frame, using a trained convolutional neural network to identify the obtained video frame, and judging whether there is a vehicle in the video according to the result, and the vehicle The specific position of the vehicle; the position of the vehicle in the image is obtained by the vehicle detection, and its value in the next frame is predicted by the Kalman filter as the vehicle tracking result; the results of the vehicle detection and the results of the vehicle tracking are correlated, according to the correlation results Modify filter parameters and update filter group information. The invention designs a convolutional neural network with less network layers and a small structure for the vehicle detection task, which has fast detection speed and high precision. The invention uses a Kalman tracking algorithm to predict the position of the detection frame, and the tracking speed is fast and stable. The invention uses a data association strategy to combine a detection algorithm with a tracking algorithm, embodies the correlation of vehicles between frames, and can reduce the missed detection rate of vehicle objects.

The invention relates to a photovoltaic power generation fault diagnosis system based on a photovoltaic physical model. The photovoltaic power generation fault diagnosis system is connected with a photovoltaic system to be monitored. The photovoltaic power generation fault diagnosis system comprises an environmental parameter collection device, a photovoltaic system parameter collection device, a signal converter and a fault diagnosis subsystem. When the photovoltaic system operates, the environmental parameter collection device collects photovoltaic system environmental parameters every a preset time, and the photovoltaic system environmental parameters are transmitted to the fault diagnosis subsystem after being converted by the signal converter; meanwhile, the photovoltaic system parameter collection device collects photovoltaic system parameters every a preset time, and the photovoltaic system parameters are transmitted to the fault diagnosis subsystem after being converted by the signal converter; the fault diagnosis subsystem carries out calculation and fault condition judgment according to the acquired parameters and outputs results. Compared with the prior art, the photovoltaic power generation fault diagnosis system has the advantages of being small in number of needed calculating parameters, simple in wire layout and reconstruction, low in cost and the like, and achieves safe and reliable monitoring and fault diagnosis on the photovoltaic system.

The invention relates to a practical method for calculating the length coefficient of a shaft pressure rod considering constraint influences of fixed spring hinges at the two ends. The method includesthe following steps of determining the rotation rigidity Ra of a bottom support and the rotation rigidity Rb of a top support of the shaft pressure rod, calculating the linear rigidity i of the shaftpressure rod, calculating the ratio ra of the rigidity of the bottom spring hinge to the rigidity of a pressure rod of the shaft pressure rod and the ratio rb of the rigidity of the top spring hingeto the rigidity of the pressure rod of the shaft pressure rod, and calculating the length coefficient of the shaft pressure rod considering constraint influences of the fixed spring hinges at the twoends according to a practical formula for calculating the length coefficient of the shaft pressure rod. The method has the advantages that the problem that an existing method for calculating the length coefficient micron of the shaft pressure rod is too simplified, so the limit bearing capacity of the shaft pressure rod is overestimated in engineering design or tests can be solved, the problem that the existing method can not directly calculate the length coefficient micron of the shaft pressure rod under designated support rigidity can be also solved, and meanwhile the method is simple in step, small in required parameter and high in precision.

The invention discloses an image defogging method based on blue channel correction, which is applied to the technical field of computer vision and image processing, and aims at solving the problems ofpoor defogging effect and large algorithm limitation of the existing histogram equalization algorithm, multi-scale Retinex (MSR) algorithm, dark channel prior defogging algorithm and the like. The method comprises the following steps: firstly, extracting information of red, green and blue channels, combining and sorting the information to obtain depth and transmissivity information, then respectively calculating the red, green and blue channels of an image to obtain a defogged image of each channel, and finally combining the images to obtain a final defogged image. The method has the advantages of few calculation parameters, simple and rapid calculation, good effect and no color cast, and is very suitable for processing outdoor fog images.

The invention provides a fly face recognition method based on a deep convolutional neural network. The method comprises the steps of optimizing a multi-task convolutional neural network MTCNN to perform face positioning and feature point detection on a fly image; obtaininig a fly face feature point image through positioning and detection, and carrying out fly face alignment; making the aligned flyface images into a data set; building a fly face deep convolutional neural network model; connecting two groups of feature vectors extracted from the thickness in the fly face deep convolutional neural network in series to form a group of feature vectors; and testing the fly face deep convolutional neural network model through the test set, and verifying the fly recognition effect. According to the method, a fly face deep convolutional neural network is adopted, contour features of an image are roughly extracted by using a large number of convolution pooling, and then specific part features such as small eyes in fly compound eyes are extracted by using Inception-ResNet and Reduction networks. According to the network, the learning target and difficulty can be simplified, and richer feature vectors can be extracted while gradient loss is prevented.

The invention provides a method for measuring the salt expansion of saline soil foundation. The method for measuring the salt expansion of the saline soil foundation can determine the salt expansion of the saline soil foundation in a winter temperature reduction period scientifically and reasonably, and can meet practical engineering requirements. The method for measuring the salt expansion of thesaline soil foundation comprises the following steps that through field investigation or data query, the salt expansion deformation calculating depth l<0> of the saline soil foundation, the temperature T<i1> of an I layer of soil before winter temperature reduction and the lowest temperature T<i2> of the I layer of the soil during winter temperature reduction period are determined; soil samples of saline soil are collected on site, and a salt expansion test, a moisture content test, a soil particle specific gravity test and a gravity test are carried out indoors, the salt expansion index C<si> of the I layer of the soil, the salt expansion force sigma<Ti> of the I layer of the soil after the temperature drops from T<i1> to T<i2>, the initial void ratio e<0i> of the I layer of the soil at the temperature of T<i1> and the effective self-weight stress sigma<y0i> are determined; and the salt expansion of the saline soil foundation is determined through the following formula, and please see the specification for the formula.

The invention discloses an image defogging method based on blue channel correction, which is applied in the technical field of computer vision and image processing, and aims at the existing histogram equalization algorithm, multi-scale Retinex (MSR) algorithm and dark channel prior defogging method and other algorithms, the problem of poor defogging effect and large algorithm limitations; the present invention firstly extracts the information of the red, green and blue channels, and combines them to obtain the depth and transmittance information, and then converts the image red, The green and blue channels are calculated separately to obtain the defogged image of each channel, and the final combined image is obtained to obtain the final defogged image; it has fewer calculation parameters, simple and fast calculation, good effect, and no color cast, which is very suitable for outdoor fog images Advantages of processing.

The invention discloses a method and a system for detecting defects of flat plate shell based on Lie-Mean. The method comprises the following steps: obtaining a flat plate shell image and carrying outimage preprocessing; manually marking the preprocessed image, marking the flat shell product with the defect, and giving the category with the defect to obtain a data sample set; mapping the obtaineddata sample set into a plum group manifold space, and calculating to obtain a mean value in the plum group of each product category; carrying out image preprocessing on the to-be-detected flat plateshell image, and mapping the to-be-detected flat plate shell image into the plum group manifold space to obtain a test sample; and calculating geodesic distances from the test samples to the obtainedmean values in the plum groups of each category, and judging the category to which the mean value in the plum group with the shortest distance belongs as the category of the test samples. The method has the advantages that (1) the method has good comprehensiveness; and (2) few calculation parameters, excellent calculation performance and small time delay are achieved, and the classification accuracy is improved while the good working efficiency is ensured.

The present application relates to a machine learning-based spectrum sensing method, device, computer equipment and storage medium in the NOMA system. The method includes: when it is necessary to use the target frequency band to send a message, sending a sensing application of the target frequency band to the corresponding base station, so that the base station issues a spectrum sensing task for the target frequency band to each user within the coverage area; Spectrum observation energy signals collected by secondary users in the sensing time slot; according to the location information of each user, the spectrum observation energy values ​​corresponding to each spectrum observation energy signal are filled into the spectrum energy observation matrix, and the obtained energy observation matrix is ​​pre-formed. Processing, the processed energy observation matrix is ​​converted into an image, and the obtained grayscale image is input to the spectrum sensing model based on the Shuffle-Dense neural network for identification, and the current channel state of the target frequency band is output, which improves the accuracy of spectrum sensing.

The invention discloses a multi-channel LSC gain adjustment method and device, a storage medium and image processing equipment. The method comprises the steps: determining the LSC gain of each channelbefore gain adjustment and the LSC gain of a first channel after gain adjustment; according to the LSC gain of each channel before gain adjustment and the LSC gain of the first channel after gain adjustment, determining the LSC gain of other channels except the first channel after gain adjustment, wherein the channel value of each channel after gain adjustment, the product of the LSC gain and thewhite balance gain are equal. According to the method disclosed in the invention, the LSC gains of other channels can be determined under the condition that the LSC gains of the first channel beforeand after adjustment are known, and therefore the color difference problem caused by inconsistent LSC gain change of each channel due to environment brightness or color temperature change is effectively prevented.

The invention provides an offshore wind turbine generator maintenance whole process optimization method and system, and the method comprises the following steps: 1, carrying out the statistics of the fault processing information of a target offshore wind power plant, and 2, carrying out the calculation of the fault processing information according to the obtained fault processing information, respectively setting an economical efficiency index objective function, an equipment failure rate index objective function and a workload balance index objective function which take economical efficiency, equipment failure rate and workload balance as optimization objectives; step 3, constructing a multi-objective optimization algorithm function; 4, setting constraint conditions for the obtained multi-objective optimization algorithm function; 5, performing multi-objective optimization calculation on the multi-objective optimization algorithm function obtained in the step 3 by combining the obtained constraint conditions to obtain multiple offshore wind turbine generator maintenance schemes with the shortest downtime, the lowest maintenance cost and the lowest equipment failure rate as indexes; according to the invention, the defects in the prior art are overcome, and the power generation loss and the maintenance cost caused by shutdown are reduced to the maximum extent.

An execution method when a CPU is applied to an artificial intelligence related program comprises the following steps that when TensorFlow is executed on electronic equipment, a corresponding artificial intelligence model is called according to program content; judging and extracting one or more sparse matrixes used by the artificial intelligence model in operation; performing a matrix simplification procedure on the one or more sparse matrixes; performing instruction conversion on an instruction set adopted by the artificial intelligence model; the artificial intelligence model issues an instruction to the weighted value central processing unit through the converted instruction set; and after receiving the instruction, the weighted value central processing unit averagely distributes a plurality of execution threads to execute a plurality of programs indicated by the artificial intelligence model according to the weighted value.

The invention discloses a method for quickly determining a third-order passive intermodulation product level of a rectangular waveguide flange, and belongs to the passive intermodulation field of microwave components. The method includes: Step 1: the rectangular waveguide flange contact surface is equivalent to a metal-insulation layer-metal structure, and the potential difference between the twosides of the rectangular waveguide flange contact surface is determined; Step 2: The third-order nonlinear current generated by the rectangular waveguide flange contact surface is determined by the rectangular waveguide size, the flange contact area, and the potential difference between the two sides of the contact surface; Step 3: the level of the third-order passive intermodulation products of the rectangular waveguide flange is determined by using the carrier frequency, the rectangular waveguide size, and the third-order nonlinear current.. The invention can obtain the third-order passive intermodulation product level of the rectangular waveguide flange by simple calculation without obtaining the passive intermodulation low-order product and the surface topography parameter measurementin advance, thereby greatly improving the rectangular waveguide method. The efficiency of the passive passive intermodulation problem is reduced and the cost is reduced.