71results about How to "Preserve precision" patented technology

The present invention provides a multi-wavelength interference real-time absolute distance measurement device on the basis of femtosecond optical comb synchronization frequency locking. The multi-wavelength interference real-time absolute distance measurement device on the basis of femtosecond optical comb synchronization frequency locking comprises: a multi-wavelength generator on the basis of the femtosecond optical comb synchronization frequency locking configured to emit a plurality of high-frequency stable object wavelength lasers; a heterodyne interferometer configured to receive lasers outputted by the multi-wavelength generator after the femtosecond optical comb synchronization frequency locking, returning measurement light and reference light after generation of interference through the heterodyne interferometer being coupled in single mode fibers by lens focusing to form mixed multi-wavelength heterodyne interference signals; and a synchronization phase demodulation module configured to separate the mixed multi-wavelength heterodyne interference signals obtained by the heterodyne interferometer through wavelength demodulation to obtain interference signals corresponding to each wavelength and perform synchronization phase measurement processing of the interference signals corresponding to each wavelength to finally obtain an accurate absolute distance value. The multi-wavelength interference real-time absolute distance measurement device on the basis of femtosecond optical comb synchronization frequency locking is high in traceability of values, high in measurement precision, large in non-fuzziness measuring range, fast in update rate and easy to measurement on site in real time, and is applicable to the fields of the industrial production, the equipment manufacture and the large-size precision measurement.

The invention relates to a millimeter wave radar and vision cooperative target detection and identification method based on a neural network. A millimeter wave radar can obtain information such as theposition and the speed of an object by transmitting and receiving millimeter waves. A camera can obtain a visual image of an object. The position information obtained by the millimeter wave radar ismapped to the image. The region of interest is delineated in the image feature map. The region of interest is sent to the deep learning neural network to be processed, and the identification and positioning information of the target can be obtained. According to the invention, a millimeter wave radar and visual processing combined mode is adopted to complete target detection and recognition tasks.By means of a deep learning neural network technology, the object positioning time is shortened, and the recognition accuracy is improved.

The invention discloses an electric power inspection image defect identification method. The method comprises the following steps: creating and cascading a target detection network and a plurality ofclassification networks; obtaining a plurality of frames of inspection image samples, labeling targets in the inspection image samples, and generating a training sample set; adopting a training sampleset to train a cascade network, wherein the quantization parameter of each network layer is related to the quantization stage number and the quantization range of the network layer where the networklayer is located; and identifying defects in the newly acquired inspection image by adopting the trained cascade network. According to the invention, an effective FPGA airborne identification system is provided for operation and maintenance of a power grid tower and an overhead line, and a corresponding quantization function can ensure that different channels of different network layers can be properly quantified, so that the precision of the network is reserved to the maximum extent; through cascading the target detection network and the classification network, the defect detection accuracy is greatly improved, and the unmanned aerial vehicle routing inspection of the power grid truly realizes automatic identification.

The invention provides a quantification method of a convolutional neural network based on knowledge distillation. The invention relates to the technical field of convolutional neural networks. The optimal cutting proportion and quantization digit of each network weight layer of the convolutional neural network is obtained by using a dynamic search mode; the trained convolutional neural network issequentially subjected to sparsification, knowledge distillation fine adjustment training by adopting a teacher network, layer-by-layer quantification and knowledge distillation fine adjustment training by using the teacher network again, so that the storage requirement and the calculation requirement of the convolutional neural network are greatly compressed on the premise of keeping the networkprecision. In the quantitative compression process of the convolutional neural network, the teacher network is used for knowledge distillation-based fine adjustment training, different quantization precisions can be adopted for different network weight layers of the convolutional neural network, and the network precision cannot be excessively lost, so that the compression degree of the convolutional neural network can be increased in the quantization process.

The invention provides a long-dynamic-process simulation method based on a PSS/E trend API. The method includes the following steps of establishing a power grid model, forming a PSS/E trend calculation interface layer, calling the interface layer to calculate a trend and building a simulation system to achieve long-dynamic-process simulation of a power grid. The method achieves control over the whole PSS/E trend calculation process of an external application program; the simulation method maintains the accuracy of PSS/E trend calculation, further extends the functions of PSS/E trend calculation, and achieves long-dynamic-process simulation of the power grid; the method supports the mode that modeling is conducted through leading-in from model files in various standard data formats, and has high flexibility and adaptability; according to the method, a Fortran API provided by PSS/E is used, the Fortran API can be directly called by C++ and the like, can be directly embedded into a developed program of a user and is suitable for development of a large-scale application program.

The invention relates to a CPIII (control point of III level) control net adjustment method; a CPIII control net acquired by using constraint adjustment method is adverse to the evaluation and regulation of internal geometric dimensions of a rail; the method focuses only on meeting the requirement for external geometric dimensions of a rail; a correction numerical matrix for direction observation and horizontal distance observation are acquired by using an error equation of direction observation and horizontal distance observation, an additional condition matrix is acquired through the correction numerical matrix of the error equation, an additional condition equation and adjustment rules, namely an adjustment model is established, and adjustment calculation is finished by solving the model; the method of the invention gives the priority to meet the requirement for internal geometric dimensions of a rail, can meet the requirement for external geometric dimensions of the rail, and is free of influence form original data errors, the CPIII control net is never distorted, the CPIII control net is introduced to a line design coordinate system through position reference and direction reference, and the CPIII control net can meet the requirement for 'triply play' on high-speed railway.

The invention provides a method for measuring the average wind velocity of a tunnel by single point statistics. The method comprises the following steps: determining sampling point positions and the number of sampling points of a to-be-measured tunnel section; determining the number of sampling of the sampling point wind velocity by setting the sampling point wind velocity sampling standard error confidence level and time-average error according to the statistic principle; arranging a wind velocity sensor in each sampling point, collecting the instant wind velocity of each wind velocity sensor, and determining the time-average wind velocity in the sampling point; determining the average wind velocity of a tunnel section according to a statistic measurement mathematic model of the tunnel average wind velocity; selecting an arbitrary point in the to-be-measured section as a single measurement point position, assembling a wind velocity sensor in the single measurement point position, and determining the wind velocity conversion coefficient of the single measurement point position; and collecting the instant wind velocity of the single measurement point position in real time, and determining the current time-average wind velocity of the single measurement point position to determine the current average wind velocity of the tunnel section. The method maintains the accuracy of a multi-point wind measurement method and has the advantages of simpleness, convenience and high efficiency of a single point method.

The invention discloses a compression method and device based on characteristics of a flat region of an image. The method specifically comprises the following steps of (1) utilizing a single frame of a video or information of a single picture to obtain a flat region and establish a pixel gradient model of the flat region, and dividing the flat region into blocks of the flat region according to the pixel gradient model; (2) utilizing pixel information in the blocks of the flat region to obtain a gradient distribution model of the blocks of the flat region and utilizing the gradient distribution model to classify pixels included in the blocks of the flat region in order to obtain a pixel set and a pixel set description operator corresponding to the pixel set; (3) compressing the pixel set and the corresponding pixel set description operator to generate descriptor code streams; (4) utilizing the descriptor code streams to obtain a prediction image of the pixel set; compressing a residual image obtained by solving a difference of an original image of the pixel set and the prediction image to obtain residual code streams; and (5) carrying out multiplexed compression on the descriptor code streams and the residual code streams to generate compressed image code streams or files.

The present invention discloses a dual-mode controller of an electrical-mechanical converter. The dual-mode controller of the electrical-mechanical converter comprises a control chip, a digital analog converter (DAC), a hysteresis comparison PWM modulator, a current signal conditioning circuit, a difference orthogonal signal conditioning circuit, a signal switching switch, an H-bridge driving module, a protection circuit, the electrical-mechanical converter, a temperature sensor, a current sensor and a digital grating scale. An input signal and a feedback signal are processed by the signal conditioning circuits and then are sent to the control chip, the control chip determines and operates to obtain a control signal, and the control signal is sent to an H-bridge driver to drive the electrical-mechanical converter to reach an expected control objective, wherein the control chip switches between a PID closed-loop control mode and a time optimal Bang-Bang control mode according to an error signal, thereby achieving the control objective of being high in response characteristic and not losing the precision.

The invention provides a convolutional neural network pruning method, apparatus and device, and a medium. The method comprises the steps of obtaining a data set of an image and dividing the data set into a training set and a verification set; after a to-be-pruned convolutional neural network model is initialized, multiple rounds of pre-training being carried out on images in the training set, and determining a sensitive layer of the network; multiple rounds of automatic pruning operation being carried out through reinforcement learning, the model accuracy of each round of pruning operation being obtained, and therefore obtaining the model pruning strategy with the highest model accuracy. According to a deterministic strategy of reinforcement learning, different compression strategies are carried out on a sensitive layer and a non-sensitive layer of a network, so that the number of filters of each network layer is pruned; and performing fine tuning operation to obtain a final convolutional neural network model. According to the method, reinforcement learning is involved from the beginning of model pruning, the pruning strategy of the model is optimized according to the environment, the pruning object is a complete filter, irregularity of the model cannot be caused, and generalization is greatly improved.

The present invention relates to an animation data processing method and device, a computer device and a storage medium. The method comprises the steps of: obtaining an original key frame data set corresponding to a current animation object to be subjected to data processing; obtaining first motion data of a forward adjacent key frame in a current transformation dimension, and obtaining second motion data of a backward adjacent key frame in the current transformation dimension; comparing current motion data with the first motion data and the second motion data; deleting the current motion dataaccording to a comparison result at present; returning the step of obtaining the key frame to be processed at present until the key frame to be processed in the original key frame data set is completely processed, and obtaining a current key frame data set; and deleting the current dimension motion data according to a current motion type, wherein the current dimension motion data is motion data of each key frame in the current key frame data set in the current transformation dimension. The method provided by the invention reduces the occupation degree of the animation motion data in the computer resource.

The invention discloses a method for detecting hidden discharge current based on a voltage extreme value, comprising the steps of in real time monitoring information according to voltage waveform in an electric transmission line; determining amplitudes, phases, wave peak and wave valley time of the voltage waveform in the electric transmission line; acquiring lengths, waveform and triggering time of traveling wave signals by a traveling wave online monitoring device; selecting proper voltage extreme value neighborhoods; comparing the triggering time of each traveling signal in the electric transmission line with the extreme value time thereof according to the condition that the traveling wave triggering time becomes into hidden waveform; determining whether hidden discharge current presents at each extreme value position and in the entire voltage waveform section or not; if not, taking detected waveform current signals as clutter signals, and totally removing the interference waveform of non-hidden discharge current. By the method, the identifying efficiency of hidden discharge current can be greatly increased, hidden troubles of a power grid are reduced in removal and maintenance and the power grid is ensured to safely and steadily operate.

The invention relates to a device for measuring an optical free-form surface, which is used for measuring a three-dimensional shape of a workpiece. The device comprises a focused type laser displacement sensor, a reflector, a gas floating measuring head, a nano two-dimensional displacement mechanism, a signal acquisition circuit of a focused type laser displacement sensor, an industrial control computer and a nano two-dimensional displacement mechanism controller, wherein the gas floating measuring head comprises a probe and a probe shaft; the reflector is arranged at the tail of the probe shaft; the probe shaft is arranged on a gas floating guide rail; the industrial control computer controls the nano two-dimensional displacement mechanism through the nano two-dimensional displacement mechanism controller to move according to a designed path; the probe of the gas floating measuring head slides on the surface of a test workpiece to enable the reflector to generate continuous axial movement; the focused type laser displacement sensor generates a focus error signal; and the focus error signal is input into the industrial control computer through the signal acquisition circuit.

According to the mobile high-definition video intelligent bidirectional detection bandwidth control method provided by the invention, a target end frame-level time delay trend extraction method and an extended RTP/RTCP driven source end rapid adaptive method are fused, so that the accuracy of the target end frame-level time delay trend extraction method is reserved; according to the method, the coding rate can fully use the available video bandwidth of the link to achieve the best subjective effect, meanwhile, the characteristic of high speed of the RTP/RTCP is fused, the packet loss phenomenon before congestion is detected due to the fact that the link cache of mobile high-definition video communication is too small is avoided, the experience of the mobile high-definition video is improved. A video bandwidth control architecture is provided on the basis, so that the video bandwidth control architecture has a better video effect for a fluctuating mobile network, is high in feasibility in practical application, is simple, efficient and high in practicability, serves as the most important link for guaranteeing the quality of a mobile high-definition video, and plays an extremely important role in mobile high-definition video communication.

An adjustable spanner with electronic strain gauge function which comprises a head; an retaining jaw extending from the head; and sliding groove being transversally formed at one upper side of the head; a handle connected to a lower side of the head; at least one strain gauge; a hollow connecting tube having one end connected to the handle; the hollow connecting tube being installed with an output display device; a clamping surface of the moveable jaw and a clamping surface of the retaining jaw being formed as a working area for clamping a screw device; and a positioning portion being installed on the moveable jaw and at a non-working area below the work area; the position portion facing toward a clamping opening between the clamping surfaces of the moveable jaw and retaining jaw; and the positioning portion being capable of resisting against the screw device.

The invention discloses a sensitive information leakage detection method. The method comprises the steps of collecting a to-be-detected network data packet; classifying the network data packets to bedetected; respectively sending the network data packet to be detected to a plurality of processors, and performing parallel TCP stream restoration processing by adopting the processors to obtain restored text information; carrying out natural language processing on the restored text information, judging whether sensitive text information exists in the restored text information or not, and calculating the leakage rate of the sensitive text information; and judging a network data packet where the sensitive text information is located as a sensitive information leakage data packet, performing traceability analysis on a source address of the sensitive information leakage data packet, and positioning a sensitive information leakage source. According to the sensitive information leakage detection method disclosed by the invention, the capture speed of the data packet can be improved, and the time for identifying and detecting the sensitive information of the network is shortened. The invention further discloses a sensitive information leakage detection device and a storage medium.

The invention relates to a far-field light-field full-vector calculating method. After near-field distribution of an optical device is obtained through other full-vector calculation methods, distribution of any point of an optical-field far field is obtained through a full-vector integral method so that precision of full-vector calculation is reserved and speed and far-field calculability of a diffraction algorithm are also realized. The calculation result is more accurate than a scalar approximation method and not only intensity of the far-field light field can be calculated accurately, the phase of the far-field light field can also be calculated accurately. At the same time, because an integral formula is significantly small in occupied memory and unrelated with a far-field distance, a far field which cannot be calculated through other full-vector methods can be calculated so that the method has the advantages of being rapid and efficient. The method is widely applicable to the field of optics.

The invention discloses a CNC numerical control equipment tool path corner smoothing processing method. The method comprises the following steps of setting a smoothing parameter in a tool path motion instruction of CNC numerical control equipment, wherein a default value of the smoothing parameter is ''Yes''; setting or reading a maximum deflection angle threshold theta and a bow high error epsilon of the tool path motion instruction of the CNC numerical control equipment so as to calculate a maximum fractional length; reading a current tool path motion instruction type, acquiring a characteristic value of a subsequent path, and carrying out segment treatment on the subsequent path according to the characteristic value of the subsequent path; carrying out end point adjustment; and carrying out line segment adjustment. By using the method in the invention, precision of a line segment and a middle portion of a long arc is retained; simultaneously a corner surrounding of a deceleration point can be adjusted so that a speed and the precision can be balanced. An experiment proves that the tool path on which the smoothing processing is performed by using the method is round and smooth. For some processing fields, such as highlight processing, a significant effect is achieved, and processing efficiency and processing quality are greatly increased.

The invention discloses a distance measuring apparatus and a distance measuring method. The distance measuring method comprises the following steps: a distance measuring optical signal is emitted on an object to-be-measured, wherein, the distance measuring optical signal comprises a pattern of at least one light spot; a distance measuring image formed by the distance measuring optical signal projected on the object to-be-measured is obtained, and the distance measuring image comprises a projection light spot formed by the distance measuring image projected by at least one light spot; when gravity centre of the distance measuring image is in a first section, the distance between the object to-be-measured and the distance measuring apparatus is calculated according to a first mapping relation, wherein the first mapping relation records several first reference gravity centre positions and several corresponding first reference distance reciprocals; when gravity centre of the distance measuring image is in a second section, the distance between the object to-be-measured and the distance measuring apparatus is calculated according to a second mapping relation, wherein the second mapping relation records several second reference gravity centre positions and several corresponding second reference distance. The method can simultaneously keep the distance measurement accuracy of long distance and near distance.

The invention provides a full-polyester sanded cloth paperless digital direct spray printing process and relates to the field. The full-polyester sanded cloth paperless digital direct spray printing process comprises the following steps: modulating sizing slurry; dipping-rolling and sizing: performing twice-dipping and twice-rolling on full-polyester sanded cloth in the prepared slurry, wherein the slurry rolling rate is 55-65 percent; primary drying: performing primary drying; digital spray printing: making a spray printing template through computer digital drawing, and then directly sprayinga drawn pattern with high-temperature dispersible ink through a digital color printing machine, wherein the environment temperature during spray printing is 22-30 DEG C, the relative humidity is 50-80 percent, a spray printing mode is a medium speed mode, the spray printing speed is 60-75 m<2>/h, the spray printing precision is 360*800 dpi, and the height of a spray head is 3-4 mm; performing secondary drying; detecting that the printing precision, the effect and the color fastness test are qualified; moreover, as no paper or water is used, the production cost is greatly reduced, and the economic benefits and the social benefits are great.

Explorative searching method, apparatus and computer programs are presented. A first explorative search is run in a first branch. A first group of entities relating to the first explorative search is presented to a user. A second explorative search is run in a second branch in parallel with the first branch. A second group of entities relating to the second explorative search is presented to the user simultaneously with the presenting of the first group of entities. The user is allowed to import one or more entities of either one of the first and second explorative searches to the remaining one of the first and second explorative searches and updating said remaining explorative search.