43results about How to "Addressing the drop in accuracy" patented technology

The invention provides an Internet of Vehicles data processing method, a server and a terminal. The method comprises obtaining historical traffic data through the server; performing training of the historical traffic data, and obtaining program scripts for analyzing user's driving behaviors; and sending the program scripts to the terminal. According to the technical scheme, the historical traffic data is obtained firstly through the server, the training is performed on the historical traffic data then, the program scripts for analyzing the user's driving behaviors are obtained, and the program scripts are finally sent to the terminal. The user' driving behaviors are dynamically analyzed according to the program scripts and the driving data, the problem that algorithm precision is reduced because the algorithm is not updated in time is solved, accuracy of analysis of the user's driving behaviors is guaranteed, and, further, extra cost brought by updating of a terminal firmware is reduced.

The invention discloses an endoscope imaging method and an endoscope imaging system. The endoscope imaging method comprises the following steps of illumination, visible light and fluorescence collection, image signal collection and image processing. Visible light reflected by a human subject and excited fluorescence are collected through two sensors located at the front end of an endoscope tube, corresponding image signals are generated, and then the image signals are transmitted to the rear end to be processed to obtain a final endoscope image. Compared with the prior art that visible light and fluorescence are transmitted to the rear end through a light path and then collected and processed, the endoscope of the invention solves the problem that the precision is reduced due to light pathpropagation, the imaging precision is improved, and the light path cost and the light path occupied space are also saved; in addition, the visible light and the fluorescence can be focused and collected respectively due to the design of independent light paths of the visible light and the fluorescence, so that the visible light image and the fluorescence image are consistent in definition; and the endoscope imaging system is binocular stereoscopic vision, and can obtain three-dimensional information to compensate fluorescence brightness of different depth areas.

The invention discloses a honing machine detecting system which comprises a two-dimension (2D) laser scanner, a sliding system and an image processing system. The 2D laser scanner is used for transmitting laser to a guide rail and obtaining an appearance light section curve of the guide rail. The sliding system is connected with the 2D laser scanner and used for fixing and moving the 2D laser scanner. The image processing system is connected with the 2D laser scanner and used for analyzing the light section curve. The 2D laser scanner is utilized to quickly detect a large-scale horizontal type honing machine in a high-efficiency mode, a light section line image is transmitted to a digital signal processor (DSP) processing circuit by a charge-coupled device (CCD) camera, so that all parameters to be measures of each section of guide rail are obtained and compared with specified parameters of the guide rail of the honing machine to confirm whether the honing machine meets machining requirement or not, well solve the problem that accuracy of the honing machine is reduced, and conform to the requirements for being high in machining manufacturing accuracy and high in speed of modern industry. In addition, the honing machine detecting system is simple in structure, low in manufacturing cost, and obvious in practical effect, and has good application value.

The invention discloses a ground object coordinate information-based rapid target positioning method. The method includes the following steps that: the attitude deviation quantity of a satellite is determined according to the position information of a reference point and the orbit information of the satellite; deviation compensation is performed on the attitude information of the satellite based on the attitude deviation quantity; and the position information of a target point is determined according to the compensated attitude information. The present invention also discloses a ground objectcoordinate information-based rapid target positioning device, a satellite-mounted device and a storage medium.

The invention provides an attitude estimation method based on a decoupling step network in the technical field of human body attitude estimation. The attitude estimation method comprises the followingsteps of S10, building the decoupling step network based on a decoupling residual module and a waterfall module; S20, acquiring a large number of human body sample images, and training a decoupling step network by using the human body sample images; and S30, inputting a to-be-tested image into the trained decoupling step network, calculating the position of each joint point in the to-be-tested image, and forming a complete human body posture based on the position of each joint point. The method has the advantage that the speed and precision of human body posture estimation are greatly improved.

The invention discloses a full-period digital time converter based on a clock calibration technology, which is characterized by comprising a single-to-differential circuit used for converting an input single-ended signal into a differential signal and inputting the differential signal into a multi-phase clock signal generation unit; the multi-phase clock signal generation unit is used for generating a multi-phase clock signal according to an input differential signal and inputting the multi-phase clock signal into the adjustable delay module; the adjustable delay module is used for delaying the multi-phase clock signal according to the output signal of the digital module and then inputting the delayed multi-phase clock signal into the multiplexer; the multiplexer is used for selecting a group of adjacent two-phase signals to output according to the switch control signal generated by the digital module; the phase interpolator is used for generating numerical control time signals according to two adjacent phase signals; the time-to-digital converter is used for calculating the time difference of rising edges of two adjacent phase signals and inputting the time difference to the digital module; and the digital module is used for generating a delay control signal of the adjustable delay module on the multiphase clock signal and a switch control signal according to the time difference of the rising edges of each group of adjacent two-phase signals.

The invention relates to a sound event detection and positioning method based on deep learning. The method comprises the following steps of: step 1, segmenting a data set; step 2, preprocessing, namely performing feature extraction on the data set containing the sound signal to obtain a Log-Mel spectrogram and GCC-PHAT; step 3, constructing a deep learning model, namely constructing a network architecture combining a ResNet framework and RNN by referring to the ResNet framework, and compounding a pooling module, a regularization module and a normalization module between layers for optimizing feature extraction and improving nonlinearity; and step 4, two-step training, namely firstly training the SED task to obtain an optimal model and inputting the training result into DOA task training asa feature; and then training the DOA task to finally obtain an optimal training model. According to the method, the characteristics suitable for task training are extracted firstly, so that the reverberation resistance is improved, a new frame structure is provided to solve the problem that the precision is reduced due to network deepening, and finally the prediction precision is improved.

The invention belongs to the technical field of images, and discloses a federated learning system for multi-type task image analysis, the federated learning system comprises a Server and a Client, the Client comprises a Decoder, an Encoder, an Encoder average value and fc, and the Server comprises the Encoder average value. According to the method, data islands among all parties are broken through by using a federated learning framework, so that modeling can be carried out by combining data of all parties on the premise of protecting image data privacy, the accuracy is higher than that of independent modeling, and image analysis of multiple types of tasks can be carried out at the same time.

The invention provides a fiber grating sensor temperature stress decoupling method in a metal hole structure. The fiber grating sensor temperature stress decoupling method in the metal hole structure comprises steps of designing a layout of the fiber grating sensor, measuring an original state of a detected basal body, applying an increasing pulling force to the detected basal body, measuring central wavelength changing conditions of two grid points in a first group, applying external force and temperature influences to the detected basal body at the same time, measuring a changing quantity of the central wavelength of each grid point, and further calculating stress strains of all other points around a hole. The fiber grating sensor temperature stress decoupling method in the metal hole structure realizes fiber grating sensor temperature and stress decoupling in the metal hole structure, respectively solves magnitudes of temperature and the stress of the fiber grating sensor, improves accuracy of a fiber grating stress sensor monitoring a structure stress, solves a problem that the accuracy of the fiber grating stress sensor is reduced due to temperature interference and benefits popularization and application of a fiber Bragg grating sensor.

The invention relates to the technical field of biological signal recognition, in particular to an electromyographic signal gesture action recognition method based on texture features. The method comprises the steps of firstly collecting myoelectricity data under different gesture actions, secondly representing original myoelectricity data by using a cyclic gray-level co-occurrence matrix, then extracting texture features of corresponding position offset robustness according to the obtained cyclic gray-level co-occurrence matrix, finally extracting time domain features on the basis of the texture features; and inputting the obtained time domain features into a classifier for model training, and finally outputting a gesture action classification result. The method provided by the invention has great flexibility, can well solve the problem that the model precision is reduced due to the position offset of the sensor in practical application, and effectively improves the robustness of the myoelectricity man-machine interaction system.

The invention provides a production mold for a micro-fluidic chip and a preparation method and an application thereof and belongs to the technical field of biochemical instrument and equipment production. In embodiments, during the preparation of the micro-fluidic chip, the production mold can greatly prolong the service life of a mold in the same shape, so that production cost of a glass micro-fluidic chip is greatly reduced.

The invention relates to a sun disc surface speed difference-based autonomous celestial navigation method. Aiming at the phenomenon that linear speeds of different points on a sun disc surface are different and the radial speeds, observed by a detector at different positions, of different sun points are different due to differential rotation of the sun, a research finds that speed information is afunction of the current position of the detector, so that the position and the speed of the detector can be estimated by using the speed information. According to the method, the radial speeds of different points on the sun disc surface are measured by using spectrometers; speed difference values between the radial speeds are calculated; a measurement model is built by taking the speed differencevalues as measured quantities; and the position and the speed of the detector are estimated by combining an orbital kinematic model with Unscented Kalman filtering. According to the sun disc surfacespeed difference-based autonomous celestial navigation method, the effects of spectrometer errors and speed changes of the sun on the navigation performance can be eliminated, so that the estimation accuracy is high.

The invention relates to the field of face key point detection, and discloses a face box selection method in a face key point detection module, which comprises the steps of obtaining a face detectionmodel and a key point model; obtaining a face box coordinate scaling ratio according to the face box coordinates of N pictures and the real face box coordinates of N pictures obtained by the face detection model; optimizing the face box coordinates of each picture in a test data set output by the trained face detection model by using the face box coordinate scaling ratio to obtain the optimized face box coordinates of each picture in the test data set; and performing coordinate conversion on the optimized face box coordinates to obtain new face box coordinates, cutting out a face box from theoriginal picture according to the new face box coordinates, and inputting the cut-out face box into the trained key point model to obtain key point coordinates. According to the method, the problem ofkey point precision reduction caused by inconsistent face box selection standards in the face detection module and the key point detection module can be effectively solved.

The invention discloses a method for reducing the precision of a continuous gyroscope clinometer under the influence of irregular vibration. The method adopts the three-axis acceleration information and combines the z-axis angular velocity information of a fiber-optic gyroscope. Meanwhile, the interference, except the instrument attitude measurement, is removed. Therefore, the measurement precision of the continuous gyroscope clinometer is comprehensively improved. According to the method, the problem that the precision of the continuous gyroscope clinometer is reduced due to the irregular collision of the continuous gyroscope clinometers with an oil sleeve during the downhole upward and downward process can be solved.