61results about How to "Improve precision and accuracy" patented technology

The embodiment of the present invention provides a method for identifying a clothes image, and a labeling method and device of the clothes image. The method for identifying a clothes image comprises a step of obtaining clothes images which are labelled with positioning key points and clothes wearing areas, a step of extracting an image area which represents a feature region in at least one of the clothes images according to the positions of the clothes wearing areas and the positioning key points, and a step of obtaining feature information which represents the feature region according to the region feature identification model corresponding to the feature region. Through labeling the positioning key points, the image which expresses clothes content in the clothes images is positioned, and the interference of background content in the clothes images is reduced. On the above basis, the image area which represents the feature region are extracted from the clothes images according to the clothes wearing areas and the labelled positioning key points, then the feature information of the feature region is obtained according to the corresponding region feature identification model, and thus the accuracy and precision of the feature information of the clothes in the clothes images are improved.

The invention relates to an apparatus and a method for detecting a partial discharge of a power transformer by utilizing a fluorescence fiber, more particular to a simulation experiment apparatus and a method thereof for a partial discharge of a needle-board electrode in a power transformer. The apparatus provided in the invention mainly comprises: an induction regulator; a dizziness-free test transformer; a partial discharge protective resistance; a standard capacitance voltage divider; a partial discharge generator, which is formed by a needle-board electrode, a fluorescence fiber and a housing; a photomultiplier; a direct current regulated power supply; and an oscilloscope and the like. According to the method provided in the invention, the apparatus is utilized; and a needle-board electrode partial discharge experiment in the air of a power transformer is simulated. According to the invention, partial discharge experiments of a transformer with different exerted voltages and in different spacings are simulated; besides, the apparatus has advantages of strong anti-interference capability, good insulating property and high detection sensitivity and the like. Moreover, the apparatus and the method can be widely applied to luminous characteristic researches of a power transformer on floating potential partial discharge, insulator surface discharge and partial discharge in a transformer oil and an SF6 of a GIS in the air medium.

The invention discloses a holographic-image-based streetscape image fragment optimization method, which comprises the following steps: 1, acquiring vehicle-mounted LiDAR (light detection and ranging) point cloud data and a holographic image, and registering the holographic image and the vehicle-mounted LiDAR point cloud data; 2, dividing the vehicle-mounted LiDAR point cloud data into a plurality of image fragments, obtaining a corresponding relationship between the image fragments and sites of the holographic image, and projecting the image fragments onto the holographic image to obtain a perspective plane image corresponding to the image fragments; 3, analyzing the perspective plane image, deleting tree points, and performing image fragment stretching. According to the method, the streetscape image fragments are optimized on the basis of the holographic image, and the precision and accuracy of the image fragments in point cloud data results are further improved.

The invention provides a detection method for black and white checkerboard image corners based on least square optimization. According to the method, a colorful checkerboard image is converted into a grey level checkerboard image, binarization is performed on the checkerboard image to obtain a binarized image, the corner region is determined according to the binarized image and the grey level checkerboard image, a method of least square fit for circles is used for fitting the circles, the centers of the circles are the checkerboard image corners, and therefore a corner coordinate set is generated. According to the method, parameters for calibrating a camera lens are not needed, and the method is suitable for detecting any distorted black and white checkerboard corner and widely applied to the fields of camera calibration, panorama parking, visual detection and monocular distance measuring.

The invention relates to a filter bank training method. The filter bank training method comprises the steps that first, a training image which has a target position mark is preprocessed to obtain a denoising training image; second, initial clustering is conducted on the denoising training image, and the image is decomposed into K training sets; third, an ideal filter output model is designed according to the target position mark in the training image; fourth, K total filter models are obtained by training according to the ideal filter output model to constitute a filter bank; fifth, whether an image sample set is convergent or not is judged, if yes, the seventh step is executed, and otherwise the sixth step is executed; sixth, whether the convergent frequency reaches a preset threshold value or not is judged at present, if yes, the seventh step is executed, otherwise, classification is conducted again to obtain K new training sets, the K new training sets replace the K training sets, and the fourth step is executed again; seventh, the filter bank is stored, and the training process of the filter bank is completed. The filter bank training method has better distinguishing performance to targets, and improves the accuracy and the precision of positioning to a certain extent.

The invention discloses a cigarette permeability on-line detection method which is applied to a cigarette making and plug assembling machine set. The method is characterized by comprising the following steps of applying positive-pressure air onto a cigarette ignition end on a detection drum wheel, respectively installing a pressure sensor on two ends of a cigarette to detect the pressure value on the air inlet end and the air outlet end, applying the identical positive-pressure air onto the filter tip end of the cigarette on a next slot, installing a pressure sensor on the cigarette ignition end to detect the pressure value on the air outlet end, transmitting the three pressure values to a controller processing unit to be calculated, judging the cigarette permeability, outputting a removal signal, and removing the cigarette with unqualified permeability on a removal drum wheel. By adopting the method, the influence of the air pressure fluctuation and different brands on the detection of the permeability can be reduced. The online detection precision and accuracy of the cigarette permeability can be improved, the quantity of the cigarettes with unqualified permeability in the market can be reduced, and the cigarette quality can be improved.

The invention discloses a target community data identification method, which comprises the steps of performing transaction relationship network construction processing on acquired transaction relationship data of a plurality of objects to obtain a transaction relationship network; performing community classification on the transaction relationship network based on a modularity graph clustering algorithm to obtain a community classification result; and determining an analysis dimension according to the attribute of the target community, and analyzing the community classification result throughthe analysis dimension to obtain target community data. The transaction relation network is subjected to community classification through the graph clustering algorithm based on modularity, the community classification result is obtained, and then the target community data is screened out, so that the target community data search precision and accuracy are improved, and the target community recognition effect is improved. The invention also discloses a target community data identification device, a server and a computer readable storage medium, which have the above beneficial effects.

The invention discloses a method and a device for correcting a defective pixel of an image, which are used for solving a problem of lower correction precision of the defective pixel in the prior art.The method specifically comprises the following steps that electronic equipment establishes an N*N pixel matrix by taking the fth pixel point in an image to be detected as a central pixel point, wherein f is a positive integer which is not larger than the total number of the pixel points in the image to be detected; the electronic equipment respectively converts the colors of the pixel points of an inner ring and an outer ring into the color of the central pixel point, and respectively determines pixel values after each pixel point is converted; and the electronic device determines the centralpixel point as the defective pixel based on the pixel value after each pixel point is converted, and corrects the central pixel point.

The invention discloses a visual detection system and method for a through hole of a transparent plate . A combined light source is controlled by a controller to perform time-sharing switching of multiple lighting modes; at least one imaging device is controlled to carry out imaging scanning in different illumination modes; multi-channel image acquisition of shapes and defects of the through holeis realized; the information amount of image acquisition is greatly enriched; the accuracy and the precision of the through hole shape information and the defect information are improved; finally, theprocessor is used for analyzing based on a large amount of collected image data to obtain the shape precision and internal defects of the through hole in the transparent plate; the detection rate andthe recognition rate of the through hole defects are ensured; and a three-dimensional distribution diagram of the through hole defects in the transparent plate is accurately constructed, so that important reference indexes can be provided for guiding defect causes, adjusting a production process and grading material finished products on line, the real-time performance is high, the reaction speedis high, the online improvement of the processing quality is facilitated, and the production operation efficiency and the product yield are greatly improved.

The invention relates to the technical field of cameras, and provides a camera focusing method and system. The camera focusing method comprises the steps that a fringe test image is acquired, and an area to be recognized is positioned in the fringe test image; whether a stripe group satisfies a preset imaging condition is determined according to the stripe width in order from large to small; if the stripe group satisfies the preset imaging condition, the stripe group of next stripe width is judged, otherwise the stripe group of previous stripe width is output as a result; whether the stripe width of the final output stripe group is within a preset stripe width range is detected; and if the stripe width of the final output stripe group is within the preset stripe width range, focus operation is completed, otherwise a lens is adjusted and re-focus operation is carried out. According to the camera focusing method provided by the invention, the accuracy rate and precision of focusing are improved without referring to the observation result of human eyes, and the focusing efficiency is improved.

The present invention provides a laser-type reflector coordinate system calibration method. The method comprises the steps of: scanning a reflector by a laser radar sensor to scan three unequally spaced reflection points in order; if the scanned three unequally spaced reflection points are collinear, taking a straight line where the two points have the largest interval distance are located as an Xaxis of a rectangular coordinate system, taking a point the farthest to the middle point as an original point, and determining a Y-axis direction of the rectangular coordinate system according to a right-hand rule; and if the scanned three unequally spaced reflection points are not collinear, allowing the three unequally spaced reflection points to form a non-isosceles triangle, taking the longest side of the non-isosceles triangle as an X-axis of the rectangular coordinate system, taking a vertex of an intersection point of the longest side and the longer side in other two sides as an original point, and determining a Y-axis direction of the rectangular coordinate system according to a right-hand rule. The method can calibrate the unique rectangular coordinate system and employs the laser for scanning to increase the location precision and reduce the navigation errors of a robot.

The invention discloses a road-level global environment map generation method for unmanned driving. The method comprises the steps of acquiring current data of the driverless vehicle, calculating the current motion state of the unmanned driving vehicle, extracting an image feature matrix, screening a similar visual template, adding a new visual template, performing empirical point matching, performing lane line detection, performing lane line and global empirical map fusion and the like to finally generate an unmanned driving road-level global environment map. The unmanned driving car utilizes the vehicle-mounted sensor to sense the surrounding environment of the car. The road, the vehicle position and the obstacle information are obtained through perception; therefore, the aim of unmanned driving is fulfilled; the problem that the traditional method is susceptible to terrain, weather conditions and precision is not accurate enough is solved; according to the road-level global environment map generation method for unmanned driving, environment information is continuously captured in the vehicle driving process to generate a global experience map, and then in combination with multi-lane line detection, the road-level global environment map with high accuracy is obtained.

Disclosed are a method and a system for detecting flock fractal features in water in different spatial dimensions. A column precipitator is provided with a mud hole at the bottom and a water inlet at the top, a sampling pipe is connected with the side of the column precipitator close to the water inlet, the sampling pipe is provided with a screw valve, a machine vision LED backlight source is arranged on one side of the column precipitator, a telecentric lens is connected with a CCD camera, the CCD camera is fixed on a horizontal testing platform, and an acquiring system composed of the horizontal testing platform, the telecentric lens and the CCD camera is arranged on the observation side of the other side of the column precipitator. The CCD camera is connected with an industrial control machine via a data line, a display is connected with the industrial control machine, one end of the sampling pipe is connected with the column precipitator, and the other end of the sampling pipe is provided with a sampling inlet.

The invention discloses a text emotion information identification method, which comprises the following steps: preprocessing a to-be-predicted text by adopting a trained Word2vec model to obtain an input vector of the to-be-predicted text; carrying out feature recognition on the input vector by adopting a trained BiLSTM model to obtain context time sequence text information features; performing feature extraction on the context time sequence text information features through a trained CNN model to obtain target features; and identifying the target features by adopting the trained neural network to obtain an emotion identification result. Through the trained Word2vec model, BiLSTM model and CNN model, the time sequence characteristics of the context of the to-be-predicted text are extracted, the local characteristics and sequence information of the text are obtained, and the precision and accuracy of sentiment analysis are improved. The invention further discloses emotion information recognition equipment, a natural language analysis device and a computer readable storage medium which have the above beneficial effects.

An ultrasonic flow rate detection system comprises a system control part, a circuit part and an assistant device, wherein the system control part comprises a single chip, a digital signal processor, and an on-site programmable gate array; the circuit part comprises a switching circuit, an automatic gaining control amplifying circuit, an analog/digital converting circuit, and a digital/analog converting circuit; the assistant device comprises a storage part, an input device, a display device a power supply and an ultrasonic transducer; the on-site programmable gate array is connected with the input device and the ultrasonic transducer respectively, and used for receiving the input instruction of the input device and conducting time control on the ultrasonic transducer; the digital signal processor is connected with the storage part, and used for data information exchange; the on-site programmable gate array is in bi-directional connection with the single chip and the digital signal processor respectively; the single chip is connected with the display device, and used for displaying data information through the display device.

The invention discloses a mountain fire detection method and system based on static and dynamic multi-feature fusion. The method comprises the following steps: firstly, performing frame sampling on aforest fire video to generate an initial image sequence, and preprocessing the initial image sequence to generate a preprocessed image sequence; dividing the preprocessed image sequence into a staticsample and a dynamic sample, and extracting static features of the static sample and dynamic features of the dynamic sample by using a convolutional neural network; fusing the static features and thedynamic features by using a kernel canonical correlation analysis method to generate fused features; inputting the fused features into a support vector machine classifier for learning and training, and generating a trained forest fire detection classifier; and carrying out mountain fire detection by adopting the trained mountain fire detection classifier. According to the method, in the process ofextracting the image features based on the video frames, the static features and the dynamic features of the preprocessed images are extracted respectively and then fused, the forest fire features can be extracted comprehensively, and the precision and the accuracy of forest fire detection are improved.

The invention relates to an MRI image heart structure segmentation method based on a multi-path convolutional neural network. The MRI image heart structure segmentation method comprises the steps: collecting cardiac movie MRI training data of a normal person and a heart patient; enabling a doctor with rich experience to manually mark the heart structure in the training data to serve as a heart segmentation marking result; training the heart region extraction model based on the training data, so as to enable the heart region extraction model to accurately extract the heart region; training theheart segmentation network according to the heart region extracted from the training data to segment each structure of the heart; and measuring the segmentation performance of the constructed heart segmentation network by taking the standard segmentation labeling result as the standard. The MRI image heart structure segmentation method adopts the heart region extraction model based on the generative adversarial network to extract the heart, so that the accuracy of heart region extraction is improved. Meanwhile, the MRI image heart structure segmentation method utilizes contextual information between adjacent layers through the multi-path convolutional neural network, so that the segmentation precision and accuracy are improved.

The invention relates to an intermittent energy mass data processing method. The method includes the following steps that (1) an intermittent energy mass data processing platform is established, wherein the processing platform comprises a main cloud central server, an auxiliary cloud central server, a first load server, a second load server, a third load server, a fourth load server, a fifth load server, a sixth load server, a seventh load server, an eighth load server, a ninth load server, a tenth load server, an eleventh load server, a terminal data bus, a load bus and a data interface bus; (2) data are acquired, wherein original data stored in a terminal data computer are acquired by the main cloud central server and the auxiliary cloud central server through data acquisition interfaces C of the servers; (3) data are stored, wherein the acquired mass data are stored in an HDFS; (4) the data are processed, wherein concurrent processing is conducted on the data through a MapReduce model, the main cloud central server only uniformly dispatches storage and processing of the data, and the specific data storage and processing steps are conducted through the first load server, the second load server, the third load server, the fourth load server, the fifth load server, the sixth load server, the seventh load server, the eighth load server, the ninth load server, the tenth load server and the eleventh load server. The method has the advantages that the processing speed is high and safety and reliability are achieved.

The invention discloses a fusion system for lane lines based on multiple sensor data. The fusion system comprises a data receiving module, a data converting module, an own vehicle lane judging module,a target vehicle lane judging module and a target vehicle lane dividing module; the data receiving module is used for receiving target vehicle data and lane line data; the data converting module is used for converting the target vehicle data and the lane line data into the data under a preset coordinate system so as to form a lane line equation under the preset coordinate system and sequencing the lane lines formed by the lane line equation under the preset coordinate system; the own vehicle lane judging module is used for judging the lane on which the own vehicle locates according to the sequence data of the lane lines; the target vehicle lane judging module is used for substituting a target vehicle coordinate under the preset coordinate system into the lane line equation and judging which two lane lines the target vehicle locates between according to the target vehicle coordinate; and the target vehicle lane dividing module is used for judging the number of the lane on which the target vehicle locates according to the number of the lane on which the own vehicle locates. The invention also discloses a fusion method for lane lines. According to the invention, lane line dividing can be realized; diversity of surrounding information acquired by intelligent driving can be increased; and higher accuracy of lane line detection can be provided.

The invention relates to a method and device for dynamically measuring the hole deviation and the toolface angle in conditions of small hole deviation, and solves the problems of the prior art that the real-time situations of the hole deviation and the toolface can not be accurately and really reflected in the underground vibration and impact interference environment. The device mainly comprises a signal collection unit for collecting an acceleration signal and converting the acceleration signal into a digital signal for subsequent processing through sampling and filtering; and a digital signal processing unit for processing the signal which is input from the signal collection unit and outputting the parameters of the hole deviation and the toolface according to the acceleration signal.

Embodiments of the invention provide a data standardization method and apparatus, computer equipment and a storage medium. The data standardization method includes the steps: acquiring data of one item of to be standardized in a medical examination report; determining a data type corresponding to the specific value of the data of the item; and performing standardization processing on the data of the item according to the determined data type, wherein different data types correspond to different standardization processing modes. The data standardization method adopts different standardized processing methods for different data types of data, can comprehensively perform standardization processing on the medical examination report, can improve precision and accuracy of the standardization processing of the medical examination report, and further enables the data after standardization processing to be further applied to model learning, and improves the consistency and accuracy of the dataof model learning.

The invention discloses an oxygen potential control system of a spheroidizing annealing furnace. The oxygen potential control system comprises an oxygen probe unit, a pressure sensor unit, an electric actuator, a data acquiring unit and a data processing unit, wherein the oxygen probe unit, the pressure sensor unit and the electric actuator are connected with the spheroidizing annealing furnace; the data acquiring unit is connected with the oxygen probe unit and the pressure sensor unit; the data processing unit is connected with output of the data acquiring unit; and through detecting and controlling the oxygen potential of the spheroidizing annealing furnace in real time, the oxygen potential in the spheroidizing annealing furnace can be detected and regulated in real time, and a situation, such as surface oxidation, decarburization or recarburization after a metal material is annealed is avoided.

The invention relates to a device for testing the magnetic property of a high-permeability magnetic material in a low-frequency weak magnetic environment. The device comprises a coil group, a magnetic property tester, a magnetic shielding cylinder and an upper computer, the coil assembly is wound on a tested sample piece and is connected with the magnetic property tester, and the tested sample piece wound with the coil assembly is placed in the magnetic shielding cylinder; the magnetic shielding cylinder provides a weak magnetic environment for a tested sample piece; the magnetic property tester is used for providing exciting current for the coil assembly, detecting the induced voltage of the tested sample piece and transmitting the measured data to the upper computer; and the upper computer is used for acquiring data tested by the magnetic characteristic tester and drawing a magnetization curve. According to the technical scheme, the problem that an existing magnetic material magnetic property testing device cannot obtain the magnetic conductivity of a high-magnetic-conductivity material in a low-frequency weak magnetic environment is solved, so that the precision and accuracy of theoretical calculation and simulation calculation of a magnetic shielding device are improved, and the error between a design value and an actual value is reduced.

The invention discloses an LED light guide plate defect detection method based on machine vision. The method comprises the steps of: collecting multi-view original images of a light guide plate in aneffective illumination structure; preprocessing the original image; positioning the light guide plate ROI area of the image after edge enhancement and segmenting the light guide plate ROI area and a background area; performing illumination influence elimination and defect part enhancement processing on the segmented light guide plate image to obtain an image after defect part enhancement; detecting the defect of the conventional size in the image with the enhanced defect part, and determining the final defect point of the conventional size; and detecting fine-size defects in the image with theenhanced defect part, and determining final fine-size defect points. The method has the advantages that an image acquisition and lighting scheme is combined with machine vision, automation of light guide plate image segmentation and defect classification detection is achieved, and detection precision and accuracy are improved through adoption of an effective image enhancement and preprocessing method while detection efficiency is not affected.

The invention discloses a weld joint identification and path extraction method for wall-climbing robot navigation. The method comprises the steps that 1, a wall-climbing robot collects a metal wall surface weld joint, marks a weld joint object in an image, and builds a data set; 2, after an identification host initializes and sets network parameters, the weld defect data set is sent to a MaskRCNNdeep learning network for training, and a weight file is generated; 3, the wall-climbing robot operates on the metal wall surface, and the industrial camera acquires a welding seam image of the current position and sends the welding seam image to the recognition host; and 4, the identification host receives the image to perform identification test, sets deep learning network parameters, loads theweight obtained by training, and sends the weight into a MaskRCNN network to perform weld joint classification, regression and pixel mask generation. 5, image processing is carried out on a Mask-RCNNrecognition result; a welding seam path image is obtained, and a welding seam path is extracted; and 6, the identification host sends weld joint path information to the control end of the wall-climbing robot, and the control end adjusts the position and speed of the robot to realize weld joint tracking operation.

The invention belongs to the technical field of black tea detection equipment, and particularly relates to a black tea fermentation process real-time detection system. According to the detection system, the matching degree between the image features of the fermented leaves and the fermentation degree and the accuracy degree of the model are improved, and the judgment efficiency and the judgment accuracy of the moderate fermentation stage of the black tea are most accurately guaranteed. The method comprises establishment of a fermentation degree digital classification label and establishment of a fermentation degree discrimination model, and comprises the following steps: 1) establishment of the fermentation degree digital classification label; 2) establishment of a fermentation degree discrimination model; and 3) performing morphological identification on the fermented leaves through the fermentation degree discrimination model so as to judge which stage the fermentation degree of the current fermented leaves is in.

The invention discloses a digital-to-analog converter and a control method thereof. The digital-to-analog converter comprises a control unit, an analog-to-digital conversion unit and at least one digital-to-analog conversion unit; the control unit is used for outputting a digital control signal to the digital-to-analog conversion unit according to a received control instruction; the digital-to-analog conversion unit for converting a digital control signal into an analog output signal to output; the analog-to-digital conversion unit is used for converting the analog output signal into a digitaldetection signal, and outputting the digital detection signal to the control unit; the control unit is further used for adjusting the digital control signal according to the digital detection signaloutput by the analog-to-digital conversion unit, thereby enabling the signal output by the digital-to-analog conversion unit to meet an expected value indicated by the control instruction. The rejection on the long drift can be realized, and the accuracy and the precision of the output are improved.

The invention discloses a method for measuring the crosslink density of an elastomer. A magnetic resonance crosslink density spectrometer is utilized to detect magnetic resonance imaging decay curves, a CPMG sequence is adopted to acquire data, and the parameters of the magnetic resonance crosslink density spectrometer are controlled to be as follows: the echo time is arranged from small to large, and arrangement of the echo time complies with one of logarithmic functions, exponential functions, Gaussian distribution, logarithmic Gaussian distribution or Fibonacci sequence distribution. The method for measuring the crosslink density of the elastomer can improve the precision and the accuracy in magnetic resonance detection of the crosslink density of the elastomer.

The invention provides a method for improving the precision and the accuracy of a capacitive touch key. The method comprises the following specific steps of: step 1), a resistor is added to the transmission path of the touch key, thus the various noise signals are attenuated to amplitudes which cannot be identified after passing through the resistor, so as to improve the identification precision, a resistor with a resistance of 1k to 3k is added to the transmission path at the rear end of the touch key, and because the amplitude of an interference signal is low, the amplitude is greatly attenuated and cannot be wrongly identified after passing through the resistor; step 2) in order to solve the problem of nonadjustable sensitivity or too high sensitivity of the touch key, a capacitor is connected aside the key in parallel, after the capacitor is added, a total capacitance of the touch area of the touch key to the ground is equivalently increased, when the key is touched by a finger, the generated capacitance change is certain, and if the total capacitance is increased, that is, a change ratio is reduced, the problem of too high sensitivity of key touch identification is solved, and when a certain distance exists between the finger and the key or the finger incidentally slides through the surface of the key, the operations of the key cannot be triggered.

The invention discloses a text abstract generation method and system for the finance field, and the method comprises the steps: collecting a text data set in the finance field, and constructing a field corpus of the finance filed; preprocessing the text data set according to the field corpus to generate a training sample, a test sample and a verification sample; creating a text abstract generation model, wherein the text abstract generation model is composed of a coding structure and a decoding structure, the coding structure comprises a bidirectional LSTM network and adopts one-hot coding, the decoding structure comprises a unidirectional LSTM network and is additionally provided with a zoom translation layer, and the zoom translation layer adopts a pointer generation network and is further provided with a coverage vector; training the text abstract generation model by using the training sample, the test sample and the verification sample to obtain a target text abstract generation model; and inputting a to-be-detected text into the target text abstract generation model, and outputting a text abstract. Precision and accuracy of the current automatic text abstract can be improved, and the text abstract generation efficiency is improved.