48results about How to "Solving recognition difficulties" patented technology

The invention discloses a pseudorandom bar code applied to a digital level bar code ruler, belonging to the technical field of manufacture of instruments of surveying and mapping. The pseudorandom bar code is formed by alternatively arranging a plurality of light and shade stripes with different widths, wherein the widths and light and shade changes of the strips are modulated by a plurality of pseudorandom sequences. The strips in the bar code comprise coarse code strips for remote measurement and thin code strips for close measurement. The thin code strips are formed by alternatively arranging light and shade strips with smaller widths, and the width changes of the light strips and shade strips are respectively modulated by a pseudorandom sequence. The coarse code strips are formed by arranging light and shade strips with bigger widths, and the widths and light and shade changes of the strips are modulated by a pseudorandom sequence. The coarse code strips are regularly embedded in the fine code strips to form a complete bar code. The pseudorandom bar code is characterized in that a plurality of pseudorandom sequences are alternatively coded for modulating the widths and light and shade changes of the strips. The pseudorandom bar code has the advantages of compatibility with remote measurement and close measurement, capability of forward measurement and backward measurement, simplicity in making a staff gauge, low requirement for a hardware system, small decoding operand, and rapidness and accuracy in identification.

The invention relates to a multi-resource remote sensing distinguishing method and system of an underground potassium- and halide-containing water resource in a salt lake area. The method comprises the following steps: acquiring data, namely, acquiring multi-resource remote sensing image data; processing data, namely, processing multi-spectral remote sensing image data and high-resolution remote sensing image data by the multi-color transformation technology, building a remote sensing stratum lithology interpretation mark and performing remote sensing stratum lithology interpretation; extracting salt-containing abnormality information from the multi-spectral remote sensing image data by the main component analyzing technology; extracting the salt information by the interpretation extracting method; processing a radar remote sensing image data, building a geologic structure interpretation mark for potassium- and halide-containing water resource and performing geologic structure interpretation; integrating data, namely, processing three information obtained in data processing by overlapping, projection transformation and matching. According to the method, the multi-resource remote sensing image data are processed, interpreted and integrated to realize accurate distinguishing of the potassium- and halide-containing water resource.

The invention discloses a stem content in leaf infrared imaging detection method based on active heating, which is characterized in that tobacco leaves are uniformly spread and thinned through a vibration disc and an acceleration belt, the uniformly spread and thinned tobacco leaves are heated by using a heating device and sent to a refrigeration apparatus for cooling, tobacco leaves are scanned to the images of leaf blades and leaf stems by an infrared heat imaging instrument, the infrared heat radiation of the target tobacco leaves can be detected through the infrared heat imaging instrument to obtain the distribution images of leaf blades and leaf stems of the target tobacco leaves can be obtained; the images segmentation of the distribution images of leaf blades and leaf stems of the target tobacco leaves can be carried out, and the segmented leaf stems images can be analyzied, the leaf stem information which accords with practical production requirement can be obtained, and the stem content rate in leaves can be calculated. According to the invention, the clear stem images of the tobacco leaves can be obtained, and the shape and size of the leaf blades and leaf stems can be accurately and rapidly identified.

The invention discloses a mountain crack detection method, comprising: analyzing collected images through an image sliding window, and determining the contrast ratio of each pixel point in the images; in dependence on contrast ratios, constructing and analyzing a contrast graph, determining the optimal segmentation threshold of the contrast graph, determining an area to be enhanced in dependence on the threshold, and performing image reinforcement processing on the area to be enhanced; analyzing processed images, and determining a gray scale characteristic graph and a variance characteristic graph; calculating optimal segmentation thresholds of the gray scale characteristic graph and the variance characteristic graph, and the binary images of contrast characteristics and variance characteristics, and performing AND operation on the two binary images to determine the binary image of racks; transforming the binary of racks to determine a refined frame, and performing area extension processing on the frame to obtain all crack latent areas on the frame; and traversing frame pixels to obtain crack length information, and calculating a crack area in dependence on the number of area pixels.

The invention discloses a CRNN-based picture table extraction method, and the method comprises the following steps: S1, carrying out the perspective transformation of a to-be-detected picture, and correcting the picture; S2, performing table skeleton extraction on the corrected picture by using a deep neural network; S3, obtaining a cell ROI (Region of Interest) from the table skeleton; S4, recognizing text contents in all the cell ROIs through an OCR recognition model; and S5, restoring the text content to a table through the table skeleton typesetting in the step S2, so that the picture table is converted into a data table, and the extraction of the picture table is completed. According to the method, perspective transformation is carried out on the to-be-recognized picture once, the picture angle is corrected, and then the deep neural network model is used for extracting the overall table skeleton, so that the situation that the edges of cells are connected by handwritten characters in the prior art or the problems that a table in the picture is not clear, the picture is light, inclined and fuzzy are solved, and the problem that time and labor are wasted because a large amount of manual parameter adjustment is needed is solved.

The invention relates to a beam line splicing method, device and equipment and a storage medium, and relates to the technical field of drawing correction. The beam line splicing method comprises the following steps: analyzing first position information of each beam line from an architectural drawing; according to the first position information, any two beam lines with the same end point are obtained, any two beam lines with the same end point serve as a combination, and the end points serve as angular points; obtaining a target angular point needing to be spliced with the angular point; connecting the angular points with the target angular points to generate spliced line segments; and generating a spliced beam line according to the beam line and the splicing line segment. The method is used for solving the problem that beam members are difficult to recognize as beam lines are not drawn according to actual conditions.

The invention relates to a finger vein recognition method. The method comprises the following steps: S1, carrying out ROI positioning on an image acquired by a high-definition camera; S2, performing gray normalization processing on different parts of the image; S3, performing mean filtering and Gaussian filtering on the image to eliminate noise; S4, judging the central positions of the lines by using curvature detection; S5, enhancing the contrast of the grey-scale map through gamma transformation; S6, judging edge pixel points by using a refinement algorithm based on an index table; and S7, carrying out fusion, expansion and smoothing treatment on the extracted finger vein skeleton. Finger vein recognition is performed through template matching, Hu invariant moment and improved Zernike moment. Particularly, under the condition of unclear image acquisition, the method has relatively high anti-interference capability and relatively high-precision identification capability.

The invention relates to a beam line processing method and device, equipment and a storage medium, and relates to the technical field of drawing correction. The beam line processing method comprises the following steps: analyzing first position information of each beam line from an architectural drawing; second position information of the contour line of each wall column is analyzed from the architectural drawing; according to the first position information and the second position information, non-standard beam lines are recognized, and the non-standard beam lines comprise at least one of the beam lines having intersection points with the contour line of the wall column and the beam lines having no intersection points with the contour line of the wall column, and the distance between the beam lines and the contour line is smaller than a preset distance value; the non-standard beam line is cut off, the cut-off part is removed, the cut-off beam line is generated, and the cut-off beam line is the standard beam line. The method is used for solving the problem that beam members are difficult to recognize due to non-standard beam lines.

According to the cold-rolled coil section bar code and defect detection device in the iron and steel industry, end detection intelligent robots are arranged on the two sides of operation and transmission near a saddle at a unit inlet, and the defect detection and size measurement functions of the end faces of the two sides of a steel coil can be completed on the two sides of the saddle at the same time. When end face defect detection is carried out, the tail end of the robot carries the three-dimensional sensor to carry out circular motion in the radius direction of the steel coil, and the three-dimensional contour surface type of the end face of the steel coil is obtained through the laser three-dimensional sensor. The sensor can output a scanning gray level image while scanning, and defects which cannot be determined by three-dimensional scanning are judged. The bar code and character recognition adopts a 10 million pixel industrial camera for imaging.

The invention discloses a leukemia cell microscopic image classification method and system based on DenseNet, and the method comprises the steps: adding a dyeing deconvolution layer and a discrete cosine transform module on the basis of a DenseNet network, so as to obtain the feature information of a cell microscopic image in a spatial domain and a frequency threshold of coloring agent absorptionamount; and acquiring a feature vector by fusing the two kinds of information and using the vector for judging whether the blood cells belong to cancer cells or not. The coloring agent absorption amount is used for representing the characteristics of the cell microscopic image, the problem that recognition is difficult in an RGB color space is solved, the average accuracy rate and the average weighted F1 score reach 89.81% and 89.64% respectively, automatic classification of leukemia cells is achieved, and therefore a pathologist is assisted in diagnosing leukemia, and the missed diagnosis rate and the misdiagnosis rate are reduced.

The invention discloses an automatic identification method and device for a whole brain range single neuron axis sudden synaptic junction. The method comprises the steps of carrying out the splittingof a neuron skeleton tree, obtaining a plurality of sub-segment fibers, and extracting the neighborhood image signals of the sub-segment fibers; carrying out the adaptive foreground segmentation on the neighborhood image signals of the sub-segment fibers, positioning the axonal expansion structures of the sub-segment fibers, and fusing the axonal expansion structures on all the sub-segment fibers;and classifying the axonal expansion structures by using a deep convolutional network, and removing the axonal expansion structures of the non-axonal sudden contact junctions in the axonal expansionstructures to obtain the final axonal sudden contact junctions. The device comprises an acquisition module and a classification module. According to the invention, the identification difficulties caused by the diverse shapes of the axonal synaptic junctions, adhesion of the axonal synaptic junctions, interference of false axonal synaptic junctions and the like can be overcome, all axonal synapticjunctions of a single nerve are quickly identified from a whole brain neuron morphological data set, and the accuracy of an identification result is kept while a whole brain oversized data set is quickly processed.

The invention discloses a football match offside judgment method and device based on an unmanned aerial vehicle, and electronic equipment, and the method comprises the steps: obtaining a real-time video of a football match, the real-time video being obtained by a visual sensor of the unmanned aerial vehicle; applying an identification algorithm to the current frame image of the real-time video to obtain a football detection area and a player detection area; performing feature extraction on the player detection area through cosine similarity, nonlinear normalization transformation and HSV transformation, and inputting the extracted features into a classification model to obtain category information of a team to which players belong; according to the football detection area and football player detection, judging whether passing occurs by judging whether an overlapping area exists between the football player detection area and the football detection area and comparing the distance between the center points of the football detection areas of the front frame and the rear frame, and if passing occurs, obtaining category information of a team to which the football player of the passing party belongs; according to the player detection area of the passing party and the information of the team, performing straight line detection, obtaining the position of an offside line, and performing offside judgment.

The invention discloses a DRM signal identification method under a non-ideal channel based on autocorrelation, and relates to a DRM signal identification method under a non-ideal channel. The objective of the invention is to solve the problem of low recognition accuracy of DRM signals transmitted through an ionized layer channel in the prior art. The method comprises the following steps: 1, obtaining a time domain signal; 2, performing normalization processing on the time domain signal; 3, calculating an autocorrelation function of the normalized signal; 4, selecting a threshold value, comparing the threshold value with the autocorrelation function, reserving values greater than or equal to the threshold value in the autocorrelation function, and setting values less than the threshold value to zero; 5, obtaining a coefficient cdt1; 6, obtaining a coefficient cdt2; 7, selecting a threshold value Thd1 and a threshold value Thd2, and if the cdt1 and the cdt2 are respectively greater than the Thd1 and the Thd2, judging that the selected time domain signal is a DRM signal; otherwise, judging that the selected time domain signal is not the DRM signal. The method is used in the field of signal classification and identification.

The invention discloses a 3D electromagnetic scanning system which can solve a problem of intuitive and visual representation of a three-dimensional spatial position relationship between a detection point and a measured object and spatial electromagnetic field distribution. The system comprises a control module, a 3D scanning module, a mobile scanning module, an electromagnetic probe and an electromagnetic signal receiving module. The control module acquires 3D contour information of a measured area and spatial position information of mark points from the 3D scanning module; the control moduleacquires position information of the electromagnetic probe on the mobile scanning module from the mobile scanning module; and the control module acquires scanning point electromagnetic signal frequency spectrum information received by the electromagnetic probe from an electromagnetic signal receiving module. And the control module converts the spatial position information of the mark points and the position information of the electromagnetic probe on the mobile scanning module to obtain a scanning point spatial position set of the electromagnetic probe, and finally fits the 3D contour information of the measured area, the scanning point spatial position set and the scanning point electromagnetic signal frequency spectrum information to output a 3D spatial electromagnetic distribution map.

The invention belongs to the field of bill text recognition, and relates to a financial bill seal erasing method based on an attention mechanism and a generative adversarial network, which is implemented by a processor and comprises the following steps: receiving an original picture of a financial bill; determining a first feature map according to the original picture by using a feature extraction module in a convolutional neural network; using the convolutional neural network to respectively extract a background color map of the original picture and an attention heat map reflecting position distribution of a stamp on the original picture according to the first feature map; using the convolutional neural network to generate a seal-erased picture of the original picture in a generative adversarial mode according to a second feature map spliced by the original picture, the background color map and the attention heat map in the channel direction; training the convolutional neural network in a generative adversarial mode. The problem that the financial bill containing the seal is difficult to identify is solved, and the purpose that the seal is erased without losing original character information is achieved.

The invention provides a micron-scale ID image recognition method and system, and the method comprises the steps: observing the overall condition of a to-be-detected object, so as to determine the number and position relation S1 of the to-be-detected object, and enabling the number of the to-be-detected object to be one or more; And adjusting a view field, sequentially identifying the ID images ofthe to-be-detected object, and sequentially recording corresponding ID information S2. According to the technical scheme provided by the invention, the work of manually identifying and recording thechip ID with naked eyes by using a microscope can be replaced, automatic identification and recording are realized, the technical requirement on an operator is reduced, the accuracy of identificationand recording is greatly improved, and the working efficiency is improved; Through the secondary imaging positioning combined by the large-view-field camera and the high-power microscope, the problemthat automatic positioning cannot be achieved due to the fact that chip packaging boxes are different and ID positions are different is solved.

The invention relates to a human body posture estimation method based on a joint relation, and the method comprises the following steps: S1, constructing a joint relation module which comprises two sub-modules: a channel-based feature relation module and an adjacent joint space relation module; s2, constructing a human body posture estimation model based on a joint relation; s3, training a human body posture estimation model based on a joint relation by utilizing the marked human body posture data; and S4, performing a human body posture estimation task based on a single image by using the trained human body posture estimation model added with the joint relation module to obtain a predicted human body posture. Compared with the prior art, the invention effectively overcomes the problem that the positions of four-limb joints with high degree of freedom, such as wrist joints, ankle joints and shielded invisible joints, are difficult to detect in the image, and is high in human body posture estimation accuracy.

The embodiment of the invention discloses an early warning method and device for network fraud, computer equipment and a storage medium. The method comprises the following steps: collecting network flow data in real time, and extracting a target access website included in the network flow data; inputting the target access website into a pre-trained fraud website early warning model, and collecting an recognition result of whether the target access website belongs to a fraud website; and if the recognition result is that the target access website belongs to the fraud website, recognizing an access party type of the target access website as a fraud party or a deceived party, and performing fraud early warning by adopting a matched early warning mode according to the access party type. According to the method provided by the embodiment of the invention, the problem of difficulty in recognizing fraud persons is solved, and the case data is processed in real time. The occurrence of network fraud cases is reduced, the property loss of people is reduced, and early warning can be carried out on the network fraud cases.

The invention discloses a method for identifying Odonata insects based on a region suggestion network, comprising the steps of: S1. Cleaning and arranging images of Odonata insects to obtain a data set of Odonata insect images; S2. Data of Odonata insect images The data set is enhanced to obtain an enhanced data set; S3. divide the enhanced data set to obtain a training set, a verification set and a test set of images of Odonata insects; S4. Construct a deep convolution based on the region proposal network Product network model; S5. use the training set and verification set to train the deep convolution network model to obtain the trained network model; S6. input the test set to the trained network model, and output the test to obtain the The classification results of the set. The invention can make the recognition process simple and fast, save a lot of labor costs, and solve the problem of difficult recognition caused by complicated backgrounds of pictures of Odonata insects in the natural environment.

The invention discloses a method for detecting cracks in a mountain body, which includes analyzing the collected image through an image sliding window, and determining the contrast of each pixel of the image; constructing a contrast map according to the contrast, analyzing the contrast map, and determining the contrast map According to the threshold, determine the area to be enhanced, and perform image enhancement processing on the area to be enhanced; analyze the processed image, determine the grayscale feature map and variance feature map; calculate the grayscale feature The optimal segmentation threshold of the graph and the variance feature map, and obtain the binary image of the grayscale feature and the variance feature, and perform the sum operation on the two binary images to determine the binary image of the crack; the binary image of the target crack The graph is transformed to determine its refined skeleton, and the skeleton is extended to obtain all the potential areas of cracks on the skeleton; the pixels of the skeleton are traversed to obtain the length information of the crack, and the area of ​​the crack is calculated according to the number of pixels in the region .

The invention relates to a text relation extraction method integrating multi-level information extraction and noise reduction, and belongs to the technical field of computer natural language processing. Including: 1. Use BERT as the encoder to vectorize the text information, and extract the hidden layer vector information of mentions, entities, sentences and texts; 2. Propose a method to integrate multi-level information to solve multiple instances For multi-label questions, multi-level information includes: mention-level information. Entity-level information, sentence-level information, and article-level information; 3. It is proposed to use the mentioned location information to roughly extract the proof sentences, and then use the noise reduction method to capture the important relational features of the proof sentences. The method can consider and solve the problems of multi-instance and multi-label in the text and the difficulty of identifying the proof sentence when extracting the relationship of the entity pair in the text. Experiments show that the method has a significant improvement in the evaluation index of F1.

The invention relates to a method for extracting chapter relation by fusing multi-level information extraction and noise reduction, and belongs to the technical field of computer natural language processing. The method comprises the steps of 1, using BERT as an encoder, performing vectorization representation on chapter information, and extracting mentions, entities, sentences and hidden layer vector information of chapters in the chapter information; 2, providing a method for fusing multi-level information to solve the problem of multiple instances and multiple labels, wherein the multi-level information comprises information referring to levels, information of an entity level, information of a sentence level and information of a chapter level; and 3, proposing to roughly extract the proof sentences by using the mentioned position information, and then capturing the important relation characteristics of the proof sentences by using a noise reduction method. According to the method, when relation extraction is carried out on entity pairs in a chapter, the problems of multiple instances and multiple labels in the chapter, difficulty in identification of proof sentences and the like can be considered and solved. Experiments show that the method is significantly improved in the evaluation index of F1.

The present invention relates to a human body posture estimation method based on joint relationship, comprising the following steps: S1: building a joint relationship module, including two sub-modules, a channel-based feature relationship module and an adjacent joint space relationship module; S2: building a human body based on joint relationship Pose estimation model; S3: use the labeled human pose data to train the human pose estimation model based on the joint relationship; S4: use the trained human pose estimation model with the joint relationship module to perform the human pose estimation task based on a single image, and obtain the prediction human body posture. Compared with the prior art, the present invention effectively overcomes the problem that it is difficult to detect the position of limb joints with high degrees of freedom such as wrist joints, ankle joints and hidden invisible joints in the image, and the accuracy of human body posture estimation is high.

The invention discloses a method for navigation, positioning and task planning of a robot for archives, which includes the following steps: (1) preprocessing: defining the coordinate axes and initial parameters of the archive shelf area; (2) starting from the initial position, the robot passes through the archives (3) judging whether the robot has passed the file shelf with the largest number in the task; (4) judging whether it is an executable task according to the task matrix to be executed, (5) decomposing the task to form a new task matrix; ( 6) Establish the objective function and optimize the new task matrix; (7) Execute the task. The invention can solve the problem that the robot is difficult to identify when the file rack is opened, so that the robot can monitor changes in the external environment in real time, effectively and quickly adapt to the changed external environment, is applicable to different maps, has wide applicability, and can realize Efficient human-computer interaction between clients and robots.

The invention provides a grating detection method for low reflection energy transmission fiber grating. The method is suitable for a fiber grating detection and measurement system. The system consistsof a broadband light source, a first high-speed optical switch, a first attenuator, a circulator, an optical cable, a fiber grating, a second attenuator, a second high-speed optical switch, a tracking filter, a demodulator, and a logic controller. By using two high-speed optical switches to perform high-precision time domain control on the emitted and reflected light, using the tracking filter tofilter the reflected light of the fiber grating in a specific band, and using the background noise suppression technology, stable identification of the wavelength of the low reflection energy transmission fiber grating and accurate distance measurement can be realized. The technology has the characteristics of simple structure, low cost and high detection accuracy.