43results about How to "High detection and recognition rate" patented technology

The invention belongs to the technical field of navigation and positioning and discloses a low-orbit navigation constellation enhanced network RTK (real-time kinematic) technology. The low-orbit navigation constellation enhanced network RTK technology comprises the following steps: design of a global navigation and positioning constellation and multiple low-orbit constellation schemes, a low-orbitsatellite collecting orbit determination and inter-satellite link autonomous orbit determination method, design of a low-orbit satellite broadcast ephemeris and a high, medium and low-orbit satellitefusion processing unified mathematical model. The problem in the prior art that the network RTK method is low in positioning precision is solved, the baseline distance can be increased, the network density can be reduced, the cost can be reduced, the precision and the ambiguity fixing rate are increased, and the convergence time can be shortened. The sub-satellite point track coverage range of the low-orbit constellation is wide and even covers polar regions, the three-dimensional orbit errors of 60 and 192-star constellation are 46 cm and 18 cm correspondingly, the forecast precision is increased by about 30 percent compared with that of a klobuchar model, and the final positioning precision is at horizontal millimeter and height centimeter level.

The invention provides a drift fault recognition method of hot-rolling strip steel based on sound signals. The method comprises the following steps of: preprocessing a signal by combining the features of a hot-rolling production line and adopting a resampling and framing technology, carrying out feature extraction on the signal by adopting a Mel frequency cepstrum technology and carrying out feature selection and recognition by adopting a principal component analysis method and a multivariate statistic process T2 control diagram so as to realize the diagnosis of drift faults of the hot-rolling strip steel. The invention has the advantages that the drift phenomenon can be judged accurately and rapidly on line by utilizing an acoustic detection method, the roller replacement time can be reasonably arranged, the roller consumption can be effectively reduced, the production cost can be controlled, and the quality defect prevention capacity and the production operability of the product can be improved.

The invention discloses a lane line confirmation method in a lane line detection system. The method comprises the following steps that: 1, a camera collects an image; 2, edge enhancement is carried out on the image collected by the camera; 3, binary processing is carried out on the image after the edge enhancement; 4, candidate edge points at the inner side of a lane line are screened out according to characteristic rules of the lane line; 5, a fitting technology is adopted for carrying out fitting on the inner edge of the lane line, and obtaining a line or curve model of a lane; and 6, confirming the lane line. The image collected by the camera in the step 1 is a gray level image or a color image, if the image is the color image, the color image is converted into the gray level image through a color image to gray level image conversion formula.

The invention relates to a wafer defect detection system and method. The system comprises a processing unit, a control assembly, a displacement assembly, a detection assembly, and a signal acquisitionunit, the displacement assembly and the detection assembly are cooperatively arranged, and the displacement assembly controls the detection signal acquisition and movement of the detection assembly through displacement. The control assembly controls movement of the displacement assembly through an instruction and indirectly controls detection signal acquisition of the detection assembly, a signalacquired by the detection assembly and instruction information of the control assembly are transmitted to the signal acquisition unit, and the signal acquisition unit transmits the acquired signal tothe processing unit for processing to obtain defect information; the sensitivity requirement of an area array detector is reduced, and the detector cost can be effectively reduced; compared with a traditional scanning mode, the detection speed of the whole system can be greatly increased; and as the area array detector does not move in the area area exposure process, the problem of dithering andblurring of a traditional scanning mode can be solved, and the defect detection and recognition rate is increased.

The invention discloses a method for recognizing natural images and computer generated images based on multi-wavelet transform, comprising the following steps of: (1) transforming training sample images and test sample images to an HSV color space, respectively obtaining corresponding hue component images, saturation images and luminance images; (2) pre-processing each component image and then performing first-order multi-wavelet transform to each component image; (3) taking 16 sub-bands of the obtained hue component images, saturation images and luminance images as objects, and calculating the mean value, the variance, the skewness and the kurtosis of each sub-band wavelet coefficient, thereby obtaining 192 characteristic values; and (4) calibrating and normalizing the characteristic values, and then substituting to an SVM classifier for training and testing, thereby obtaining classes of the images. The method is high in detection recognition rate and low in calculation complexity.

Provided is an image processing apparatus for detecting pedestrians. The image processing apparatus includes a lane detecting module configured to extract a lane coordinate value from an input image and a pedestrian detecting module configured to set, as a pedestrian region of interest (ROI), a region between a first line passing through ends of first left and right lanes and a second line passing through ends of second left and right lanes which are respectively disposed above the left and right lanes, and search for the pedestrian ROI by using a predetermined window to detect a pedestrian region having a pedestrian feature.

The invention discloses a method for detecting and identifying road surface disaster targets under the interference of reinforcement high-reflection echo. The method mainly comprises target detection and identification algorithms and a reinforcement echo suppression algorithm. The method for detecting and identifying the road surface disaster targets under the interference of the reinforcement high-reflection echo comprises the following steps of: firstly, under the background of interference of the reinforcement high-reflection echo, pre-processing time-space two-dimensional echo data which are received by a ground penetrating radar antenna, detecting targets, determining the transverse position of each target, extracting one-dimensional echo data of each target, and detecting the targets to determine the transverse position of reinforcement; and finally, by using a wave-field separation theory, separating reinforcement echo from target echo, and performing further suppression, re-detection and re-identification to fulfill the aim of detecting and identifying disasters. The method is free from influence of noise and the reinforcement echo, high in detection and identification rate, and the like.

The invention provides a detection and recognition device for sprayed codes on the cylindrical surfaces of copper rings. The detection and recognition device comprises an industrial display screen, a left conveying belt mechanism, a servo sliding table mechanism, a right conveying belt mechanism, an industrial computer, an article grasping clamping air cylinder, an image collecting mechanism, light source equipment, a rejected product receiving box, a qualified product receiving box, a removing mechanism and a rotating mechanism. According to the detection and recognition device for sprayed codes on the cylindrical surfaces of copper rings, the machine vision technology is adopted to detect and recognize sprayed code character information on the cylindrical surfaces of copper rings, and the defects of low precision, high subjectivity, low judgment efficiency, high missed detection rate and false detection rate and the like of traditional manual detection are overcome; the automation degree is high, and automatic recognition and automatic sorting of sprayed codes on the cylindrical surfaces of copper rings can be completed. The invention further provides a detection and recognition method for sprayed codes on the cylindrical surfaces of copper rings. By means of the detection and recognition method, irregular sprayed code information on the cylindrical surfaces of copper rings on a production line can be continuously and automatically collected and distinguished, and the detection and recognition method is especially suitable for handling recognition of fuzzy and irregularly-shaped characters on the cylindrical surfaces and is high in detection and recognition rate.

The invention discloses a crowd abnormal event detection method based on a three-dimensional pyramid image generation network, and relates to a method for identifying graphs. The method includes: generating an image by using a three-dimensional pyramid image generation network; and detecting the crowd abnormal event by comparing the difference between the generated image generated by the three-dimensional pyramid image generation network and the to-be-detected image. The defects that in the prior art, motion information between image sequences, local characteristics of different scales and reconstruction errors between normal images and abnormal images are ignored and crowd abnormal behaviors are difficult to judge in an abnormal detection method based on generated images are overcome.

The invention discloses a traffic sign board recognition method and device and computer equipment. The recognition method comprises the steps of acquiring a gray level image of a to-be-recognized image, generating an image pyramid according to the gray level image, the image pyramid comprising a plurality of images, and the sizes of the images are different; for the image of each size, traversingthe image of each size according to the target detection window and the target sliding step length, and obtaining an N-dimensional gradient direction feature vector of the image of each size; and classifying the N-dimensional gradient direction feature vector of the image of each size through a pre-trained signboard detection model to detect whether a traffic signboard exists in the original imageor not. According to the method, the detection and recognition speed of the traffic sign board can be increased, and the detection and recognition rate is increased.

The invention provides an intelligent oncomelania detection, recognition and statistics system. The system comprises a CT image acquisition module, an image automatic detection module and a data analysis and processing module. The CT image acquisition module is used for acquiring a to-be-detected high-definition soil sample three-dimensional image containing oncomelania, namely an oncomelania CT image, and sending the oncomelania CT image to the image automatic detection module; the image automatic detection module is used for carrying out image preprocessing on the collected CT image of the oncomelania and then detecting the oncomelania by utilizing an automatic screw recognition and detection algorithm; and the data analysis and processing module is connected with the image automatic detection module and is used for counting the number and types of the detected oncomelania according to the detection result of the image automatic detection module and giving out the distribution condition of the oncomelania. According to the invention, the recognition and detection rate of oncomelania in the collected image can be improved, and the whole set of system is based on portable equipment, so that the general personnel who do not receive blood defense education can complete image collection and oncomelania detection and statistical operation.

The invention provides a license plate detection method based on a blocked license plate area regression technology. The license plate detection method based on a blocked license plate area regression technology includes the steps: a) performing vehicle detection, obtaining a target vehicle, and determining a whole license plate detection area by means of the target vehicle; b) dividing the license plate detection area into n small blocks, wherein each small block is partially overlapped; c) using a first deep neural network learning model to perform fitting to obtain a general license plate area, and obtaining the credibility that the area is a real license plate at the same time; and d) according to the relationship between the position of the license plate area and the credibility, obtaining the final license plate area through fusion. The license plate detection method based on a blocked license plate area regression technology utilizes the blocked license plate area regression technology to reduce the time of detecting a license plate. The license plate detection method based on a blocked license plate area regression technology can improve the accuracy for detecting a license plate by introducing the deep learning model.

The invention relates to a video stream based face tracking identification method and belongs to the technical field of mode identification. A video sample is obtained, and continuous frame images ina video are processed; a face detection module locates a face position in the preprocessed frame image and takes the face position as an initial tracking window of face tracking; and an Adaboost algorithm is combined with a camshift algorithm in a tracking algorithm, a morphological method, a target state prediction method and an interframe difference method are used, and the problem that the traditional camshift algorithm is not suitable for complex conditions including high speed target movement, complex background and shielding due to the initial tracking window needs to be selected manneris solved. An improved face identification algorithm in which a weighted average face is combined with a 2D major constituent analysis characteristic space is used to reduce difference of face characteristic within the type, and improve the accuracy of face identification.

The invention discloses a hidden network camera identification method based on network communication data. The hidden network camera identification method comprises the following steps: accessing real-time data of a monitoring network; analyzing, restoring and identifying a network communication session; counting active network equipment objects; filtering and cleaning the network equipment objects; identifying and confirming by the network camera equipments; and performing recording and displaying. According to the invention, a bypass deployment passive identification mode is adopted, no burden is added to the network environment, and the real-time performance is high; for the remote control behavior of a common electronic equipment integrated network camera, the detection and identification rate is high, accurate and fast; matching identification is carried out on the hidden camera adopting the non-standard communication protocol; a network behavior matching mode of a network objectis adopted to carry out network camera identification and alarm.

The invention provides a multi-view camera exposure method and device, equipment and a medium. The multi-view camera exposure method comprises the following steps: acquiring at least two frames of images acquired by the same target object under the same or different illumination conditions by using a multi-view camera; and when it is detected that the face of the target object cannot be obtained in any image, re-determining a face area in the any image according to face positions obtained by other images, and re-exposing the face area to obtain a face image. According to the invention, throughmutual reference of human face positions, all cameras can be ensured to be normally exposed to obtain human face images; meanwhile, the exposure effect of the camera under strong light, backlight anddark environment is greatly improved; and when the method is applied to in-vivo detection, the identification rate of in-vivo detection is also greatly improved.

The invention discloses a banknote and bill detection device, comprising: a glass plate, a banknote passing channel, a light source, a first camera, a second camera and a main board: the glass plate is arranged directly above the banknote passing channel; There are multiple light sources; the second camera is located on the lower surface of the glass to collect reflected images; the first camera and circuit board are located directly below the banknote-moving channel to collect light-transmitting images; the detection sensor is set at the entrance of the banknote-moving channel, It is used to detect banknotes or bills that enter the banknote-feeding channel; the light source, the first camera, the second camera, and the detection sensor are respectively connected to the circuit board. The invention has strong adaptability to currency types, low hardware cost, high detection and recognition rate, and stable performance. Reliable and other advantages.

The invention discloses a banknote number identification and management method, comprising: collecting the number image of the moving banknote through a linear array image sensor; converting an analog signal into a digital signal and inputting it into a DSP processor; The number image is preprocessed; the banknote number image after the pretreatment is analyzed, and the recognition result is stored in a designated memory. The invention also discloses a banknote number recognition and management system. By adopting the scheme of the invention, banknote numbers can be automatically collected and identified and recorded and stored during the banknote output process, and can also be connected with printing equipment to facilitate printing.

The invention belongs to the field of deep learning and image processing, and discloses a Mask R-CNN-based cotton tip identification and detection method, which comprises the following steps: collecting cotton tip images under different conditions; carrying out image enhancement on the collected cotton top end image and forming a sample set; dividing the sample set into a test set and a training set; constructing a cotton tip identification neural network model based on Mask R-CNN (Region Convolutional Neural Network); inputting the training set into an initial Mask R-CNN model for training, and learning cotton tip target features; and inputting the test set into the trained Mask R-CNN model for testing, and adjusting to model convergence to obtain a Mask R-CNN-based cotton tip identification and detection model. According to the invention, cotton tip position information can be accurately identified, and preparation is made for next mechanical cotton topping.

A gesture recognition method and device, a computer storage medium and electronic equipment belong to the field of image recognition. The method is characterized by comprising the following steps: segmenting a hand area from a video frame containing a hand image through a threshold segmentation method, and detecting a gesture action; obtaining a centroid coordinate of the hand area; when the detected gesture action is matched with a preset starting action, starting to record a mass center coordinate of the hand area until the detected gesture action is matched with the preset ending action; and taking the recorded centroid coordinate of the hand area as effective track coordinate, performing track type identification on the effective track coordinate, and providing an identification resultto preset finger-reading equipment for identification response. The problem of gesture recognition errors caused by gesture track changes caused by execution of different executors of the same gesture can be effectively solved, the detection and recognition rate of the dynamic gesture can be greatly improved, the method has universal applicability and real-time performance, and the method can beapplied to the industries of actual life, industry, control, smart home and the like.

The invention discloses a security inspection device and an image detection method thereof. The detection method comprises the steps that the security inspection equipment obtains at least one three-dimensional scanning image; the security inspection equipment projects the at least one three-dimensional scanning image to obtain a two-dimensional low-resolution image; the security inspection equipment carries out super-resolution reconstruction on the two-dimensional low-resolution image to obtain a two-dimensional high-resolution image, and the super-resolution reconstruction comprises reverseiteration projection; and the security inspection equipment displays the two-dimensional high-resolution image. In this way, under the condition that the hardware of the security inspection equipmentis not changed, the resolution of the security inspection equipment image is improved, the foreign matter detection and recognition rate of the security inspection equipment is increased, meanwhile,the false alarm rate is reduced, and the security inspection efficiency is improved.

The invention discloses a man-machine identification method and device and a computer readable storage medium. The man-machine identification method comprises the following steps of: receiving a logininstruction of an object to be logged in; in response to the login instruction, obtaining operation data of the to-be-logged-in object in a set time period before the moment when the login instruction is received; and identifying the operation data by using a pre-constructed identification model to obtain an identification result. The embodiment of the invention is based on operation data in a set time period before an object to be logged in logs in. Statistical features and nonlinear features corresponding to a series of operations executed before login are extracted by using the recognitionmodel, so that an execution body of the login operation is judged, the man-machine detection recognition rate is effectively improved, and illegal behaviors performed by using scripts without root orinformation tampering and the like can be effectively recognized.

The invention discloses a face detection and recognition method, device and system. The face detection and recognition method includes: detecting a face area from an acquired two-dimensional image; combining the two-dimensional image data of the face area with the acquired Project the depth image data into the same coordinate system to obtain three-dimensional image data; use the three-dimensional image data to perform overall concave-convex curvature detection and multiple local concave-convex curvature detections, and when the overall concave-convex curvature is within a first predetermined range , determining whether the local weighted concave-convex curvature obtained from a plurality of local concave-convex curvatures is within a second predetermined range, and if the local weighted concave-convex curvature is within the second predetermined range, then it is determined that the face area is real face. The present invention does not require the customer to do additional cooperation during the entire detection and recognition process, and can ensure a high detection and recognition rate under the premise of a high recall rate, and avoid detecting and recognizing a picture face as a real human face.

The invention discloses a target detection method for bubbles in a plate heat exchanger, and aims to solve the technical problems that a traditional detection method is complex in detection process, large in calculated amount and poor in detection effect. Two-phase flow flowing videos in a lower plate heat exchanger are shot through a high-speed camera to make a bubble data set; a Darknet learningframework is utilized, a YoLov3-tiny model is adopted, parameters before a full connection layer are frozen through transfer learning, and the model is utilized as a feature extractor; a target detection problem in the scene is taken as a regression prediction problem of the target; an improved three-frame difference method is added, so that the recognition rate is improved; and repeated candidate boxes are removed by using an IoU score screening algorithm, and then a final detection result is obtained. According to the method, the calculated amount of image features after transfer learning is used is reduced, the detection process is simplified, the detection speed is high, the network recognition precision is improved, and the detection result is more comprehensive.

The present application discloses an image processing method and apparatus for mammography target calcification detection, and a server. The method comprises the following steps of: obtaining a firstresidual image from a target image through a reconstruction network; Passing the first residual image through a T. detect loss training to get detection model; inputting an image to be recognized intothe detection model to obtain a second residual image; determining whether there is an area in the second residual image larger than a preset threshold value; and if it is judged that there is an area larger than a preset threshold value in the second residual image, the area is used as a detection result of a calcified area in the mammography target. The present application solves the technicalproblem of poor detection and identification effect. With the method of the present application, it is possible to use an image having a large reconstruction error as a calcification point for detecting breast molybdenum target calcification.