136results about How to "Guaranteed detection speed" patented technology

The invention provides a target detection method based on SSD improvement, and the method selects ResNet-101 to replace VGG-16 as a basic network of a whole model, and provides richer feature map information. A feature fusion strategy is adopted, so that the detection effect of multi-scale prediction of a network prediction layer on a small target is improved to a certain extent; an SE-block module is added to a classification branch and used for capturing global environment information of the feature map and outputting the feature map with channel weight, so that classification is more accurate; and a Centerness parallel to the classification prediction layer is added to suppress a low-quality bounding box so as to improve the detection precision. According to the method, anchor-frame-free detection is adopted, all hyper-parameters related to the anchor frame are avoided, the hyper-parameter amount is greatly reduced through the adopted prediction bounding box strategy, the network design complexity is reduced, and the training stage becomes very simple. According to the method, a loss function adopts a focalloss function, so that the model detection precision is improved while the detection speed is kept.

The present invention is computer vision detecting device and method for real-time on-line detection based on the characteristics of article. The device consists of one capsule arranging and conveying unit, one computer vision recognizing unit and one capsule eliminating unit; and features the computer vision recognizing unit with CCD camera mounted over the capsule arranging and conveying unit and the capsule eliminating unit mounted beside the conveying chain of the capsule arranging and conveying unit. The capsule arranging and conveying unit includes one vibrating feeder and one conveying chain mainly; the capsule eliminating unit eliminating soft capsules in unqualified weight and shape consists of air source, solenoid valve and nozzle connected successively; and the computer vision recognizing unit consists of video camera system, lighting box, photoelectronic sensor, control module and corresponding software.

The invention discloses a pedestrian detection method in unmanned driving based on an improved YOLOv2. Firstly, a YOLOv2 network is trained by a KITTI dataset, and a training model is obtained. Then,the video is captured by a car camera, each frame in the video is used as the input of the YOLOv2 network, and the training model is imported into the YOLOv2 network. Then the network is run and the location information and probability of pedestrians are obtained. At last, after suggestion boxes which can not contain the pedestrian target area are screened out, the final pedestrian detection boxesare obtained by using the non-maximum suppression. The method can be effectively applied to pedestrian detection in unmanned driving.

The invention relates to a method for positioning and detecting a QFP (Quad Flat Package) chip, comprising the following steps of: a collecting and pretreating an image of the QFP chip; b extracting four extreme points of coordinate axes x and y on the contour of the QFP chip to generate a extreme quadrangle; c calculating the coarse deflection angle of the QFP chip according to the extreme quadrangle on the contour of the QFP chip; and d determining four straight lines according to the coarse deflection angle and the four extreme points, fitting edge points into four straight lines by adopting a least square method after extracting the edge point of the tail end in the length direction of a pin at each side according to the set threshold value, and finally accurately calculating the deflection angle and the central coordinate of the QFP chip, thereby accomplishing the positioning and detection of the QFP chip. The invention improves the positioning and detecting precision and speed of the QFP chip and can accurately and efficiently accomplish the positioning and detection of the QFP chip on a high-speed chip mounter.

A power line foreign body detection method based on light convolutional neural network in low altitude aerial images belongs to the field of computer vision, and a real-time detection method for powerline foreign body in aerial images of unmanned aerial vehicles is studied. Firstly, a light power line detection model is constructed by using convolution neural network, and the depth characteristics of power lines in aerial images are calculated; then a multi-target power line foreign body detection model is constructed by using convolution neural network, and the prediction value of multi-scale targets is calculated by using depth features of convolution layers with different lengths and widths; finally, the power line detection model is used to filter the video frame without a power line,and the multi-target power line foreign body detection model is used to realize the real-time detection of power line foreign body in low-altitude aerial images on the video where a power line is detected.

The invention provides a Lasso (Least absolute shrinkage and selectionator operator)-based anomaly detection method and system. The method comprises the steps of establishing an anomaly detection model; determining model parameters through a Lasso algorithm; inputting to-be-detected data and obtaining a predicted value; comparing the predicted value with a preset threshold; and judging whether anomaly data exists or not. According to the method and system, the accuracy of judging a network anomaly behavior is improved on the basis of ensuring detection speed in combination with excellent characteristics of quick parameter estimation and accurate regression fitting of an Lasso; a sparse representation method is used in a data processing process, so that data dimensions are greatly reduced, model detection time is shortened, higher detection speed is achieved, and real-time online detection can be realized; and network data and host data can be both monitored, the data can be processed in batches in a matrix form, and hardware is adopted for realizing a linear regression method, so that the algorithm execution speed is greatly increased and quick, efficient and accurate anomaly detection is realized.

The invention discloses a connected domain-based natural scene text detection method, and mainly solves the problem of low accuracy of an existing text detection method. The method is implemented by the steps of 1) performing grayscale transform on an input original image to obtain a grayscale image IG; 2) extracting character candidate regions from the IG, and obtaining a character candidate region image Im; 3) filtering a few candidate regions containing no characters in the Im to obtain a preliminarily filtered image I1; 4) filtering a few candidate regions containing no characters in the I1 to obtain a final image I2; 5) combining the residual character candidate regions in the I2 into text line regions; and 6) inputting the text line regions to a convolutional neural network text detector in sequence, and filtering the text line regions containing no texts to obtain final text-containing text line regions. According to the method, by filtering the candidate regions containing no texts for multiple times, the text detection accuracy is improved; and the method can be used for automatically extracting the texts in the image.

The invention discloses an optical remote sensing image ship detection method based on knowledge distillation, belongs to the field of remote sensing image target detection, and solves the problem of low ship detection speed caused by large model parameter quantity and high calculation complexity in the prior art. According to the technical scheme, the method comprises the following steps: collecting or downloading a disclosed optical remote sensing image ship detection data set; building a ship detection network framework based on knowledge distillation, and enabling a student network to fully imitate behaviors of a teacher network in a training process by adopting Hint loss; jointly training the teacher network and the student network, and storing model parameters of the trained student network; and testing by using a student network to obtain a final ship detection result. The teacher network with high detection accuracy is used for guiding training of the student network, the original detection speed of the student network is kept, the detection accuracy of the student network is improved, and therefore rapid and accurate optical remote sensing image ship detection is achieved.

The invention discloses a visual detection method for the quality of liquid medicine on a high-speed pharmaceutical production line. The visual detection method is characterized by comprising the following steps of: firstly, acquiring images, i.e., acquiring five continuous sequence images of liquid medicine under a halted state of a bottle after revolution at a high speed, wherein the shot images are the gray-level images; secondly, initially denoising the images, i.e., removing the dynamic interference of the bottle wall by an offset background reduction method based on a characteristic point; thirdly, detecting a target, i.e., removing the static interference by a difference method and a threshold-based division method; fourthly, recognizing the target, i.e., judging whether the target is a foreign body or not according to the continuation of the motion trail of the target, and recording the foreign body; and fifthly, judging the target, i.e., judging whether the foreign body is the visible foreign body or not according to whether the size of the recorded foreign body exceeds the specified range or not. The method has the advantages of high detection precision and high detection speed.

The invention provides a human face detection method, comprising the steps of: importing: importing an image to be processed; sub-image extracting: traversing the image to be processed by using windows of different scales to extract parts of the image to be processed in the windows to serve as sub-images; preprocessing: specific to the sub-images, calculating a Sobel image of each sub-image by using a horizontal Sobel template, and excluding the sub-image if the ratio of the sum of Sobel response values of a predetermined region in the Sobel image to the total Sobel response value of the sub-image is smaller than a preset ratio threshold; detecting: specific to the sub-images passing the preprocessing, excluding non-human-face images by using an Adaboost cascade classifier to obtain candidate human-face sub-images; and verifying: specific to the candidate human-face sub-images, verifying the candidate human-face sub-images by using an Adaboost classifier established based on Haar-Sobel features so as to exclude the non-human-face sub-images, wherein the rest sub-images are taken as the human-face sub-images. The invention further provides human-face detection equipment correspondingly.

The invention discloses a fast face detection method based on a deep cascade convolutional neural network. The method includes creating a face data set, constructing a deep cascade convolutional neural network, testing the network model, using positive and negative samples to compose training set and verification set to train the depth cascade convolution neural network for deep learning, adding gender classification auxiliary task in the depth learning process, and adopting fine-tuning training at the same time. At the same time, the training method and network structure are optimized. The invention is based on a deep cascade convolutional neural network, the convolution layer is optimized to increase the network depth, and at the same time, the on-line difficult-to-negative sample miningmethod in the phase of auxiliary task training and fine-tuning training is introduced, which improves the classification accuracy of the network, reduces the false detection rate, and guarantees thedetection speed and accuracy of the method in practical application.

The invention discloses a high-definition image small target detection method based on an auto-encoder and a YOLO algorithm. The method mainly solves the problem that the accuracy and the speed of high-definition image small target detection cannot be considered at the same time in the prior art. The method comprises the following steps: 1) collecting and marking a high-definition image to obtaina training set and a test set; 2) performing data expansion on the marked training set; 3) generating corresponding Mask data according to marking information; 4) building an auto-encoder model; 5) training by using the training set; 6) splicing the trained coding network of the auto-encoder with a YOLO-V3 detection network to obtain a hybrid network, and training the hybrid network by using the training set; and 7) carrying out target detection on the test set by using the trained hybrid network. The calculation amount of target detection is reduced, the detection speed is improved, the detection precision of small targets in high-definition images is improved under the condition that the detection speed is guaranteed, and the method can be used for target recognition of aerial images ofunmanned aerial vehicles.

The invention discloses a gas storage well shaft detection system based on an ultrasonic phased array technology and a detection method thereof. The system comprises surface equipment such as a system host, optical fiber communication equipment and an encoder as well as underground equipment such as a centering device, an ultrasonic phased array detector and an annular phased array probe. The detection method includes the steps that the system host is started, and the centering device is adjusted to keep the underground equipment in coaxial parallel with a shaft; electric signals, for detecting control information, of the system are converted into optical signals through an optical transceiver and are transmitted to the ultrasonic phased array detector through optical fiber; the annular phased array probe is stimulated by the ultrasonic phased array detector according to the focusing law to transmit ultrasonic waves to scan the shaft wall; after being converted into optical signals, collected echo electric signals are transmitted to the system host through optical fiber to be stored and processed into images. Real-time S display and C display can be achieved by the detection system, the scanning coverage rate can be flexibly adjusted by changing the stepping magnitude of an ultrasonic deflecting angle, a corroded part is located and scanned carefully, and detection speed and detection precision are high.

The invention relates to a real-time video face key point detection method based on deep learning, and the method employs a convolutional neural network to carry out the key point detection of a single frame, employs a depth separable convolution to improve the model detection rate, employs a boundary heat map as an additional subtask of an original network to improve the constraint of a global face structure of the original network. The method improves the detection accuracy of an original network, is used for solving a data imbalance loss function of a heat map, improves the generalization capability of a model for a large attitude sample under a limited sample, and improves the inter-frame smoothness through an optical flow loss function. In the detection process, for a frame of which the confidence is lower than a key point confidence threshold due to an extremely large angle, fitting is carried out by utilizing 3DMM to obtain dense key point coordinates, 68-point sampling is carried out on the obtained dense key points according to a projection error between minimum frames, and the consistency with the previous frame is kept. The method has the advantages of real-time performance, capability of utilizing global inter-frame information, high detection accuracy of a face large posture condition and the like.

The invention discloses an environment control method based on a brain computer interface. By the method, an environment control device can operate in a synchronous / asynchronous control mode under different conditions; the asynchronous control mode is used for starting / stopping the whole environment control device or used for selecting some equipment (such as a television) so as to achieve autonomous control of the environment control device; after the equipment is started, the environment control device operates in the synchronous mode so as to increase accuracy and detection speed of the brain computer interface. When the environment control device operates in the asynchronous mode, system misoperation rate can be controlled effectively through introduction of pseudo keys and authentication mechanisms, so that the environment control device employing the method is practical. The environment control method is applied to the practical environment control device, so that seriously disabled persons are assisted in daily living, and their self-help abilities for daily living are improved.

The invention belongs to the field of computer vision, particularly relates to a single-stage small sample target detection method for decoupling measurement, and aims to solve the problems that an existing small sample detection target detection method is insufficient in detection precision, classification and regression interfere with each other in a non-decoupling mode, and detection network training is prone to over-fitting in the case of small samples. The method comprises the following steps: acquiring a to-be-detected image as an input image; obtaining a category and a regression box corresponding to each to-be-detected target in the input image through a pre-constructed small sample target detection network DMNet; wherein the DMNet comprises a multi-scale feature extraction network, a decoupling expression conversion module, an image-level metric learning module and a regression frame prediction module. According to the method, the overfitting problem during detection network training is avoided, the mutual interference of classification branches and regression branches is eliminated, and the precision of small sample target detection is improved.

The invention discloses a railway roadbed disease radar map real-time detection method based on a convolutional neural network. The railway roadbed disease radar map real-time detection method comprises the following steps: marking and dividing roadbed disease radar images into a training set and a test set; and expanding the training set, sending the expanded training set to a convolutional neural network, outputting a disease type, a disease type, a disease position coordinate and a disease confidence coefficient, and carrying out iterative calculation through a gradient descent method to obtain a railroad bed disease detection model; and adopting mean value average precision and the frame number per second as indexes for evaluating the quality of the model. According to the railway roadbed disease radar map real-time detection method, the multi-scale prediction network is fully utilized, and no candidate region is generated in the whole process, so that the detection time is greatlyshortened while the precision is ensured, and real-time detection is realized.

The invention discloses a colloidal gold test strip for rapid detection of Human enterovirus 71 (EV71) IgM (immunoglobulin m), belonging to the field of medical detection consumables. The colloidal gold test strip is characterized in that a colloidal gold-labeled cellulose pad is coated with colloidal gold-labeled EV71 antibody; a control line C is coated with anti-mouse IgG polyclonal antibody; a test line T is coated with anti-Mu chain monoclonal antibody; and a sample cellulose pad is coated with EV71 recombinant antigen. The inventive test strip adopts antibody capture principle and colloidal gold technology to detect EV71 IgM in serum / plasma / whole blood samples simply, sensitively, precisely and rapidly; and can be used for the rapid detection of EV71 IgM and diagnosis of hand-foot-mouth disease.

The invention discloses a highway pavement disease detection method based on improved YOLOv4, which can quickly detect various types of pavement diseases. A camera is triggered to shoot through a mileage sensor, and a highway pavement image is obtained in a vehicle driving process; and the pavement disease in the pavement image is detected by using the pavement disease detection model based on the improved YOLOv4. In the CSPDarknet-53 backbone network of YOLOv4, the depth separable convolution is adopted to replace the common convolution, so that the calculation amount of network parameters is reduced. In a loss function calculation stage, a loss function of YOLOv4 is improved based on a Focal loss function, and the problem of low detection precision caused by imbalance of positive and negative samples in a network training process is solved; and mileage position data is added to a detection result to form pavement disease comprehensive information. The method provided by the invention can realize automatic and real-time detection of highway pavement diseases, greatly reduces the cost, improves the detection efficiency, and has a better detection effect compared with a traditional target detection technology.

The invention discloses a rapid face detection method based on combination between skin color segmentation and AdaBoost. The rapid face detection method based on the combination between skin color segmentation and the AdaBoost comprises the following steps that (1) an N-layer Gaussian pyramid is built; (2) the layer number Ev is determined, wherein skin color segmentation face detection is carried out on the Evth layer in the Gaussian pyramid; (3) skin color segmentation face detection is carried out on an image of the Evth layer of the Gaussian pyramid, and rectangular face candidate areas are calibrated; (4) the layer number Ev' is determined, wherein AdaBoost face detection is carried out on the Ev'th layer in the Gaussian pyramid; (5) AdaBoost face detection is carried out on an image of the Ev'th layer of the Gaussian pyramid; (6) a face area is obtained. According to the rapid face detection method based on the combination between skin color segmentation and the AdaBoost, skin color segmentation and the AdaBoost are combined, the Gaussian pyramid is introduced, the Gaussian pyramid is created for an image to be detected, and then skin color segmentation face detection and AdaBoost face detection are carried out on the selected appropriate layers of the Gaussian pyramid respectively, so that face detection efficiency is greatly improved.

The invention discloses an AdaBoost cascade classifier rapid detection method. The method comprises the following steps of: S1, training an AdaBoost-based primary cascade classifier; S2, testing detection performance of the primary cascade classifier after each stage of the one-stage cascade classifier is combined; S3, determining the number X of starting layers using a zoom window according to the detection performance of the primary cascade classifier after the combination; S4, detecting a sample library by using X layers of primary cascade classifiers and the combination operation, so as to determine detected positive samples and falsely detected negative samples; S5, retraining an AdaBoost secondary cascade classifier by using the detected positive samples and falsely detected negative samples; and S6, carrying out detection by adoption of a manner of combining a primary cascade classifier zoom classifier and a secondary cascade classifier zoom detection window. According to the method disclosed by the invention, a manner of combining the primary cascade classifier and the secondary cascade classifier is adopted, so that the detection performance of systems is further improved while the detection speed is ensured.

The invention discloses a high-strength bolt connecting gusset plate compression detection method. The method includes the following steps: conducting an ultrasonic detection on a splice plate which has unscrewed nut by using an ultrasonic lossless defecting detector, computing an ultrasonic echo loss rate r' when the nut is not screwed; screwing the nut so as to generate a pressure among the splice plate, conducting an ultrasonic detection on the splice plate which has nut screwed by using the ultrasonic lossless defecting detector, computing to obtain an ultrasonic echo loss rate r when screwing the nut; based on the r' and r, obtaining an actual contact area among the splice plates; and based on the actual contact area among the splice plates, computing the compression of the splice plate. According to the invention, the method can measure the compression of the high-strength bolt connecting gusset plate in a direct, quick and accurate manner, and obtains accurate state of gusset plate.

The invention discloses a video target detection method based on a motion history image, and aims to improve the speed and accuracy of video target detection. The video target detection method comprises three aspects: (1) for an input video frame sequence, calculating motion history images of the video frame sequence, and performing feature extraction on video frames and the motion history imagesthereof through a residual network; (2) fusing the extracted two parts of features, and inputting the fused features into a convolutional neural network to extract candidate boxes; and (3) obtaining avideo target detection result according to a bounding box regression algorithm and the constructed classifier. According to the video target detection method, the motion history images are added intothe model training process, so that the feature information of the video frames is provided for the model, and the association information between the video frame sequences is increased, and the accuracy of video target detection can be improved.

The invention relates to the underwater testing field, especially relates to a underwater digital ultrasonic flaw detector based on the underwater robot (ROV), the flaw detector is composed of a marine device and a underwater device, the two are connected by a communication cable, wherein the underwater device includes a water tight probe, an ultrasonic signal transmitting and receiving circuit, the marine device includes a ultrasonic waveform oscilloscope, the signal detected by the water tight probe is transferred to the ultrasonic waveform oscilloscope by the ultrasonic signal transmitting and receiving circuit and is displayed, the device is suitable for divers to carry and can acquire high quality images, the cost is low, which can guarantee quality and testing speed of visual investigation, and can be used in muddy water and various waters.

The invention discloses a machine vision two-dimension detecting platform device which comprises an imaging system, a supporting system and a control system which controls the imaging system and the supporting system. The imaging system comprises a photoelectric imaging module and a data collecting module. The supporting system comprises a two-dimension workbench and an actuating mechanism, and the control system comprises a controller and a computer. The controller and the computer are in both-way connection, the controller receives a control instruction of the computer, and the computer and the data collecting module are in both-way connection. The controller controls the actuating mechanism and the photoelectric imaging module respectively, the actuating mechanism and the two-dimension workbench are in driving connection, the two-dimension workbench is connected with the photoelectric imaging module, and the photoelectric imaging module and the data collecting module are in both-way connection. The entire process of the machine vision two-dimension detecting platform device is rarely or does not need to participate by people, the automatic degree is high, online measurement can be basically achieved, and the measuring speed is based on the complexity of measured objects.

The invention discloses a silkworm cocoon sorting method. According to the silkworm cocoon sorting method, silkworm cocoons are subjected to quality detection and separation by using a silkworm cocoon sorting machine; the silkworm cocoon sorting machine comprises a rack and a discharging device, wherein a rolling stripping roller is arranged on the rack and in front of the discharging device; a plurality of conveyor belts with positioning grooves at interval are arranged in front of the rolling stripping roller; a separation scraping roller with the advancing direction opposite to that of the conveyor belts is erected above the plurality of conveyor belts; the rack is provided with a working platform, a plurality of conveying chains and a box body; a camera and a lamp are arranged in the box body; and a sorting device is arranged on the rack and behind the box body. The silkworm cocoon sorting method comprises the following steps: A, blanking; B, separating silkworm cocoons; C, putting the silkworm cocoons in place; D, detecting; and E, sorting. With the adoption of the silkworm cocoon sorting method, the silkworm cocoons are subjected to quality detection and sorting one by one.

The invention relates to a method for quickly detecting pathogenic bacteria carried by plant seeds of bacterial black rot, and relates to the biological detection of plant pathogenic bacteria. The method comprises the following steps of: (1) enriching the pathogenic bacteria of the bacterial black rot in soak solution of the plant seeds to be detected by adopting immunomagnetic beads; and (2) performing PCR amplification detection by taking enrichment solution obtained by the step (1) as a template, wherein antibodies coated on the immunomagnetic beads are antibodies of anti-pseudomonas syringae pv. maculicola; PCR amplification primers comprises a forward primer T1: 5'TGCTTTGCACACCCGATTT 3' and a reverse primer T2: 5'CCCCAAGCAATCTAGGT 3'; and an amplification product is 454bp. The detection method combines the adsorption technology of the immunomagnetic beads and real-time fluorescence quantitative PCR; and compared with detection methods of single planting, selective media and ELISA, the detection method of the invention is time-saving, space-saving, peculiar and sensitive, and is particularly important to the integrated control of plant bacterial black rot of plants. The detection method can provide technical guarantee for safety production, seed import and export trade, and is particularly suitable for the detection of crucifer seeds.

The invention discloses a mileage measuring device and method of an in-pipeline detector, and belongs to technical field of pipeline detection. The mileage measuring device is arranged on the in-pipeline detector, and comprises a magnetic field sensor unit, a signal conditioning module, an A / D (analog-to-digital) conversion module and a central processing unit. The method comprises the steps: step 1: acquiring pulse electrical signals corresponding to field changes of various mileage wheels; step 2: carrying out filtering and amplification processing on the pulse electrical signals respectively; step 3: carrying out A / D conversion on the pulse signals being processed in the step 2; step 4: carrying out secondary filtering processing on the digital pulse signals subjected to A / D conversion in the step 3; step 5: carrying out judgments of mileage wheel abnormality and pipeline turning; step 6: selecting and outputting an optimal mileage pulse signal currently; step 7: calculating the mileage of the in-pipeline detector. By applying the mileage measuring device and method of the in-pipeline detector, the stability of the detection speed and the mileage pulse signal of a system can be guaranteed, the anti-interference of the system is improved, and the mileage measurement accuracy and the output of the optimal mileage are guaranteed.

The invention discloses an operation simulation system of power grid information equipment based on an improved AABB-OBB (Axis Aligned Bounding Box-Oriented Bounding Box) collision detection algorithm. A user inputs working environment information and interaction control information from an input / output terminal, the working environment information and the interaction control information are transmitted to a working environment simulation server which carries out three-dimensional scene generation and corresponding interaction control processing on a working environment where the power grid information equipment operates by using the working environment information and the interaction control information, processed data is transmitted to a monitoring server for verification and then stored in a storage server, and a virtual view of the working environment is displayed in the input / output terminal for the user for interaction operation at the same time. A collision detection process fully exerts advantages of an AABB and an OBB, and the detection precision is greatly improved under the condition that the detection speed is ensured.

The invention discloses a harmonic wave detection method based on discrete Fourier transform (DFT). The harmonic wave detection method based on DFT is characterized by comprising steps that a three-phase current of a power grid side is acquired in a set period, and discrete processing on the acquired three-phase current is further carried out; fundamental wave active current amplitude of the three-phase current and a calculation model of fundamental wave reactive current amplitude are constructed; DFT positive transformation of the acquired three-phase current of the power grid side is carriedout through a window sliding iteration method to respectively acquire the fundamental wave active current amplitude of the three-phase current and the fundamental wave reactive current amplitude; fundamental wave active current positive sequence components of the three-phase current are calculated through employing coordinate transformation; a detection three-phase harmonic wave current is calculated according to the acquired three-phase current and the fundamental wave active current positive sequence components.