2076results about How to "Guaranteed detection accuracy" patented technology

The invention belongs to the field of deep learning and intelligent vehicle road detection, and specifically relates to a vehicle detection method based on improved YOLOv3. The method comprises the following steps: redesigning a convolutional neural network structure between a Darknet layer and three yolo layers, and designing a YOLO-TN network by referring to an idea of TridentNet weight sharing;carrying out model pruning on the YOLO-TN convolutional neural network; constructing a vehicle detection data set, and labeling vehicle position information in the data set; respectively training vehicle detection models based on the YOLO-TN and the YOLOv3, completing a vehicle detection task, and comparing detection results of the YOLO-TN and the YOLOv3. The method has high average precision while guaranteeing real-time detection, and is lower in omission ratio and more accurate in positioning when used for detecting distant vehicles and small-scale targets.

The invention discloses a mura defect detection method based on sample learning and human visual characteristics, which belongs to the TFT-LCD display defect detection field. According to the invention, the method comprises the following steps: firstly, utilizing the Gaussian filter smoothing and Hough transform rectangle to preprocess the TFT-LCD display image, removing a large amount of noise and segmenting the image areas to be detected; then, using the PCA algorithm to conduct learning to a large amount of defect-free samples; automatically extracting the differential characteristics between the background and the target and re-constructing a background image; and then, thresholding the differential characteristics between a testing image and the background; through the reconstructing of the background and the threshold calculating, jointly creating a model. According to the invention, based on the training sample learning, a relationship model between the background structure information and the threshold value is established; and a self-adaptive segmentation algorithm based on human visual characteristics is proposed. The main purpose of the invention is to detect different mura defects in a TFT-LCD, to raise the qualification rate and to increase accuracy for the detection of mura defects.

The invention discloses a video image processing-based fire detection system and detection method. A video acquisition module, a fire detection module, a fire analysis module, a fire judgment module and a fire alarm module are connected with one another so as to form the video image processing-based fire detection system; the video acquisition module acquires video images in the coverage area of a camera; the fire detection module performs background modeling and background updating on the video image and detects a foreground moving object; the fire analysis module analyzes the video images to identify a suspected flame region and extracts static characteristics and dynamic characteristics of the suspected flames; the fire judgment module adopts the static characteristics and dynamic characteristics of the suspected flames extracted by the fire analysis module as the input of a BP neural network so as to judge whether flames exist in a video; and the fire alarm module judges whether to give an alarm according to the output result of the fire judgment module. With an existing monitoring system used in combination, the video image processing-based fire detection system can decrease detection cost and greatly reduce a false alarm rate with fire detection accuracy ensured.

The invention discloses a face key point detecting method and device. The method comprises the steps that a face image to be detected is obtained; initial key points of N specific organs are determined in the face image by using a first face key point detecting algorithm, wherein N is larger than or equal to 1; the region, where each specific organ in the N specific organs is located, in the face image is determined with the corresponding initial key point as reference; a final key point of each specific organ in the N specific organs is determined in the region, where the specific region is located, in the face image by using a second face key point detecting algorithm. By means of the method, the number of the searching regions detected through the second face key point detecting algorithm can be reduced, and the detecting efficiency of face key points can be improved obviously.

The invention relates to deep learning based defect detection method of a power transmission line. A detection target is cut from an image, of a random resolution, shot by a unmanned aerial vehicle (UAV) during tour inspection of the power transmission line, or short by a mobile phone during manual tour inspection, or cut from a video shot by a fixed camera on a power transmission pole, an image of a fixed resolution is generated and serves as a training sample, sub-images of positive and negative training samples with different defects of the power transmission line are input to a target detection deep neural network for learning, a unified detection model with all defects of the power transmissions line is generated, the unified deep neural network model is utilized to detect all defectsof the input power transmission images of the random resolution, and an output image includes all defect types and marks defect positions.

The invention discloses a scene text end-to-end recognition method based on boundary point detection, and the method comprises the steps: extracting text features through a feature pyramid network, and generating candidate textboxes through a region extraction network; detecting a more accurate multi-directional bounding box of the text instance through a multi-directional rectangular detection network; secondly, detecting an upper boundary point sequence and a lower boundary point sequence of the text in the multi-directional bounding box; and finally, converting the text in any shape into ahorizontal text by utilizing the detected boundary point sequence for the subsequent attention mechanism-based sequence recognition network to performing recognizing, and finally, finding out the mostmatched word of the prediction sequence in the given dictionary by utilizing a cluster search algorithm to obtain a final text recognition result. According to the method, the scene text in any shapein the natural image can be detected and recognized at the same time under the condition that character-level labeling is not needed, the scene text comprises the horizontal text, the multi-directiontext and the curved text, and end-to-end training can be completely carried out.

The invention relates to a circuit and a method for detecting a direct-current power ground insulation resistor. The circuit comprises resistors R1 to R5 and switches S1 to S2, wherein the resistors R1 and R3 have the same resistance value; the resistors R2 and R4 have the same resistance value; one end of the resistor R2 is connected to an anode of a direct-current power supply through the resistor R1, and the other end of the resistor R2 is grounded through the switch S1; the resistors R1 and R2 and the switch S1 form a first detection branch circuit; one end of the resistor R4 is connected to a cathode of the direct-current power supply through the resistor R3, and the other end of the resistor R4 is grounded through the switch S2; the resistors R3 and R4 and the switch S2 form a second detection branch circuit; and the resistor R5 is connected with the first detection branch circuit or the second detection branch circuit in parallel. The circuit has the advantages as follows: when resistance values of positive and negative ground insulation resistors R+ and R- of the direct-current power supply are reduced in a uniform amplitude way or close to each other, the resistance values of the resistors R+ and R- can be still detected, and detection precision of the resistors R+ and R- can be also guaranteed.

It involves an intelligent air condition control method and system, whose method is: A. infrared detection devices detect infrared signal of human body, and the presence information of human been detected is sent to air-conditioning controller through the wireless signal; B. the controller of air condition adjust the work state based wireless signal. Step A includes: A1. Infrared detection devices will be installed in the upper space to be detected, A2. Judging whether human existence, when human exists, produce the wireless signal to launch air conditioning controller; in A2 step, which further test the battery voltage, if the battery voltage is insufficient, a corresponding alarm signal is generated. The system includes air conditioning controller and also includes infrared detection devices to detect infrared signal, and the human presence information been detected is sent to the air condition controller through the wireless signal. The air condition controller adjusts its work state based on the wireless signal it received. The invention is of high accuracy, the easy implementation and more energy-saving.

The invention relates to an automatic wheel hub appearance sorting device based on machine vision. The automatic wheel hub appearance sorting device comprises a detection body. The detection body comprises a positioning assembly, at least three light source camera assemblies and a detection frame, wherein the positioning assembly is used for fixing a wheel hub through a clamping device and driving the wheel hub to make Z-axis movement and rotary movement; the light source camera assemblies are correspondingly arranged on the upper portion, the lower portion and the side edge of the fixed wheel hub; and the detection frame is used for fixing the positioning assembly and the light source camera assemblies through connecting devices, and the light source camera assemblies are connected with an upper computer through network interfaces. Through the automatic wheel hub appearance sorting device, the functions of wheel hub type recognition, fixed centering and automatic appearance detection are achieved, and the automatic wheel hub appearance sorting device has the advantage of being high in detection efficiency and accuracy rate.

The invention discloses a photographing and sorting device for egg separation. A conveying system is arranged on a mounting bottom frame and comprises a conveyor belt and a conveying tray; bearing through holes for placing eggs are formed on the conveying tray; a photographing system and a sorting system are arranged on the mounting bottom frame; the photographing system comprises a light source and a camera which are respectively positioned on the upper side and the lower side of the bearing through holes during use; a cross sliding table structure which is used for correspondingly and movably adjusting the camera in the transverse direction and the longitudinal direction is arranged on the mounting bottom frame; the sorting system comprises a pneumatic suction disc mechanism for sucking the eggs; the pneumatic suction disc mechanism comprises a suction disc and a gas circuit control system which is communicated with the suction disc and is used for extracting or inflating gases in the suction disc; and the mounting bottom frame is provided with a cross sliding table structure which is used for correspondingly and movably adjusting the suction disc in the transverse direction and the longitudinal direction. By the photographing and sorting device, the eggs can be photographed and sorted, so that the operations of manually observing and selecting abnormal eggs in the prior art are effectively replaced.

The invention discloses a target detection method, a target detection model and a target detection system based on a cascade detector. The target detection method comprises the following steps of S1,training the target detection model by adopting a training data set with a target label; S2, inputting a to-be-detected picture into the trained target detection model, and performing feature extraction, upsampling and feature fusion to obtain feature maps P2, P3,..., Pn with different sizes; S3, predicting according to the feature map Pn to obtain a target position Bn and a category Cn; extracting corresponding features from the corresponding feature map Pi-1 according to the target position Bi for target prediction to obtain a target position Bi-1 and a category Ci-1, i = 3-n; S4, taking thetarget position B2 as a final prediction result. According to the invention, through the multiple iterative regression prediction of the the target candidate box, the target position with more accurate prediction is obtained, the target detection precision is improved, and the detection accuracy is higher.

The disclosed detection method for defect and magnetic leakage of a large-area steel plate comprises: local magnetizing the steel plate, detecting flux density on pole to calculate the inner magnetization saturation state by a formula; then, using multiple magnetic sensitive elements to form a detection probe to detect leakage flux density by equal interval, processing the output signal, comparing the detected value with dynamic-set threshold for decision. This invention improves cover range and space resolution, and ensures detection precision.

A camshaft signal panel angle detection gauge comprises a supporting base, a compressing mechanism, an angle positioning mechanism and a detecting mechanism. The compressing mechanism is installed on a supporting base and located on the outer side of the angle positioning mechanism. The angle positioning mechanism comprises a top base, a rear support and a front support, wherein the top base, the rear support and the front support are sequentially arranged on the supporting base. An angle positioning pin is installed on the top base. The detecting mechanism comprises a sliding rail, a detecting base and a displacement sensor. The sliding rail is installed on the supporting base. The detecting base is installed on the sliding rail. A detecting shaft is installed in the detecting base. An angle detecting block is installed at one end of the detecting shaft. An angle detecting pin, a reference groove and a reference surface are arranged on the angle detecting block. The displacement sensor is fixedly installed on one side of the detecting base. The displacement sensor makes contact with the reference surface of the angle detecting block. The camshaft signal panel angle detection gauge is simple in structure and easy and convenient to operate, calibration is easy, the detection precision of a camshaft is ensured, and the product percent of pass is improved.

A remote sensing image fast target detection method based on YOLOv2 belongs to the technical field of image processing and pattern recognition. The invention realizes the fast detection of important targets in the remote sensing image. Firstly, the remote sensing image dataset is constructed for the training and performance testing of the model. Then a convolution neural network structure suitablefor remote sensing image classification is proposed for feature extraction, and then a target detection network is constructed. Aiming at the problem that the detection capability of the convolutionneural network to the small target is poor, the invention adopts the methods of increasing the training scale and the batch regularization to improve the performance of the network. The invention defines the offset factor to correct the target position, and utilizes the SVM classifier to carry out the secondary classification of the target background on the detection result, which ensures the detection precision and the detection speed, and realizes the end-to-end detection. Most importantly, the model allows the new data detection results to be extended to the training dataset, thus updatingthe training target detection network and improving the generalization ability of the model.

The invention discloses a low-altitude target detecting system of a three-coordinate continuous wave one-dimensional phase sweeping unmanned aerial vehicle, and is aimed at providing a detecting device high in data detecting rate and anti-interference capacity. By means of the technical scheme, the system measures a target distance through a frequency modulation continuous wave ranging system, anantenna feedback subsystem converts radio frequency signals into electromagnetic wave signals to be radiated to the space, a receiving subsystem conducts amplifying, frequency varying and SFC controlprocessing on echo signals output by an emission and wave control subsystem, a signal processing subsystem generates the working time sequence of the whole system, measures the three-coordinate vectors of the distance, direction and pitching of a target position and calculates frequency modulation continuous wave ranging data of target signals, and a data processing and communication control subsystem parses the target angle, establishes the trajectory of target movement, controls the working mode of a radar and detects three-coordinate target information according to the target signals calculated by the signal processing subsystem and the antenna angle information output by the antenna control subsystem.

The invention discloses a self-adaptively started ddos defense method based on a trust value in an SDN. The self-adaptively started ddos defense method comprises the following steps: a switch receives an unmatched packet, and sends a packet_in data packet to a controller; a counter is deployed in an SDN controller; the counter pre-sets a packet_in number value; when an arrival packet_in number is equal to a set value every time, the packet_in arrival rate in the time period is calculated; then, whether the packet_in arrival rate is abnormal or not is judged; the flow on the switch corresponding to packet_in is precisely detected by utilization of a reverse neural network classifier; a neural network, an intermediate layer of which is 50*50, is established; six characteristic values are calculated; training is carried out by taking the six characteristic values as input of the classifier; and, if the output value of the neural network is between 0.5 and 1, the fact that the flow is a ddos attack is determined. By means of the self-adaptively started ddos defense method and system based on the trust value in the SDN disclosed by the invention, the characteristics of the SDN, such as flow table information of a data layer and control of the controller in a control layer to the whole network, are sufficiently utilized; and thus, the ddos attack in the SDN can be effectively defensed.

The invention discloses an advanced persistent threat APT attack detection method. A semi-supervised learning algorithm is used to mark data having similar characteristics, and a small quantity of marked data are used to generate a large-scale training data set, and then an information gain ratio is introduced to determine the degrees of the influences of the different characteristics on detection. The information gate ratio is used for characteristic extraction of every sub-data set in a detection model, and accurate identification of unknown attacks is realized. An improved k-means algorithm is used to mark the data having the similar characteristics, and on the basis of the small quantity of marked data, the accurate marking of a lot of training data sets is realized, and therefore the detection accuracy of the model is guaranteed; and by introducing the information gain ratio in the model, the degrees of the influences of the different characteristics on the detection are determined, and influences of redundancy and noise characteristics in the data are reduced, and therefore important flow characteristics are selected, the generalization capability of the detection model is improved to detect the unknown attacks.

The invention discloses a file data replication method for quick deduplication. The method comprises the following steps of: before a first target file in a storage medium is wholly replicated to a target file directory for the first time, performing fingerprint calculation on metadata information of the first target file to form identity (ID1) by a processor, extracting contents of a plurality of file data segments from the first target file according to an interval with preset size and performing the fingerprint calculation to form ID2, performing the fingerprint calculation on the metadata information of a second target file to form ID3, and extracting contents of a plurality of file data segments from the second target file according to an interval with preset size and performing the fingerprint calculation to form ID4; comparing the ID1 with the ID3; if the ID1 is the same as the ID3, saving replication; if the ID1 is different from the ID3 and the ID2 is the same as the ID4, updating metadata of the first target file; and if the ID2 is different from the ID4, wholly replicating the second target file.

The invention discloses an online detection method based on a laser radar, and the method comprises the following steps: S1, point cloud acquisition: establishing a global coordinate system, a vehiclebody coordinate system and a measurement coordinate system at an initial position of an intelligent inspection robot, and converting detection point coordinates, measured by a lidar, into position coordinates in the global coordinate system; S2, point cloud preprocessing: calculating an orbital plane by point cloud data of all the detection points located in the global coordinate system; S3, early warning judgment: projecting a real-time measured scanning point of the lidar to the orbital plane extracted at step S2, and calculating the projection distance from the scanning point to the orbital plane, determining that a foreign object invades the boundary if the projection distance is less than a set threshold value, or determining that the foreign object does not invade the boundary if the projection distance is greater than the set threshold value. The method provided by the invention enables a problem of a three-dimensional space into a one-dimensional space for judgment, reduces the calculation burden under the condition that the detection precision is guaranteed, and achieves the real-time boundary invasion detection in a tunnel of rail traffic.

The present invention discloses a method for detecting and locating insulator defects in power transmission line image via a drone, and belongs to the technical field of image extraction and recognition. The detection and location method for the insulator defects comprises the steps of: S1. converting an aerial image from an RGB color space into an HSI (Hue-Saturation-Lightness) space, and carrying out binarization on extracted H component image and S component image separately to obtain a binary image of a preliminary contour of an insulator string; S2. using an ant colony algorithm based on a particle swarm optimization parameter to extract a contour of an insulator monomer; and S3. using a least square method to perform ellipse fitting on the contour of the insulator monomer, and locating the insulator defects by detecting the space between the contours of insulator monomers. By using the technical scheme provided by the invention, the detection precision of the insulator defects can be significantly improved, background interferences are reduced, and the operation speed is relatively high.