163results about How to "Solve the low detection accuracy" patented technology

The invention provides a vehicle appearance damage identification method based on deep learning. The vehicle appearance damage identification method comprises the steps of obtaining an actual vehicleappearance damage image and marking a damage type and a position; Building a deep convolutional neural network; Carrying out model training to obtain a trained model; And vehicle appearance damage identification and model evaluation are carried out by using the trained model. According to the vehicle appearance damage identification method based on deep learning, the damage type and degree of thevehicle appearance part are identified in a complex environment based on the deep convolutional neural network model, and the algorithm operation speed is improved while the algorithm precision is ensured.

An infrared array number-of-personnel sensor-based counting method and apparatus comprises the following steps: processing an output signal of an infrared array sensor arranged above a door, identifying the contour of human bodies, and tracking the movement direction of the contour; and judging personnel goes in or out according to the movement track of the contour to realize counting of personnel access to the door. A counting apparatus comprises a single-chip microcomputer system, and the infrared array sensor, a PIR sensor and a power supply circuit which are connected with the single-chip microcomputer system, and the single-chip microcomputer system acquires the infrared image of a detection area through the infrared array sensor, and detects whether personnel passes by the door or not through the PIR sensor. The counting method and the counting apparatus solve the problem of low detection precision of infrared emission and reception sensors in the process of multiple persons passing through the door, has the advantages of multi-target tracking, high identification precision and low power dissipation, is difficult to damage due to high installing position, has substantial advantages, and adapts to application promotion.

The invention provides an improved safety belt detection method. A convolutional neural network (CNN) is used as a training model and is used for solving the problem of an existing deep learning safety belt detection method that the detection accuracy is low. The detection precision of the convolutional neural network is improved through utilizing a novel feedback increment type convolutional neural network training method and a novel multi-branch final evaluation value acquisition method; meanwhile, a method for randomly selecting a safety belt target candidate region in a multi-scale manner is used, and the selecting rate of a safety belt region is increased; and finally, a method for setting a fault-tolerant threshold value by a user is utilized so that the flexibility of detection operation is improved. The improved safety belt detection method is successful application of a CNN structure to safety belt detection; and compared with an existing algorithm, the detection accuracy is improved.

The present invention discloses a pedestrian detection method based on semantic segmentation information, and relates to the field of a pedestrian detection method based on a neural network. The pedestrian detection method based on semantic segmentation information comprises the steps of: 1: inputting original RGB images in training set samples into a backbone network, inputting corresponding semantic segmentation images into branch networks, and setting a loss function of the whole network to complete training; 2: inputting the original RGB images in THE training set samples into the backbonenetwork which completes training to perform convolutional feature extraction and generate a multi-layer feature map; 3: inputting the multi-layer feature map into an area generation network which completes training to perform pedestrian candidate frame extraction and generate a pedestrian candidate area; and 4: performing classification and location of the pedestrian candidate area by a classification regression network which completes training to output detection result images including a pedestrian position surrounding frame. The problem is solved that pedestrians and backgrounds are difficult to be distinguished in a low-resolution condition to cause low detection precision in the current pedestrian detection, and the precision of pedestrian detection is improved in the low-resolutioncondition.

The invention discloses a VAD dynamic parameter adjusting method and device, and the method comprises the steps: extracting an emotion feature vector of a voice signal of each sentence in a training corpus; enabling the emotion feature vector of the voice signal of each sentence to serve as the input feature of a neural network, enabling a pre-determined optimal VAD parameter sequence of the voice signal of each sentence to serve as the expected output of the neural network, employing a set neural network training algorithm, and carrying out the training of the built neural network; and carrying out the voice end point detection of a current sentence during voice processing through a VAD parameter which is outputted by the trained neural network through taking the emotion feature vector of a former sentence of the current sentence as an input feature. The method finds out the rule between the emotion information in voice and related parameters of a VAD model, obtains a VAD effective optimal parameter model, employs the optimal parameter model to carry out the dynamic pre-estimation of the VAD parameters during voice end point detection, and achieves an effect of optimizing the VAD in a special scene.

The invention provides a remote sensing target detection method based on boundary constraint CenterNet, which is used for solving the technical problems of relatively low detection precision and recall rate of dense small targets in the prior art. The method comprises the following implementation steps: obtaining a training sample set; the method comprises the following steps: constructing a boundary constraint CenterNet network; obtaining a prediction label and an embedded vector of the training sample set; calculating the loss of the boundary constraint CenterNet network; carrying out the training of a boundary constraint CenterNet network; and obtaining a target detection result based on the trained boundary constraint CenterNet network. Through performing maximum pooling in the constrained pooling area through the corner constraint pooling layer, the fine features around the target are extracted, the detection precision and recall rate of dense small targets are effectively improved, meanwhile, the boundary constraint label generated by the boundary constraint convolutional network is utilized to constrain the prediction box, a more accurate target prediction box is obtained, and the detection precision of the target is further improved.

The invention discloses a method for detecting and identifying traffic signal lamps in real time, and belongs to the field of traffic light detection and identification. The method is provided for solving the problem that the YOLOv4 algorithm is insensitive to small target detection, so that the traffic signal lamp detection precision is low. A shallow feature enhancement mechanism is provided, two shallow features in different stages in a feature extraction network are fused with high-level semantic features obtained after two times of up-sampling respectively, so that the scale of two detection layers is increased, and the positioning and color determining capability of the network for small targets are improved. A bounding box uncertainty prediction mechanism is introduced, output coordinates of a prediction bounding box are modeled, a Gaussian model is added to calculate the uncertainty of coordinate information, and the reliability of the prediction bounding box is improved. A LISA traffic signal lamp data set is used for carrying out detection and identification experiments, and the AUC value of an improved YOLOv4 algorithm in the detection experiments is 97.58% and is increased by 7.09% compared with VVA. The mean value of the average precision of the improved YOLOv4 algorithm in the recognition experiment is 82.15%, which is 2.86% higher than that of the original YOLOv4algorithm. The improved YOLOv4 algorithm improves the detection and identification precision of the traffic signal lamp.

An embodiment of the invention discloses a near-duplicate image detection method and belongs to the computer technical field and solves the problem of low accuracy of near-duplicate image detection inthe prior art. The method comprises the steps of: determining a first feature and a second feature of a target image in an input image pair by a multi-task network model respectively, wherein the first feature comprises image features reflecting inter-class differences and intra-class differences, and the second feature comprises image features reflecting the intra-class differences; constructinga fusion feature of the target image according to the first feature and the second feature of the target image; and according to the fusion feature, determining whether the target image is a near-duplicate image. The near-duplicate image detection method disclosed in the present application further improves the comprehensiveness and accuracy of the image fusion feature by combining the intra-class information and inter-class information of the image to construct the fusion feature of the image, thereby improving the accuracy of the near-duplicate image detection.

The invention relates to a ceramic wall and floor tile surface defect detection device and method based on multi-feature information fusion. The method is characterized by adopting two brightness-adjustable LED white backlight sources to illuminate the ceramic tile surface in an illuminating mode of being inclined to the tile surface at a low angle to obtain a ceramic wall and floor tile with uniform light distribution; collecting image data of the ceramic wall and floor tile through a CCD camera; extracting a suspected defect area of the image data; adjusting the brightness, height and angleof the LED white backlight sources and controlling the height of the CCD camera; collecting image data of a clearer suspected defect area; and finishing multi-feature calculation and information fusion through a fuzzy classifier based on a fuzzy C-mean algorithm, and identifying the defect type corresponding to the suspected defect area. The method can fuse various characteristic quantities, distributed in a grayscale feature space and a morphological feature space, of the suspected defect area of the ceramic wall and floor tile, can accurately and effectively determine existence of the defectand the type of the defect, and is good in defect detection effect.

The invention discloses a video color bar detecting method and a video color bar detecting device. In order to solve the problem of low accuracy in color bar detection, the method disclosed by the invention comprises the following steps of: performing boundary search on a video frame to be detected so as to determine boundary lines of each area of the video frame to be detected; acquiring boundary points obtained by intersecting each boundary line with the scanning line by taking a straight line which intersects with each boundary line as a scanning line; and determining the video frame to be detected to be the color bar based on that an analysis result of the position information of boundary points on the scanning line is in accordance with predetermined color bar characteristics. Therefore, by determining the boundary line and determining the boundary point obtained by intersecting the boundary line with the scanning line by using a boundary search algorithm, and determining the video frame to be detected as the color bar based on that the analysis result of the boundary points on the scanning line is in accordance with the predetermined color bar characteristics, high accuracy of the detection is ensured.

The present invention discloses a video flame smog detection method of multi-information fusion. The method comprises the following steps: firstly, a changed pixel detection method is designed for extracting foreground pixels, and the method has little influence of video jitter; secondarily, all the pixels having large changes are subjected to color feature analysis to obtain pixels according with the flame smog color; the pixels are subjected to communication area segmentation; and finally, the areas are subjected to a series of shape, area and position logic determination to remove areas which are obviously not the flame smog areas to finally obtain a flame smog detection result. The video flame smog detection method of multi-information fusion improves the accuracy of a traditional flame smog detection method, reduces the false drop rate, can be applied in the complex video monitoring environment and has high robustness.

The invention discloses a system bug attack detection method and apparatus. The system bug attack detection method is provided according to one aspect of embodiments of the invention. The system bug attack detection method comprises: obtaining operation information of an operation currently performed by an account with a non highest system privilege in a system; and determining whether the operation indicated by the operation information is used for modifying a privilege of the account to a highest system privilege from the non highest system privilege or not, and if the operation indicated by the operation information is used for modifying the privilege of the account to the highest system privilege from the non highest system privilege, determining that a bug attack event exists in the system. With the adoption of the method and the apparatus, the problem of low detection accuracy of detecting a privilege escalation bug attack in the prior art is solved and the effect of detecting the system bug attack event timely and accurately is achieved.

The invention provides a monitoring video abnormity detection method based on unsupervised learning. According to the method, firstly, a motion block in a video is extracted, then abnormity detectionis carried out from two different angles of local and global, and a detection result is more accurate through diversified detection angles. In local anomaly detection, firstly, a motion block in a video is expanded, then the expanded motion block is used as a basic detection unit, and the difference between the motion block and a neighborhood motion block of the motion block is compared from the time dimension, the space dimension and the space-time dimension; In global anomaly detection, firstly, moving blocks in a video are clustered to extract moving targets, then a sliding window is used on a moving target sequence, the difference between the two moving targets in the window is compared, and finally, a detection result is optimized based on the consistency. The method is suitable for abnormal detection of the monitoring video, low in calculation complexity, accurate in detection result and good in robustness. The method has wide application in the technical field of video analysis.

The invention relates to an image change detecting method based on principle component general inverse transformation in the technical field of image processing. The method comprises the following steps of: carrying out transformation in two characteristic spaces after reorganizing data of an image to be detected; carrying out waveband updating differential treatment in the transformed characteristic spaces to obtain a changed component; and extracting a change area by an automatic threshold value confirming method to realize image change detection. The invention carries out differential detection on an image in the characteristic spaces obtained after principle component general inverse transformation, effectively inhibits noise, reduces the influence of image fuzzy distortion, and the like on the reduction of detecting precision and improves the detecting precision.

The invention discloses a road section average travelling time calculation method based on an electronic license plate. The method includes that an electronic license plate is installed on each vehicle, and an electronic license plate card reading device is installed at the position of an entrance / exit of a road. When vehicles pass through the card reading device, the card reading device reads the electronic license plate of each vehicle and transmits the read electronic license plate, the reading time, the reading position and other information to a center server, and the center server removes vehicle samples with the travelling time obviously difference from the average travelling time according to the read information of the vehicles, and comprehensively calculates the average traveling time of the road section in multiple directions. Compared with travelling time calculation based on an electronic license plate method, the method has the obvious advantages of low cost and high accuracy. Reading of the electronic license plates is not affected by weather and illumination, the problem of low detection accuracy caused by factors such as the weather and the illumination when a video mode is adopted to conduct license plate recognition is solved, and shortest travelling time calculation can be conducted accurately.

The embodiments of the present invention provide an information processing method, an apparatus, a cloud processing device, and a computer program product, which relate to the field of data processingtechnologies, and improve the efficiency of detecting whether a pavement region exists in a road surface. The information processing method provided by the embodiment of the present invention includes: acquiring a depth image; processing the depth image to obtain a line average graph, determining a road surface region in the depth image according to the row average map; determining the road surface region The suspected pitted area; determining the suspected pothole area according to the pothole threshold to determine whether the pitted area is included in the depth image.

The invention belongs to the field of face recognition, deep learning and image forensics, particularly relates to a face deep tampered image detection method, system and device based on multi-scale depth feature fusion, and aims to solve the problem of low detection accuracy of a face deep tampered image detection method. The method of the system comprises the following steps: acquiring a to-be-detected face image as an input image; performing normalization processing on the input image, and obtaining a detection result through a pre-trained face detection model; the face detection model is constructed based on a convolutional neural network, and a convolutional layer of the face detection model is composed of a deep convolutional network and a hole convolutional network. According to theinvention, the detection rate of the human face deep tampered image is improved.

The invention provides a vehicle remaining oil quantity detection method and device, a vehicle and a storage medium. The method comprises the steps of: determining the current acceleration and the current angle of a vehicle according to current vehicle state parameters; obtaining an oil quantity voltage value output by the oil quantity sensor in a set time period according to a preset period so asto obtain a first preset number of oil quantity voltage values, and determining a current oil quantity voltage mean value based on the first preset number of oil quantity voltage values; according tothe current acceleration and the current angle, determining whether a current oil quantity voltage mean value is reserved or not; when it is determined that the current oil quantity voltage mean value is reserved, determining the current oil quantity resistance value according to the current oil quantity voltage mean value; determining a current oil quantity resistance mean value according to a second preset number of oil quantity resistance values and the current oil quantity resistance value which are sequentially determined and obtained from the current moment to the front; and based on the corresponding relation between the pre-stored residual oil quantity of the oil tank and the oil quantity resistance value, determining the current residual oil quantity of the vehicle according to the current oil quantity resistance mean value.

The invention discloses a tumble detection method based on video articulation points and a hybrid classifier. A traditional video-based fall detection algorithm depends on manual extraction of fall features, fall is detected by means of a linear discriminant classifier method, the model is simple, but the accuracy is low. The method comprises the following steps: 1, extracting each frame of imageof a detected video clip; 2, acquiring a human joint data matrix; 3, establishing a plurality of behavior matrixes; and 4, calculating a time characteristic parameter and a space characteristic parameter; 5, carrying out primary classification; 6, carrying out secondary classification. According to the method for extracting the human skeleton joint points, the problem that the human posture cannotbe accurately estimated by extracting the human aspect ratio, the projection area and the like through a traditional method is solved. The behavior matrix is constructed by adopting the sliding window with the fixed size, modeling can be performed on time and space axes at the same time, and falling behavior characteristics are fully expressed.

The invention discloses a multi-dimensional distributed abnormal transaction behavior detection method. The method comprises the following steps: firstly, mining multi-dimensional original feature examples in network transaction behaviors before and in transactions; secondly, providing an automatic feature learning and fusion algorithm MSDAE based on deep learning to remove redundancy and noise inoriginal features and automatically learn implicit and representative features; and finally, a parallel distributed integrated framework SpaEnsemble based on Apache Spark is provided to realize efficient and rapid analysis and detection of large-scale abnormal transaction behaviors. The method has a wide application prospect in the field of network security.

Embodiments of the invention provide a slow disk detection method and apparatus, which relates to the technical field of computers and can solve the problem of low slow disk detection precision. The method comprises the steps of obtaining average service time of input / output (I / O) requests corresponding to different types of hard disks in a current preset period; obtaining slow disk thresholds corresponding to the different types of hard disks in a next preset period according to a relationship between the average service time of the I / O requests corresponding to the different types of hard disks and a preset value; and determining slow disks in the next preset period according to the slow disk thresholds corresponding to the different types of hard disks and the average service time of the I / O requests corresponding to the hard disks in the next preset period. The method and the apparatus are used for dynamically adjusting the slow disk thresholds.

The invention discloses a multi-size shelf commodity detection method. The method comprises the following steps: (1) obtaining a to-be-detected shelf image; (2) preprocessing the shelf images in the step (1), detecting shelf edge positions in the images through edge detection, Hough transform and linear screening, segmenting the images with excessive shelf layers, and correcting distortion of theimages with inclined shelves; (3) inputting the image obtained in the step (2) into a feature extraction network to obtain five layers of feature maps with different depths; (4) performing feature fusion on the obtained feature maps of each layer; (5) performing regional nomination on the feature map through a regional nomination network to obtain candidate boxes; and (6) further correcting the candidate box by using the target detection network and reasoning the type and accurate coordinate position of the to-be-detected commodity. The multi-size shelf commodity detection method based on image segmentation and the deep neural network is high in detection precision.

The invention provides an array substrate and a display device. The array substrate and the display device are used for solving the problem that due to the fact that the difference between the feature of a thin film transistor in a detection area and the feature of a thin film transistor in a display area is large, the detection accuracy is low. The array substrate comprises a display area and a pseudo pixel area located on the periphery of the display area. A second pixel unit is arranged in the pseudo pixel area which comprises at least one detection unit. Each detection unit comprises a second pixel unit. Each second pixel unit is correspondingly provided with one thin film transistor. All electrodes of all the thin film transistors are respectively connected with external testing equipment through testing wires.

The invention relates to the technical field of medical apparatus and instruments, in particular to a navigation monitoring device for a C-arm X-ray machine. The device comprises a positioning detection assembly, and the positioning detection assembly comprises two or more positioning detectors; the positioning detection assembly is integrated in the C-arm X-ray machine and used for collecting position information of target equipment in real time, and can transmit the position information to the C-arm X-ray machine. According to the navigation monitoring device, the positioning detection assembly is integrated in the C-arm X-ray machine, the collected information can be directly transmitted to the C-arm X-ray machine and analyzed and processed, so that the intra-operative operation is navigated and monitored in real time, the situation is avoided that a cart of navigation equipment also needs to be additionally arranged in an operating room, the space of the operating room is effectively saved, response of image transmission is quick, and the navigation monitoring accuracy is effectively improved.

The invention discloses a detection method and an apparatus for an operation event. The detection method for the operation event includes: operation data of a target command parser is obtained, wherein the target command parser is an interface for a user client to log in a target machine, and the operation data is data for executing the operation event via the target command parser by the user client; target data is extracted from the operation data, wherein the target data is used for detecting the data of the operation event; and the target data is processed to detect whether the operation event is legal. According to the detection method and the apparatus, the problem of low detection precision of the operation event in the prior art is solved, and the detection precision and the detection efficiency are improved.

The invention discloses a fine granularity classification-based unmanned aerial vehicle identifying and locating method. Based on fine granularity classification after object coarse granularity detection, specific external structure information of an unmanned aerial vehicle is searched out according to an identified unmanned aerial vehicle type and unmanned aerial vehicle type library information;in combination with internal parameters of a camera, two-dimensional coordinates of the unmanned aerial vehicle are mapped into three-dimensional coordinates to determine the position, in a three-dimensional space, of the unmanned aerial vehicle; and through continuous three-dimensional coordinate information of frame pictures, trajectory information of the unmanned aerial vehicle can be obtainedin the three-dimensional space. The method solves the problem of inaccurate identifying and locating of the unmanned aerial vehicle in the prior art.

The invention relates to an ambient illumination intensity detection method and device and a terminal, and belongs to the technical field of communication. The ambient illumination intensity detectionmethod comprises that first illumination intensity detected by a first light sensor and second illumination intensity detected by a second light sensor are obtained; whether both the first illumination intensity and the second illumination intensity are lower than a first preset illumination intensity threshold is determined; and if NO, a higher one of the first illumination intensity and the second illumination intensity is used as the ambient illumination intensity. The problem of low precision of ambient intensity detection due the fact that the ambient light intensity of the terminal cannot be detected accurately when the terminal is in a backlight scene or one light sensor is shielded is solved, and the detection precision of the ambient light intensity is improved.

The invention relates to an artificial intelligence technology, and discloses an abnormal work order identification method. The method comprises the following steps: classifying a marked work order set according to marking labels of marked work orders to obtain a classified work order set; screening the classified work order set by using an isolated forest algorithm to obtain a screened work order set; calculating feature similarity of work orders in the screened work order set based on a text identification model; calculating abnormal similarity of the work orders in the screened work order set by using an atomic rule model; calculating abnormal values of the work orders in the screened work order set based on the feature similarity and the abnormal similarity; obtaining abnormal work orders according to the abnormal values. In addition, the invention also relates to a blockchain technology, and the marked work order set can be obtained from nodes of a blockchain. The invention further provides an abnormal work order identification device, electronic equipment, and a computer-readable storage medium. The problem of low accuracy of abnormal work order detection is solved.