39results about How to "Solve the problem of low accuracy" patented technology

The invention discloses a visual position-pose measurement method based on point-line combination characteristics, which belongs to the technical field of computer vision measurement and relates to a method applicable to measurement on the position and pose of a split-type cylinder-like object moving at high speed. The surface point-line combination characteristics of the cylinder-like object moving at high speed are acquired by a binocular visual system; a target axis and a local coordinate system can be obtained through five steps of calibrating by a high-speed camera, segmenting a point-line characteristic image, extracting characteristic marks, matching the characteristic marks, and acquiring the position and the pose, and then target position-pose information is acquired according to the target axis and the local coordinate system; the point-line combination characteristics consists of circle reflection marks with two axes respectively formed at the front segment and the rear segment of a target object coinciding with target axes, and four point reflection marks which are uniformly distributed at the circumference of the middle segment of the target object and are restrained from different distances to the circle at the front end. The problems that the axis of the cylinder-like object is not easy to acquire, the precision of the acquired axis is low, and the surface mark can not be precisely positioned due to the repeated assembly of the split-type object are solved by utilizing straight-line characteristics of strong anti-noise capacity and large information amount which are matched with point characteristics.

The invention provides a deep learning based video crowd emotion analysis method, and mainly relates to classification for crowd emotions in the video by using a multi-stream neural network. The method comprises the steps of building a multi-stream neural network (pixel, optical flow and saliency), concurrently extracting features in pixel information, superposition optical flow information and saliency information of a video sequence by using the network, and finally integrating the three types of features to obtain classification for the crowd emotions. The self-learning ability of deep learning is given into full play, the limitation of manual feature extraction is avoided, and the method is enabled to be higher in adaptability. Training and prediction are performed by using structural features of the multi-stream deep learning network, classification results of multi-stream sub-networks are integrated finally, and the accuracy and the work efficiency are improved.

The invention provides a vehicle characteristic data processing method and a vehicle risk prediction model training method. The vehicle feature data processing method comprises the steps of obtainingat least one kind of original data, determining a plurality of required first feature factors from the original data, then selecting feature points from the track points, performing hierarchical storage on the feature points, determining connection lines between the feature points corresponding to different levels, and performing associated storage on the plurality of first feature factors and thefeature points or the connection lines between the feature points; The vehicle risk prediction model training method comprises the steps of training a vehicle risk prediction model by using a plurality of pieces of sample data, wherein each piece of sample data comprises a sample mark used for representing whether the vehicle has an accident or not; and at least one first feature factor and / or atleast one second feature factor stored on the associated feature points or the connecting lines between the feature points corresponding to a certain track level, so that the problem of low accuracyof an existing vehicle risk prediction model is solved.

The invention provides a face sequence expression analysis method based on deep learning and mainly relates to classification of face sequence expressions through a multi-scale face expression recognition network. The face sequence expression analysis method comprises the steps that the multi-scale face expression recognition network (comprising three channels for processing face sequences with different resolution ratios of 128*128, 224*224, 336*336 and the like) is constructed, features in the face sequences with the different resolution ratios are parallelly extracted through the network, finally, the three kinds of features are fused, and thus classification of the face sequence expressions is obtained. According to the face sequence expression analysis method based on deep learning, the self-learning capability of deep learning is given to full play, limitation of manual feature extraction is avoided, and thus the adaptability of the method is higher; and parallel training and predicting are conducted through the structural features of a multi-stream deep learning network, finally, the classification results of a plurality of sub-networks are fused, the accuracy rate is increased, and the working efficiency is improved.

The invention discloses an integrated fatigue detection and driving record system and a fatigue detection method. The integrated fatigue detection and driving record system is characterized by comprising a processor, a fatigue detection module, an alarm module, a GPS driving record module, a communication module and a storage module, wherein the processor is connected with the fatigue detection module, the alarm module, the GPS driving record module, the communication module and the storage module. According to the integrated fatigue detection and driving record system and the fatigue detection method of the invention, through the arrangement of the fatigue detection module, the system well adapts to the physical characteristics of different drivers, the accuracy of judgment is improved, and a driver is reminded through an alarm system or a vehicle is stopped in a forced manner if that the driver is fatigued is judged; and through the arrangement of the GPS driving record module, if an accident happens to a vehicle and the vehicle is stopped abnormally, a program uploads information to the cloud, and the family of the owner, an insurance company or a car service provider is reminded in time via the Internet according to the setting.

The invention discloses a method for detecting and positioning of cancer regions with small samples or unbalanced samples. By combining the characteristics of the histopathological image containing cancer cells, the method enhances the data set by using a method of adding noise, rotating, increasing or reducing brightness, expands the data set, balances the specific gravity of the label type of the training set, and improves the training effect of the classifier. The method is characterized in that an Inception V2 network is used as a basis; aiming at the conditions of few samples and unbalanced samples, iterative training is carried out in combination with a small number of calibrated cancerous tissue image blocks and WSI pathological images; the method is advantaged in that WSI image-level pathological image cancer area detection and positioning can be accomplished with high accuracy; problems of over-fitting, local optimization and gradient disappearance of the over-deep neural network easily caused by too few labeled training samples are solved; and accuracy of the training result and usability of the network under the same level are improved.

The invention provides a small-current grounding fault line selection system, which comprises zero sequence current transformers, a voltage transformer, a grounding transformer, a grounding loop and a grounding line selection device, wherein the zero current transformers are arranged on each feed line, the voltage transformer is arranged on a bus, the grounding transformer is connected with the bus, the grounding loop is connected between a high-voltage side neutral point of the grounding transformer and the ground, and is connected in series with a grounding resistor and a contactor used for controlling the on-off state of the loop, the grounding line selection device is respectively connected with the zero sequence current transformer, the voltage transformer and the contactor, and is used for controlling the conduction of the grounding loop when the voltage transformer detects the zero sequence voltage, and then, a fault line is judged according to the current data collected by the zero sequence current transformer.

The invention discloses a pointer instrument automatic reading method based on key point detection. The pointer instrument automatic reading method includes the steps: collecting a front RGB image ofthe pointer instrument, and constructing a key point detection network based on an improved mask region convolutional neural network; inputting the front RGB image of the pointer instrument with knowninstrument readings and the readings thereof into a key point detection network for training, and predicting a to-be-detected object to obtain a positioning position and key points of scale lines andpointers on a dial plate; fitting an arc formed by the scale lines and a straight line where the pointer is located based on the detected key points; and finally, automatically calculating the reading of the pointer type instrument according to the relative angle between the straight line where the pointer is located and the arc formed by the scale lines. The pointer instrument automatic readingmethod can achieve the automatic reading of the pointer instrument, is good in precision and robustness, can be suitable for various types of pointer instruments, and is important in engineering valueand significance.

The invention discloses an RFID tag relative position positioning method based on depth learning. The method comprises the following steps: S1, arranging a series of RFID tags uniformly and horizontally according to a random order, keeping an RFID antenna moving parallel to the RFID tags, and continuously reading phase information data and signal strength information data between the RFID tags byusing the RFID antenna; S2, setting a data collation structure, based on the data collation structure, making the phase information data and the signal strength information data into training data, and training the training data by using a depth neural network to initially obtain the relative positional relationship between all the RFID tags; S3, sequencing the relative positional relationships among the RFID tags by a specified sequencing algorithm, and determining the final relative positional relationships among all the RFID tags; the method of the invention can accurately analyze the arrangement order of a series of densely arranged RFID tags.

The invention provides an electric quantity data recovery method and device. The method comprises the steps that the source data of initial data is preprocessed, abnormal data is obtained, wherein the abnormal data comprises at least one electricity customer identification and electric quantity data corresponding to the at least one electricity customer identification, the electric quantity data contains electricity consumption dates and daily electricity consumptions corresponding to the electricity consumption dates, and each daily electricity consumption is a normal daily electricity consumption or an abnormal daily electricity consumption; for each electricity customer identification, the corresponding electric quantity data is subjected to period division, and the electric quantity data in at least one cycle is obtained; when the electric quantity data in the at least one cycle is not one, the electric quantity of each abnormal daily electricity consumption in the first cycle is determined; according to the electric quantity, adjacent to the electric quantity data in the first cycle, in other cycles, the abnormal daily electricity consumption is recovered. Based on the electric quantity data recovery method, the abnormal data is analyzed and recovered through the data periodicity, and as the recovered data can meet the change rule of real data, the accuracy of the recovery result can be higher.

The invention discloses a detecting method and a device of a vehicle logo. The method comprises that a region can be confirmed as a license plate region on a vehicle grayscale image, wherein the region can meet edge density reserved requirements after edge segmentation operation; a vehicle logo region is positioned with using license plate information and the license plate region and the vehicle logo is detected by using edge detection and watershed segmentation in the vehicle logo region. The accuracy of detecting the vehicle logo is improved through the detecting method and the device of the vehicle logo.

The embodiment of the invention discloses a banknote adhesive tape detection method and device, equipment and a storage medium. The method comprises the steps that the banknote thickness detection data of a banknote to be detected are acquired through the thickness sensors of the preset number of passages according to the preset detection number of detection points in case of passing of the banknote; the detection points of each passage are divided into the preset number of detection segments; the average value of the banknote thickness detection data in the detection segments acts as the banknote segmented thickness data of the detection segments; the banknote segmented thickness data are ordered, and the average value of the banknote segmented thickness data in the middle of the order acts as the banknote thickness reference value; and when the banknote segmented thickness data greater than the banknote thickness reference value preset threshold meet the preset judgment condition, existence of the adhesive tape on the banknote to be detected can be determined. According to the technical scheme, the effect of enhancing the detection efficiency and enhancing the detection accuracycan be realized.

The invention relates to an optical element surface defect detection method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of optical detection. Fresnel patterns are etched on the to-be-detected surface of the optical element. The method comprises the following steps: acquiring a grayscale image of the to-be-detected surface; obtaining a polar coordinate image of a target area where Fresnel fringes are distributed in the grayscale image; removing a Fresnel pattern corresponding to the Fresnel pattern in the polar coordinate image; determining a polar coordinate defect area according to the pattern area of each remaining pattern in the polar coordinate image; and marking a defect area corresponding to the polar coordinate defect area in the grayscale image.

The embodiment of the invention discloses a data processing method. The method comprises the following steps: acquiring a plurality of historical data acquisition positions corresponding to a production flow of a production system and an operation index type corresponding to each historical data acquisition position; constructing a first neural network model based on the plurality of historical data acquisition positions and the operation index type corresponding to each historical data acquisition position; acquiring historical sample data from a historical database, wherein the historical sample data comprises acquisition data corresponding to each operation index type and the product yield of the production system; and carrying out model training on the first neural network model by adopting the historical sample data to obtain a target neural network model. The embodiment of the invention further provides electronic equipment and a computer readable storage medium.

The invention discloses an automobile logo detection method and device. The method comprises the steps that an area with edge density which meets preset requirements after edge segmentation operation on a gray image of an automobile is determined as a license plate area; an automobile logo area is positioned by using license plate information and the license plate area; and in the automobile logo area, automobile logo detection is conducted through edge detection and watershed segmentation. Through the automobile logo detection method and device, the automobile logo detection accuracy is improved.

The invention discloses a strength detection device for steel. The detection device comprises a supporting frame, the top of the supporting frame is fixedly connected with a beam. A baffle and a reinforcing plate are fixedly arranged on the inner side of the supporting frame. An operation box is welded to one side of the supporting frame. A motor is fixedly mounted in the supporting frame througha bolt; a driving gear is fixedly arranged at one end of the motor; a bearing and a sliding rail are embedded into the supporting frame. A totating shaft is movably connected to the inner side of thebearing; the outer wall of the rotating shaft is movably connected with the sliding block through threads. A driven gear is welded on the outer wall of the rotating shaft; one side of the sliding block is fixedly connected with a connecting rod; one side of the connecting rod is fixedly connected with a clamping plate; an overhaul plate is embedded into the front face of the beam, a hydraulic machine is fixedly installed in the beam through bolts, a hydraulic rod is movably connected to the bottom of the hydraulic machine, a detection head is fixedly connected to the bottom of the hydraulic rod, and the problem that the detection result accuracy of the detection equipment is low is solved through the detection head, the operation box and the hydraulic machine.

The invention discloses a fault diagnosis method for a high-speed train bogie, proposes a full vector sample entropy feature extraction algorithm, and combines the full vector sample entropy feature extraction algorithm with noise-assisted multi-element empirical modal decomposition to perform high-speed train bogie failure diagnosis. The fault diagnosis method comprises the steps of: acquiring vibration signal data during train operation by means of sensors; decomposing the signals by means of noise-assisted multi-element empirical modal decomposition to obtain a series of multi-element intrinsic mode functions (IMFs); selecting IMF components most relevant to an original signal by adopting a correlation coefficient method; performing full vector sample entropy feature extraction on the valid IMF components to obtain feature vectors, and finally, using the feature components as input of a support vector machine to identify a running state of a train. The fault diagnosis method fuses the information of two homologous channels, introduces the full vector concept to obtain a fusion result, fully considers the information fusion between signals, and can better reflect the features ofthe signals of each channel, thereby acquiring higher identification rate.

The invention discloses a hybrid method for correcting sequencing errors in third-generation sequencing data under heterozygous variation. According to the method, input data is second-generation sequencing data and third-generation sequencing data, and are processed by using existing comparison software and assembly software; and based on Bayesian classifier principle, heterozygosity of gene lociis judged, and reading segments in the third-generation sequencing data are corrected in combination with a heterozygosity judgment result, so that the problems that the existing correction algorithmis low in accuracy and ineffectiveness when heterozygosity variation is processed are solved. According to the invention, heterozygosity variation is considered when sequencing errors are corrected,a series of probability models are designed to judge and classify heterozygosity, different correction strategies are adopted for different heterozygosity classifications, and the problem that correction errors occur when the existing correction method encounters heterozygosity variation is solved.

The invention discloses a body movement recognition method and device, computer equipment and a storage medium, and belongs to the field of image recognition. The method comprises the following steps: acquiring a target time sequence picture set, wherein the target time sequence picture set is a picture set of a group of limb actions with a time sequence; sequentially inputting the target time sequence picture set into a first network to obtain a target tensor output by the first network, wherein the first network is used for processing the target time sequence picture into a feature matrix, and the target tensor is used for representing a feature matrix set corresponding to the target time sequence picture set; inputting the target tensor into a second network to obtain a limb movement recognition result; extracting at least one sub-tensor from the target tensor, inputting the sub-tensor into the second network to obtain an action category probability corresponding to each sub-tensor, and obtaining a limb action recognition result according to weighted summation. According to the technical scheme, the problem that the accuracy is low due to the fact that the continuity of actions is inconsistent during limb recognition can be solved.

The invention discloses an automobile logo detection method and device. The method comprises the steps that an area with edge density which meets preset requirements after edge segmentation operation on a gray image of an automobile is determined as a license plate area; an automobile logo area is positioned by using license plate information and the license plate area; and in the automobile logo area, automobile logo detection is conducted through edge detection and watershed segmentation. Through the automobile logo detection method and device, the automobile logo detection accuracy is improved.

The invention provides a similarity determination method and device. The method comprises the steps that for each picture in N pictures, M first objects and M second objects in the picture are recognized, the M first objects and the M second objects are placed at preset positions in the picture, N is an integer larger than 1, and M is an integer larger than 0; first center positions of M first objects and second center positions of the second objects are obtained, and M position coordinates of the first center positions of the M first objects relative to the second center positions are obtained; and an M-dimensional feature vector of each picture is determined according to the M position coordinates, and the similarity of every two pictures in the N pictures is determined according to theM-dimensional feature vector of each picture. By adopting the technical scheme, the problems of low accuracy and the like in the process of judging the similarity of some objects in related technologies are solved.

The invention relates to an on-line detection method for ceramic zero resistance insulators on an alternating current transmission line based on infrared thermal imaging, and belongs to the field of power transmission. The invention aims to provide an on-line detection method for ceramic zero resistance insulators with safety, economy, high efficiency and high accuracy. The invention adopts the technical proposal that when a ceramic insulator string has a zero resistance insulator, the spread voltage of other insulators is boosted, the reactance of the whole insulator string is reduced, the leakage current of the insulator string is increased, and the heating difference of the good insulators and the zero resistance insulators is increased. By shooting thermal field images of the insulator string through a high-precision infrared imaging instrument, adopting image processing technology to extract axial temperature distribution curves of the insulator string, and utilizing the difference of the axial temperature distribution curves of the adjacent insulators, the method achieves accurate judgment of the state of the insulators. The method can be used for zero resistance detection for the ceramic insulators on the alternating current transmission line in grade of 110KV and above voltage, and is applied to lightening workload of detection greatly and improving secure operation level of a power system.

The invention provides a system and a method for unmanned monitoring, and belongs to the technical field of monitoring. The system comprises an acquisition module, a classification module, a plurality of prediction modules, a comprehensive module and a comparison module. According to the invention, environment images and monitoring real-time images of a to-be-monitored area are collected from multiple angles in different time periods, multiple prediction modules are respectively trained by using the environment images from multiple angles in different time periods, and a standard result is obtained through processing by a comprehensive module. And the monitoring real-time image is input into the prediction module and is processed by the comprehensive module to obtain a prediction result. And performing benchmarking on the prediction result and a standard result to judge whether the to-be-monitored area is abnormal or not. According to the method, video training is carried out on a specific area, the method is suitable for dangerous place contact prediction, unattended place protection, specific scene prediction, face recognition card punching, face recognition lock control and other scenes, and the problem that the accuracy of current manual video recognition is low is effectively solved.

The invention discloses a method and a device for training a model, and a method and a device for generating information. The implementation scheme of the method for training the model comprises the steps of acquiring a training sample set, and a machine learning algorithm being utilized to take behavior feature information, other feature information and object feature information included in training samples in the training sample set as input data; taking a user behavior vector corresponding to the input behavior characteristic information, an object vector corresponding to the input objectcharacteristic information, and preference values between the user corresponding to the other input area characteristic information, the user behavior vector and the object vector and each to-be-predicted object as expected output data; and training to obtain a vector and a user preference generation model. According to the scheme, the user behavior embedding, the to-be-predicted object embeddingand the user portrait preference estimation are carried out at the same time by using one model, so that the prediction of the user preference information has higher accuracy and wider coverage.

The invention discloses a method for locating relative positions of RFID tags based on deep learning. The method includes steps: S1. Arrange a series of RFID tags uniformly and horizontally in a random order, keep an RFID antenna parallel to the movement of the RFID tags, and use the The RFID antenna continuously reads the phase information data and the signal strength information data between the RFID tags; S2, setting a data arrangement structure, based on the data arrangement structure, the phase information data and the signal strength information data are prepared into training data, and adopt the deep neural network to train the training data, and initially obtain the relative positional relationship between all the RFID tags; The relationship is further sorted to determine the final relative positional relationship between all the RFID tags; the method of the present invention can accurately analyze the sequence of a series of densely arranged RFID tags.