124results about How to "Balance accuracy" patented technology

The invention discloses a locomotive license automatic identifying method and a realizing device thereof for shooting video in real time under a complex background; wherein, the identifying steps include: (1) locating a movement multi-target locomotive body area; (2) capturing a candidate license area image; (3) capturing a real license edge point; (4) capturing the smallest license area; (5) dividing characters and removing non-character images;(6) normalizing the character images and extracting the original coarse gridding characteristics of the character images; (7) using a trained BP neutral network for realizing the accurate identification of the license characters. The device for realizing the locomotive license automatic method includes a pick-up device and a computer; the pick-up device is connected with the computer by an IEEE1394 cable. The invention is suitable for the automatic identifying method and the realizing device thereof for dividing and identifying the locomotive licenses which are not interfered by the shoot traffic video image quality under a complex background with good identifying effects to the characters.

The invention discloses a vehicle license plate character recognition method based on rough grid feature extraction and a BP neural network. The method comprises the following steps: (1). preprocessing a vehicle license plate image to eliminate various interferences and obtain a minimum vehicle license plate region; (2). segmenting the vehicle license plate characters by combining vertical projection and a drop-fall algorithm; (3). screening the segmentation results to eliminate interferences of vertical frames, separators, rivets and the like; (4). normalizing the characters according to the position of a centre of mass; (5). taking each pixel of a normalized characters dot-matrix as a grid to extract the original features of the characters; (6). designing the BP neural network with a secondary classifier according to the situation of the vehicle license plate; and (7). reasonably constructing a training sample database to train the neural network, and adjusting the training samples by the recognition effect to realize accurate recognition of the network. The method effectively eliminates the noise interference, quickly and accurately segments the characters, steadily and effectively recognizes Chinese characters, and realizes the balance between real-time property and accuracy in the whole recognition course.

A method for incrementally coding and signalling motion information for a video compression system involving a motion adaptive transform and embedded coding of transformed video samples comprises the steps of: (a) producing an embedded bit-stream, representing each motion field in coarse to fine fashion; and (b) interleaving incremental contributions from said embedded motion fields with incremental contributions from said transformed video samples. A further embodiment of a method for estimating and signalling motion information for a motion adaptive transform based on temporal lifting steps comprises the steps of: (a) estimating and signalling motion parameters describing a first mapping from a source frame onto a target frame within one of the lifting steps; and (b) inferring a second mapping between either said source frame or said target frame, and another frame, based on the estimated and signalled motion parameters associated with said first mapping.

A multi-level anomaly detection method based on exponential smoothing, sliding window distribution statistics and an integrated learning model comprises the following steps of a statistic detection stage, an integrated learning training stage and an integrated learning classification stage, wherein in the statistic detection stage, a, a key feature set is determined according to the application scene; b, for discrete characteristics, a model is built through a sliding window distribution histogram, and a model is built through exponential smoothing for continuous characteristics; c, the observation features of all key features are input periodically; d, the process is ended. In the integrated learning training stage, a, a training data set is formed by marked normal and abnormal examples; b, a random forest classification model is trained. The method provides a general framework for anomaly detection problems comprising time sequence characteristics and complex behavior patterns and is suitable for online permanent detection, the random forest model is used in the integrated learning stage to achieve the advantages of parallelization and high generalization ability, and the method can be applied to multiple scenes like business violation detection in the telecom industry, credit card fraud detection in the financial industry and network attack detection.

The invention provides a real-time vehicle flow detecting and tracking method based on aerial photography data. The method comprises the following steps that 1, a pre-training model of a YOLO networkis trained on the basis of a pre-training part of weak supervised learning by using a weak supervised learning mode; and 2, the pre-training model of the YOLO network is improved on the basis of a real-time vehicle flow detecting part of aerial photography data by adopting a full convolutional neural network and a multi-target box detection method with prior information to obtain a YOLO detectingnetwork. The method has the advantages that improvement is achieved on the basis of a YOLO algorithm, the full convolutional neural network and the multi-target box detection method with the prior information are adopted, training is conducted by effectively utilizing the multi-view and multi-resolution image characteristics of an aerial photography data set of an unmanned aerial vehicle, the detecting accuracy rate of the algorithm is increased on the condition of not losing too much detecting time, and the balance of the accurate rate and the time is achieved.

The invention provides a video transmission method and a device with high fault tolerance and low latency which is applied to deep space communication, wherein, the method comprises a video coding step, a base-layer data transmission step, an enhancement-layer data transmission step and a video decoding step. In the video coding step, acquired video data is processed through scalable video coding, and a scalable video coding code stream includes a base-layer code stream and at least an enhancement-layer code stream; in the base-layer data transmission step, the base-layer code stream in the scalable video coding code stream is extracted and transmitted by using an automatic retransmission request method, and the network packet loss rate is calculated when the transmission of the base-layer code stream is finished; in the enhancement-layer data transmission step, the enhancement-layer code stream is extracted, the obtained parameter is calculated for channel coding and transmission according to the packet loss rate by combining the channel coding technology for Athlon code, and the received enhancement-layer code stream is processed through channel decoding; and in the video decoding step, the received base-layer code stream is processed through video decoding, and the enhancement-layer code stream which is processed through channel decoding is processed through video decoding by combining the received base-layer code stream.

The invention discloses a smart community big data-based commodity recommendation method, and relates to the technical field of internet information processing. The method comprises the following steps of: firstly, carrying out data pre-processing; secondly, building a sample set; carrying out feature engineering processing; carrying out model fusion; and finally recommending individualized commodity information for users, wherein the step of building sample set comprises sub-steps of extracting primary interactive behavior samples and secondary behavior samples in the last n days, extracting primary behavior samples in the last n+m days, and forming sampling samples; and features of sample set data comprise fundamental features and intersectional features. The method disclosed by the invention has the beneficial effects of reducing the content of negative samples in the sample set, improving the forecast accuracy of commodity purchase and providing more accurate and credible commodity recommendation services for the users. Meanwhile, the intersectional features are adopted to describe the data features of the sample set and forecast the user behaviors with specific meanings.

The invention discloses a multi-UAV (Unmanned Aerial Vehicle) collaborative map construction method oriented to data sharing, and aims to solve the problems of long time consumption and slow map return optimization in a map fusion process. According to the technical scheme, the multi-unmanned aerial vehicle cooperative SLAM system oriented to data sharing is improved, a global map overlapping detection module in the system detects whether matched key frames exist between maps or not within a period T, the number of the matched key frames is recorded, and a local map pair is generated; and a global map fusion module in the system clusters the maps according to the local map pairs, dynamically and adaptively selects a fusion sequence between the local maps in the local map pairs, and fuses the local maps with the maximum number of matched key frames in the same class. According to the method, map fusion can be adaptively carried out according to the change of the local map constructed bythe unmanned aerial vehicle and the sequence which is most beneficial to improving data sharing, so that map construction has correct fusion time, the fusion sequence can be adaptively adjusted, andthe map construction accuracy and efficiency are improved.

The invention provides an information releasing method and a releasing system. The method comprises the following steps of acquiring a seed crowd of a product corresponding to information; through an analysis model, analyzing the seed crowd so as to generate a plurality of distinguishing characteristics and weights of the distinguishing characteristics; based on the plurality of distinguishing characteristics and the weights of the distinguishing characteristics, expanding the seed crowd so as to form a plurality of candidate crowds; calculating an integral click rate, and / or transaction history data of the candidate crowds respectively; and according to the integral click rate, and / or the transaction history data, determining an information releasing crowd from the candidate crowds so as to carry out information releasing. In the invention, the seed crowd is analyzed so as to expand and generate the candidate crowds so that precision of information releasing is increased and a conflict between crowd precision and a crowd scale can be balanced.

The invention belongs to the technical field of unmanned aerial vehicles and bridge detection, and particularly relates to an automatic bridge detection method and system based on the cooperative operation of multiple unmanned aerial vehicles. The invention aims to solve the problems that the traditional bridge detection operation is difficult, the time required for a single drone is long, and theerror rate is high. The automatic bridge detection method and system based on the cooperative operation of multiple unmanned aerial vehicles comprises the following steps of: analyzing bridge environmental information and dividing the bridge detection task into a plurality of sub-tasks; obtaining a task clustering optimization function and dividing sub-task groups; planning unmanned aerial vehicle tracks corresponding to the sub-task groups separately and assigning the sub-task groups to the corresponding unmanned aerial vehicles; and transmitting photos taken by the unmanned aerial vehiclesto an image processing unit to extract the characteristics of bridge diseases and generate a bridge disease report. According to the invention, different shooting schemes are designed according to different bridge tasks; multiple unmanned aerial vehicles cooperate together to perform the bridge detection, and each unmanned aerial vehicle performs different sub-tasks; and anomaly remedy is performed by using the redundancy fault tolerance technology; therefore, the efficiency is high, the task continuity and integrity are good, the accuracy is high, and the error rate is low.

The invention discloses an offline map matching method based on a complex trajectory network division model. The method comprises the steps of: establishing the complex trajectory network division model, and generating a single trajectory set according to an intersection trajectory division model; inputting the single trajectory set into an optimized weight adaptation Hidden Markov model, and determining a hidden actual location sequence through an observable GNSS location sequence; and performing line segment level accurate matching and location point level accurate matching on an output result of the Hidden Markov model. The invention further discloses an offline map matching system based on the complex trajectory network division model. Through the above manner, the offline map matchingproblem of a complex trajectory network is solved, the relationship between efficiency and accuracy is effectively balanced, the practicability is higher, and the accuracy is higher.

The invention discloses a human face key point tracking method, an application and a device thereof. The method comprises the following steps of acquiring a current video image frame; positioning a human face key point from the current video image frame; judging whether a human face in the current video image frame is moved relative to a human face in a previous video image frame of the current video image frame or not; if the human face in the current video image frame is moved relative to the human face in the previous video image frame, determining the human face key point positioned in thecurrent video image frame as an effective human face key point in the current video image frame; if the human face in the current video image frame is not moved relative to the human face in the previous video image frame, determining the weighted sum result of the human face key point positioned in the current video image frame with an effective human face key point at a corresponding position of the previous video image frame as the effective human face key point in the current video image frame. According to the invention, the shaking of human face key points in a video during the trackingprocess is completely avoided.

The invention discloses a three-dimensional model to two-dimensional image space registering method of a human body bone joint system. Firstly, system position parameters at a bi-plane X-ray image collecting phase are obtained to construct a registering virtual environment; secondly, limb somatosensory motion information is utilized to initialize and align a bone joint three-dimensional model in the space; thirdly, space sampling points are set, and two-dimensional contour line sampling and extraction are carried out in sequence on the three-dimensional model; fourthly, deviation comparison is carried out on three-dimensional model contour lines and X-ray image contour lines, and optimal contour lines and corresponding sampling points are screened; and fifthly, the bone joint three-dimensional model is adjusted until registering is finished through the spatial position information of the optimal sampling points. According to the method, the rapid registering between the three-dimensional model and two-dimensional images of a bone joint system is realized under the condition of less manual intervention, complicated manual manipulation can be replaced, the efficiency and the registering precision are improved, and labor cost and time cost are saved.

The invention discloses an abnormal event detection method based on a fixed-position camera monitoring video, and the method comprises the steps: A, obtaining a frame of image shot by a camera under the normal condition, calibrating an invariant region, and enabling the image of the invariant region to serve as an initial reference image; B, acquiring a real-time image shot by the camera after a preset sampling interval time, and extracting an image of an invariant region of the real-time image as a judgment image according to the invariant region; C, comparing the similarity between the judgment image and the reference image, if the similarity is within a preset normal similarity range, replacing the current reference image with the judgment image, and then turning to the step B; Otherwise, judging that an abnormal event occurs. The invention further discloses an abnormal event detection device based on the monitoring video of the fixed-position camera. According to the invention, themethod can achieve the accurate and automatic recognition of the interference of the camera and the abnormity in a monitoring scene, is better in accuracy and real-time performance, and is low in software and hardware requirements.

The invention discloses a lithium battery SOH estimation method based on internal resistance detection. The lithium battery SOH estimation method is characterized in that the internal resistance of alithium battery is detected by a direct current discharge method, related health factor characteristic parameters are obtained, the characteristic parameters can effectively represent the change trendof a health state of the battery, a RBF neural network model of the health factor of the battery and the actual health state is established, the establishment of a complex equivalent circuit model isavoided, and the accuracy and the generalization of SOH estimation can be balanced.

The invention discloses a smart accumulator and a remote management system thereof. The smart accumulator comprises multiple accumulator monomers; any accumulator monomer comprises a battery upper cover; a selector switch and a communication interface are arranged on the battery upper cover; multiple accumulator monomers are in cascade connection through the communication interface so as to form an accumulator pack; the interior of any accumulator monomer contains an integrated chip, and the integrated chip is used for online monitoring the voltage, the internal resistance and the temperatureof the accumulator monomer in real time; the integrated chip comprises an A / D chip, a CPU processor, a voltage measurement circuit, an internal resistance measurement circuit, and a temperature measurement circuit; the A / D chip is connected with the CPU processor; and the CPU processor is connected with the communication interface. The smart accumulator disclosed by the invention can reflect the own operation working condition of the accumulator in real time so as to perform the voltage balance; and meanwhile, a fully functional accumulator professional information processing and comprehensiveanalysis internet information platform is provided so as to effectively improve an operation status of the accumulator, obviously improve the power supply reliability of the accumulator, and prolongthe life of the accumulator.

The invention discloses a time domain prediction-based saliency extraction method, which predicts a next frame saliency map from a current frame or at least one adjacent previous frame saliency map by using the time domain correlation of the saliency map. The time domain prediction-based saliency extraction method comprises the following steps: firstly, extracting features and saliency to obtain saliency submaps; secondly, performing time domain prediction on the saliency submaps; and finally combining the predicted saliency submaps to obtain the saliency prediction map of the next frame map. Through regional prediction technology of the saliency, the time domain prediction-based saliency extraction method greatly reduces the computational complexity of a saliency model greatly, and solves the difficult problem that the computational complexity of the prior saliency extraction method is too high to aim at the real time application of video coding.

The invention relates to a three-freedom helicopter fault tolerance control device based on a support vector machine and a method thereof, which belongs to an aircraft fault tolerance control field. The fault tolerance control device comprises five parts: an interference and fault injection device, a helicopter system platform, a power amplifier, a data acquiring card and a computer fault diagnosing device. The three-freedom helicopter fault tolerance control device is based on a generated gesture ring non-linear mathematical model. A PD controller, a support vector machine self-adaptive fault tolerance controller and a dynamic inverse controller are adopted for settling multiple system failures and a fault tolerance under multiple interferences. The three-freedom helicopter fault tolerance control device has advantages of high operability and easy realization. The three-freedom helicopter fault tolerance control device can be used in checking test for actuator faults of a three-freedom helicopter.

The invention discloses a target detection model training method and a rapid target detection method. The training method comprises the following steps: adding a target region feature enhancement layer during training; performing feature extraction on the training sample by using a feature extraction unit, and averaging channels of a feature map output by a previous-level feature extraction unit to obtain a first feature value matrix with a normalized channel number; traversing each pixel point in the training sample to generate a second eigenvalue matrix; multiplying the element values of thefirst eigenvalue matrix by the element values of the second eigenvalue matrix to obtain a third eigenvalue matrix; multiplying the third eigenvalue matrix by a preset adjustment function, then performing element value addition on each channel eigenmatrix of the eigenmap to obtain a target enhanced eigenmap, and inputting the target enhanced eigenmap into a next eigenextraction unit. According tothe invention, the network is fully trained, the relationship between the network depth and the detection precision is balanced, the background perception capability of the feature map is enhanced, the detection precision is high, the calculation is simple, and hardware platform transplantation is facilitated.

The invention relates to a tunnel section measurement device and a measurement method thereof. The measurement device comprises a measurement mechanism and a shooting mechanism, wherein the measurement mechanism comprises a moving component, as well as a rotating component, ranging devices, a transmission device, a control device and a power supply device which are arranged on the moving component; the rotating component comprises a displacement rotary table arranged on the measurement mechanism and a rotating shaft controlled by the displacement rotary table; the end part of the rotating shaft is provided with a mounting seat and extends out of the measurement mechanism; a plurality of ranging devices are arranged on the mounting seat according to a set distribution state, and can generate ranging light spots during ranging; the transmission device is used for generating arc profile light rays on the inner wall of a tunnel; and the shooting mechanism comprises a tripod and a camera arranged on the tripod, and is arranged at a certain distance to the measurement mechanism. Compared with the prior art, the tunnel section measurement device has the characteristics of relatively highprecision, relatively high speed, low cost, convenience in operation and the like, and can meet the requirement of quickly detecting a tunnel under the condition of ensuring certain precision.

Embodiments of the present specification provide a data encryption method and apparatus, and a data analysis method and apparatus. Converting the raw data into intermediate data by a stochastic algorithm, wherein the intermediate data takes a value as one of a plurality of discrete values, wherein the stochastic algorithm causes an expected value of the intermediate data to correspond to the raw data, wherein the plurality of discrete values are binary values having the same number of bits; And performing a random flip between 0 and 1 on each bit of the intermediate data to obtain encrypted data of the original data, wherein the random flip satisfies differential privacy.

The embodiment of the invention provides a federated learning-oriented privacy protection method and device. The federated learning-oriented privacy protection method comprises a parameter setting step, a data partitioning step, a first training step, a second training step, a first calculation step, a second calculation step and an adversarial sample generation step. According to the embodiment, the idea of confronting samples is adopted, a certain amount of noise is added in parameter updating to disturb the distribution characteristics of the parameters, the privacy inference result is randomly output according to the probability distribution expected by the user after the noise passes through the privacy attribute inference model, so that the privacy attribute inference attack is resisted, and the problem of privacy attribute leakage of federated learning is relieved.

The invention discloses a deep neural network model compression method based on pruning threshold automatic search, and belongs to the field of deep neural network model compression. The method comprises the following steps: training a model to obtain an initial model for pruning; performing adaptive grid search on the model parameters to obtain a first pruning threshold; further searching for a threshold interval corresponding to the first pruning threshold by combining a binary search method, and searching for a better threshold to obtain a second pruning threshold; performing iterative pruning processing on the original network model based on the second pruning threshold; and carrying out sparse storage on the pruned model to obtain an available compressed network model. According to the deep neural network model compression method based on pruning threshold automatic search, an existing main deep neural network model can be compressed, the technical problem that the deep neural network model cannot be deployed to embedded equipment due to the fact that the model is large is solved, and the application range of the deep neural network model is expanded.

The invention relates to a vehicle parking place management method, device and system. The vehicle parking place management method comprises the steps that images, collected by an image collection device, of an area including one or more parking places are obtained, and by judging the parking place occupation states in the two images and by judging whether corresponding vehicles in the same parking place are the same or not, the parking starting and stopping time of the vehicles in the multiple parking places can be determined simultaneously. The vehicle parking places can be managed through asmall number of the images, the requirements for equipment computing force can be greatly reduced, the flow and the hardware cost are lowered, the time interval of the two images is obtained throughrandom adjusting, the accuracy rate of parking charging, the image uploading time and the computing time are balanced conveniently, and in addition, the management efficiency of the multiple parking places is further improved.

The invention discloses a classroom quality evaluation system and method based on deep learning target detection, and the system comprises: a data obtaining module which is used for obtaining a real-time monitoring video stream of a classroom and a corresponding timestamp; a video preprocessing module that is used for preprocessing the real-time monitoring video stream; a detection and calculationmodule that is used for inputting a pre-trained video frame into the model according to the preprocessed real-time monitoring video stream so as to detect the attendance number and / or the head-up number of classroom monitoring video pictures; and a statistical evaluation module that is used for generating an evaluation result according to the attendance number and / or the head-up number. Accordingto the system, the deep learning target detection model is used for obtaining the real-time attendance condition and the head-up rate of a classroom; the multi-scale features are detected by using the feature pyramid to improve the detection effect, so that the good balance between the calculated amount and the accuracy is realized, the situation of the whole class can be counted, the teaching mode adjustment of teachers is facilitated, and the optimal teaching effect is realized.

The invention provides a simulation data generation method for simulating a real physical environment. The simulation data generation method includes the steps: 1) receiving a series of real environment data; 2) dividing an acquisition time slot of the real environment data into a plurality of time intervals; 3) fitting space distribution of physical environment parameters based on a Gaussian process model according to data values of a plurality of different positions at the same time for each time interval to obtain the value of the physical environment parameter at an optional designated position at the time corresponding to the time interval; 4) modeling time trend according to the obtained data values at each position at different time to obtain a time trend function at the position; 5) generating simulation data of the physical environment parameters according to the designated time and position. Data points at different positions at each time interval are regarded as data points acquired at the same time. Massive simulation environment data can be generated at small overhead, and the simulation data generation method is high in configurability and universality.

The invention provides an obstacle detection method, an obstacle detection device and a robot. The method and the device are applied to the robot provided with a binocular sensor. The method comprises steps: a left image and a right image photographed by the binocular sensor arranged on the robot towards the outer part of the robot are acquired; with the left image and the right image as reference respectively, feature points in the left image and feature points in the right image are matched, and a first matching point pair set and a second matching point pair set are obtained; according to overlapped feature point pairs in the first matching point pair set and the second matching point pair set, a matched feature point pair between the left image and the right image is determined; and according to distance measurement parameters of the binocular sensor and the safety movement range of the robot, whether the actual position point corresponding to the matched feature point pair is an obstacle is judged. Thus, the problems that an obstacle detection method applied to the robot in a related technology is hard to acquire balance between obstacle detection accuracy and cost performance can be solved.

The invention relates to a topology construction and internet access method for an aviation ad-hoc network. In the method, a flight process is divided into five stages including a dormancy stage, a waiting stage, a take-off stage, a cruising stage and a landing stage; a corresponding topology construction and data transmission strategy is provided for a flight node in a specific flight stage, wherein a waiting node is used as a connection point between an ad-hoc network in the air and a ground network, passively accepts network access requests of other nodes, and is mainly responsible for data relay; on one hand, a take-off node rapidly establishes a link information table; on the other hand, the take-off node accepts a request of the waiting node, carries data of the ground network, and rapidly diffuses data of the ground network in the ad-hoc network in the air; a cruising node regularly updates the link information table, and selects the optimal and backup link to perform data exchange in the air according to a link distance and the link duration; and a landing node rapidly forwards the data of the network in the air to the waiting node, and finally forwards the data into the ground network. By means of the topology construction and internet access method for the aviation ad-hoc network disclosed by the invention, fully connected topology can be rapidly formed; and communication between the aviation ad-hoc network and the ground network is realized.

The invention relates to a parallel method and device for ultrasonic standard sections, computer equipment and a storage medium. The method comprises the following steps: acquiring an ultrasonic videoof a scanning part; extracting features of each ultrasonic image in the ultrasonic video to obtain a feature image of each ultrasonic image; performing multi-scale target parallel detection on each ultrasonic image according to the feature image of each ultrasonic image to obtain a global confidence coefficient and a local confidence coefficient of each ultrasonic image; evaluating each ultrasonic image according to the global confidence coefficient and the local confidence coefficient, and selecting the ultrasonic image with the highest evaluation as the optimal standard section of the scanning part. The method can balance the accuracy and speed.

A method and a system for retrieving a fake-licensed vehicle are disclosed. The method comprises: pre-training a classification model for vehicle filtering, obtaining the type of target vehicle, color, sub-brand, license plate number and feature area information, removing the license plate number information of the target vehicle to obtain a second image, removing a license plate number information of an original test set vehicle from that vehicle passing record to obtain a second test set, filtering a second test set is associated with the target vehicle type, colors, sub-brands and differentvehicle images to get a third test set, selecting a vehicle image set according with the second image in similarity demands from the third test set, restoring the license plate number information ofthe vehicle in the image set, and matching and verifying the license plate information of the vehicle in the image set and the license plate information of the first image, so as to achieve the purpose of matching and verifying. The vehicle image set in the third test set and the vehicle image in the second image meet the similarity requirement. The technical scheme of the invention solves the technical problems of precision retrieval and verification that the prior art can not realize for the brand-set car.