1298results about How to "Small amount of calculation" patented technology

The invention discloses a license plate detection method based on deep learning, comprising the following steps: using a fast-rcnn algorithm to train an RPN convolution neural network and a fast-rcnn convolution neural network; building an image library with marks and tags as a sample set; using the trained RPN convolution neural network to process the images in the sample set to get a rough license plate area; sending a rough license plate box to the trained fast-rcnn convolution neural network for judgment; and judging whether the rough license plate area is an optimal license plate area according to the output vector of the fast-rcnn convolution neural network, and if the rough license plate area is an optimal license plate area, taking the rough license plate area as a final license plate area. According to the invention, a multi-scale and multi-proportion reference box is adopted in training of the RPN convolution neural network, so detection of license plates with unconventional scale and proportion is promoted effectively. The RPN convolution neural network and the fast-rcnn convolution neural network share convolution layer parameters, so the whole system is simpler, less amount of calculation is needed, and the rate of missed detection is lower. Moreover, the real-time requirement of the system can be satisfied.

The invention belongs to the field of the body area network and relates to a gait recognition method based on an inertial sensor. The method includes four phases. In the data acquisition phase, four sensor nodes attached to legs of a testee and a sensor node attached to the waist of the testee are used to acquire motion angular speeds, accelerations and magnetic field strengths of the legs and waist of the body of the testee and wirelessly transmit data to a base station, and then the data are transmitted to the upper computer for processing. In the period dividing phase, gait periods are divided according to angular speed change curves of the shanks so as to obtain segment data corresponding to gaits. In the feature extraction phase, waveform features and behavioral features are extracted from the segment data and used to describe fluctuation of sensor signals and sport performance of gait behaviors. In the gait classification phase, the gait features are transmitted to a classifying model, types of the gaits are acquired by computing, and thus gait recognition is achieved. The method is suitable for use in gait recognition in the fields, such as sports training, medical care and game design.

The invention discloses an LSTM deep learning model-based hydropower unit fault diagnosis method and system. The method comprises the steps of obtaining sampling sequences of N different signal channels of a hydropower unit, and performing VMD on each time sequence to obtain K IMF components; constructing corresponding training sets and to-be-diagnosed sets; building an LSTM model for the trainingset of each IMF component, and performing feature extraction on each IMF component through two LSTM layers; connecting outputs of K LSTM layers of the same signal channel to a Dense layer; through aSoftmax layer, performing feature classification on outputs of multiple Dense layers; and training a deep learning neural network model through an RMSProp gradient descent algorithm, and diagnosing the to-be-diagnosed sets by the trained model. According to the method and the system, the relatively good signal-noise separation effect of the VMD is combined with the processing advantage of an LSTMto the time sequences, so that the hydropower unit fault diagnosis accuracy is effectively improved.

The invention discloses an imaging method by a large squint angle airborne SAR spotlight mode based on non-uniform sampling. Based on an analysis of azimuth signals characteristics of targets at different azimuths on the same range gate under the condition of a large squint angle, the method enlarges the azimuth imaging range output by effective focuses by a non-uniform interpolation resampling processing method. In the method, a data acquisition model under the large squint angle is equated with a right side looking data acquisition model of increment motion compensation. The method comprises the following steps: firstly, performing cubic phase compensation in an original data field, then performing range walk correction on a time domain and range migration correction on a 2D frequency domain, completing azimuth compression by the non-uniform interpolation resampling method, and finally, completing autofocus processing by a phase gradient autofocus algorithm. Compared with the prior art, the method can help process a larger squint angle, and has the advantages of small calculation amount and easy real-time processing. The method can be applicable to the conventional radar systems without a function of tracing Dechirp demodulation by a scene center.

The invention discloses a mobile terminal cloud resource scheduling method based on a Hadoop framework, and belongs to the field of cloud computing platform data resource access. The mobile terminal cloud resource scheduling method is characterized in that a task queue is defined by nodes in a cloud computing platform based on the Hadoop framework, tasks are distributed into the task queue, where a data copy is located, of the working nodes, and when the length of the task queue, where the copy is located, of the nodes is larger than the waiting length threshold value, the tasks are removed from the task queue of the copy and added into the task queue of the shortest node in the task queue inside the framework where the copy is located to wait for being scheduled. According to the mobile terminal cloud resource scheduling method, the characteristics of being small in requested calculated amount, high in concurrency and high in user real-time requirement, of a mobile terminal are combined, the localized executing probability of the operation is greatly improved, and under the situation of a large amount of concurrent requests and the generation of hotspot data, load balancing of all the nodes in a cluster can be kept.

The invention relates to an unmanned vehicle real-time positioning method based on laser reflection strength. The method comprises the following steps of S1, photographing two sides of a city road by an on-vehicle laser radar, obtaining a plurality of frames of point cloud data, extracting road edge points and converting the road edge points into a current vehicle coordinate system; S2, selecting the road edge points of which the z-axis coordinate value is in a preset range and obtaining coordinates of the road edge points in a GPS coordinate system, using the current GPS coordinate point of the unmanned vehicle as an original point, dividing a coordinate space for obtaining a grid map; S3, matching the gridded high-precision map with the grid map which is obtained in the step S2, thereby obtaining position of the unmanned vehicle on a high-precision map; and S4, predicating vehicle posture by means of a Kalman filter. Compared with prior art, the unmanned vehicle real-time positioning method can realize real-time accurate positioning in a complicated environment and can effectively improve driving safety of the unmanned aerial vehicle.

The invention provides a commodity recommendation method and system based on video content to solve the problem that the accuracy of commodity recommendation results is low. The method comprises the steps that keyframes are extracted from target video; object detection is conducted to each keyframe according to objectivity of objects, and the detected objects serve as candidate objects of each keyframe; commodity identification is conducted to the candidate objects of each keyframe through a convolutional neural network model, and commodities of each keyframe are obtained; keyword mapping is conducted to the commodities of each keyframe, and corresponding commodity tags are obtained to serve as the commodity tags of each keyframe; the commodity tags of each keyframe are determined as commodity recommendation information of the target video, and the commodity recommendation information is output. According to the commodity recommendation method and system based on the video content, the video content is taken into consideration when commodity recommendation is conducted, the video content is analyzed, and the obtained commodity recommendation information is closely bound up with the content of the target video, so that the accuracy of the the commodity recommendation results is improved.

The invention discloses a vehicle path planning method based on a storage unmanned vehicle. The method comprises steps of firstly determining the operation node of an unmanned vehicle in a storage environment, and making a topological map based on the environment; then through the improvement of the traditional A* algorithm, calculating the shortest path from a starting point to a target point of the unmanned vehicle in an offline way; in the process of tracking the path of the unmanned vehicle, detecting whether there are obstacles on the path by a self-brought sensor, if there are obstacles and the obstacles do not completely obstruct the path, then switching to the artificial potential field method to perform online real-time obstacle avoidance; and if the obstacles completely obstruct the path, switching to the A* algorithm to replan the path until the unmanned vehicle reaches the target point. This method not only can make full use of known information to generate the global optimal path, but also can effectively avoid the random moving obstacle on the path.

The invention discloses a moving object detection and tracking method. The method comprises the following steps: firstly, obtaining a frame image from a video sequence, converting the image into a gray scale image, and filtering and denoising are performed; then, the moving target region is obtained by using three-frame difference method, and the hole in the difference image is filled by morphological filter closure operation; the pixel points with non-zero gray value in the image after morphological filter closure operation are calculated by optical flow calculation, and the moving target isextracted; then the Kalman filtering algorithm is used to predict the centroid position of the moving object in the current frame image, the predicted value is compared with the centroid position tracked by a Camshift algorithm, and the centroid coordinates of the moving object are determined by the comparison results, and the size and position of the search window are determined when searching the next frame image; the model is updated to get the next frame image, the whole tracking process is repeated, and finally target tracking is achieved. The invention reduces the error of moving targetdetection and tracking, and improves the accuracy.

The invention belongs to the technical field of communication radar. An equilateral triangular array is selected as a rough measurement array in a circular array, the phase difference between two baselines in a triangular matrix is calculated, and possible actual phase differences are exhaustively listed according to the maximum ambiguity degree; the possible actual phase differences are substituted to a circular array phase difference solving formula to build a phase difference vector sample library; the ambiguity degree of the actual circular array phase difference is obtained by a related operation, and further an actual circular array phase difference vector is obtained; and finally, an incident angle value as an accurate DOA (Direction of Arrival) estimate is inversely solved by finding the least squares solution of direction cosine. The novel phase interferometer two-dimensional direction finding algorithm for ambiguity resolution by extension baselines is applied, so that higher precision can be obtained compared with a direction finding algorithm based on a related operation, and the required amount of computation is greatly reduced. The method is a high-performance angle measurement algorithm.

The invention discloses a motion noise interference eliminating method suitable for a wearable heart rate monitoring device based on a photo plethyamo graphy. The method can effectively eliminate motion noise interference and improve heart rate value monitoring precision. According to the method, an NLMS self-adapting filter and a Mallat algorithm are combined, and six-axis acceleration signals generated by a triaxial accelerometer and a gyroscope are selected as motion reference signals of the self-adapting filter to eliminate part of motion noise interference. The method takes full consideration of complexity and calculation amount of the algorithm, the characteristic that motion noise and heart rate noise are equal in frequency, and other noises and the heart rate noise are not overlapped in frequency is utilized, and through two efficient and low-complexity algorithms, different noises are eliminated step by step, and the purpose of calculating the heart rate in real time is achieved. The low complexity of the method guarantees practicability of the method on the wearable heart rate monitoring device.

The invention provides a method for extracting a road in an SAR image, which comprises the following steps of: carrying out early stage processing on an SAR image and adjusting contrast ratio and luminance; carrying out binarization on the SAR image after early stage processing so as to carry out filter processing in the following step; carrying out mathematic morphological filtration on the binarized SAR image so that the SAR image is more smooth and clean; carrying out road geometrical filtration on the SAR image after mathematic morphological filtration to remove a false-alarm area; carrying out Canny operator processing on the SAR image after road geometrical filtration to search the edge of the road; and finally extracting the road from the SAR image by detecting road through Hough conversion. By processing in all the steps, the road area in the SAR image can be extracted, the calculated amount is smaller, and the SAR image can be more rapidly processed; and a main road can be more rapidly extracted basically by automation and less manual intervention.

A method for detecting voice end- point used in voice identification system includes carrying out framing process on inputted voice data, carrying out FFT operation on voice data by using frame as unit, dividing the whole voice spectrum to be sub band with different S ¿C N ratio in high and low and calculating out noise threshold for each sub band, carrying out preliminary judgment of voice end ¿C point according to noise threshold of each sub band and carrying out accurate judgment of voice end ¿C point according to resonant peak value.

The embodiment of the application provides a robot repositioning and environment map construction method, a robot and a storage medium. According to the robot repositioning and environment map construction method, firstly, when performing repositioning on robot, the environmental image captured by a visual sensor is fast matched in a visual map to obtain a preliminary estimated pose by combining alaser sensor and the visual sensor, and the environmental data collected by the laser sensor on a grid map are locally matched to obtain accurate pose by combining the preliminary estimated pose. Since the matching range of the grid map is reduced by using the preliminary estimated pose, the calculation amount in the grid matching is reduced, which is beneficial to the improvement of the relocation efficiency.

An embodiment of the invention discloses an advertisement putting method and system. The method and system aim to resolve the problems that calculating pressure of an advertisement putting server is large, response time is long, and advertisement putting speed is low. The method comprises the steps that historical click rates of various dimension characteristics of all candidate put advertisements are calculated, and weight of all the dimension characteristics is calculated according to the historical click rates of all the dimension characteristics; the dimension characteristics is less than advertisement putting characteristics; click rate prediction calculation is conducted on the historical click rates of all the dimension characteristics and the weight of all the dimension characteristics to obtain click rate prediction values of all the candidate put advertisements; the click rate prediction values of all the candidate put advertisements are mapped into advertisement click rates of all the candidate put advertisements, and advertisement putting is conducted according to the advertisement click rates of all the candidate put advertisements. As overmany advertisement putting characteristics do not need to be considered, calculating amount of the advertisement putting server is reduced, processing pressure of the advertisement putting server is relieved, and the response speed of the advertisement putting server is improved.

The invention relates to testing apparatus, in particular to a test simulator for displacement and extraction of natural gas hydrates through CO2. The testing apparatus comprises a reaction kettle, a constant temperature control system, a gas injection and liquid injection system, a liquid loop heating system, a return pressure control system, a vacuumization and gas-liquid separation system, a gas generation online detection system and a data acquisition system. Real-time monitoring of saturation of the hydrates is achieved by the adoption of the time domain reflection technique, the testing apparatus is easy and convenient to use and small in calculated quantity, and data supports are provided for accurate discussion of the replacement reaction rate; the gas generation pipeline of the reaction kettle is connected with a gas chromatograph, so real-time measurement of gas generation components can be achieved; the simulation environment is more approximate to the undersea extraction environment.

The invention relates to the technical field of speech recognition and discloses an end-to-end speech recognition method. The method includes acquiring a first mixed speech signal of a plurality of speakers and tag sequences as training samples; building an Encoder-Decoder Architecture based neural network model; training the neural network model; acquiring a second mixed speech signal including aplurality of speakers to be recognized; inputting the second mixed speech signal into the trained neural network model to output text information corresponding to each speaker respectively. The invention outputs the pronunciation content corresponding to each speaker respectively for the mixed speech source input formed by the simultaneous vocalization of a plurality of speakers, and does not need to include an obvious speech segmentation stage to generate a plurality of independent outputs from the mono-channel mixed speech, thereby simplifying the speech recognition process and reducing thecomputational load. The invention also discloses an electronic device and a computer-readable storage medium.

The invention relates to a face recognition based method for authenticating identity. In the method, recognition characteristic data and a corresponding recognition characteristic region are extracted through face recognition; a drawn portrait painting is generated by a face image; an authenticated person is prompted to adjust the distance and angle with an image acquisition device by the portrait painting when the face image is acquired; when the acquired image is aligned with the portrait painting, a terminal extracts the recognition characteristic region and calculates the recognition characteristic data; and the recognition characteristic data are associated with the saved corresponding recognition characteristic data to authenticate user identity. The method can greatly reduce the calculated amount and the data communication amount of face recognition, and is particularly applicable to identity authentication application of mobile terminal equipment.

The invention discloses different system measuring method and equipment. The method comprises the following steps: a network side informs the up and down conversion information of a terminal different system neighbor cell and/or the network side informs whether the terminal different system neighbor cell is synchronous with the cell; a terminal acquires the data of a time slot or subframe to be measured in a different system according to the up and down conversion configuration information; and measurement is carried out and/or the terminal determines whether synchronous search is required when a neighbor cell is measured according to the synchronous condition of the different system neighbor cell. The invention can reduce the calculation amount of terminal measurement, decrease complexity and improve measuring speed and accuracy when the terminal carries out different system neighbor cell measurement, and also can avoid blindness on operating the terminal.

A two-dimensional transmission cross coefficient is obtained based on a function representing a light intensity distribution formed by an illumination apparatus on a pupil plane of the projection optical system and a pupil function of the projection optical system. Based on the two-dimensional transmission cross coefficient and data of a pattern on an object plane of the projection optical system, an approximate aerial image is calculated by using at least one of plural components of an aerial image on an image plane of the projection optical system. Data of a pattern of an original is produced based on the approximate aerial image.

The invention discloses a patent document abstract-based automatic patent classification method. The method comprises the following steps: dictionary construction, generation of various layers and classes or feature vectors of an IPC, patent text feature selection, patent text vectorization, SVM-based classification model construction and to-be-classified patent classification. In the patent text feature selection, the information contained in the patent titles and abstracts are fully utilized so that the calculation amount of the patent feature selection is remarkably decreased; in the calculation process of the patent text vectorization, a TF-IDF calculation method is improved, and the part of speech weights and position weights of words are added, so that the classification effect is further improved; and in the patent classification, a layer-based classification method is adopted, and the method fully utilizes the advantages of an SVM classifier and a KNN classifier, so that the disadvantages that the SVM classifier is overmuch in training and the KNN classifier is large in calculation amount are overcome.

The invention relates to a three dimensions indoor positioning method based on visible light signal intensity, belonging to the visible light communication technical field. The method comprises: first, according to frame header intensity information, utilizing the frame header intensities of LED signals received by a receiver to determine the height information Z of a light detector by establishing a fingerprint database; and second, utilizing the RSS measurement method to determine horizontal information X and Y. The method has a simple structure, is easy to operate and realize, does not need additional sensors, realizes height coordinate positioning under the condition of maintaining the hardware of an original two dimensional visible light positioning system, and thereby realizes indoor space positioning. Compared with a pure fingerprint database indoor space positioning method, the method substantially reduces the calculated amount in a whole process, and increases a positioning speed.

The invention relates to an identification method of image targets of a synthetic aperture radar based on noise independent component analysis, belonging to the technical filed of synthetic aperture radar image target identification. The method comprises the following steps: firstly, preprocessing input images of an original training sample of the synthetic aperture radar to ensure that the images conform to whitening and zero mean normalization and the probability density of image noise complies with Gaussian distribution; secondly, performing logarithm noise independent component analysis on the preprocessed training sample and a real-time measurement sample to be identified, extracting the independent component characteristics of the images to be identified; and finally, identifying the sample to be identified by an independent component analysis method. The independent component characteristics extracted by the method of the invention are more suitable for classification, thus eliminating the image denoising process necessarily to be carried out in the preprocessing stage in an ideal ICA algorithm, improving the robustness of the algorithm on abnormal data, improving the instantaneity and reliability of SAR image target identification, and being suitable for SAR image characteristic extraction and automatic target identification.

There is provided a template matching method having a small amount of calculation and high detection accuracy. The amount of calculation is reduced while minimizing the degradation of the detection accuracy by calculating the similarity between a template image region and a reference image region at rough position intervals and estimating by interpolation of the similarity at a position where it is not calculated from the similarity calculated at a peripheral position. At the time, even if the gradient of a similarity greatly changes at the minimal point, estimation can be realized with high accuracy by executing estimation making use of the directional gradients of the similarity obtained in plural different directions independently. Additionally, occurrence of wrong detection is avoided by the restriction to disable the difference between the estimated similarity and adjacent similarity to exceed a threshold, which is defined based on the similarity between the template image region and the image region obtained by moving the template image region in the same direction as or in the opposite direction to the direction of estimation.

The invention relates to a target tracking method based on the fusion of a forward looking camera and a millimeter wave radar. The method comprises the following steps: 1) jointly calibrating the forward looking camera and the millimeter wave radar; 2) tracking the dynamic target according to the millimeter wave radar measurement, using the Kalman filter to update the state of a detected target state, and obtaining the tracking track of the millimeter wave radar to the target; 3) obtaining the position and the speed information of the target through the forward looking camera, tracking the dynamic target, using the Kalman filter to update the state of the detected target state, and obtaining the tracking track of the forward-looking camera to the target; 4) fusing the tracking target stateof the millimeter wave radar and the forward looking camera. According to the target tracking method based on the fusion of the forward looking camera and the millimeter wave radar, compared with theprior art, the defects of single sensor miss detection, false detection, tracking failure, and state detection inaccuracy are compensated by tracking the targets detected by two different sensors, and the safety of an intelligent car is increased by using the redundant information.

Disclosed is a decomposition calculating method of the optimal power flow of a power system with a unified power flow controller. The method is characterized by comprising the following steps of firstly obtaining steady-state operating parameters required by the power flow calculation of the power system; secondly, determining the mounting position and the control parameter range of the unified power flow controller; thirdly, based on network equivalence, equalizing the branches of the unified power flow controller to two PQ nodes containing control targets; fourthly, utilizing taboo research to perform multiple optimal power flow calculation to determine an active control target of the unified power flow controller; and fifthly, calculating the control parameters according to the obtained control target of the unified power flow controller and the operation state of the system. The decomposition calculating method of the optimal power flow system with the unified power flow controller can equivalently decompose the optimal power flow calculation problem of the power system with the unified power flow controller into a two-layer optimization problem through network, can perform the optimal power flow calculation of the power system more effectively and more reliably, and has engineering use value.

The invention provides a video image target tracking processing method and system. The video image target tracking processing method comprises the following steps: modeling a background of a current video image, conducting foreground target detection on foregrounds in the current video image to obtain a foreground block mass in the current video image, determining tracked objects, extracting a color histogram at the position of the foreground block mass where the tracked objects are located, after the tracked objects are determined, using size information and historical position information of the foreground block mass of the tracked objects of the current video image to conduct initial training on a Kalman predictor corresponding to the tracked objects, updating the color histogram of the tracked objects in the current video image, and tracking the tracked objects in a follow-up continuous video of the current video image according to the number, detected according to the current video image, of the tracked objects. The video image target tracking processing method and system are simple, high in efficiency, high in adaptive capacity and capable of accurately tracking and processing the video objects.

The invention provides a commodity recommendation optimizing method based on collaborative filtering. The commodity recommendation optimizing method comprises the steps that tag types corresponding to commodities are set according to received setting instructions; the internet behavior data information of a user is acquired, and the attention value of the user on each commodity in the commodities is determined according to the internet behavior data information; the tag types corresponding to the commodities are determined to determine the values of the attention of the user to the tag types corresponding to the commodities; according to the attention values, a user recommendation group which the user belongs to is determined; commodity recommendation is carried out on the user according to the attention values of the other users in the recommendation group on the commodities. Correspondingly, the invention further provides a commodity recommendation optimizing system based on collaborative filtering. According to the technical scheme, tags are introduced in the commodity recommendation optimizing method based on collaborative filtering, due to the fact that the number of the tags is far smaller than the number of the commodities, calculation of the pertinence of the user and the commodities is greatly reduced, the calculation quantity is reduced, and meanwhile the recommendation accuracy is guaranteed.

The invention discloses a simple method for detecting the phase sequence of three-phase industrial-frequency alternating current and a detection device. In the invention, the zero-crossing rising edges of the three-phase voltage are respectively sampled, measured at a single point and compared by using the principle of phase time difference of the three-phase industrial-frequency alternating current, and the phase sequence and phase position of the three-phase industrial-frequency alternating current are judged by calculating the phase time difference. The method comprises the following steps: measuring the voltage rise zero-crossing point of any phase (defined as A) of the three-phase alternating current, and starting to calculate time with a clock by using the rising edge zero-crossing point of the Phase A as the reference point; then sequentially measuring the voltage rise zero-crossing points of the other two phases, and measuring the time differences T1 and T2 between the phase position rise zero-crossing point and the reference point of each of the two phases; and judging Phase B and Phase C through calculation. The method can measure three-phase industrial-frequency alternating current. The integrated multifunctional detection device has the advantages of miniaturization, systematization, simple structure and small calculated quantity, and can reduce the labor intensity of electrical installation and maintenance personnel.

The present invention discloses an audio classifying method, which comprises preprocessing an input audio signal and then calculating the linear predictive coding coefficient of the processed audio signal; getting the spectral envelope of the signal according to the linear predictive coding coefficient and then determining the amplitude difference value of the coefficient by an index spectrum obtained by calculation; setting a threshold according to the statistical results of the amplitude difference values and then classifying the audio signal according to the threshold. The present invention can significantly reduce calculation amount brought by the classification of audio signals and, at the same time, have high accuracy in audio signal classification. In addition, when being applied to signal processing flow in extended bandwidth self-adaptive multi-rate coding standards, the present invention can reduce the calculation amount of audio signal classification to extremely low and, in addition, can ensure that the signal processing flow codes directly using corresponding coding modes without the need of pre-coding procedures, thereby improving the coding efficiency.