34results about How to "Moderate computational complexity" patented technology

The invention provides a human behavior identification method based on mobile equipment. The human behavior identification method comprises the steps of: acquiring data in real time by utilizing various kinds of sensors arranged inside the mobile equipment; carrying out a series of data preprocessing operations such as correction, filtering, data calculation and generation as well as data segmentation on the data obtained by the sensors; extracting features of the preprocessed data, inputting extracted corresponding feature vectors into an equipment position classification model to obtain equipment position classes; and selecting corresponding behavior classification models according to the obtained equipment position classes, and inputting the extracted corresponding feature vectors into the behavior classification models to obtain a final behavior recognition result. The human behavior identification method based on the mobile equipment provided by the invention has good universality and high accuracy.

The invention relates to a passive multi-target detecting and tracking method based on wireless sensor networks. The technical scheme of the passive multi-target detecting and tracking method based on the wireless sensor networks comprises the following steps of utilizing a scanning circle windowing detection method to create a sliding scanning circle model of a passive multi-target according to received signal strength of different wireless links in the wireless sensor networks, and utilizing a HAC (Hierarchical Agglomerative clustering) algorithm to perform a clustering analysis so as to extract a detection result; and obtaining a multi-target tracking result of target number changes by utilizing a PHD (Probability Hypothesis Density) passive multi-target particle filter tracking algorithm and according to the detection result. The passive multi-target detecting and tracking method based on the wireless sensor networks is reasonable in design, has high accuracy and robustness in detecting and tracking algorithm, can detect and track multiple targets in a complicated multipath environment, meanwhile, is moderate in computation of target detecting and tracking algorithm, can guarantee the real time of operation of a detecting and tracking system.

The invention discloses a hopping beam resource allocation method and system based on deep reinforcement learning, a storage medium and equipment, and belongs to the technical field of communication. In order to solve the problem that the time delay performance of different service volumes is poor due to the fact that an existing beam-hopping satellite communication system lacks continuity when a service scene changes continuously during resource allocation, ground service requests are divided into a real-time data service and a non-real-time data service, and optimization functions are established respectively; then the maximum effective time length Tth of data in the satellite buffer is divided into M equal-length segments, and the M equal-length segments correspond to M hopping beam time slots; a ground cell service volume request formed by data packet time delay, the number of real-time data packets and non-real-time data packets is taken as an environment state S, a satellite beam is taken as an intelligent agent, cell illumination is taken as an action, and an optimization problem of resource allocation in a satellite beam hopping technology is taken as a Markov decision process; and hopping beam resource allocation is conducted based on the deep Q network. The method and system is mainly used for allocating hopping beam resources.

The invention relates to a battery temperature estimation method based on a thermal-neural network coupling model, and belongs to the technical field of battery management. The method comprises the following steps: S1, selecting a to-be-tested battery, collecting and sorting the specification and key geometric parameters of the battery, and obtaining an experimental data set required by battery model establishment and temperature estimation; s2, a low-order thermal model of the battery is established based on a Chebyshev Galerkin approximation method by considering the thermal effect of the tab, parameter identification is carried out to obtain unknown parameters of the thermal model, and the key temperature of the battery is estimated in real time in combination with an extended Kalman filter (EKF) algorithm; s3, establishing and training a battery data driving model based on a long-short-term memory neural network, and determining a mapping relation between battery heat production, a state of charge (SOC) and an environment temperature and a battery key temperature; and S4, coupling the physical thermal model and the neural network model through an integrated learning algorithm adaboost, and optimizing the fusion weight of the physical thermal model and the neural network model, thereby realizing accurate battery temperature estimation.

The invention discloses a multi-fractal feature aircraft target classification method based on principal component analysis, which belongs to the technical field of radars, relates to an aircraft target classification method based on multi-fractal features, and mainly solves the problem of low recognition rate of aircraft target classification caused by the factors such as low pulse repetition frequency and short irradiation time of a low-resolution radar. The method comprises the implementation processes as follows: preprocessing original radar echo data; performing fractional Fourier transform on the processed radar echo data; analyzing the multi-fractal features of the radar echo data in an optimal fractional Fourier domain and extracting the multi-fractal features to form feature vectors; performing principal component analysis on the feature vectors after normalization, and performing classification and identification on the aircraft targets by using the extracted effective features; training a classifier by training sample feature vectors; and inputting the testing sample feature vectors into the classifier for classification. The multi-fractal feature aircraft target classification method based on principal component analysis still has good classification effect under the condition of low pulse repetition frequency and short irradiation time, and can be used for classification identification of aircraft targets.

The invention provides a semi-Markov decision process-based task unloading method for a vehicle-mounted fog computing system. According to the semi-Markov decision process-based task unloading methodprovided by the invention, various time delays can be comprehensively considered according to the actual situation of the task unloading process. An unloading strategy which better conforms to the actual situation is obtained, so that the system obtains more long-term benefits. The method comprises the following steps: S1, defining a state set of a system based on a semi-Markov decision model; S2,defining an action set of the system; S3, defining a reward model of the system; S4, defining the transition probability of the system; S5, solving an optimal unloading strategy in the vehicle-mounted fog computing system; the method is characterized in that in the step S3, a system reward can be expressed as a difference value of immediate income and expenditure; calculation of the immediate income is carried out through different time delays, including the time delay needed by a local processing task. The transmission time delay sent to the computing unit by the vehicle is requested. The system unloads the task to the computing unit to process the needed time delay.

The invention discloses a method for extracting the features of a two-dimensional image. The method comprises the following steps of S1, carrying out Gabor filter difference decomposition on texture images, and obtaining a binary image set; S2, extracting textural features from decomposed binary images by using statistical operators; S3, carrying out dimensionality reduction on the extracted textural features by using a principal component analysis method, and carrying out texture classification and retrieval on feature vectors existing after dimensionality reduction is carried out. The method for extracting the features of the two-dimensional image has the advantages that texture elements do not need to be defined, the computation complexity is appropriate, and the five newly proposed statistical operators more comprehensively describe the arrangement rules of texture elements and elements existing after the decomposition; in a recognition process, because the principal component analysis algorithm is introduced, the computation speed of the method is increased, and the method for extracting the features of the two-dimensional image is suitable for real-time retrieval and classification of the texture images.

The invention discloses a data block selection transmission method for a radio frequency passive TDOA positioning system. The method comprises the steps: observation stations extract timing data blocks from an IQ data stream output by a radio frequency receiver and caches the timing data blocks into a queue; the observation stations evaluate the quality of the timing data blocks based on the carrier-to-noise ratio and the zero-crossing point number quality index, and report the timestamp and the quality index of each timing data block to a central station; the central station receives the observation station report, caches the observation station report into a queue, selects candidate data blocks from the report queue according to timestamps and quality parameters, and notifies the observation station to upload specified time data blocks; the observation stations receive a central station command, search data blocks at specified time from the cache timing data block queue and uploads the data blocks to the central station; and the central station receives the timing data blocks uploaded by the observation station and caches the timing data blocks to a timing data block queue to serve as input data of a TDOA estimation algorithm. The method is used for solving the problem of burst and bandwidth time-varying signal data block optimization and solving the problem of contradictionbetween network bandwidth and sampling rate.

The invention provides a control method for self-adaptation accurate matching of a tag. The control method includes the steps that a, target data and data of at least one tag are obtained; b, a natural language similarity index is determined based on the target data; c, a scene similarity index is determined based on the target data; d, weighting treatment is performed on the natural language similarity index and the scene similarity index, and the match-degree index between the tag and the target data is obtained. The invention further provides a control device for the self-adaptation accurate matching of the tag in a computer processing system. The control method can achieve calculation of the match degree between the tag and specific key information in a self-adaptation mode, calculation efficiency is improved, and accuracy of an application program is improved. The implementation is easy, calculation complexity is moderate, and the control method is suitable for being applied in a mobile mode.

The invention belongs to the field of global navigation satellite system (GNSS), and specifically relates to a non-fuzzy tracking method for a sine BOC modulation signal, wherein the method is used for a GNSS satellite navigation signal, i.e., a BOC (binary offset carrier) modulation signal. The method comprises the steps: receiving the BOC signal; enabling a locally generated pulse signal to be multiplied by a locally generated PRN signal; carrying out the integral accumulation of two intermediate frequency signals with an advanced local auxiliary signal and a lagged local auxiliary signal, i.e., convolution operation, and obtaining the integral accumulation results of all branch circuits; inputting a phase demodulation result Epsilon into a loop filter, converting the phase demodulation result Epsilon, outputting an adjustment amount, and adjusting PRN.

A human behavior recognition method based on mobile devices, which uses a variety of sensors built in mobile devices to collect data in real time; performs a series of data preprocessing operations such as correction, filtering, calculation and data generation, and data segmentation on the data obtained by the sensors; Feature extraction is performed on the processed data segment, and the extracted corresponding feature vector is input into the device location classification model to obtain the device location category; the corresponding behavior classification model is selected according to the obtained device location category, and the extracted corresponding feature vector is input into the behavior classification model, Obtain the final behavior recognition result. The invention provides a mobile device-based human behavior recognition method with good versatility and high accuracy.

A beam-hopping resource allocation method, system, storage medium and device based on deep reinforcement learning belongs to the field of communication technology. In order to solve the problem that the existing beam-hopping satellite communication system has poor delay performance due to the lack of continuity when the service scene is constantly changing during resource allocation, the present invention divides ground service requests into real-time data services and non-real-time data services. There are two types of real-time data services, and the optimization functions are established respectively; the maximum effective time length of the data in the satellite buffer is T th It is divided into M segments of equal length, corresponding to M beam-hopping time slots; the ground cell traffic request composed of data packet delay, number of real-time data packets, and non-real-time data packets is regarded as the environmental state S, and the satellite beam is regarded as the agent , taking the illumination of the cell as an action, considering the optimization problem of resource allocation in the satellite beam-hopping technology as a Markov decision process, and performing beam-hopping resource allocation based on a deep Q-network. It is mainly used for allocation of beam hopping resources.

The present invention relates to the technical field of underwater acoustic signal processing, and in particular to a robust adaptive beamforming method based on variable uncertain set constraints. The method includes: calculating the sample covariance matrix of the received array signal, performing eigenvalue decomposition, And calculate the projection vector corresponding to each scanning angle; use the projection vector to obtain K target orientations by the LSMI method, and then estimate the target power and noise power by the CBF method; calculate the covariance matrix error by K target orientations, target power and noise power Estimating the upper bound of the norm; calculating the upper bound estimation of the steering vector error norm of the current scanning angle; solving the WCPO problem under the uncertainty set constraint of the upper bound estimation of the error norm upper bound estimation of the parameter covariance matrix and the upper bound estimation of the steering vector error norm, Obtain the best weighted vector and the estimation of the spatial power spectrum; modify the estimation of the spatial power spectrum according to the best weighted vector; until the solution of all scanning angles is completed.

The invention provides a control method for self-adaptation accurate matching of a tag. The control method includes the steps that a, target data and data of at least one tag are obtained; b, a natural language similarity index is determined based on the target data; c, a scene similarity index is determined based on the target data; d, weighting treatment is performed on the natural language similarity index and the scene similarity index, and the match-degree index between the tag and the target data is obtained. The invention further provides a control device for the self-adaptation accurate matching of the tag in a computer processing system. The control method can achieve calculation of the match degree between the tag and specific key information in a self-adaptation mode, calculation efficiency is improved, and accuracy of an application program is improved. The implementation is easy, calculation complexity is moderate, and the control method is suitable for being applied in a mobile mode.

The invention discloses a guided filtering remote sensing image fusion method based on Log-Gabor transformation and direction region entropy, and belongs to the field of remote sensing image processing. The method comprises the following steps: firstly, carrying out IHS transformation on a multispectral remote sensing image, and carrying out Log-Gabor decomposition on an intensity component I; secondly, carrying out self-adaptive guide filtering on the high-resolution panchromatic remote sensing image by utilizing an edge guide matrix based on direction region entropy, and extracting rich edge texture details contained in the panchromatic image according to information entropy changes of different regions of the panchromatic remote sensing image; secondly, constructing a fusion decision template by taking energy distribution of a Log-Gabor transformation coefficient as guidance, and determining detail injection intensity of an intensity component I in a fusion process according to the fusion decision template; and finally, performing inverse IHS transformation on the intensity component after detail enhancement, the original H component and the original S component, thereby obtaining a final fusion image.

The invention relates to a passive multi-target detecting and tracking method based on wireless sensor networks. The technical scheme of the passive multi-target detecting and tracking method based on the wireless sensor networks comprises the following steps of utilizing a scanning circle windowing detection method to create a sliding scanning circle model of a passive multi-target according to received signal strength of different wireless links in the wireless sensor networks, and utilizing a HAC (Hierarchical Agglomerative clustering) algorithm to perform a clustering analysis so as to extract a detection result; and obtaining a multi-target tracking result of target number changes by utilizing a PHD (Probability Hypothesis Density) passive multi-target particle filter tracking algorithm and according to the detection result. The passive multi-target detecting and tracking method based on the wireless sensor networks is reasonable in design, has high accuracy and robustness in detecting and tracking algorithm, can detect and track multiple targets in a complicated multipath environment, meanwhile, is moderate in computation of target detecting and tracking algorithm, can guarantee the real time of operation of a detecting and tracking system.

The invention belongs to the field of time-frequency analysis in signal processing, and in particular relates to a multi-component linear frequency modulation signal denoising technology based on Lu distribution. Aiming at multi-component linear frequency modulation signals, the present invention proposes signal reconstruction based on Lu distribution, and combines signal detection technology to perform denoising processing on signals under strong noise pollution, and the result is a denoising signal close to the original signal in the time domain. Since the Lu distribution has the characteristics of high energy aggregation and strong noise suppression for multi-component linear frequency modulation signals, the present invention can effectively denoise and reconstruct signals under negative signal-to-noise ratio (that is, the noise power is greater than the signal power), and the calculation The complexity is moderate, and the mean square error value between the denoised signal and the original signal is better than that of the prior art.

The invention discloses an FPGA (Field Programmable Gate Array)-based infrared remote sensing image real-time target detection method. The method mainly aims at solving the problems of high omission ratio, multiple false alarms, poor robustness and the like of the existing ship detection algorithm. The method is mainly composed of a coarse detection module based on a gradient convolution model anda fine detection module based on Fourier spectrum and SVM. According to the invention, the ship target and typical false alarm characteristics in the infrared remote sensing image are fully analyzed,and available features capable of effectively distinguishing ship targets and typical false alarms are selected, and an accurate and simple target classification model is constructed in combination with a support vector machine classification algorithm, and quick and accurate elimination of the false alarms is realized in combination with two-dimensional Fourier transform. Meanwhile, consumptionof FPGA logic resources and storage resources can be reduced, and the advantage of FPGA parallel control is brought into play, so that high-precision target detection is completed under the conditionthat few FPGA resources are occupied.

The invention discloses a feature extraction method of a two-dimensional image. The method includes the following steps: S1, obtaining Gabor filter differential decomposition for a texture image, and obtaining a set of binary images; S2, using statistical calculation for the decomposed binary image Sub-extracting texture features; S3, using principal component analysis to reduce dimensionality of the extracted texture feature vectors, and performing texture classification and retrieval on the feature vectors after dimensionality reduction. The present invention has the advantage of not needing to define texture primitives, and the calculation complexity is moderate. The newly proposed five statistical operators more comprehensively describe the decomposed texture primitives and the arrangement rules of primitives, and in the recognition process , the introduction of the principal component analysis algorithm increases the calculation speed of the method and is applicable to the real-time retrieval and classification of texture images.