405results about How to "Reduce memory usage" patented technology

The invention relates to a lightweight parking detection method based on a multi-scale attention mechanism, and belongs to the technical field of image processing and target detection. The method comprises the following steps: acquiring a parking space data set, and preprocessing the data set; according to the method, a lightweight YOLOv8 network structure is constructed, and a Backbone layer at least comprises a plurality of MobileNetV3 modules and a plurality of C2fScConv modules; an EMA attention mechanism is introduced between a C2f module and a Deect module of the Neck layer; establishing a loss function WIoU v3 for the lightweight YOLOv8 network structure; and training the network to obtain a lightweight parking detection model, and detecting the parking space state in real time through the lightweight parking detection model. According to the invention, while the target detection precision and robustness are improved, the operation cost and operation time of model training and detection are reduced, and lightweight and accurate parking space state detection are realized.

The invention belongs to the technical field of model training, and particularly relates to a large model increment training method and system based on dynamic sparsification. The method comprises the following steps: S1, evaluating the parameter importance of a large model by adopting a multi-dimensional evaluation method, and reflecting the contribution of parameters to the performance of the large model; s2, performing dynamic sparsification control on the large model structure by adopting a self-adaptive structure adjustment method; s3, performing incremental training management on the large model by constructing a capacity-adjustable empirical playback buffer pool and proposing a multi-objective loss function; and S4, carrying out model compression optimization on the large model by adopting a mixed precision quantification strategy. The method has the characteristics that the network structure can be dynamically adjusted by evaluating the importance of model parameters in real time, so that the efficient training is realized while the model performance is ensured.

The invention discloses a parallel iteration Runge-Kutta method for simulating a second-order nonlinear differential equation. The parallel iteration Runge-Kutta method comprises the steps that a simulated ordinary differential equation is converted into a target nonlinear implicit differential equation, and a system differential algebraic equation is established; judging whether a system matrix and a decomposition iteration matrix in the system differential algebraic equation are updated or not; a four-order extrapolation method is adopted to obtain a predicted differential variable derivative, and if a prediction result is not converged, parallel iterative operation is carried out through a four-stage four-order Runge-Kutta method; after the iteration process is converged, judging whether an operation result accords with a preset error range, if not, updating a system matrix according to a step size reduction strategy and re-decomposing an iteration matrix, and if yes, updating the operation result; and storing and displaying an operation result when judging that simulation is finished. The algorithm stability and the calculation efficiency of the rigidity problem can be considered.

The embodiment of the invention provides a seismic wave detection method and system, electronic equipment and a storage medium, and belongs to the field of seismic wave engineering detection. The method comprises the steps that seismic sources and detectors are arranged in advance according to actual working conditions, seismic waves are excited at the positions of the seismic sources, and seismic wave data are collected through the detectors; based on seismic wave data, solving a time fractional order viscoelastic acoustic wave equation by using a short memory fast algorithm so as to simulate the propagation behavior of seismic waves in a real stratum medium; and performing Bayesian full waveform inversion and reverse time migration imaging based on the seismic wave data to obtain a stratum structure and stratum characteristics. The memory effect of a time fractional order viscoacoustic wave equation is reasonably simplified through a short memory fast algorithm, so that the storage requirement for wave field historical data is reduced, efficient numerical solution is achieved, on the premise that precision is guaranteed, memory occupation is greatly reduced, and the cost is reduced. And the time fractional order viscoelastic acoustic wave equation has higher operability and calculation efficiency in engineering application.

The invention relates to the technical field of farmland identification, and discloses a remote sensing image classification and farmland identification method based on deep learning, comprising the following steps: S1, data collection: acquiring remote sensing image data covering a target farmland area; s2, data preprocessing and labeling: preprocessing the remote sensing image data, and labeling the preprocessed data; s3, model construction: constructing a convolutional neural network model, and training the model by using the preprocessed remote sensing image data and annotation data; and S4, remote sensing image classification and farmland identification: inputting a to-be-identified remote sensing image into the trained convolutional neural network model, and outputting a classification result by the convolutional neural network model. According to the invention, transfer learning and fine tuning technologies are utilized, and a pre-trained deep learning model is applied to a remote sensing image farmland identification task, so that the remote sensing image farmland identification task can adapt to farmland identification tasks of different regions, different seasons and different crop types.

The invention discloses a cross-modal target tracking method based on a fully convolutional twin network. The method comprises the following steps: inputting a visible light image and a thermal infrared image into a cross-modal target tracking network model to obtain a tracking structure; the cross-modal target tracking network model comprises a feature extraction backbone network, a self-adaptive complementary fusion module and a target classification regression network, and the feature extraction backbone network comprises a visible light backbone network and a thermal infrared backbone network which are the same in structure; the visible light backbone network and the thermal infrared backbone network respectively extract features of a template and a search area through convolution and perform adaptive convolution to extract time sequence information; and the self-adaptive complementary fusion module performs channel attention enhancement and space attention enhancement on the visible light features and the thermal infrared features extracted by the feature extraction backbone network and fuses the visible light features and the thermal infrared features. The invention also discloses a cross-modal target tracking system based on the fully convolutional twin network for realizing the method. The method can improve the stability and robustness of tracking in different environments.

The invention provides a method for training an end-to-end automatic driving strategy, and the method comprises the following steps: S1, data collection and preprocessing, S2, defining a loss function and an optimizer, and constructing a model, S3, data division, S4, model training, S5, evaluation on a verification set, S6, model adjustment and optimization, and S7, test set evaluation. S8, weight pruning and evaluation: calculating and evaluating the importance of each weight by adopting a gradient norm calculation mode, evaluating the model after weight pruning by using the test set in the step S7 according to an importance evaluation result, starting from a relatively small proportion, gradually increasing and removing unimportant weights, and if the performance of the model does not reach the standard at the moment, evaluating the model after weight pruning by using the test set in the step S7. According to the method, through weight pruning, the applicability of a model system is wider, and the training efficiency of the system is higher.

The invention relates to an X-ray image contraband detection method based on multi-scale feature fusion, and belongs to the technical field of image processing. The method comprises the following steps: acquiring an X-ray contraband data set, and dividing the data set into a training set and a test set; constructing an X-ray image contraband target detection model based on multi-scale feature fusion, wherein the model comprises an image branch, a self-adaptive high-low pass filter module, a text branch, a Neck layer and a Head head; images in the training set are input into the model to train the model; optimizing the model by adopting a loss function to obtain a trained model; and inputting the images in the test set into the trained model to obtain a contraband detection result. According to the method, through image and text joint training, the performance is improved in real-time task monitoring, the efficiency is higher, the calculation amount and memory occupation are reduced, and the problem that image boundary details are fuzzy in a complex scene is solved.

The invention discloses a method for designing a tree data storage mode in a relational database by using preorder traversal, which comprises the following steps: S1, basic node data of a tree structure is constructed in the relational database, each node comprises id, name, lvalue and rvalue attributes, and the lvalue and the rvalue are used for identifying the position relationship of the nodes in a tree through an interval representation method; s2, setting a query condition according to a node ID needing to be queried, and determining a descendant node range of the node through lvalue and rvalue attributes of the node; s3, the query results are subjected to ascending sort of the left value lvalue, and a sort list (sort List) is constructed; and S4, by traversing the sortedList, in combination with the temporary set level List and the cache level NodeCache, calculating and setting a father node ID (pid) of each node and whether the father node ID (pid) is a leaf node isLeaf attribute or not according to the logic of pre-order traversal, thereby completing the rapid construction of the tree structure.

The invention discloses a large model reasoning acceleration method and device and a medium, and the method comprises the steps: constructing a training set and a test set for training a large model according to scene data; expanding a prediction head in the large model to obtain an expanded target large model, and freezing original prediction head parameters in the large model so as to update the prediction head parameters corresponding to the expanded prediction head in the training process of the target large model; training and evaluating the target large model through the training set and the test set to obtain a trained target large model; constructing a calibration data set formed by partial scene data, and inputting the calibration data set into the target large model for forward reasoning to obtain a calibration matrix corresponding to each linear layer; and determining an input matrix and an output matrix corresponding to the linear layer according to the calibration matrix, and quantizing a weight corresponding to the linear layer based on the input matrix to obtain a quantized model weight.

The invention provides a multi-stage structure topological optimization method based on a potential diffusion model, and relates to the technical field of engineering design and structure design optimization. Creating a structure topology optimization image data set by using a traditional SIMP method and performing preprocessing; mapping the structure topological optimization image to a low-dimensional potential space for representation so as to reduce the data dimension and improve the calculation efficiency; a forward diffusion process and a reverse diffusion process are carried out on the initial potential variable in the low-dimensional potential pair, a denoising network is added in the reverse diffusion process, and a cross attention mechanism and a condition mechanism are introduced, so that a topological structure is adjusted and designed according to physical and functional requirements, and the structural fidelity is ensured; after training is completed, potential variables of a low-dimensional space are mapped back to a high-dimensional space, a topological structure is generated, few-step SIMP direct optimization is carried out on the topological structure, the structure is rapidly refined, physical information is directly injected into the design so as to adapt to given boundary conditions, and generation of the optimal topological structure can be completed through few-step iteration.

The invention provides a graph rendering method and system based on a GLSL shader, and relates to the technical field of three-dimensional model rendering, and the rendering method comprises the steps: loading a 3D model, extracting data of a control point of the 3D model in a VBO of a GPU, generating a first geometric primitive, extracting vertex data in the first geometric primitive, generating an index array of the vertex data, and creating an EBO, based on an index array stored in the EBO, a vertex shader is used for sequentially processing vertex data, obtaining the position and interpolation data of each vertex, cutting the vertex data, regenerating a second geometric primitive, converting the second geometric primitive into a fragment, and a fragment shader calculates color data of each fragment; and the GPU performs parallel rendering on the color data to obtain a final output image, so that resource consumption and memory occupation are reduced while the 3D model is converted into the 2D image, and the rendering efficiency is improved.

The invention discloses a medium metasurface reverse optimization method based on a space coupling mode theory, and relates to the field of optics, and the method comprises the following steps: building a cut-off waveguide array based on a metasurface, calculating a guided mode and a propagation constant of each waveguide according to the cut-off waveguide array, and coupling adjacent waveguides through a coupling matrix, a waveguide array is obtained; an input light field is coupled into a waveguide array, propagation of light in waveguides is simulated based on a spatial coupling mode theory model, and near-field intensity distribution is converted into far-field intensity distribution by using a Rayleigh-Sommerfeld diffraction theory; and calculating the difference between the target intensity and the far-field intensity through a loss function, evaluating the effect of the current metasurface design, and gradually adjusting the geometric parameters of the waveguide unit through an optimization algorithm. According to the method, the problems of efficient modeling and optimization in metasurface design are solved through the space coupling mode theoretical model.

The invention relates to the technical field of intelligent interaction control, and discloses a dynamic loading method and device for an intelligent interaction control unit of a television box. The method comprises the following steps: performing data mining on a user operation log to obtain a function module classification list, a use frequency statistical table and a dependency graph; and performing modular reconstruction on the software system based on the classification list to generate a function module package. And constructing a dynamic priority queue by using the frequency statistical table and the dependency graph. And carrying out predictive module loading in combination with the priority queue and the function module packet to obtain a loaded module list and a resource allocation state. And unloading the intelligent module according to the loading condition, and updating a module list and a resource use state. And executing incremental updating based on the version information in the module packet, updating the functional module packet and reconstructing the dynamic priority queue to obtain a target priority queue. According to the method, the resources can be flexibly allocated according to actual requirements, resource waste caused by static loading is avoided, and the resource utilization efficiency is improved.

The invention relates to a quality standard retrieval tool and method based on a large language model. The method comprises the following steps: storing a quality standard file, embedded vector data corresponding to the quality standard file and a mapping relation between the quality standard file and the embedded vector data into a quality standard file knowledge base; question vector data converted from a question described by a natural language is received, embedded vector data most similar to the question vector data is retrieved from a quality standard file knowledge base, and a quality standard file text corresponding to the embedded vector data is a related support document; recording the historical retrieval question and the corresponding retrieval result as historical question and answer information, and summarizing a preset response rule, the current retrieval question, the related support document and the historical question and answer information into cue words; and adopting a pre-trained large language model to generate a retrieval result according to the cue word.

The invention discloses a database query plan optimization method and system, and belongs to the technical field of databases and artificial intelligence. AI model training: training the model by using the preprocessed historical query data according to a specific model architecture, a training mode and resource and time arrangement, and dynamically adjusting the model according to the change of the database; query analysis: evaluating the complexity of the query and the possibly involved data volume, and allocating a complexity level and a data access volume range for the query; query plan generation: generating and evaluating a plurality of candidate query plans based on prediction and query analysis results of the AI model, and selecting an optimal query plan as a final execution scheme; and performing query execution and monitoring. According to the method, the query execution time can be greatly shortened, and the database response speed is increased; system resources are utilized more effectively; dynamic changes of the database can be adapted, and the query plan is automatically updated according to structure and data distribution adjustment.

The invention relates to an aerodynamic force-thermal-structure multi-field coupling intelligent prediction method for a high-speed aircraft, and belongs to the technical field of high-speed aircraft design. According to the flight path of the aircraft, sample parameters used for calculating aerodynamic force and aerodynamic heat states are extracted; modeling the aircraft, and carrying out analog calculation to obtain an aerodynamic force and aerodynamic heat data set; calculating an aircraft temperature field along the trajectory; training the aerodynamic force data set to obtain an aerodynamic force prediction model; calculating structural stress displacement field data of the aircraft model; extracting concerned aircraft feature point corresponding data from the aerodynamic data set, the aircraft temperature field and the structural stress displacement field data of the aircraft model to obtain a feature point data set; training the feature point data set to obtain an accurate prediction result; according to the method, after artificial intelligence algorithm offline learning modeling is carried out on the aerodynamic force-heat-structure coupling data of analog simulation, the flight path is input, and the data, changing along the flight path, of the feature point data of the output aircraft is rapidly predicted.

The invention discloses a mobile platform coal quality remote detection system and method, and solves the problem that the analysis precision and analysis efficiency of a model are affected by a spectral data acquisition error rate due to the fact that parameters of pulverized coal cannot be preprocessed when a model between a spectral line and a calorific value is established and a coefficient matrix and an error term of the model are obtained in an existing method. The method comprises the following steps: detecting stored coal in real time by adopting mobile coal quality detection equipment, pre-constructing a data processing model based on the combination of adaptive interactive multiple models and deep learning, preprocessing spectral sample data by the data processing model, and establishing a coal quality index quantitative model based on the combination of a convolutional neural network and a lightweight detection structure; according to the method, noise reduction, normalization and feature extraction preprocessing are performed on the real-time spectral data by pre-constructing the data processing model based on the combination of the adaptive interactive multiple models and deep learning, so that the spectral data acquisition error rate is reduced, and the analysis precision and analysis efficiency of the coal quality index quantitative model are ensured.

The invention discloses a disaster relief unmanned aerial vehicle path planning method based on an APF-IRRT fusion algorithm, and the method comprises the following steps: initializing parameters, expanding nodes, optimizing and smoothing a path, carrying out the smoothing of the path through cubic spline curve fitting or polynomial interpolation, and finally, carrying out the planning of a disaster relief unmanned aerial vehicle path. And outputting the optimized and smoothed flight path of the unmanned aerial vehicle by setting an algorithm termination condition. According to the disaster relief unmanned aerial vehicle path planning method based on the APF-IRRT fusion algorithm, the attraction force and the repulsive force are combined with adaptive step length and sampling angle optimization, the high efficiency of path planning and the obstacle avoidance safety are realized, and meanwhile, the smoothness and the performability of the path are improved through path smoothing processing; the method is suitable for efficient path planning of unmanned aerial vehicles in complex disaster relief environments.

The invention relates to the technical field of computer vision and deep learning, and provides an SLAM (Simultaneous Localization and Mapping) dynamic disturbance suppression method based on fuzzy processing and target detection, which comprises the following steps of: obtaining a fuzzy remover; performing blurring processing on the image by using a blurring remover, and extracting ORB feature points; obtaining a dynamic target detector; detecting the clear image through a dynamic target detector to obtain a target detection frame; classifying motion states of objects in the clear image according to the target detection frame and the ORB feature points, and obtaining a static feature point set by using an overlapping frame strategy and epipolar geometric constraint; and performing system pose estimation and trajectory prediction according to the static feature point set. According to the method, the adverse effect of the dynamic target on other parts of the system is reduced through deblurring, the dynamic target can be accurately recognized through the target detection function, comprehensive elimination of the dynamic feature points is achieved in combination with the overlapping frame strategy and the epipolar geometric constraint, and the overall performance of the system is well improved.

The invention relates to the technical field of data encryption, in particular to a rapid data encryption method and system based on a national cryptographic algorithm, and the method comprises the steps: obtaining a plaintext, and employing a counter mode of the national cryptographic algorithm SM4 to carry out encryption processing on the plaintext to obtain a ciphertext; wherein the encryption process of the counter mode of the national cryptographic algorithm SM4 is accelerated by adopting the following mode: reconstructing an encryption function, and encrypting a message to be encrypted through the reconstructed encryption function to obtain an encrypted partial ciphertext and a residual plaintext; and constructing a new encryption function based on different lengths of the residual plaintexts, and encrypting the residual plaintexts with different lengths based on the new encryption function to obtain ciphertexts of the residual plaintexts. The SM4 encryption method has the advantages of being universal, small in memory occupation, high in efficiency, safe in implementation and the like.

The invention discloses a dynamic window-based graph optimization GNSS (Global Navigation Satellite System) and IMU (Inertial Measurement Unit) integrated navigation method, which improves the adaptability, calculation efficiency and positioning accuracy of a system in a dynamic environment by introducing a dynamic window strategy and a key frame mechanism. According to the method, the size of a sliding window is dynamically adjusted according to a real-time motion state, environment change and computing resources: when the environment is complex or high precision is needed, the window is enlarged to contain more historical information; when the environment is stable or computing resources are limited, windows are reduced to improve computing efficiency. Meanwhile, threshold values of position and posture changes are set, and only when the variable quantity of the current state and the latest key frame state exceeds the threshold values, new key frame nodes are added, key motion information is captured, and redundant calculation is avoided; according to the method, the optimal balance between the positioning precision and the calculation efficiency in a complex dynamic environment is realized, the method is suitable for an embedded platform with limited resources, and the requirement of high-precision real-time navigation is met.

The embodiment of the invention provides a widget display method and electronic equipment, and relates to the technical field of terminals. The method comprises the steps that a first trigger operation of a user is received, the first trigger operation is used for adding a first static widget, and the first static widget corresponds to a plurality of components; and rendering a plurality of components corresponding to the first static widget. A first static image can be generated and displayed according to the rendered components, and the first static image is used for displaying the first resource content corresponding to the first static widget. One static image is used for replacing a plurality of rendered components, so that on the basis of ensuring normal display of one static widget, continuous operation and refreshing of a plurality of components corresponding to the static widget can be avoided, the memory occupation of the widget is reduced, and the power consumption of the electronic equipment is reduced.

The invention provides an overcurrent prediction method, medium and system for a liquid-cooling over-charging pile, and belongs to the technical field of electrical variable measurement, and the method comprises the steps: designing a time-sensitive HyperLogLog (TS-HLL) structure through collecting the charging parameters and heat exchange parameters of the liquid-cooling over-charging pile, and coding the parameters; statistical characteristics of charging parameters and heat exchange parameters are extracted by using a time sliding window technology, and a correlation matrix between the charging parameters and the heat exchange parameters is calculated. Based on the correlation matrix, a relation graph of the charging current and other parameters is constructed, a graph neural network is used for learning the relation graph, and a complex nonlinear relation between the charging current and other parameters is extracted. And finally, combining the learned features with historical charging data, inputting the combined features and historical charging data into a recurrent neural network, and predicting the charging current in a period of time in the future, thereby solving the problem of rapid over-current prediction of the liquid cooling over-charging pile, which is difficult to solve in the prior art.

The invention discloses a method for automatically monitoring structural seismic response by using a monitoring video, and relates to the technical field of structural health monitoring. The invention aims to solve the problems existing when a monitoring camera is used for earthquake response monitoring. The method comprises the following steps: respectively identifying natural feature points in each frame of image of a pre-earthquake video and an earthquake time video, and describing the natural feature points by adopting a BRIEF descriptor; matching the natural feature points in the pre-earthquake video and the post-earthquake video, and extracting two-dimensional pixel coordinates of the natural feature points during the earthquake; and based on the two-dimensional pixel coordinates of the natural feature points during the earthquake, the displacement of the monitoring equipment during the earthquake relative to the lower-layer ground is calculated by adopting a PnP algorithm, and the monitoring equipment is fixedly connected with the upper-layer floor, so that the structural earthquake response is represented.

The invention belongs to the technical field of artificial intelligence, and relates to a junk call identification method and device, computer equipment and a storage medium, and the method comprises the steps: obtaining initial audio data corresponding to a target call when the target call is accessed; preprocessing the initial audio data to obtain specified audio data; performing feature extraction on the specified audio data based on a pre-training language model to obtain a specified audio feature vector; calling a junk call recognition model; performing identification processing on the specified audio feature vector based on a junk call identification model to obtain a probability that the target call belongs to each preset junk call category; and screening out a target probability with the highest value from all probabilities, and generating a junk call recognition result corresponding to the target call based on the target probability. In addition, the invention also relates to a block chain technology, and the junk call recognition result can be stored in a block chain. Based on the use of the junk call recognition model, the junk call recognition efficiency and recognition accuracy are improved.

An incomplete multi-view face image depth anchor image clustering method based on manifold relationship maintenance belongs to the field of face image data processing in multimedia information processing, and comprises the following steps: firstly, constructing a subspace feature for each view through a graph neural network, and then maintaining an original feature and the subspace features by adopting a manifold relationship; meanwhile, anchor points specific to the view angle are randomly selected to be combined with the common subspace, the similarity between the result and original data is compared, gradient descent optimization is carried out on the network for many times, spectral clustering is carried out after a common subspace anchor map with discrimination is obtained, and finally the clustering result and accuracy are obtained. Compared with other methods, the method is higher in accuracy and more stable in performance.

The invention discloses a 3D visual energy consumption management method. The method comprises the following steps: (1) constructing an Internet of Things platform and installing a sensor; (2) creating a 3D model; (3) data integration and processing: according to actual display requirements of the 3D model, associating the acquired energy consumption data and environmental data with the 3D model by adopting a color coding method, a height change method, a label and chart method, an animation display method and a symbol and icon method; corresponding energy consumption data and environment data are matched with corresponding areas, floors, rooms and devices in the 3D model, integration processing is carried out, energy consumption conditions of different areas, floors, rooms and devices are represented by color codes, height changes, charts, animations, symbols and icons, and the energy consumption conditions of the areas, floors, rooms and devices are stored on an Internet of Things platform; (4) a user requests visual display of a corresponding area according to needs; (5) data analysis and optimization; and (6) real-time monitoring and reporting.

A portable electronic nose used for beer identification and based on a convolutional neural network is characterized in that a vertical annular gas sensor array is arranged, the gas sensor array is connected with a preprocessing module through an ADC analog-to-digital converter, and the preprocessing module is connected with a beer identification module; the gas sensor array is used for collecting beer gas signals and transmitting the beer gas signals to the preprocessing module; the preprocessing module is used for preprocessing the beer gas signal and then transmitting preprocessed beer gas data to the beer identification module; and the beer identification module is a two-dimensional convolutional neural network model CC-2DCNN based on a channel attention mechanism and a cosine annealing hot restart strategy, and is used for carrying out beer identification according to the beer gas data and outputting a beer identification result. The method has the advantages that accuracy and reliability of beer identification are improved, and a reliable solution is provided for detecting counterfeit beer and preventing beer deterioration.

The invention discloses a lightweight visual inertia fusion positioning method and system, and relates to the technical field of autonomous positioning. The problems that a traditional visual inertial odometer is not robust enough under the condition of data damage, a sensor fused with many deep learning models in an existing deep learning method is huge in size and needs more computing resources during operation, so that deployment on an embedded system or mobile equipment with limited resources becomes difficult, and the cost is low are solved. And further application of the visual inertial odometer is limited. The method comprises the following steps: acquiring image data and IMU data; extracting visual feature information from the image data by using a FlowNetSimple network, and extracting inertial feature information from the IMU data by using a one-dimensional convolutional network; fusing the visual feature information and the inertial feature information by adopting a hard attention mechanism; a recurrent neural network is used for time modeling, and an output transformation network is used for converting the output of the model into a final prediction result. The method is applied to automatic driving.