237results about How to "Reduce storage requirements" patented technology

The invention relates to the technical field of data storage management, in particular to a data storage method based on a large model. The method comprises the following steps: performing input class cluster fragmentation optimization on a target pre-training large model to obtain storage modulus fragmentation data; constructing an intelligent fragmentation storage strategy by using the storage modulus fragmentation data; obtaining target model training data; performing intelligent fragmentation storage on the target model training data by using an intelligent fragmentation storage strategy, and performing redundancy compression optimization to obtain dynamic redundancy storage index data; and training data storage space optimization is carried out according to the dynamic redundancy storage index data, and an optimized data storage strategy is obtained. According to the method, intelligent fragmentation storage and refined data redundancy compression are performed on the data, so that efficient and stable model training data access is realized, the problem of low data storage and access efficiency in a large model training process is solved, and the model training efficiency and reasoning real-time performance are remarkably improved.

The invention discloses a lifelong target re-identification method based on descriptive text prompt prototype compensation, relates to the field of computer vision, and aims to display and solve the problem of feature drift caused by model increment update by introducing a prompt prototype compensation module. The module generates a multi-modal vision and text prototype for each target category through a vision language pre-training model, and maps old task features to a new task feature space through a cross attention mechanism so as to realize fine-grained semantic drift compensation. In addition, a boundary perception prototype correction module is designed to solve the problem of boundary confusion caused by updating of old prototypes. The module dynamically samples false features in compensation distribution to correct old prototypes with fuzzy boundaries, focuses on categories easy to confuse to maintain clear decision boundaries, optimizes cross-task and intra-task correction loss to reduce category confusion, effectively alleviates the problem of disastrous forgetting, and improves the adaptability and generalization ability of the model.

The invention discloses a three-phase permanent magnet synchronous motor full-speed-domain driving control system and a control method thereof. The system is composed of a torque limiting module, a full-speed-domain control neural network module, a current control module, a coordinate transformation module-I, a coordinate transformation module-II, a coordinate inverse transformation module, an SVPWM module, a SiC MOSFET driving module, a full-speed-domain position rotating speed observation module, a weak magnetic feedback module and a three-phase permanent magnet synchronous motor. The control method is used for driving the control system. According to the method, accurate tracing of the optimal dq-axis reference current working point is realized based on deep learning, and accurate control of three operation modes of MTPA, field weakening control and MTPV is realized; an improved mobile vision field estimation algorithm is adopted, and the rotor position and the rotor rotating speed are accurately estimated within the full-speed domain range; the SiC MOSFET is used as an inverter switching tube, and a brand new drive circuit is designed to adapt to the operation condition of the electric vehicle.

The invention relates to the technical field of video compression processing. The invention provides a monitoring video compression processing method and device, and the method comprises the steps: extracting the video features of a to-be-processed image in a monitoring video in a target scene region through a video feature extraction model, and obtaining the feature vectors of a plurality of video features in different image regions; according to the obtained feature vectors of the various video features in the different image areas, performing following classification on the to-be-processed image to obtain classification results: image area classification, scene type classification and importance level classification; determining an initial compression ratio of different image areas in the to-be-processed image according to the classification result; decomposing the to-be-processed image into sub-images of different scales by adopting an image decomposition method based on wavelet transform, and determining an optimal decomposition layer number according to image quality; and determining a final compression ratio and a coding mode of different image areas according to the scale and content characteristics of the sub-images of the image to be processed. According to the method, the compression ratio is greatly improved on the premise that the image quality is not obviously reduced.

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 discloses a method for performing efficient lossless text compression based on a large language model, which relates to the technical field of text compression and comprises the following steps of: preprocessing to-be-compressed text data to obtain an initial token sequence; selecting and loading a pre-trained big language model, analyzing the current context information by using the big language model, obtaining a probability value of each token, forming a probability distribution list, and recording a ranking sequence number of the current token; the context window is moved backwards by one token, and context information is updated until all the tokens are processed; and recording the ranking sequence numbers of all the tokens in sequence to form a sequence number list so as to obtain a final text compression result. According to the method, the compression rate is remarkably increased, the requirements for storage space and transmission bandwidth are reduced, the high processing speed is kept, and the method is suitable for real-time compression and decompression of large-scale text data.

The invention relates to the field of switching power supply high-side current detection, in particular to a switching power supply high-side current detection method, which comprises the following steps of: acquiring current signals to form a sequence, acquiring an intrinsic mode by using empirical mode decomposition, analyzing each mode to construct a difference amplitude-frequency index, and calculating a noise index in combination with a similarity index and heterogeneity. According to the method, the de-noising window is intelligently adjusted, and the local outlier factor is utilized to detect the anomaly, so that accurate current anomaly identification is finally realized, the detection precision and the system intelligence level are remarkably improved, the resource utilization is optimized, and a powerful guarantee is provided for stable operation of a power system.

The invention discloses a ship remote sensing image detection method based on a multi-dimensional attention mechanism, and the method comprises the steps: constructing a remote sensing image detection model based on YOLOv10, replacing an SPPF module in a Backbone network with an SPPFCMSA module, and enabling the SPPF module to be fused with an attention mechanism CMSA to form the SPPFCMSA module; a PSA module in the Backbone network is replaced by an MSGE module; the method comprises the following steps: replacing the last two C2f modules in a Neck network with a C2fSev2 module, wherein the C2fSev2 module is formed by introducing an improved SELayer2 module into a BottleNeck of the C2f module; and training a remote sensing image detection model by using the data set, and performing remote sensing image ship detection by using the trained remote sensing image detection model. The method has the advantages of high detection precision and small model calculation amount, and can meet real-time remote sensing image detection requirements.

The invention relates to a table detection and content extraction method and system based on image processing. The method comprises the following steps: constructing a table detection model and a table structure recognition model; using the public data set to train the table detection model and the table structure identification model, and using the self-made data set to finely adjust the table detection model and the table structure identification model; converting the input file into a picture format and preprocessing the input file; inputting the preprocessed file picture into a table detection model for feature extraction, and obtaining frame information and category information of a table in the picture according to a feature graph of the input picture; extracting the table in the picture according to the obtained frame information and category information of the table; inputting the table picture extracted by the table detection model into a table structure detection model, and detecting and identifying a table unit structure; and identifying the text information in the table picture by using an optical character recognition (OCR) technology, and matching the obtained table unit structure with the text information content to generate a target table.

The invention relates to an end-side cloud collaborative intelligent optimization and resource scheduling method and system for an industrial Internet of Things. The method comprises the steps that sparse processing and quantization are conducted on model parameters on end equipment through accelerator hardware, the storage and transmission requirements of a model are reduced, and meanwhile the accuracy of the model is kept; through the knowledge extraction technology, the end equipment can extract key information in the data, and high-quality input is provided for subsequent data processing and model training; by deploying the lightweight reasoning model and the routing model in the edge device, the uploaded data can be preliminarily classified, and only the data which is difficult to process is sent to the cloud, so that the overall delay is reduced to the greatest extent while the accuracy is kept; performance monitoring, bandwidth allocation and task segmentation are carried out on end equipment and edge equipment in real time through a cloud end, and global model optimization and resource scheduling are responsible for realizing optimal performance.

The invention relates to the technical field of computers, in particular to a mobile application large model risk automatic assessment method based on improved BERT, and the method comprises the following steps: S1, a test preparation stage; s2, realizing an automatic question answering whole process of the large model of the mobile application version based on an interactive automatic question asking method for simulating click by integrated group control equipment; s3, realizing batch acquisition of test result text data and picture information based on a dynamic and static data acquisition engine of the integrated group control equipment; s4, based on the risk study and judgment model of the improved BERT, performing safety risk judgment and classification on the test result, and performing evaluation and scoring according to a safety risk evaluation rule; and S5, uniformly and properly storing the assessment result and the test result screenshots. The method improves the accuracy and reliability of an output result discrimination algorithm, and establishes a comprehensive security risk assessment system.

The invention provides a single-channel speech enhancement method based on an improved convolutional recurrent network, and the method comprises the steps: 1, downloading a data set for model training and testing, carrying out the preprocessing of the speech of the data set through employing a VoiceBank + DEMAND data set, and extracting the amplitude spectrum of a speech signal through short-time Fourier transform (STFT); 2, inputting the amplitude spectrum into an encoder, and extracting high-dimensional features from the amplitude spectrum; step 3, using the Channel-S4D block as a recursion module of the model, inputting a result output by the Channel-S4D block into a decoder, and recovering the result to an original dimension; 4, constructing a joint loss function; and 5, reconstructing and evaluating the enhanced voice signal. According to the invention, the Channel-S4D module constructed based on the diagonalization state space model efficiently processes a voice sequence, captures channel-level voice key information, better models the time dependence relationship of voice signals, and improves the adaptability of the model to a complex noise environment while reducing the parameter quantity and the calculation quantity of the model.

The invention belongs to the technical field of wind power short-term prediction, and particularly provides a short-term wind power prediction method of a Bi-MGLSTM model based on VMD decomposition and Bayesian optimization, and the method comprises the steps: preprocessing data; variational mode decomposition is carried out on the wind power data; adding one forgetting gate and one input gate in the front and back LSTM of the BiLSTM model, constructing a Bi-MGLSTM model, and inputting the decomposed modal components into the Bi-MGLSTM model for power prediction; a Bayesian algorithm is used to optimize the hyper-parameters of the Bi-MGLSTM model, and an optimized Bi-MGLSTM model is obtained; predicting by the optimized Bi-MGLSTM model to obtain a prediction result of each modal component, and superposing and reconstructing to obtain a preliminary power prediction result; performing differencing on the preliminary power prediction value and a true value, and performing training and prediction on an error sequence by using the optimized Bi-MGLSTM model to obtain an error prediction result; and adding the power prediction value and the error prediction value to obtain a prediction result. According to the method, the wind power prediction error can be remarkably reduced, and the accuracy of short-term wind power prediction is improved.

The invention belongs to the technical field of energy management of hybrid power systems, and discloses a quantum strategy learning energy management method and system for a multi-stack fuel cell hybrid power system, and the method comprises the steps: building a multi-stack fuel cell hybrid power system model which comprises a fuel cell power model, a lithium battery equivalent circuit model and power balance; constructing a quantum hybrid neural network, interacting with the established hybrid power system environment, distributing output power of different power sources, and constructing a state-action-reward set; the strategy is evaluated and updated on the basis of the collected state-action-reward data set, and then the energy management strategy is optimized. By fusing the variable component sub-circuit and the traditional neural network, efficient operation of the fuel cell system is ensured while the state of charge of the lithium battery is maintained in an ideal interval; and the gradient of the quantum circuit parameters is accurately calculated by using a parameter transfer rule, so that the optimization process of the quantum strategy learning energy management strategy is more accurate and efficient.

The invention discloses a respiratory rehabilitation training data management method and system based on artificial intelligence, and relates to the technical field of data processing, the respiratory rehabilitation training data management method based on artificial intelligence comprises the following steps: obtaining training data of a respiratory rehabilitation person, and preprocessing the training data to obtain a multi-source data set; constructing a teacher network based on the high-performance model, training the teacher network through the multi-source data set, constructing a student network, and training the student network based on the trained teacher network; optimizing the trained student network by using a differential evolution algorithm to obtain a student network optimized model; and training a reinforcement learning agent by using data generated by the model after student network optimization, and automatically adjusting rehabilitation training parameters of the respiratory rehabilitation personnel through the reinforcement learning agent. According to the invention, computing resource and storage requirements are reduced, data-driven decision is realized, and a personalized rehabilitation training scheme is provided.

The invention discloses a block chain node consensus method and system based on block chain fragments, and the method comprises the steps: enabling a block chain node added into a committee fragment to be called as a committee node, collecting transaction data through the committee node, and carrying out the Hash processing of the transaction data, and obtaining a Hash value; the committee node configures itself to a committee Merkel tree according to the Hash value; the committee Merkel tree is realized by adopting an improved Merkel compressed prefix tree structure; meanwhile, the block chain node added into the global fragment is called as a global node, and the global node receives the root hash of each committee fragment; configuring the global node into a global Merkel tree according to the root hash; the global Merkel tree is also realized by adopting an improved Merkel compressed prefix tree structure; consensus operation is completed on the committee nodes of the committee fragments; and completing a consensus operation on the global node of the global fragment.

The invention provides a multi-target tracking method and system based on a cross Transform data association algorithm, and belongs to the field of artificial intelligence. The problems that an existing traditional MTT algorithm based on radar data is high in storage requirement and low in calculation speed, a learning-based algorithm needs to restrain the number of targets, network design considers typical features of position data, two cross tracks cannot be distinguished, and missing detection or false detection may exist are solved. According to the method, the optimal estimation in the aspects of precision and efficiency of radar-based position measurement is obtained by using a cross Transform-based data association network and adopting a BLUE filter; furthermore, a score-based trajectory management strategy is proposed to be based on the associated historical measurements; simulation experiments on various data sets show that the correlation and tracking method provided by the invention is superior to the existing method in complex multi-target tracking scenes including track intersection, a large number of false alarms and missing detection.

The invention relates to the technical field of logistics monitoring, in particular to a logistics service digital management system which comprises an environment sensor, an electronic nose and a visual sensor which are all arranged in a refrigeration logistics transportation device for transporting fruits and vegetables. The acquisition module is used for acquiring environment data of refrigerated fruits and vegetables at an environment acquisition frequency, acquiring smell data of the fruits and vegetables at an electronic nose acquisition frequency and acquiring surface images of the fruits and vegetables at a visual acquisition frequency; the state judgment module is used for identifying internal and external rotting states, pesticide residue states and appearance abnormal states of the fruits and vegetables; the frequency adjusting module is used for adjusting the environment acquisition frequency, the electronic nose acquisition frequency and the visual acquisition frequency; and the block chain storage module is used for storing the environment data, the smell data and the surface image which are acquired for multiple times in an account book of a block chain. According to the invention, the collection frequency of the sensor and the data volume stored in the account book of the block chain are controlled through the judgment results of various fruit and vegetable states, and the high processing speed of the block chain is realized.

The invention relates to the technical field of data processing, in particular to a data compression and block storage processing method based on deep learning. The method comprises the following steps: acquiring integrated data to be compressed; analyzing the effective to-be-compressed integrated analysis data according to the to-be-compressed integrated data; constructing a heterogeneous data feature analysis tree model; extracting to-be-compressed feature data of the effective to-be-compressed integrated data through the heterogeneous data feature analysis tree model; designing a recursive sparse coefficient evaluation objective function based on the to-be-compressed feature data; performing adaptive sparse compression processing on the effective to-be-compressed integrated data based on the recursive compression sparse coefficient evaluation objective function to obtain compressed data; performing block processing on the compressed data to generate block compressed data; and performing compressed data block storage optimization processing on the partitioned compressed data to generate an optimized storage compressed data block. According to the method, data features are analyzed through a deep learning algorithm, so that efficient processing of data compression and block storage is realized.

The invention provides a large language model-based rotation position coding method, which comprises the following steps of: carrying out iterative calculation on a trigonometric function through a preset iterative calculation formula, and obtaining a current round of trigonometric function value through a previous round of trigonometric function value; and obtaining at least one input vector, and carrying out element-by-element multiplication and addition operation on the input vector and the trigonometric function value of the round to obtain a vector code of the rotation position code. The invention further provides a coding device of the rotation position coding based on the large language model, a storage medium and electronic equipment. Therefore, according to the method, the balance of calculation precision, calculation efficiency and hardware overhead is realized, and the on-chip storage requirement is remarkably reduced, so that powerful support is provided for acceleration of a large language model.

The invention discloses a vehicle-mounted camera shooting management method and system, and particularly relates to the technical field of image communication, and the system comprises an intelligent image collection processing module which is used for synchronously collecting high-quality image data through a plurality of cameras and carrying out the preprocessing of the image data, and a vehicle-mounted camera shooting management module which is used for carrying out the real-time analysis and recognition of the collected image data, and the real-time image analysis and identification module is used for extracting useful information and providing real-time feedback according to an analysis result. Through efficient image acquisition, real-time analysis, data management and intelligent auxiliary driving functions, the driving safety, the driving stability and the data management efficiency are remarkably improved. The system utilizes a multi-camera synchronization technology and a convolutional neural network denoising algorithm to ensure that high-quality images are captured in various environments, and a dynamic range adjustment module optimizes the image quality under different illumination conditions.

The invention discloses a bridge support displacement online probability prediction method, which comprises the following steps: acquiring historical monitoring data of main beam temperature and support displacement, and introducing a group of false points to construct a probability prediction model based on a flow sparse Gaussian process, according to the approximate posterior distribution and the real posterior distribution, constructing a variational free energy boundary as a lower bound of a logarithmic marginal likelihood function, deriving an optimal variational free energy boundary and optimal approximate posterior distribution, and obtaining a specific expression of a probability prediction model; when new monitoring data is obtained, new monitoring data and old historical monitoring data are summarized in combination with new false points and old posterior distribution, new approximate posterior distribution is derived, new real posterior distribution is derived by using the Bayesian theorem, then model parameters and posterior distribution are updated online, and a new specific expression of the probability prediction model is obtained. According to the method, massive health monitoring flow data of the bridge is processed based on the flow sparse Gaussian process model, online probability prediction of bridge support displacement is achieved, and the method has high engineering application value.

The invention provides a model data processing method and device, and the method comprises the steps: obtaining original model data of a target model; determining repeated vertexes, deleting vertex data corresponding to the repeated vertexes to obtain de-duplicated vertex data, and performing quantization processing on the de-duplicated vertex data to obtain processed vertex data; according to the data type of the map data, compressing the map data to obtain processed map data; according to the material data, model components with the same material attribute are determined and combined, combined model components are obtained, and the material data of the combined model components serve as processed material data; and compressing and uploading the processed vertex data, the processed mapping data and the processed material data as new model data. By reducing redundant data, optimizing a data structure and adopting an efficient compression algorithm, light weight of model data is realized, so that the rendering efficiency and the transmission rate are improved.

The invention discloses a subway track surveillance video lightweight transmission method, which comprises the following steps: acquiring surveillance video data through an edge server, and performing corresponding frame extraction according to the region type of the surveillance video data to obtain first video surveillance data of different region types; decomposing the first monitoring video data through a first-level server to obtain a plurality of frames of images, performing image recognition according to the type of the located region to obtain monitoring event data, and dynamically adjusting the frame rate of the first monitoring data according to the type of the located region and the monitoring event data to obtain second video monitoring data; performing queue storage on the second video monitoring data according to different area types and the priority of the monitoring event data, and transmitting the second video monitoring data to a secondary server according to the priority of the video monitoring data and the condition of network resources; and dynamically compressing the second video monitoring data through the secondary server according to the duration, priority and different frame rate distribution proportions of the second video monitoring data to obtain third video monitoring data.

The invention relates to a content video-based hypercompression storage method. The method comprises the following steps of S1, acquiring video stream data of a camera through a network; s2, for the video stream data, performing corresponding decoding according to the code stream type to generate YUV data; s3, analyzing the YUV data, determining an ROI (Region of Interest), and extracting content feature values of a plurality of moving targets and position data of images where the moving targets are located as element data of the video; the ROI is logically separated from the non-ROI, secondary high-fidelity compression is carried out on the ROI, high compression is carried out on the non-ROI, and the high compression rate of the ROI under the high-fidelity condition is achieved; s4, combining the code streams of the ROI and the non-ROI into the same frame of code stream; s5, taking the ROI metadata of each region of interest as structured description data of the frame data; and S6, sending the coded code stream and the ROI metadata to a video storage engine for intelligent video storage.

The invention belongs to the related technical field of neuromorphic visual sensors, and discloses an eagle-eye bionic dynamic visual sensor and a preparation method and application thereof, the visual sensor comprises a dielectric layer substrate, a functional layer and a cross array structure composed of a bottom electrode, an insulating layer and a top electrode which are arranged from bottom to top; wherein the functional layer is of a core-shell heterojunction structure formed by noble metal nanoparticles and a two-dimensional metal chalcogenide. According to the obtained core-shell heterojunction structure, an electron capture and slow release phenomenon caused by local surface plasmon resonance effect excitation and interface defects of precious metal nanoparticles is constructed on a device, and the sensor can realize an eagle-eye-imitating dynamic visual function; and the visual sensor can perform eagle-eye-imitating dynamic visual processing on image information due to a unique electronic synapse function while realizing an image optical sensing function, so that integration of sensing and preprocessing functions is realized.

The invention relates to the crossing field of artificial intelligence and bird ecology, and particularly discloses a large-scale bird sound recognition method based on a deep learning technology, and the method comprises the steps: collecting a bird sound audio through a project region recording device, training a bird sound event detection model through a Bird audio detection challenge 2018 data set, carrying out the class prediction through field recording data, and carrying out the recognition of a bird sound event. Effective bird sound and background noise are separated; constructing a bird sound recognition data set in combination with a bird species catalog of the Chinese bird watching recording center and an audio file of Xen-Canto, training a bird sound recognition model by using background noise and the bird sound recognition data set, and performing verification; and finally, inputting the effective bird sound data into the model to carry out label prediction so as to identify the bird sound. The invention aims to carry out wide bird voice recognition on a project area under construction through a deep learning technology, so that the data processing cost is saved, and the recognition accuracy in a complex scene is improved.

The invention discloses a large language model fine tuning method based on LoRA improvement, and the method comprises the steps: obtaining a weight file of a large model, and obtaining a training data set according to a needed vertical field and a downstream task; building a fine adjustment platform; the method comprises the following steps: acquiring an open source LoRA code, and modifying the open source LoRA code to obtain a modified code; and performing fine adjustment on the selected large model by utilizing the training data set and the modified code, and loading the output LoRA module onto a weight matrix of the original model after fine adjustment is completed, so that the large model has the capability of better completing specific downstream tasks. According to the method, the training parameter quantity of the LoRA in the fine tuning process is reduced.

The invention provides an optical orbital angular momentum identification method and system based on a binary neural network model, and is used for solving the technical problems that an OAM identification method based on a spatial light modulator deep learning method is high in calculation complexity, large in memory occupation and difficult to meet the application requirements of light weight and real-time performance of space laser communication. According to the optical orbital angular momentum identification method based on the binary neural network model provided by the invention, the identification of different topological charges OAM is realized by constructing the binary neural network model, and compared with a traditional OAM identification method based on a spatial light modulator deep learning method, the method provided by the invention adopts a binary weight and a binary activation function, so that the identification accuracy is improved. And the parameter storage requirement is obviously reduced. Meanwhile, according to the binary neural network model, PopCount processing is adopted to replace convolution operation, the calculation amount is reduced, the real-time performance is enhanced, and the requirement of application scenes such as inter-satellite communication for rapid recognition can be met.

The present invention relates to the technical field of video compression processing. A monitoring video compression processing method and device are provided, the method comprising: using a video feature extraction model to extract video features of an image to be processed in a monitoring video of a target scene area, and obtaining feature vectors of multiple video features in different image areas; performing the following classifications on the image to be processed according to the feature vectors of multiple video features in different image areas, and obtaining classification results: image area classification, scene type classification, and importance level classification; determining the initial compression ratio of different image areas in the image to be processed according to the classification results; using an image decomposition method based on wavelet transform to decompose the image to be processed into sub-images of different scales, and determining the optimal number of decomposition layers according to the image quality; determining the final compression ratio and encoding mode of different image areas according to the scale and content characteristics of the sub-images of the image to be processed. The present invention greatly improves the compression ratio without significantly reducing the image quality.