8765 results about "Training data sets" patented technology

In embodiments, systems and methods for improving machine-learning systems are disclosed. In embodiments, a system includes a data pool system that is configured to receive data from a plurality of different data sources and maintain a training data set that is used to train a specific machine-learning model based on the data from the plurality of different data sources. In embodiments, the system further includes a data scoring system that determines a data reliability score corresponding to the new data based on a set of intrinsic features of the new data and a data scoring model, wherein the data pool system selectively adds the new data to the training data set based on the reliability score of the new data. The system also includes a machine learning system that trains the specific machine-learning model based on the training data set.

An example method can include providing a natural language instruction and user interface image data to a machine-learned sequence processing model that is configured to process image data and generate commands for controlling the target computing device, wherein the machine-learned sequence processing model has parameters learned using an interface recognition objective based on an evaluation of an interface recognition output generated based on processing a rendered training interface from a pre-training dataset and an interface navigation objective based on an evaluation of a user interface command generated based on processing a rendered training interface from a fine-tuning dataset; receiving, from the machine-learned sequence processing model, a command indicating an interaction with the user interface to implement the natural language instruction; and generating, based on the command, a control signal configured to initiate the interaction.

The invention relates to a model training method and device, equipment, a storage medium and a product. The method comprises the following steps: constructing a training data set according to a target text reasoning chain obtained by converting multi-modal data; according to the training data set, performing supervision fine tuning on the pre-trained multi-modal large language model to obtain a basic reasoning model; performing optimization processing on the basic reasoning model according to reinforcement learning training of long thinking to obtain a target reasoning model; the target reasoning model is used for outputting a target answer containing a reasoning process according to the input multi-modal data. Therefore, the long text constraint can be directly used for reinforcement learning, and the training efficiency is greatly improved; and by adopting long-thinking reinforcement learning training, the model can easily learn a correct thinking process in training, so that the reasoning ability of the multi-modal large language model for processing a complex visual reasoning task is improved, and the correct thinking process is displayed in the reasoning process.

The invention relates to the technical field of structural health monitoring and intelligent diagnosis, in particular to a structural fatigue damage identification method based on acoustic emission and deep learning, and the method comprises the steps: collecting a structural response signal under a fatigue load through an acoustic emission sensor array, inputting the structural response signal to a CNN-BiLSTM-Attention mixed deep learning model, and carrying out the recognition of the structural fatigue damage through the CNN-BiLSTM-Attention mixed deep learning model; the model extracts local time domain features through a dynamic adaptive convolution kernel, captures long time sequence dependence by using a bidirectional long-short-term memory network, focuses key damage features through a bimodal space-time attention mechanism, divides damage stages based on a nonlinear dynamic threshold algorithm of fracture opening amount, constructs a training data set of physical-data fusion, and performs dynamic time domain feature extraction. The learning rate is optimized by adopting a gradient sensitive cosine annealing algorithm, and the robustness of the model is improved in combination with an anti-noise and anti-loss function. The method integrates physical characteristics and an intelligent algorithm, and has the advantages of adaptive noise suppression, strong cross-domain generalization ability, high real-time performance and the like.

The invention provides a road bridge crack detection method and system, and the method comprises the steps: collecting a bridge surface multi-view image, and constructing a training data set containing crack feature labeling through quality screening and standardized labeling; preprocessing the image by using a multi-scale feature fused deep convolutional neural network and carrying out semantic segmentation, initially identifying a suspected crack region and generating a segmentation mask; and constructing a BeNNS proxy model based on the mask, and establishing a mapping relationship between the detection result and the bridge structure topology, the stress flow field and the service function chain so as to evaluate the result reliability. And inputting an evaluation result into a hybrid evaluation mechanism, performing online real-time detection and offline batch verification to optimize precision, and outputting a verified crack region. Finally, morphological analysis is conducted on the area, geometric parameters and danger levels of cracks are extracted and integrated to a bridge health monitoring system, a crack evolution tracking algorithm and an early warning mechanism are established, and dynamic tracking early warning is achieved. The problem of low detection precision in a complex environment can be solved.

The invention discloses a long-tail image recognition method based on multi-modal semantic generation and image-text fusion. The method comprises the following steps: extracting structured semantic description from a tail image; carrying out semantic rewriting and enhancement based on a multi-modal visual language model, and generating an image semantic extension description; the image semantic extension description is optimized based on semantic duplicate judgment and a style alignment mechanism, and an optimized text description set is obtained; inputting the optimized text description set into a text graph model, generating a tail class image sample, performing semantic and visual quality screening, and constructing to obtain an enhanced image set for training; constructing a training data set based on the original long-tail data set and the enhanced image set, and training an image-text fusion classification model; and inputting a to-be-identified image into the trained image-text fusion classification model, and outputting classification results of all categories. According to the method, the discrimination capability in a long-tail distribution scene is enhanced, and the method has a stronger generalization characteristic.

The invention discloses an improved YOLOv11-based high-reflection and high-transmittance material surface defect detection method, which comprises the following steps: S1, acquiring a mobile phone screen surface defect image, preprocessing and labeling, and generating a training data set containing three defects of scratches, edge breakage and cracks; s2, constructing a defect detection model; s3, averagely dividing an input image into four independent detection regions, adopting a bounding box overlapping area weight distribution strategy for cross-region defects, and performing partition counting on the image by using a defect detection model; s4, constructing a surface defect detection network meeting the real-time requirement of the industrial production line, and outputting a detection result and a processing frame rate; s5, pixel-level mask segmentation is carried out on the scratch defect, and contour coordinates and geometric features are extracted; and S6, fitting the actual damage size of the scratch defect on the surface of the high-reflection and high-transmittance material according to the segmentation result. According to the method, the detection speed is high under the condition that the accuracy is ensured, and a reliable solution is provided for surface flaw detection of the high-reflection and high-transmittance material.

The invention provides an infectious disease early warning and monitoring method and system based on big data and deep learning, and relates to the technical field of infectious disease early warning, and the method comprises the steps: obtaining infectious disease historical data, crowd movement track data, medical treatment data and environment monitoring data as a training data set; establishing a space-time propagation link diagram and carrying out dynamic segmentation coding to generate a propagation characteristic sequence; training a deep learning model based on the propagation feature sequence to predict an infectious disease propagation inflection point and a high-risk area propagation probability; and when the propagation probability exceeds a preset threshold value, generating early warning information and pushing the early warning information to a monitoring terminal. According to the invention, accurate prediction and timely early warning of the transmission trend of infectious diseases can be realized, and the epidemic prevention and control efficiency is improved.

The invention provides a self-adaptive polarization modulation-based inter-satellite laser communication and signal scheduling method, which relates to the field of laser communication, and comprises the following steps of: acquiring polarization state information of emitted and received light beams, calculating a mismatch degree, generating a compensation parameter, constructing a polarization state prediction model by using a deep learning neural network for pre-correction, and calculating the mismatch degree; a polarization modulator is adopted to perform real-time modulation and establish a link quality evaluation model, carrier synchronization and channel compensation are realized based on an adaptive optical carrier recovery technology of coherent detection, a link state is monitored in real time, and a training data set is updated.

The invention discloses a power transformer fault auxiliary decision-making method and system based on a knowledge graph and a large language model, and the method comprises the steps: obtaining power transformer fault text data, carrying out the preprocessing, obtaining fault-related text and table data, preliminarily defining an ontology, selecting a part of text data for marking, and carrying out the recognition of the ontology; a plurality of named entity recognition and relation extraction models are trained, an optimal model is determined, then triple extraction is carried out on unlabeled text data, and a knowledge graph is constructed; predicting a potential entity relationship in the knowledge graph based on a link prediction model, and complementing the knowledge graph under the large language model and human assistance; constructing a semantic search model training data set based on the large language model and training a semantic search model; based on the complemented knowledge graph, the trained semantic search model is utilized to retrieve knowledge graph sub-graphs according to the problem, and auxiliary decision making is completed. According to the method, the problem of low potential knowledge utilization level in the existing power transformer fault auxiliary decision-making based on the knowledge graph is solved.

A method, system, and device for wind speed prediction and layout optimization in wind power generation are provided. The method includes: obtaining a basic wind resource dataset of a target region; constructing a physics-informed neural network model based on the basic wind resource dataset; obtaining wind speeds data at a specific location in a velocity field based on the physics-informed neural networks and constructing a training dataset; training the physics-informed neural network model based on the training dataset; reconstructing a wind speed distribution within the velocity field and predicting wind speeds for a next time period with a wind farm using the trained physics-informed neural network model; and optimizing a layout of a wind turbine cluster based on a reconstructed wind speed distribution within the velocity field. The present application reconstructs a two-dimensional velocity field of the wind farm by training the PINN and enables accurate ultra-short-term wind speed prediction.

The invention discloses an intelligent detection method and device for fusing a medical image learning image, and relates to the technical field of medical image processing. The method comprises the following steps: acquiring and preprocessing a bimodal medical image, and extracting a feature map through multi-scale decomposition; constructing a cross-modal correlation model, and setting a modal attention mechanism (embedding anatomical structure prior guidance feature complementation) and a morphological attention mechanism (setting lesion morphological constraint weight); the method comprises the following steps: collecting multiple types of image samples, pairing according to a focus form and an imaging mode to construct a bimodal joint data set, and correlating and labeling to generate a training data set with modal attributes; after a multi-stage iteration training model, inputting the preprocessed image to carry out feature fusion so as to obtain a fused image; and generating a lesion probability graph according to the fused image, positioning a lesion area through multi-threshold segmentation, and outputting a detection result. The system comprises a data acquisition module, a preprocessing module and the like. The method improves the accuracy and reliability of medical image detection, and is suitable for clinical multi-modal image analysis.

The invention relates to an infrared and visible light image fusion method based on a cross-domain Transform, and belongs to the field of computer image processing. The method comprises the following steps: respectively carrying out preprocessing operation on an infrared image and a visible light image to obtain a training data set; an end-to-end image generator network is designed, an encoder module is used for extracting deep semantic features of an infrared image and a visible light image, a fusion module introduces an axial attention mechanism to enhance the global modeling capability of the features, and feature fusion is carried out in combination with information of a spatial domain and a frequency domain; the fused features are gradually recovered to an image space through a decoder module, and a fused image is generated; constructing a fusion loss function module, and guiding the network to focus a significant feature difference between the source image and the fusion image based on a comparative learning idea; and finally, inputting the infrared and visible light image Y channel into the network model, generating a fusion image, completing a training process, and realizing unified optimization of fusion performance and visual quality.

The invention discloses a deep counterfeiting detection model training method, a deep counterfeiting detection method and a deep counterfeiting detection system, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a training data set containing a real image and a plurality of counterfeit images, and enabling the image to be provided with a label for representing the authenticity; in the training process, the deep forgery detection model can be in contact with various types of image samples, so that wider and more complex image features and forgery modes can be learned, a forgery reason is further marked for a forgery image, and the deep forgery detection model can be helped to deeply understand essential features of forgery content in the training process. Therefore, the problem of insufficient detection capability for well-designed and high-quality counterfeited contents can be solved, and the technical effect of improving the accuracy of counterfeited detection is achieved.

The invention discloses a coal mining subsidence area surface crack intelligent extraction method based on an improved Transform model, and belongs to the technical field of remote sensing image processing. Firstly, an unmanned aerial vehicle carrying a high-resolution optical camera is used for collecting images, the image overlapping rate of 70%-80% is guaranteed, and a training data set is constructed through professional labeling, cutting screening and data enhancement. The encoder of the innovative model is very distinctive, and the adaptive multi-scale patch mapping layer can dynamically adjust the patch size according to the local complexity of the image and efficiently extract features; double-attention fusion is combined with optimization position coding, and long-distance dependency capture is enhanced; and the calculation amount and the overfitting are reduced by the dynamic sparse connection full-connection layer. Residual attention enhancement pyramid pooling and a space-channel attention bottleneck mechanism are adopted, key features are highlighted, and noise is suppressed; and a breakpoint detection and connection rule determination module is utilized to realize complete restoration of the ground fracture. After the data set is used for training a model, deployment is carried out, and through preprocessing, encoding and decoding and post-processing, ground fracture information can be accurately obtained, and a data foundation is built for mining area safety management and geological disaster prevention and control.

The invention discloses a rolling bearing fault diagnosis method based on a multi-scale residual attention network and an adaptive Transform encoder. The rolling bearing fault diagnosis method comprises the following steps: acquiring original vibration data in the running process of a rolling bearing; segmenting the collected original vibration data into samples with specified lengths, and dividing the samples into a training data set and a test data set; inputting the training data set into a multi-scale residual attention network to perform preliminary multi-scale feature extraction; inputting the feature information extracted by the multi-scale residual attention network into an adaptive Transform encoder to obtain time sequence features; finally obtained feature information is subjected to GAP processing and then is input into a Softmax layer for fault diagnosis; the forward propagation calculation and the back propagation calculation are repeatedly executed to optimize model parameters until the diagnosis accuracy and loss of the training data set reach a stable level; and inputting the test data set into the trained model for fault diagnosis, and determining the health condition of the rolling bearing. According to the method, the adaptability and the diagnosis accuracy in time sequence dependence scenes such as rolling bearing fault diagnosis are enhanced.

The invention provides a multi-modal image fusion method based on modal self-adaption and modal interaction compensation, and the method comprises the following steps: S1, obtaining a multi-modal image fusion data set, and obtaining a training data set through preprocessing; S2, analyzing the modal difference characteristics of infrared and visible light images, and evaluating the correlation characteristics of image pairs in different scenes; s3, capturing a cross-modal feature dependency relationship through a self-attention mechanism; s4, a differential feature extraction strategy is adopted, model parameters are optimized through iterative training, and multi-modal image fusion is completed; s5, a modal interaction compensation module is additionally arranged, unit dynamic balance common features and modal exclusive features are fused, feature complementation is achieved in channel and space dimensions, parameters of the modal interaction compensation module are optimized, the model is made to learn the optimal fusion weight of the multi-modal features in a self-adaptive mode, and multi-modal fusion image generation optimization is achieved through the model; according to the invention, multi-modal image fusion can be accurately and effectively carried out.

The invention discloses a disaster intelligent Internet of Things multivariable time sequence anomaly detection method and device, and relates to the technical field of Internet of Things anomaly detection, and the method comprises the steps: S1, constructing an initial anomaly detection model; s2, acquiring a training data set; s3, performing optimization training on the initial anomaly detection model by using the training data set to obtain an optimized anomaly detection model; s4, acquiring real-time monitoring data; s5, analyzing the real-time monitoring data by using the optimized anomaly detection model to obtain a detection result; the dynamic gated expansion convolutional network DGDC solves the problems of rigid structure and parameter explosion of a traditional TCN. Dynamic expansion rate scheduling enables a receptive field to expand in an exponential level along with the number of layers, and second-level burst and week-level periodic characteristics can be captured at the same time; the parameter quantity is reduced by 60%-70% through depth separable convolution, and the efficiency and precision of local feature extraction are both superior to those of an existing convolution module by combining the suppression effect of a gated linear unit GLU on noise features.

The invention discloses a cross-cultural customer service dialogue quality automatic evaluation method in combination with sentiment analysis, and relates to the technical field of natural language processing, and the method comprises the steps: carrying out the alignment of voice and text based on a transmission matrix in real time, extracting a speech, a metaphor and polarity, and generating a speech tag; constructing an emotion channel and a polite channel, and fusing expression and shielding intensity through sharing attention; comparing and aligning with the same language prototype in a regional culture baseline library to obtain a calibration representation and updating a language offset record table; the potential upgrading probability is represented and recurred according to round aggregation calibration, and a risk vector and a high-risk position are formed; fusing risk and business indexes by a capacity integral kernel, outputting a comprehensive quality score, and giving factors and round attributions; sample recovery is triggered according to score and feedback difference, a micro-weight training data set is constructed, gradient increment training is carried out under low-rank adaptation, and cross-language consistency, early recognition of upgrading risks and interpretable evaluation are achieved through a closed loop.

The invention discloses a renewable energy power generation power prediction and power dispatching method and system, and the method comprises the steps: collecting the historical power generation data and real-time meteorological data of renewable energy power generation, carrying out the linear interpolation of the historical power generation data and the real-time meteorological data, and carrying out the missing value filling and box plot anomaly detection, obtaining a normalized training data set; constructing a hybrid prediction model by using the normalized training data set and adopting a neural symbol acceleration technology with time logic constraints, extracting medium and long term space time features, and generating a renewable energy power generation power prediction result; and according to the renewable energy power generation power prediction result and the system constraint condition, adopting a linear one-dimensional projection constrained distribution robust control method to formulate a scheduling strategy, and utilizing the scheduling strategy to solve an optimal scheduling scheme through mixed integer linear programming. According to the method, the renewable energy power generation power prediction precision and the power dispatching robustness are remarkably improved.

The invention relates to the technical field of traffic scene target detection, in particular to a traffic scene target detection method based on feature fusion and an attention mechanism, and the method comprises the steps: obtaining a public traffic scene target image training data set; training a traffic target detection model by using the traffic scene target image training data set; and inputting a to-be-detected traffic scene image into the trained traffic scene target detection model to obtain an output detection result graph. A hierarchical receptive field is constructed through a feature fusion module, a fine structure of a target edge is reserved in a deep convolution stage, cross-channel semantic information is dynamically aggregated through point convolution, an attention module is put forward to dynamically adjust a convolutional receptive field and a space attention weight, interference of background noise is suppressed, and a target is obtained. The local detail feature expression of the target is better enhanced, a PIoU loss function is adopted, and the positioning precision and robustness are further improved by combining a target size adaptive penalty factor and an anchor frame quality-based gradient adjustment strategy.

The invention discloses a multi-granularity visual reasoning model construction method and device based on reinforcement learning. The method comprises the following steps: constructing an'image-reasoning query-bounding box 'triple as a training data set; designing a composite reward function including positioning precision, target counting precision and format reward; training the multi-modal large language model by adopting a GRPO algorithm; the trained model can output a region-level bounding box, and a pixel-level mask is generated and a contour-level result is extracted in combination with the segmentation model. According to the method, the problems that in the prior art, a reasoning path depends on manual annotation, multi-granularity tasks cannot be expanded, and generalization is insufficient are solved, and the autonomous decision-making ability, task expansibility and generalization in a distribution offset scene of the model are improved.

The invention relates to the field of water conservancy and hydropower engineering, in particular to a multi-objective optimized hydropower ecological scheduling decision-making system and method, and the system comprises a data collection module, an ecological model module, an intelligent decision-making engine module, a scheduling execution module, an effect evaluation module and a knowledge base module. The data acquisition module collects multi-source data, the ecological model module generates a training data set, the intelligent decision engine carries out ecological process modeling and probability prediction based on a deep learning architecture and spatial-temporal feature extraction, and generates a scheduling decision, the scheduling execution module controls hydropower engineering operation, and the effect evaluation module monitors ecological and economic effects. The knowledge base module stores historical experience and provides optimization suggestions, and the system improves the simulation accuracy of the ecological system, especially when the flow changes suddenly; and through uncertainty quantification, the system reliability and the ecological safety guarantee rate are enhanced, and high efficiency, accuracy and sustainability of ecological scheduling of the hydropower engineering are realized.

The invention relates to the technical field of motor state monitoring and fault prediction, and provides a motor residual life analysis method and system based on a support vector machine, and the method comprises the steps: building a multi-dimensional degradation feature set of a motor, and calculating a health index degradation rate based on multi-source sensor data and a failure threshold; constructing a support vector machine regression model of an adaptive kernel function, predicting a health index change track by using the model and real-time data, and generating a residual life evaluation result if a prediction deviation is within an allowable error range; otherwise, starting an incremental learning mechanism to update the training data set, dynamically adjusting kernel function parameters, and recalculating the trajectory; and if the error requirement is still not met, model regularization parameters are optimized in combination with the working condition data until the residual life evaluation result is converged. According to the method, the accuracy and dynamic adaptability of motor residual life prediction can be improved, and the robustness of the model to complex working conditions is enhanced.

The invention provides a meteorological downscaling method based on space-time fusion and physical constraint, and belongs to the technical field of meteorological downscaling, and the method comprises the steps: carrying out the preprocessing of multi-source meteorological related data and a high-resolution meteorological truth value, and constructing a training data set; an improved U-Net model is constructed, spatiotemporal features and multi-source auxiliary features are obtained through a multi-branch feature extraction unit, high-resolution information is recovered through fusion and decoding, and an attention enhancement module is embedded to highlight a key area; a model is trained through a training data set, parameters are optimized by adopting a loss function fusing topographic features and physical rules, and prediction error differentiation constraint on a complex area and violating the physical rules is achieved; and preprocessing target low-resolution data, inputting the preprocessed target low-resolution data into the model, and outputting high-resolution meteorological data and a physical attribution result. The problems that in the prior art, multi-source meteorological data fusion is insufficient, downscaling precision of a complex terrain area is insufficient, and prediction errors violating physical laws are lack of effective constraints are solved.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes of financial science and technology, medical health and the like, and discloses an image detection method, device, equipment and medium based on adversarial generation. Performing global feature modeling by adopting an image feature extraction module of a self-attention mechanism, generating a forgery probability graph in combination with a forgery region recognition module, constructing a joint loss function based on a detection loss value and an adversarial loss value, and optimizing model parameters of a generator, the feature extraction module and the recognition module through the joint loss function to obtain a forgery probability graph; and finally, a counterfeit detection model for identifying counterfeit information in the image is formed. According to the method, the diversity of training data is improved through adversarial sample generation, the image feature modeling capability is enhanced through a self-attention mechanism, multi-dimensional loss optimization is realized through fusion of counterfeit region difference information, and the detection precision and robustness of the model to a counterfeit image are improved.

The invention discloses an event information guided image deblurring and high-frame-rate reconstruction method. The method comprises the following steps: step 1, obtaining a blurred image and an event stream corresponding to the blurred image to construct a training data set; 2, constructing a deblurring and frame insertion combined reconstruction network based on a dynamic cross-modal fusion module; step 3, training to obtain a trained joint reconstruction network based on the cross-modal attention mechanism and the UNet architecture; and step 4, inputting a to-be-reconstructed blurred image and a corresponding event stream into the trained deblurring and frame insertion combined reconstruction network based on the dynamic cross-modal fusion module to obtain a final multi-frame reconstruction image. According to the method, a higher weight dynamic state is given to an event mode in a high-speed motion scene, an image data weight is added in a static scene, and a scale feature fusion strategy is adopted for a seriously fuzzy scene, so that the problems of semantic difference and noise interference are reduced, and the image recovery quality of a fuzzy region is improved.

The invention discloses an online prediction method and device for node voltage of a time-varying topology direct-current power distribution network, and the method comprises the steps: building a simulation model based on the historical operation data of the direct-current power distribution network in combination with the real topology of a power grid, and generating a time sequence training data set containing voltage, power and line states through dynamic topology simulation; based on the training data set, constructing a dynamic space-time diagram neural network fused with topology dynamic change, and learning a voltage time sequence evolution rule in combination with a time sequence module; node power data and topology information are collected in real time by using a dynamic space-time diagram neural network, network topology input is dynamically updated, and a future short-time domain voltage predicted value is output; and for a power grid topology change scene, graph connection parameters associated with change nodes are updated, rapid fine adjustment of the model is realized, and stable prediction is carried out. According to the method, network topology information can be dynamically input under the condition that the topology of the direct-current power distribution network is changed, and rapid and stable prediction of the node voltage of the direct-current power distribution network under the time-varying topology is achieved.

The invention discloses a lower limb exoskeleton gait track prediction method based on an LSTM-KAN fusion model, and the method comprises the steps: collecting human motion data through a sensor assembly, and carrying out the filtering, missing value processing and normalization of the human motion data; then constructing an overall architecture of a prediction model based on an LSTM-KAN network, optimizing parameters of the prediction model by using a particle swarm optimization (PSO) algorithm, extracting key features in the preprocessed data as a training data set, and inputting the training data set into the prediction model for training; and finally, collecting current human body motion data, pre-processing the current human body motion data, inputting the pre-processed current human body motion data into the trained prediction model, and outputting future human body gaits and tracks by the prediction model. And the controller takes a future gait track generated by the prediction model as a reference track to generate a driving signal and sends the driving signal to the actuator so as to realize accurate control of the actuator. By adopting the gyroscope sensor, the acceleration sensor and the pressure sensor for gait estimation, exoskeleton motion control can be effectively improved, and man-machine interaction experience is effectively improved.

The invention discloses a multi-fingered dexterous hand grabbing posture generation method and system based on CVAE and a Ball Query algorithm, and belongs to the technical field of robot grabbing control. The method comprises the following steps: (1) a data sampling step; (2) a data preprocessing step: carrying out standardization processing on the collected data, generating enhanced point cloud data and constructing a training data set; (3) a model training step: through a multi-scale feature extraction module, combining global semantics and local geometric features extracted by a Ball Query algorithm, generating a grabbing attitude by using a conditional variation auto-encoder, and optimizing model parameters through reconstruction loss and KL divergence; and (4) real-time deployment: integrating the trained model to a physical platform, screening candidate grabbing postures based on parallel collision detection, and realizing real-time grabbing control in combination with inverse kinematics verification. The problem that a traditional method is insufficient in generalization ability in complex object grabbing is effectively solved, and self-adaptive grabbing of unknown objects is achieved while grabbing stability is guaranteed.