1371 results about "Gradient descent" patented technology

The invention relates to a task scheduling optimization method and device based on reinforcement learning, equipment and a medium. The method comprises the steps that firstly, system resource state data are collected in real time, dynamic environment characteristics are determined through preprocessing and time sequence analysis, task characteristic data are analyzed at the same time, and a task priority sequence and a resource demand vector are generated through a priority ranking algorithm and a resource evaluation model; and then a state space and an action space are constructed by adopting a reinforcement learning algorithm, an optimal task allocation scheme is generated through strategy iteration and reward function optimization, and if the scheme meets a resource balance threshold, scheduling is executed, and performance indexes are collected. And finally, fusing real-time indexes with historical data, and updating parameters of the reinforcement learning model through experience playback and gradient descent to form a closed-loop optimized improved scheduling strategy. By adopting the method, the accurate mapping of the resource state and the task requirement can be realized, and the problem of insufficient adaptability of the traditional static scheduling to a complex scene is solved.

The invention discloses a solar radiation space-time prediction method and system based on physical information constraint and a neural network, and the method comprises the steps: collecting multi-dimensional time sequence meteorological data, extracting high-dimensional time sequence dynamic features, converting geographic space data into a fuzzy set, and carrying out the defuzzification of the fuzzy set through an inference rule, thereby obtaining geographic space features; and a gating mechanism is adopted to realize deep fusion of the space-time features to generate high-dimensional space-time fusion features. In a model training stage, an energy conservation equation is introduced into an optimization process, a physical residual error is constructed by calculating a time derivative and a space derivative of a predicted value, a physical constraint total loss function is formed in combination with a data loss item, and model parameters are updated by using a gradient descent method. According to the method, the accuracy and reliability of a prediction result are remarkably improved while the calculation efficiency is ensured, and the method is particularly suitable for solar radiation prediction under complex meteorological conditions; according to the method, abnormal prediction caused by data noise can be effectively corrected, and a solution with physical rationality and data adaptability is provided for the fields of solar resource evaluation, photovoltaic power generation power prediction and the like.

The invention provides a multi-modal heterogeneous model retrieval enhancement method and system, and the method comprises the steps: building a knowledge and application example double-corpus based on user multi-modal query, and designing a joint retrieval mechanism to obtain a result set; mapping and scheduling to obtain feature representation through special processing channels for texts, images and audios and a Spiking neural network with a segmented trapezoidal topological structure; constructing a three-stage cascade architecture of a basic model, an advanced model and human experts, and obtaining a decision path and answer candidate set in combination with a recursive and discarding decision mechanism; a Hamiltonian graph network is used for representing a multi-modal relation, and a gradient-free descent method is used for rapidly training and optimizing model parameters; an enhanced retrieval result is obtained through cross-modal semantic alignment and dynamic retrieval window adjustment; and high-quality response is obtained through context-aware sorting and retrieval enhanced reasoning. According to the method, the multi-modal information retrieval processing efficiency and the heterogeneous model reasoning response quality are improved.

The invention discloses a multi-source-domain multi-teacher knowledge distillation method and system based on reinforcement learning, and the method comprises the steps: obtaining target domain sample data, inputting the data into N pre-trained teacher models, and generating the output features of all teacher models; inputting the target domain sample and all teacher model outputs into a reinforcement learning strategy network, generating a dynamic weight of each teacher model, and calculating a knowledge distillation loss function based on the dynamic weights; constructing a total loss function according to the knowledge distillation loss function and the cross entropy loss output by the student model; student model parameters are updated through gradient descent; and calculating a reward value according to student model performance change, and updating reinforcement learning strategy network parameters. According to the method, the reward function based on student model performance improvement is constructed, the strategy network is continuously updated in a strategy gradient optimization mode, the distillation efficiency is effectively improved, knowledge conflicts among teachers are relieved, and the robustness and generalization performance of the student model in a multi-source complex environment are remarkably improved.

The invention relates to the technical field of electronic detection, and discloses an electronic component dispensing detection method. The method comprises the following steps: acquiring dispensing image data of the surface of the electronic component, and generating a standardized dispensing data matrix containing glue position, thickness and uniformity characteristics through standardized preprocessing; constructing a dispensing correction matrix based on an adaptive window frame, and performing spatial reference dynamic correction on the standardized data matrix to obtain a spatial correction dispensing data matrix; inputting the data into a multi-layer sensor fusion network for feature fusion, and outputting a multi-source fusion dispensing data set; constructing a multi-dimensional abnormal feature incidence matrix based on the data set, and identifying abnormal dispensing data nodes by using a dynamic threshold detection algorithm; performing parameter optimization iteration on the multi-source fusion data set by using a gradient descent optimization algorithm to generate an optimized dispensing parameter set; and finally, constructing a three-dimensional visual dispensing quality model, and establishing a dynamic mapping relationship between model parameters and glue physical characteristics. The method can more comprehensively detect the glue quality.

The invention discloses a dynamic self-adaptive recommendation strategy optimization method for a business handling failure scene, and relates to the technical field of business handling recommendation and intelligent decision making, and the method comprises the steps: firstly collecting various types of data of a whole business handling process, and guaranteeing the integrity and real-time performance at a frequency of 100 milliseconds per time; a decision tree and Bayesian network fusion algorithm is used for attribution, and direct and indirect reasons are clarified; integrating data to construct a user portrait, and mining potential and subsequent demands; generating a recommendation scheme set based on attribution and portraits, and adjusting priorities and forms in combination with scene features; feedback data is introduced, a strategy weight is optimized by using a gradient descent algorithm, and the scheme is updated regularly; a multi-dimensional index weighted evaluation effect is set, and emergency optimization is carried out if the evaluation result does not reach the standard; and establishing a distributed strategy library, and reusing the optimal strategy of the similar scene by using a K-nearest neighbor algorithm. According to the method, failure reason accurate positioning and personalized recommendation are realized, the recommendation effect is continuously optimized along with data accumulation, and the method is adaptive to multiple service types and user groups.

The invention provides a slope displacement monitoring data processing system based on an unmanned aerial vehicle laser radar, and relates to the technical field of data processing, and the system comprises the steps: carrying out the spatial interpolation processing of a topographic feature data set, and constructing a digital topographic surface model; the displacement field calculation module is used for performing iterative optimization based on a digital terrain surface model through spatial similarity analysis and fusion with a gradient descent algorithm, calculating a slope surface displacement vector field, identifying a potential sliding surface and a deformation abnormal region, and generating a displacement field calculation result; and the evaluation module is used for inputting a displacement field calculation result into a risk evaluation model and carrying out slope stability quantitative evaluation through a multi-source data fusion analysis platform. According to the invention, the practicability and operability of the monitoring result are improved.

The invention relates to the technical field of robot joint control, and discloses a robot dog mechanical joint control method. The method comprises the following steps: acquiring a whole-body joint real-time sensing data set which covers the torque reading of a multi-axial force sensor, the angle data of a joint encoder and the body attitude data of an inertial measurement unit; motion state compensation is carried out on the sensing data based on a dynamic window mechanism, and a dynamic compensation matrix containing kinematics compensation parameters is generated; and inputting the multi-modal motion control data set into a gradient descent model under space-time constraint to complete multi-modal motion feature fusion so as to obtain a fusion motion control data set. Joint behavior abnormity is detected through a Lyapunov stability analysis algorithm, and abnormal nodes are identified; and iteratively optimizing control parameters through a fuzzy PID optimization algorithm to generate an optimized control parameter set, finally constructing a three-dimensional virtual motion space model, and establishing a dynamic mapping relationship between a virtual track and an actual mechanism.

The invention belongs to the technical field of signal synchronization, and discloses a multi-camera anti-shake time sequence synchronization control system and method based on an FPGA. The method comprises the following steps: taking an FPGA as a core, generating an FSYNC signal as a camera exposure reference, and obtaining a delay value through a timestamp and differential measurement; acquiring I MU data and obtaining a global motion vector through a gradient descent algorithm; analyzing by combining the two to obtain phase deviation and feeding back the phase deviation to the FPGA, and adjusting the FSYNC signal by combining the global motion vector by the FPGA according to the phase deviation; according to the method, nanosecond timestamp recording is realized through a homologous clock reference and hardware-level signal binding mechanism, and the limitation of dependence on static time difference compensation in the prior art is broken through in combination with global motion vector analysis and dynamic phase offset adjustment.

The invention discloses a brain deep nerve stimulation method and system based on adaptive adjustment, and relates to the technical field of brain deep nerve regulation, and the method comprises the steps: collecting a local field potential signal of a brain deep target region of a target patient, and extracting a beta frequency band power spectrum density and a gamma frequency band phase synchronization index as neural activity characteristic parameters; determining an individual baseline value and a preset threshold value based on historical data, and outputting a stimulation adjustment trigger signal when the beta frequency band power spectral density exceeds the individual baseline value and the gamma frequency band phase synchronization index is lower than the preset threshold value; in response to the trigger signal, calculating an optimal stimulation parameter combination through a gradient descent optimization algorithm and executing nerve regulation; and monitoring the signal change after regulation and control, calculating a relative change rate and updating a threshold value. Through a two-parameter joint judgment mechanism and a threshold updating strategy, individualized adaptive adjustment of stimulation parameters is realized, and the stimulation accuracy and the treatment effect are improved.

Methods, systems, and computer-readable storage media for providing an initial version of a prompt template, the prompt template including dynamic input and first static input, generating a prompt using the initial version of the prompt template at least partially by populating the dynamic input with training data, receiving, from a large language model (LLM), an output that is responsive to the prompt, providing an evaluation at least partially based on the output, and selectively updating the prompt template to provide an updated version of the prompt template by prompting the LLM at least partially based on the evaluation, the updated version of the prompt template including second static input that is generated by the LLM and that is different from the first static input.

A hybrid neural network architecture is disclosed that integrates matrix multiplication-free (MatMul-free) transformation layers with spiking neural network (SNN) layers for efficient, low-power computation. The system includes an interface module configured to convert intermediate continuous-valued data from MatMul-free layers into a spike-compatible format using encoding techniques such as rate coding, phase coding, or threshold-based conversion. The SNN layers process the spike-encoded data in an event-driven manner, enabling sparse, temporal inference. Training is supported by a hybrid optimization strategy combining backpropagation in MatMul-free components with surrogate gradient descent or spike-timing-dependent plasticity (STDP) in SNN layers. The architecture reduces computational complexity, supports real-time adaptability, and enables deployment in energy-constrained environments such as edge devices and neuromorphic platforms. The system may be implemented in hardware, software, or a co-designed pipeline optimized for dynamic sensor data, control signals, or continuous inference tasks.

A method for updating a tabular search large language model (LLM) includes performing data pre-processing on new data associated with the tabular dataset to obtain a set of sequences, applying a first fine-tuning operation on the tabular search LLM using the set of sequences, applying a second fine-tuning operation on the tabular search LLM using training data to obtain a set of final loss results and a set of updatable gradients, wherein the training data comprises at least the set of sentence predictions, applying an optimization function on the set of final loss results and the set of updatable gradients to obtain optimized gradient descent parameters, and applying the updated tabular search LLM to a new input associated with the new data to produce a new output.

The invention provides a novel magnesium alloy refining control system which comprises the steps that according to melt velocity field and concentration field data, a real-time data set is combined, melt state feature vectors are extracted, and an initial state feature set is generated and used for follow-up prediction and optimization processing; aiming at different sensor sampling frequencies and data formats, applying a timestamp aligned multi-source data synchronization algorithm, fusing the real-time data sets, and generating a fused data set with a unified format; according to the fusion data set and the initial state feature set, applying a long short-term memory network time sequence prediction algorithm to generate future change trend data of melt component distribution and flow state; and according to the change trend data and the initial state feature set, a refining parameter optimization model based on gradient descent is constructed, an optimized refining process parameter set is generated, and the melt state is adjusted.

The invention discloses a photovoltaic digital twin operation and maintenance platform oriented to multi-source collaborative diagnosis, and particularly relates to the technical field of photovoltaic operation and maintenance, normalization, time-frequency feature extraction and adaptive filtering are performed on component health degree and environmental parameters through edge nodes, and a lightweight encrypted performance abstract is generated, and is subjected to differential compression and synchronized to a cloud end; the cloud executes security federal fusion and dynamic reputation weighting on the abstract, and outputs a robust global performance vector; the collaborative optimization engine calls reinforcement learning based on the vector and a historical strategy to generate an original power distribution strategy, and the original power distribution strategy is issued after being corrected through a projection gradient descent mapping function; an edge actuator loads an environment snapshot to perform parallel simulation verification and risk assessment in a digital twin environment, so that the problems of multi-source data isolation and privacy disclosure are effectively solved; a self-adaptive multi-resolution grid, heterogeneous acceleration parallel and distributed message scheduling are provided, and high efficiency and real-time performance of large-scale multi-physics coupling simulation are achieved.

The invention belongs to the technical field of mobile mechanical arm grabbing, and provides a mobile mechanical arm grabbing control method and system based on machine vision, and the method comprises the steps: collecting an image in the moving process of a mobile mechanical arm according to a preset grabbing path, recognizing a dynamic obstacle in the image through an RT-DETR algorithm, and generating a prediction track of the dynamic obstacle, collision detection is conducted on a preset grabbing path of the movable mechanical arm, an obstacle avoidance track switching instruction is generated by combining the current time point of the mechanical arm with the predicted collision time point of the dynamic obstacle, trajectory tracking is conducted on the obstacle avoidance track of the mechanical arm, a model reference self-adaptive control algorithm is adopted, a reference model is established, and obstacle avoidance is achieved. According to the method, mechanical arm vibration caused by obstacle avoidance track switching is effectively restrained, interference of vibration on grabbing stability is reduced, and the operation precision and reliability of the mechanical arm in a dynamic obstacle avoidance scene are remarkably improved.

The invention belongs to the field of industrial thermal engineering and intelligent modeling, and discloses a heat exchanger heat exchange coefficient prediction method and system based on a physical information neural network. The method comprises the following steps: acquiring multi-dimensional operation data through a signal acquisition system, cleaning abnormal and blank values, standardizing, and segmenting into time sequence samples by adopting a sliding window method; a double-layer physical information long-short-term memory network is constructed, and a time sequence feature and a physical equation residual error are combined to generate a space-time fusion feature matrix. And a composite loss function including data loss, physical equation loss and physical consistency loss is designed, physical and data driving influences are balanced through hyper-parameter tuning, and accurate prediction of the heat exchange coefficient is achieved based on a gradient descent optimization model. The method combines field physical laws and data features, improves the reliability and physical interpretability of prediction, and is suitable for operation optimization of the heat exchanger of the desulfurization wastewater treatment system of the thermal power plant.

The invention discloses a multi-agent combat task cooperation method for a structure entropy guided graph neural network, and the method comprises the steps: S10, each combat agent interacts with an environment according to an action generated by a strategy network, the environment comprises environment information, task parameters and a preset task target, and the strategy of each combat agent is completely executed in a decentralized manner; collecting complete empirical trajectory data; s20, using the collected data for centralized training; performing value evaluation on the global state of each time step by using a value network; s30, calculating strategy loss and value loss by using a multi-agent near-end strategy optimization algorithm in combination with the output of the strategy network and the value estimation of the output of the value network; updating parameters of the strategy network and the value network by using a gradient descent method; and S40, performing loop iteration. The problems that in a traditional method, the battlefield game dynamic structure sensing ability is insufficient, the hierarchical strategy learning and generalization ability is limited, the adaptability of a model in a small sample area is poor, and the migration efficiency is low are solved.

The invention relates to the technical field of artificial intelligence, in particular to an injection molding process parameter optimization method and system based on hybrid algorithm and model fusion, and the method comprises the steps: optimizing a parameter combination of a support vector regression model through a simulated annealing algorithm, building a weighted fusion model based on the optimized support vector regression model and a random forest, and optimizing the model; constructing a hybrid model of an adaptive selection weighted fusion model and an optimized support vector regression model; constructing a three-objective optimization model including buckling deformation, volume shrinkage and production energy consumption, and searching a Pareto optimal solution set in a process parameter space by adopting a multi-objective genetic algorithm by taking the hybrid model as a target value evaluation tool; carrying out local correction on the key process parameters by adopting a gradient descent method until the deviation falls back to be within a preset threshold value, and obtaining optimized process parameters; the defect rate of products can be reduced, and meanwhile production energy consumption is reduced.

The invention provides a zero-sum game-based distributed elastic consensus optimal control method under denial of service attack of a multi-agent system, and the method specifically comprises the steps: firstly, constructing a multi-agent formation model in which a leader and a plurality of followers cooperatively move through graph theory knowledge and a multi-agent second-order state equation; secondly, in order to reduce the influence of denial of service attack on communication topology, a time-varying weight distributed elastic observer is provided to estimate the state of a leader, and the attacked condition of the leader is considered; then, by constructing an augmentation system, a distributed consistency tracking problem with a leader is converted into a local tracking problem between each follower and a virtual leader thereof; and finally, in order to solve the zero-sum game problem, introducing a Hamiltonian-Jacobian-Ansaxophone equation to realize optimal control input under maximum external disturbance, and realizing algorithm design by using single-evaluation reinforcement learning with experience playback and combining a gradient descent method.

The invention discloses an unmanned aerial vehicle transformer substation inspection path planning method and device, equipment and a medium. The method comprises the following steps: firstly, generating a grid map; dividing the substation inspection space into a plurality of sub-regions; establishing an electromagnetic exposure intensity model, and setting an electromagnetic exposure intensity safety threshold; randomly sampling a plurality of nodes in each divided sub-region, constructing a random geometric graph, and generating an initial inspection path by adopting an RRT algorithm; optimizing the generated initial inspection path by using a path pruning optimization algorithm and a gradient descent convex optimization algorithm, and smoothing the path into a continuous curve by using a B-spline curve fitting technology to obtain a final optimized inspection path; according to the invention, through an overall strategy of first region division and then path planning, random geometric graph tree building, path pruning optimization, gradient curve smoothing and key region guiding sampling mechanisms are fused, and the efficient, stable and high-environment-adaptability unmanned aerial vehicle three-dimensional path planning method for the substation complex scene is realized.

The present invention relates to the technical field of natural ventilation in buildings, and in particular to a computationally assisted decision-making method and system for climate-adaptive building cavity design. The method comprises: using a Delaunay triangulation method to generate an initial mesh, using a Laplace operator-based mesh refinement method to adaptively refine the initial mesh, so as to obtain an adaptive mesh system, wherein the adaptive mesh system is used for dynamically adjusting the mesh density of conditional PINNs; constructing a multi-task learning framework within the conditional PINNs, wherein the multi-task learning framework is used for jointly predicting a plurality of physical field variables within the conditional PINNs; and iteratively training the conditional PINNs, and using a gradient descent algorithm to minimize a loss function until a predetermined number of training iterations or loss convergence is reached, thereby generating a key physical field variable prediction model in building cavity design. The present invention can implement efficient and accurate prediction of key physical quantities in building cavity design, and can be adapted to different building layouts and functional space characteristics.

The invention discloses a virtual synchronous direct drive fan subsynchronous oscillation suppression method based on two-stage lagging link phase compensation control. The method comprises the steps that 1, a phase compensation control module formed by connecting two stages of lagging links in series is constructed; 2, performing parameter setting on the phase compensation control module of the two-stage lagging link by using a momentum gradient descent optimization algorithm and a Newton-Rafson iteration method to obtain a set optimized phase compensation control module; and 3, analyzing damping characteristics of the system, and further selecting a proper compensation point to configure an optimized phase compensation control module for suppressing subsynchronous oscillation control of the virtual synchronous direct-drive fan grid-connected system. Through the design and reasonable configuration of the damping compensation module, the damping characteristic correction of the system can be realized, so that the subsynchronous oscillation of the grid construction type direct-driven fan grid-connected system is effectively inhibited, and the system stability is improved.

The invention discloses an automobile part assembly precision intelligent compensation method and a self-adaptive regulation and control system, and relates to the field of intelligent compensation, and the method comprises the steps: synchronously collecting part geometric parameters, tool poses and environment data, and constructing an associated data matrix; analyzing data by using an improved random forest-attention model, and outputting a deviation factor contribution degree sequence; a compensation calculation model is designed accordingly, initial compensation amounts are generated in a segmented mode in combination with a precision margin threshold value, and correction is conducted through historical data similarity matching; selecting an execution path according to the compensation amount and the core deviation type; an actual precision value is obtained through laser detection after assembly, and a compensation error is calculated; based on an error triggering model optimization mechanism, model parameters are iteratively updated by using a gradient descent algorithm, and the deviation identification and compensation precision is improved. The method has the advantages that the model attribution deviation is improved, the precise compensation amount is calculated in combination with historical data, flexible execution, real-time monitoring and model self-optimization are matched, and the assembly precision and the production efficiency are efficiently improved.

The invention relates to the technical field of big data analysis, and discloses a prospecting target area positioning method and system based on big data, and the method comprises the steps: collecting multi-source exploration data in real time through distributed nodes, completing coordinate normalization, semantic alignment and time synchronization through a spatial heterogeneous data flow engine, and generating a standardized incremental data block; performing local feature sensitivity analysis based on the historical model library, identifying a newly added feature dimension, and performing parameter increment updating by adopting a sliding window gradient descent method; inputting the updated model into a target evolution model driven by a Bayesian space-time probability field, and dynamically calculating the metallogenic probability of each space grid in combination with a stress field, an element migration path and historical verification data; and generating high, medium and low three-level target area maps according to probability sorting, and pushing the high, medium and low three-level target area maps to a three-dimensional visual decision terminal. According to the method, minute-level dynamic response of the target region under triggering of newly-added data is realized, computing resource consumption is reduced to be less than 5% of that of an original system, and prospecting efficiency and abnormal region identification timeliness are improved.

The invention relates to the technical field of information, and discloses a radar dynamic anti-interference method and system based on interference source positioning. The method comprises the following steps: acquiring current signal data and a historical signal sequence, and determining an initial deviation value; extracting a historical deviation sequence according to the initial deviation value to obtain a deviation change trend vector; calculating a trend slope and a fluctuation amplitude, and determining an adjustment trigger signal; extracting feature interaction influence according to the adjustment trigger signal, and determining a weight coefficient update increment; iterative optimization is carried out in combination with the convergence rate parameter, and an optimized convergence rate value is obtained; adjusting deviation compensation model parameters according to the convergence speed value, and outputting a deviation compensation result when the compensation residual error is lower than a preset threshold value; and carrying out positioning calculation according to a deviation compensation result, and determining a corrected positioning coordinate. The model is dynamically optimized through technologies such as a sliding window and gradient descent, the problem of positioning deviation caused by complex environment signal interference is solved, and the positioning precision and the response speed are improved.

The invention relates to the field of gas concentration detection, and particularly discloses a real-time feedback and adaptive learning method for natural gas infrared spectrum measurement, which comprises the following steps: S1, acquiring infrared spectrum information of a natural gas body by using an infrared spectrometer, and constructing a historical sample set; s2, preprocessing the spectral data of the historical sample set; s3, selecting an optimal algorithm and a hyper-parameter by adopting XGBoost and Bayesian optimization; s4, constructing a qualitative model to identify gas types and match data; s5, calculating the similarity between a field sample and a historical sample through a Siamese network, and setting a threshold value to screen local data; s6, improving the KNN to construct a local dynamic quantitative model to predict the concentration; s7, processing low-similarity abnormal data by the global dynamic model, and improving the reliability by combining moving average and abnormal calibration; and S8, introducing reinforcement learning and online gradient descent to adjust parameters in real time to optimize the precision. According to the technical scheme, high-accuracy natural gas detection can be carried out in a complex environment.

A method includes performing a gradient descent mutation of a current generation of prompts by an evolutionary algorithm framework engine. The gradient descent mutation includes sending a prompt to a large language model (LLM) with an evaluation input-output pair and instructing the LLM to generate a modification recommendation for the prompt. The prompt is modified according to the modification recommendation. The modified prompt is processed by the LLM with the evaluation input output pair, causing the LLM to generate a response matching the output of the evaluation input-output pair. The modified prompt is added to a next generation of prompts.

The invention relates to the field of twitter recognition, in particular to a twitter classification method based on harmonic enhancement and time-frequency semantic joint modeling, which comprises the following steps: collecting twitter samples and carrying out noise reduction and standardized preprocessing, carrying out multi-scale convolution operation on Mel spectrograms by utilizing a layered acoustic encoder, extracting time-frequency features in combination with a channel attention mechanism, and classifying twitter classification results. The method comprises the following steps of: generating adaptive position codes through a dynamic time-frequency joint coding module, carrying out time-frequency mode modeling by combining a global-local interaction mechanism, introducing a semantic fusion module which comprises a frequency band pyramid unit, a harmonic enhancement unit and a time-frequency gating unit, realizing dynamic weighted fusion of multi-layer features, and carrying out time-frequency mode modeling through a global-local interaction mechanism. And inputting the fusion features into a classification layer, training a network by adopting a cross entropy loss function and a gradient descent algorithm, and outputting bird categories through a full connection layer, thereby solving the key problems of insufficient description of a non-stationary time-frequency mode, insufficient modeling of a harmonic structure, reduction of recognition performance in a complex noise environment and the like in the prior art.

The invention provides a 6D attitude rapid estimation method and device for a target object, and relates to the field of electrical digital data processing in the technical field of machine vision and perception, and the method comprises the steps: obtaining an RGB image and a depth image of the target object, inputting the RGB image into a segmentation model, and obtaining a binary mask of the target object; acquiring a camera internal reference matrix, and constructing a curved surface point cloud matrix of the target object according to the depth image, the binary mask and the camera internal reference matrix; constructing a cubic package of the target object, and constructing a fitting penalty function according to the curved surface point cloud matrix and the cubic package; according to the curved surface point cloud matrix, the cubic package and the fitting penalty function, gradient vectors of the center point and the Euler angle of the target object are obtained through calculation; and performing iterative solution on the gradient vectors of the central point and the Euler angle through a gradient descent algorithm to obtain optimal solutions of the central point and the Euler angle, and taking the optimal solutions of the central point and the Euler angle as the 6D attitude of the target object.