5388 results about "Learning methods" patented technology

The invention provides an intelligent task alarm rule self-learning method and system based on support priority, and relates to the technical field of intellectualization, and the method comprises the steps: constructing a bidirectional feature extraction network to collect task operation data, and combining historical alarm data to generate a task exception association model; and carrying out migration mapping on the model and a business scene knowledge base by utilizing a migration learning algorithm, generating a task priority quantitative index, and constructing a self-adaptive priority scoring system. And training an alarm rule generator based on the model and a scoring system, generating a multi-level alarm strategy tree, and realizing differential alarm. And the incremental learning engine monitors a task state in real time, dynamically adjusts an alarm response strategy and feeds back data to optimize a rule. A federal learning mechanism is adopted to cooperatively train a model, and the generalization ability and adaptability of alarm rules are improved, so that more accurate and efficient task alarm is realized, the false alarm rate is reduced, and the operation and maintenance efficiency is improved.

The invention discloses a fine-grained multi-modal prompt learning method based on a visual language pre-training model, and relates to the technical field of visual language pre-training models.The method comprises the steps that a multi-modal prompt learning vector used for learning visual features and text features is constructed, and a fine-grained multi-modal prompt learning model is constructed based on the fine-grained multi-modal prompt learning model; global differences, subtle distinguishing visual features and category difference text features among different visual categories are generated, and a multi-modal prompt learning vector is optimized based on a cosine similarity loss function. According to the invention, through combination of dual-granularity visual prompt and text prompt, the problems that in an existing prompt learning tuning method, cooperative reasoning cannot be carried out in a dual-mode prompt representation space, and fine distinguishing features are difficult to capture are effectively solved; and superior migration generalization ability and few-sample learning ability are verified on a plurality of main image identification reference data sets, and the performance is remarkably improved in downstream tasks.

The invention discloses a predictive maintenance method for an optical storage and charging integrated power station based on deep learning, and the method comprises the steps: constructing an efficient equipment state evaluation and prediction model based on multi-source data fusion, an intelligent prediction algorithm and a closed-loop optimization feedback mechanism, collecting multi-source data, and carrying out the fusion processing, an improved Attention-LSTM model is utilized to evaluate and predict the state of equipment, a transfer learning method is adopted to improve generalization ability, Bayesian optimization and an adaptive sliding window technology are combined at the same time, dynamic threshold adjustment is performed, a deep reinforcement learning algorithm based on a Markov decision process is adopted to optimize a maintenance strategy, and the maintenance efficiency is improved. Weibull distribution is introduced for failure probability modeling, the maintenance cost and the fault risk are balanced, continuous optimization and dynamic adaptive adjustment of a predictive maintenance scheme are realized through a closed-loop feedback mechanism, the prediction accuracy and the intelligent level of maintenance decision are remarkably improved, planned maintenance and sudden fault maintenance are reduced, and the maintenance efficiency is improved. And the reliability of the charging station is improved.

A method is provided for improving image classification accuracy in few-shot learning scenarios, where only a limited number of training examples are available. The method combines the use of Gabor filters and convolutional neural networks (CNNs) to extract detailed texture and orientation features from images. These features are then enhanced through global average pooling, aggregated into comprehensive feature vectors, and refined using an attention mechanism that identifies and emphasizes the most relevant features for classification. Masks generated from this attention process selectively enhance critical features, which, after optional re-encoding, are used to train a classifier via a metric learning approach. This method aims to increase feature separability and classification performance, facilitating more accurate classification of new images with minimal training data.

The invention provides an intelligent temperature adjusting method and system for an injection molding mold. The method comprises the steps that real-time multi-source sensor data flow is generated by collecting temperature data and pressure data of all areas of the injection molding mold; performing feature extraction on the real-time multi-source sensor data flow to obtain temperature field distribution features; a temperature field distribution prediction model is obtained through training based on the temperature field distribution characteristics and a differential private Bayesian optimization algorithm, and then a mold temperature prediction result is output through the model; performing feature fusion on the mold temperature prediction result and the pressure data, constructing input features of a neural network controller, training based on the input features and a positive definite curvature learning method to obtain the neural network controller, and generating a power control signal of a heater and a flow control instruction of a water cooling system through the neural network controller, and based on the control signal and the instruction, the temperature field of the mold is dynamically controlled through the adjustable clearance compensation mechanism. According to the invention, the precision and response speed of temperature adjustment are improved.

The invention relates to the technical field of engineering data monitoring, in particular to a hydraulic engineering data intelligent monitoring method and system based on digital twinning, and the method comprises the following steps: collecting real-time operation data through physical sensors disposed at different nodes of a hydraulic engineering; a self-adaptive digital twinborn model capable of being automatically adjusted is generated through a meta-learning method, and the self-adaptive digital twinborn model dynamically adjusts the internal structure of the self-adaptive digital twinborn model according to the change of real-time data; and on the basis of the adaptive digital twinborn model, constructing a simulation module of a plurality of target scenes, and performing decision simulation in a virtual environment in real time. According to the method, the accuracy and the stability of the digital twinborn model in a complex environment are improved, the model can quickly adapt to real-time changes in the operation process of a water conservancy project through the self-adaptive mechanism, and the prediction precision and the response of the model in a dynamic complex scene are improved.

The invention relates to a traffic flow prediction optimization method based on an intelligent optimization algorithm, which is applied to the technical field of traffic flow prediction. The system comprises a multi-modal data acquisition and fusion module, a hybrid deep learning prediction module, an intelligent parameter optimization module, an integrated learning module, a multi-scale error correction module, a prediction result analysis and visualization module, a distributed learning and privacy protection module and the like. Wherein the hybrid deep learning prediction module adopts an attention mechanism to fuse models such as CNN, Transform and LSTM, and introduces an adaptive error correction mechanism, so that the prediction precision is remarkably improved; the intelligent parameter optimization module accelerates the model search efficiency by using technologies such as quantum computing and swarm intelligence optimization; the integrated learning module and the adversarial training strategy enhance the robustness and generalization of the model; and the multi-scale error correction module adopts a time sequence decomposition and self-adaptive online learning method to improve the accuracy of long-term and short-term prediction.

The invention discloses an adaptive differential privacy federal learning method, a client, a server, a storage medium and a product, and the method is applied to the client, and comprises the steps: determining a gradient cutting threshold based on a current training round and a cutting threshold adjustment strategy, the cutting threshold adjustment strategy is constructed according to the data distribution similarity and the gradient change rate corresponding to each training round; cutting the gradient of each sample to obtain a target gradient of each sample; adding noise disturbance to the target gradient of each sample according to an adaptive differential privacy algorithm and the target privacy budget to obtain adaptive noise; and performing model updating in combination with the target data set and the global loss function to obtain local model parameters, sending the local model parameters to a server, and updating the parameters by the server according to the local model parameters to obtain a global model. Through the above mode, the robustness of the model to noise and attack is enhanced, and the capability of protecting data privacy, the accuracy of the model and the generalization capability of the model of the client are improved.

The invention discloses an integer programming automatic modeling method based on a large language model, and the method comprises the steps: obtaining an integer programming problem data set, carrying out the processing through employing a data enhancement and dynamic complexity balance method, and obtaining a balance data set of multi-modal representation; processing by adopting a model architecture search and combinatorial optimization method to obtain an optimized neural network model structure; processing by adopting a dynamic sparse training and self-adaptive difficulty course learning method to obtain a model subjected to first-stage training; processing by adopting hierarchical reinforcement learning and a multi-target Pareto frontier exploration method to obtain a final model subjected to second-stage training; processing by adopting a dynamic neural symbol inference method to obtain an integer programming model with enhanced logic consistency; and processing by adopting a self-adaptive antagonism test generation method to obtain a comprehensive evaluation result. According to the method, the modeling efficiency, accuracy and adaptability of the integer programming problem are improved.

The invention belongs to the technical field of smart home control, and particularly relates to a smart home generation strategy optimization system based on adaptive learning, and the system comprises the steps: collecting the operation history, sensor data and equipment use modes of a user in a smart home system, and determining an evaluation standard and an optimization direction; extracting features according to user behaviors, environment changes and equipment states, and constructing a feature matrix; selecting a reinforcement learning method for modeling according to an optimization problem, designing an adaptive learning mechanism, and enabling the model to be continuously adjusted according to real-time data in an operation process; according to the method, historical data is used for training the model, a reward mechanism in reinforcement learning is used for guiding the implementation of an optimization target, a control strategy adapting to the current environment and demand is generated based on model prediction, and the method has the effects of better meeting the use demand of a user for the smart home and improving the intelligent level of the system and the user experience.

The invention discloses an image label noise learning method based on granular ball calculation and contrast learning, and the method comprises the steps: obtaining image data, carrying out the preprocessing of the image data, inputting the preprocessed image data into a trained image label noise learning model, and obtaining an image classification result. The training process of the image label noise learning network comprises the following steps: acquiring an image training data set, and performing preprocessing and data enhancement on the image training data set; respectively extracting feature representations of the strong data enhancement sample and the weak data enhancement sample by adopting a contrast learning framework SimSiam, inputting the feature representation of the weak data enhancement sample into a granular ball calculation layer, inputting a feature matrix of the granular ball sample and a label thereof into a classifier after clustering to obtain an image classification result p, and calculating classification loss; predicting the feature representation of the strong data enhancement sample through a prediction layer, and calculating comparison loss by taking an image classification result p as an indicator function; and the network is optimized through back propagation of the comparison loss and the classification loss. According to the invention, the network performance can be obviously improved.

The invention relates to the related technical field of digital data processing, in particular to a personalized federated learning method and system for a heterogeneous multi-source industrial internet, and the method comprises the steps: connecting a client, evaluating a load, time delay and modal similarity to generate a dynamic association table, deploying a hierarchical encryption protocol, and constructing a priority queue; a cache mechanism is set to coordinate distributed iterative optimization, so that the technical problem that network oscillation and computing resource waste are aggravated due to overhigh load of part of nodes caused by frequent access and exit of equipment and data volume difference in the industrial internet and repeated migration of clients and nodes is caused is solved, cross-equipment shared knowledge base vectors are extracted, and the computing efficiency is improved. The technical effects of reducing the influence of model isomerism on aggregation, dynamically scheduling high-frequency parameter local aggregation and low-frequency parameter cloud synchronization, optimizing the association weight of a client and a fog node in real time, realizing privacy protection and efficient personalized federated learning, and ensuring the privacy and security of user data in the training process are achieved.

The invention provides a student adaptive auxiliary learning method and system based on artificial intelligence, and the method comprises the steps: constructing a subject knowledge graph, and carrying out the correlation and structuralization of knowledge points; the knowledge points are associated with learning resources and test questions in the subject knowledge graph; collecting learning behavior data of students, and constructing student portraits; the student portrait comprises three dimensions of learning style, knowledge level and hobbies and interests; wherein the knowledge level is a mastering probability of each knowledge point acquired according to the subject knowledge graph; personalized learning paths, learning resources and learning strategies are recommended to the students according to the student portraits and the subject knowledge maps; and learning results of the students are automatically evaluated and fed back. According to the characteristics and requirements of each student, a personalized learning scheme is provided, the learning efficiency is improved, the limitation of time and space is broken through, and the students can obtain high-quality learning resources which are more personalized for themselves anytime and anywhere.

The invention provides a hybrid representation learning-based omni-directional motion control method for a quadruped robot in a three-dimensional environment, aiming at solving the problem that an existing motion control method for the quadruped robot is limited by a narrow forward view angle and external perception noise of the robot, so that the omni-directional motion capability in the three-dimensional environment is difficult to obtain. Comprising the steps of designing various input and output contents of a first-stage end-to-end neural network learning framework, designing a training process including mixed representation learning based on a reinforcement learning framework, constructing a specific neural network model of each module, and constructing a quadruped robot simulation motion control training environment. The method has the advantages that network input and output content beneficial to efficiently learning the relation between actual observation of the robot and the environment truth value is provided, and a mixed representation learning method including supervised learning and comparative learning is introduced, so that the robot can effectively deduce environment information beyond the visual field range of the robot in the training process.

The invention relates to the technical field of federated learning, and discloses a federated learning method and system based on multiple agents and knowledge distillation, and a medium, and the method comprises the steps: S1, training a local model; s2, knowledge distillation based on decoupling; s3, decentralized knowledge sharing based on a block chain: packaging the standardized distillation knowledge fragments extracted in S2 and metadata thereof into block chain transactions, submitting the block chain transactions to a block chain network, verifying the legality of the transactions, and writing the transactions into a block chain account book to realize decentralized distribution; s4, performing multi-agent collaborative knowledge management based on a large language model, and updating a student model; and S5, repeatedly executing the steps S1 to S4 until a training termination condition is met. According to the method, personalized selection, adaptive filtering and efficient sharing of knowledge in the federated learning process are realized, the client drift phenomenon caused by data non-independent identical distribution characteristics is effectively relieved, the convergence speed, the overall performance and the personalized level of the model in a heterogeneous data environment are remarkably improved, and the communication overhead between clients is optimized.

The invention discloses a holographic display digital human speech recognition enhancement method based on multi-modal interactive learning, and the method comprises the following steps: S1, collecting and preprocessing data, and generating a multi-modal feature matrix; s2, inputting the multi-modal feature matrix into a self-adaptive multi-modal interaction kernel to generate multi-modal interaction features; s3, inputting the multi-modal interaction features into an emotion convolution block to generate semantic-emotion mapping features; s4, inputting the semantic-emotion mapping features into a cross attention generator, and outputting multi-modal fusion features; s5, inputting the multi-modal fusion features into a multi-modal fusion-decomposition two-way network, and generating multi-modal features after decoupling optimization; s6, inputting the multi-modal features into a holographic-speech symbiosis optimization module to generate speech recognition and holographic display data; and S7, performing parameter optimization by applying self-supervised learning and comparative learning strategies to realize self-adaptive adjustment. According to the invention, the enhancement of holographic display digital human speech recognition is realized by using a multi-mode interactive learning method.

The invention discloses a multi-unmanned aerial vehicle collaborative adversarial learning method based on reinforcement learning, and the method comprises the steps: designing a communication mechanism based on multiple agents, combining a feature aggregation extraction module of a long-short-term memory network, and accurately and efficiently extracting battlefield situation features; a hierarchical reinforcement learning method is adopted, a bottom maneuvering control module and an upper combat decision module are introduced, bottom control and upper decision are achieved respectively, training is performed based on an efficient environment parallel module, and combat strategy training efficiency in a complex combat scene is improved; in addition, the generalization ability of the combat strategy is improved through a multi-agent game training module based on population evolution diversification. The method can effectively extract battlefield situation features, carries out efficient cooperative combat training, adapts to a complex and changeable battlefield environment, and improves the generalization ability of unmanned aerial vehicle autonomous cooperative combat.

The invention discloses a personalized federated learning method based on parameter decoupling. The personalized federated learning method comprises the following steps: S1, constructing a federated learning framework including model parameter decoupling and personalized cooperative training; s2, in a model parameter decoupling module, a local model of the end side equipment is decoupled into global shared model parameters and local personalized model parameters; s3, calculating a difference coefficient between the personalized model parameters and the global shared model parameters, wherein the difference coefficient is used for guiding updating of the personalized model; s4, in the personalized collaborative training module, the end-side equipment performs local model training based on the difference coefficient, and performs dynamic weighting according to the similarity with personalized models of other equipment to realize selective fusion of knowledge; s5, the cloud server aggregates the global shared model parameters uploaded by the end side devices, and finally a complete personalized model is formed. According to the method, the performance of the personalized model is improved, the collaborative effect of personalized training is enhanced, and the generalization ability of the whole model is improved.

The invention relates to the technical field of microprocessors, and discloses a safety debugging method of an embedded microprocessor and the embedded microprocessor, the method comprises the following steps: implanting a random module in the microprocessor, constructing a resistance network array through random distribution of silicon substrate lattice defects to generate a disordered sequence, and if microprocessor data access is carried out, starting the random module; if physical intrusion is detected, a debugging session is established, three types of physical characteristics are synchronously generated to construct a composite key, a triple dynamic confusion layer is constructed in the microprocessor, the dynamic confusion layer and the composite key construct physical logic linkage protection, a probe access instruction logic address is called, if physical intrusion is detected, a probe triggers an invisible check point, and the probe accesses the invisible check point. The integrity of an execution environment is verified, if abnormal debugging is detected, a hardware defense mechanism is triggered, an access attack model is established, confusion parameters are trained and optimized, and a module interconnection mode is changed, so that the learning cycle of an attacker is broken, and the risk of attacking a processor in a machine learning mode is reduced.

The invention relates to the technical field of intelligent scheduling, in particular to a logistics storage task distribution system based on AI. According to the method, through fine calculation of energy consumption data of the equipment in the storage area and dynamic adjustment of the path energy consumption gradient, the material carrying path is optimized, the use frequency of the high-energy-consumption path is reduced, and the energy efficiency ratio of equipment operation is increased. In combination with density analysis of task execution time, task execution conditions in different time periods are monitored, the concentration degree of tasks in peak periods is identified, potential congestion risks are predicted based on trend analysis of task density, a task execution window is reasonably adjusted, and operation delay caused by excessive concentration of tasks is reduced. And analyzing the execution sequence of the storage tasks, and adjusting the tasks with close dependency relationships so that the tasks can be executed at reasonable time points. In the dynamic task allocation process, simulation optimization is carried out by using a reinforcement learning method, so that task allocation is more intelligent, and the overall operation cooperation capability of the warehouse is improved.

Disclosed in the present invention is a federated learning method against a backdoor attack. Firstly, a few benign model updates having a relatively high confidence are selected by means of clustering-based voting, then the selected models are considered as benign modes of model updates, and on the basis of an anomaly detection method of a variational auto-encoder, more benign model updates are progressively selected from among the remaining candidate model updates, so that the scale of a group of the selected benign model updates is continuously expanded. The present invention does not depend on differential privacy, weight pruning, and learning rate adjustment and merely sightly changes a protocol of original federated learning, and therefore, on one hand, the impact on the accuracy of a global model is relatively small, and on the other hand, integration with existing federated learning systems is facilitated. Compared with existing federated learning methods against backdoor attacks, the present invention has relatively high adaptability.

The invention belongs to the technical field of federated learning, and particularly relates to a federated learning method and device based on feature comparison optimization and classifier dynamic integration. The method comprises the steps that a global model is initialized through a server and issued to # imgabs0 clients participating in federated learning, the clients conduct local model updating based on the received global model and train the updated local model, then feature extraction is conducted on a local private data set through a trained local feature extractor, and a local private data set is obtained; the method comprises the following steps: constructing a local model to construct a local feature prototype set, uploading the local model, the local feature prototype set and the total sample size of a client to a server, using a contrast learning technology to improve the quality of a global feature prototype when the server globally aggregates, and finally issuing a global prototype and a global feature prototype set which are obtained by aggregation to each client. And executing the next round of learning until the local model converges or reaches the set communication round.

The invention belongs to the technical field of robot learning, and particularly discloses a robot continuous imitation learning method and system based on a double diffusion model. Comprising the following steps: generating old task data in a submerged space by utilizing a pre-trained latent diffusion model, generating a model through guidance of a task identifier in combination with conditional diffusion, generating an image and state data related to an old task, modeling to construct a diffusion model, realizing multi-modal mapping from a state to an action, and generating a diversified action sequence; high-quality images and state data are generated through a pre-trained variational auto-encoder and a conditional diffusion model, and the consistency of the generated data in time and the accuracy of global semantics are ensured through time sequence continuity constraint and semantic consistency constraint; and performing feedback optimization training on the latent diffusion model. According to the method, the loss weight is dynamically adjusted, and the model is ensured to show good generalization ability and stability in a complex task in combination with collaborative training of an imitation learning strategy and a generator.

The invention relates to a personalized federal learning method based on self-adaptive local model initialization and double knowledge distillation, and belongs to the technical field of edge computing. The method comprises the following steps: performing local model initialization on each client according to cosine similarity between a global model and a local model to obtain a self-adaptively initialized local model; respectively performing knowledge distillation on each client according to the global model and the historical local model, and updating parameters of the local model through double knowledge distillation; and the central server receives local models uploaded by all clients participating in training, and obtains a final global model based on dual dynamic aggregation weights of data quality and client contribution degrees. According to the method, the precision and convergence speed of the model can be remarkably improved when the data heterogeneity problem is processed, and good adaptability and robustness are shown in various edge calculation scenes.

The invention discloses a short-cut denitrification-anaerobic ammonia oxidation system and an intelligent regulation and control method thereof, and belongs to the technical field of intelligent regulation and control of sewage treatment. Comprising the following steps: constructing and processing a machine learning data set to obtain a training set and a test set; based on the training set, a prediction model is established through a machine learning method, and prediction performance is evaluated to select an optimal prediction model; performing importance analysis on the data in the data set as feature data, and identifying key features; and taking the key features as optimization regulation and control parameters, and constructing an optimal control strategy through a particle swarm optimization algorithm. Compared with the prior art, the method has the beneficial effects that a system denitrification performance prediction model is constructed and optimized by utilizing a machine learning method by utilizing a large number of system operation parameters, microbial community structures and other multi-dimensional data, the operation characteristics of the system are fully reflected, and a unique optimal control strategy is designed and formed aiming at a specific scene; and the adaptability of the strategy is improved.

The invention discloses a dynamic weight aggregation federated learning method and system for credit risk assessment, and the method comprises the steps: firstly, calculating the cosine similarity of each client model and the last gradient of a global model, so as to assess the consistency of model parameters; secondly, the distribution difference is quantified by calculating the data statistic difference between each client data and the global data, so that the diversity of the data is evaluated; further, a contribution degree comprehensively considering model parameter similarity and data distribution difference is obtained by combining a hyper-parameter with cosine similarity and data distribution difference; and finally, dynamically adjusting the weight of each client in global model updating according to the contribution degree of each client, so that the model pays more attention to the client with higher data quality, thereby improving the overall model performance. The federated learning method is especially suitable for processing non-independent identically distributed data, optimizes the generalization ability of the model, and is suitable for credit risk assessment in the financial field.

The invention discloses a power grid dispatching anti-error method and system based on topology analysis, and relates to the technical field of power grid dispatching anti-error, and the method comprises the following steps: building a remote signaling credibility evaluation model based on remote signaling evaluation data, and obtaining high-credibility remote signaling data; performing sequential logic verification based on high-credibility remote signaling data and multi-source data fusion to generate a power grid topology model; carrying out partition division on the power grid topology model based on spectral clustering and a state-driven edge weight method, and carrying out boundary equivalence processing to obtain a plurality of topology modules; and performing simulation scheduling operation on the plurality of topology modules based on a scheduling instruction dispatching order objectification analysis result, performing path traversal based on a depth-first search method, and performing anti-error analysis on the traversed path in combination with a reinforcement learning method. According to the power grid dispatching anti-error method combining remote signaling credibility evaluation, dynamic topology partition and reinforcement learning algorithm, the anti-error verification capability and the intelligent level of a power grid dispatching system are comprehensively improved.

The invention discloses a lightweight bearing fault diagnosis method based on uniform quantization and anti-distillation. A knowledge distillation and antagonistic learning method is adopted, a student model obtains valuable information by learning a soft label provided by a teacher network, and a smaller student model can also have diagnosis accuracy similar to that of a deep network under the help of the teacher network which is well learned. And meanwhile, uniform quantification is combined with the distillation process, so that the scale of the student model is remarkably compressed. According to the method, calculation is simple, the result is effective, the lightweight network is more suitable for actual equipment with limited calculation and storage resources, the method can be applied to some small equipment such as mobile phones and embedded equipment, and bearing fault real-time diagnosis is achieved.

The invention discloses a subway tunnel structure service performance analysis method and device, and the method comprises the steps: building a prediction model database based on physical experiments and numerical simulation, and determining main factors which affect the service process of a tunnel structure through an interpretable machine learning method; through coupling with a physical equation, the service performance of the tunnel structure is accurately evaluated under the small sample condition, and the characteristic that in an existing machine learning method, the prediction precision is not high under the small sample condition, and a black box is avoided; according to the tunnel structure service performance prediction model constructed by the invention, the subway service performance (safety performance) can be qualitatively evaluated objectively, so that the service state of the lining structure is monitored, evaluated and pre-warned, and the service performance of the tunnel structure can be predicted under the evaluation framework of monitoring-analysis-evaluation-pre-warning-disposal-decision making. Problems existing in a current structure are identified and processed, and a solid technical guarantee is provided for tunnel safety and continuous service.

The invention relates to a small focus segmentation and feature learning method based on Transform and CNN fusion. The method comprises the following steps: preprocessing an input three-dimensional image to obtain original input features, and inputting the original input features into a segmentation network model of an encoder-bottleneck-decoder structure; after passing through an encoder based on channel-Transform fusion, obtaining fusion features through a bottleneck used for expanding a receptive field; wherein the encoder comprises a channel convolution module for local feature capture and a local attention mechanism module for global modeling; the preprocessed input features pass through a channel convolution module and a local attention mechanism module in sequence, and then pass through a bottleneck containing cavity convolution to carry out feature fusion; and inputting the fused features into a decoder based on a jump connection mechanism to gradually restore the low-resolution features to the same size as the input image, and generating a final segmentation result or classification result. According to the method, efficient feature extraction and fine prediction of three-dimensional images such as medical images can be realized.