64 results about "Knowledge module" patented technology

A decentralized artificial intelligence platform is disclosed that enables secure sharing and execution of AI modules in sensitive application domains. The platform comprises a knowledge module registry for discovering specialized AI modules, a security framework enforcing structural integrity of AI modules, a zero-knowledge proof validation service for verifying usage conditions without exposing private data, and a blockchain-based value chain for managing smart contracts and payments. In operation, an AI agent can fetch a required AI module from the knowledge module registry, prove the right to use it, execute the AI module on local data, and automatically pay the AI module's provider through a blockchain transaction.

An AI-driven, cloud-based system (100) for real-time biomedical and pharmaceutical compliance and risk management, including: (a) a compliance knowledge module configured to ingest, interpret and structure regulatory data using natural language processing (NLP) and generate machine-readable compliance rules; (b) a real-time monitoring and event recording module configured to collect and normalise operational data from distributed biomedical and pharmaceutical systems, including laboratory information management systems (LIMS), manufacturing execution systems (MES) and IoT-enabled devices; (c) an intelligent risk assessment and prediction module configured to correlate operational data with compliance rules, calculate dynamic risk scores and predict potential compliance violations using machine learning models; (d) an automated policy and workflow enforcement module configured to initiate remedial actions, assign tasks and log activities based on predefined standard operating procedures (SOPs); (e) an audit readiness and reporting module configured to generate compliance logs, audit trails and standardised regulatory reports in real time; and (f) an adaptive learning and feedback optimization module configured to refine rule sets and predictive models based on feedback, historical data and regulatory updates; g) the modules are integrated into a cloud infrastructure to enable real-time, scalable and predictive compliance and risk management across biomedical and pharmaceutical processes.

The invention discloses an LLM central multi-agent tool arrangement and virtual-real closed-loop evolution scheduling system. The system comprises an intelligent sensing module, a knowledge module, an intelligent scheduling decision module and a closed-loop execution module. The intelligent sensing module collects real-time and multi-modal data of a physical environment and outputs structured sensing data, the knowledge module is used for constructing a knowledge graph and a historical task case library, and the intelligent scheduling decision-making module takes a large language model (LLM) as a core scheduling center, receives and analyzes task instructions, fuses the structured sensing data and graph and case information, and performs task scheduling on the knowledge graph and the historical task case library. Generating a decision scheme according with a constraint condition by using knowledge enhanced reasoning, and continuously optimizing in multiple iterations to generate a decision instruction; and the closed-loop execution module receives and drives the execution end to execute the decision instruction. According to the method, an intelligent closed-loop system of perception-decision-execution-evolution is constructed, and global optimization and high-reliability self-adaptive scheduling of multi-machine collaborative welding are realized.

According to the student personalized learning path generation method based on the neural network, a data acquisition module is deployed on a teaching platform, a learning terminal and an interactive interface, and multi-dimensional data such as student learning duration, interactive frequency and wrong question records are acquired in real time. A knowledge graph is constructed based on a course outline and a prior textbook structure, and a graph neural network GNN is utilized to dynamically adjust association weights of nodes and edges of the graph neural network GNN to reflect weak knowledge areas of students. A Bi-LSTM encoder is constructed for the collected data to generate a behavior feature vector, and the behavior feature vector and a knowledge graph node embedded vector are aligned and fused through an attention mechanism to obtain student personalized multi-modal representation. And taking the representation and the knowledge graph node weight as an environment state, selecting a next knowledge point to be learned as an action space, designing a reward function by integrating multiple indexes, updating a target by using a deep Q network, and training a strategy network to generate a personalized learning path. The trained strategy is deployed as micro-service, a path is displayed according to knowledge module sorting, and related parameters are iteratively updated according to student feedback to improve the recommendation effect.

The invention relates to the technical field of data privacy protection, in particular to a cross-domain agent knowledge migration and privacy barrier system, which comprises a migration request analysis and sensitive marking unit, an original migration module extraction and identification, authority control and domain exclusive elements marking unit, an operation log containing a domain label and a timestamp generation unit, and a privacy barrier unit. The dynamic privacy barrier processing unit replaces an identity label with a generalization identifier, deletes violation elements according to a target domain sovereignty rule to generate a compliance knowledge module, separates, encrypts and stores a domain label and a timestamp, performs hash processing on a log operation type, and stores the log operation type; the privacy disclosure blocking unit scans residual identity labels and undeleted permission elements through secondary verification, and inputs a compliance block into a migration channel after verification is passed, and the system realizes accurate desensitization and compliance verification of cross-domain knowledge migration, prevents privacy disclosure, and ensures cross-domain collaborative security of agents.

The invention discloses a multi-agent debate-based content auditing optimization method, which comprises the following steps of: 1, acquiring and preprocessing multi-source content data: integrating to-be-audited content and associated context data, and constructing a comprehensive and reliable input data basis through standardization processing and quality verification; 2, intelligent agent role design based on domain knowledge: designing a specialized intelligent agent role in combination with audit scene characteristics, and injecting a domain knowledge module to enhance the understanding ability for complex semantics and rules; and step 3, a multi-agent debate and dynamic optimization framework: adopting a debate stream and arbitration stream parallel multi-agent debate optimization framework, mining deep semantics through multiple rounds of structured debate, and combining a dynamic arbitration mechanism to integrate multi-party viewpoints. According to the method, accurate identification, fair judgment and efficient processing of complex contents are realized through multi-agent structured debate and a full-process optimization mechanism.

The invention relates to the technical field of auxiliary diagnosis and treatment systems, in particular to an intelligent auxiliary diagnosis and treatment system for doctors. Comprising a system initialization and role service registration module, a user input analysis and intention recognition module, an intelligent routing and professional tool set selection module, an SSE streaming response initialization and workflow starting module, a knowledge graph retrieval and reasoning module, a diagnosis and treatment logic reasoning and scheme generation module and a streaming result output and node state updating module. A dynamic knowledge update triggering and verifying module; according to the invention, on the system integration level, efficient integration of multi-professional medical resources is realized based on role-as-a-service architecture, and complete coverage of multi-professional knowledge modules of internal medicine, surgery, oncology and the like is ensured through a role service registration validity evaluation mechanism.

The invention discloses a multi-article binding recommendation method based on graph contrast learning, and relates to the technical field of e-commerce application. According to the method, through heterogeneous binding relation modeling, a user-article-binding ternary interaction graph is constructed, direct preference of a user to an article and indirect association of binding driving are jointly learned, and the problem of relation splitting caused by multi-graph splitting in AMCBR and MGCRL is solved; according to the method, light-weight projection head optimization is carried out, double-layer linear projection is adopted to replace a traditional multi-layer perceptron (MLP), the over-smoothing risk is reduced while the contrast characterization distinction degree is enhanced, and the redundancy problem caused by fusion or knowledge modules in MultiCBR and KEMB-Rec is solved; the method introduces a multi-objective loss function, combines main task recommendation loss, cross-view comparison loss and binding comparison loss, and collaboratively optimizes local and global embedding representation, is more comprehensive than EBRec single view loss, and effectively alleviates sparsity and over-smoothing problems.

The invention provides a multi-dimensional knowledge extraction construction method and system applied to operation and maintenance technology services, and relates to the field of operation and maintenance technology services, and the method comprises the steps: firstly obtaining an operation and maintenance service full record generated in an operation and maintenance technology service process, and then carrying out the association network construction processing of the operation and maintenance service full record, and obtaining a target association network structure. And based on the structure, identifying the core record cluster, extracting complementary operation and maintenance information, and fusing to form an operation and maintenance knowledge module set. And then performing dynamic knowledge system construction processing on the operation and maintenance knowledge module set to form a target operation and maintenance knowledge system with a self-adjustment capability. And finally, based on the target operation and maintenance knowledge system, aiming at the newly received operation and maintenance service request, generating operation and maintenance service response content containing the resource optimization suggestion, thereby effectively improving the operation and maintenance service quality and efficiency.

The invention relates to the technical field of code generation, in particular to a knowledge-adaptive code retrieval model, method and system. The knowledge adaptive code retrieval model provided by the invention comprises a code language selection module, a code knowledge module, a general knowledge module and a feature fusion module, the code language selection module and the feature fusion module are respectively connected with the code knowledge module and the general knowledge module; the code language selection module selects appropriate code languages for different user questions, the various code knowledge modules learn specific knowledge of each code language, and the universal knowledge module captures user intentions or universal knowledge of various code languages, so that under the interaction of the universal knowledge module and the various code knowledge modules, the user questions can be learned through the code language selection module. And the learned feature embedding is enabled to be richer and more professional, so that the accuracy of a code retrieval task is improved. And the code language selection module adaptively selects a proper code language for the user problem, and considers the applicability and the applicability degree of each code language, so that the code retrieval accuracy can be improved.

The invention discloses an innovative agent, a device and a computer readable medium, and relates to the technical field of artificial intelligence. The innovative agent depends on an innovative middleware comprising an innovative knowledge module and an innovative method module, an integrated man-machine interaction module, a search enhancement module and a decision generation module; in combination with innovation knowledge and an innovation method library, a man-machine interaction module understands user demands and plans tasks, a search enhancement module is utilized to generate diversified candidate schemes, a decision generation module comprehensively evaluates and optimizes the schemes, it is ensured that achievements meeting innovation degree requirements are output according to an Azilshuller test standard, and the innovation process is completed. The key problems that the innovation process depends on manpower, the efficiency is low, the effect is difficult to quantitatively evaluate and the self-adaptive iteration capability is lacked are solved, the problems that the innovation capability of a general agent is lacked, independent innovation cannot be carried out, the innovation capability evaluation cannot be realized and the like are also solved, and the intelligent level and the output quality of the innovation activity are remarkably improved.

The invention discloses a dialogue system based on multiple agents, which comprises an arbitration module, a car knowledge module, a car skill module and a chat module, and is characterized in that the arbitration module responds to an input current problem arbitration unit to generate cue words according to historical arbitration pairs, and adopts a large model to output a falling-domain arbitration result based on the cue words; calling a preset knowledge model to generate an answer based on the current question in response to the vehicle knowledge falling-domain arbitration result; in response to the vehicle skill falling region arbitration result, a plurality of alternative intentions are output through a preset language model according to the current question, an autonomous judgment result is output according to the alternative intentions and the current question, the autonomous judgment result comprises vehicle skill or chat, and the vehicle skill module extracts a final intention and slot position parameters in response to the autonomous judgment result of the vehicle skill; the chat module generates questions and answers in response to the chat autonomous arbitration result.

The invention discloses a federated domain incremental learning method based on pre-training ViT and parameter adaptation, and belongs to the technical field of federated learning. The method comprises the following steps: firstly, freezing a basic weight of a pre-trained Vision Transform model, establishing an independent domain expert knowledge network for each domain, and designing a domain knowledge extractor and a federated training strategy; when a new task appears, dynamically adding a new domain expert knowledge network in a global model structure of the server, sending the new domain expert knowledge network to a client participating in new task training, and locally training the domain expert knowledge network consisting of a domain expert knowledge module and a classification head by the client; and the cosine feature center of local data is trained to participate in the construction of a global model domain knowledge extractor. The constructed domain knowledge extractor can autonomously select the domain expert knowledge network in the reasoning stage according to the cosine similarity between the sample features and the domain center in the reasoning stage. According to the method, the performance of the global model in a domain increment scene is improved, and the communication overhead between the server and the client is greatly reduced.

The invention discloses an NL2SQL question and answer method and system based on an entity recognition small model and a large language model, and the system comprises an intention recognition module which is used for searching a related knowledge base according to a user question, inputting the large language model through constructing a cue word, judging the natural language intention of a user, and triggering a subsequent branch module; the entity recognition module is used for extracting key entities and related knowledge in a natural language of a user as components of cue words through a Bert model-based fine-tuning trained entity recognition small model in an NL2SQL model; the table selection module is used for vectorizing a natural language input by a user, retrieving keywords of a data table, matching similar questions and corresponding SQL examples in a sample library, and constructing cue words for a large language model to select a target data table; the large model cue word construction module is used for designing and optimizing a cue word template, and introducing a related knowledge module and a reference example module into a traditional role, task and constraint structure; the SQL statement execution and check error correction module is used for processing the generated SQL statement, if an error is reported, a new prompt word is constructed based on error information, a user problem and the SQL, check error correction is performed through a large language model, and execution is performed again for at most three times; according to the invention, more efficient and accurate NL2SQL service is provided for the user.

The invention discloses an experimental explanation real-time generation method, which comprises the following steps that: a training database is constructed, a knowledge enhancement language model is trained, and the training database comprises an experimental title, a video, an audio, voice transcription text information and sensor time sequence data; the training process specifically comprises the following steps: splicing and fusing data in a database to obtain a structured input sequence, inputting the structured input sequence into a knowledge enhancement language model, and generating a corresponding basic operation instruction; the knowledge enhancement language model judges whether a knowledge module needs to be called for content supplementation or not, the content comprises scientific principles or operation specifications, and an experiment explanation text is output; and generating an experimental explanation speech corresponding to the experimental explanation text by adopting an end-to-end real-time speech generation mechanism based on semantic driving. The system has a real-time response capability, and can output voice content which is moderate in information amount, clear in structure and rich in guiding significance in real time according to sudden changes, dynamic events and the like in the experiment process.

The invention relates to the technical field of model voice interaction, and particularly discloses a multi-agent collaborative medical stroke follow-up visit voice dialogue system and a construction method thereof. According to the method, the voice interaction model, the planingagent module, the knowledge retrieval agent module, the quistirewriting agent module and the messagagent module are set as system components, the knowledge retrieval agent module establishes a database by crawling real and reliable stroke professional knowledge, the illusion of compiling of a large model can be avoided, the quistirewriting agent module and the planingagent module are matched to act, the knowledge retrieval agent module, the quistirewriting agent module and the messagagent module are matched, and therefore the interestingness of the system is improved, and the interestingness of the system is improved. According to the method, the long problem generated in the interaction process can be analyzed into a plurality of sub-problems, errors in the problem are corrected, deviation of the system in interaction information analysis is avoided, and the analysis capability and stability of the system are effectively improved.

The intelligent apple planting management and expert decision-making system comprises a sensor data acquisition module, an image data acquisition module, an edge calculation module, an expert knowledge module, an intelligent analysis module, an expert decision-making module, a wireless communication module, an equipment control module, a growth data tracing analysis module and a user interaction control module. Environmental and growth data are collected through a sensor and an image data acquisition module, and are transmitted to an edge calculation module through wireless communication. The system combines artificial intelligence technology analysis and expert knowledge, evaluates the growth condition of fruit trees, provides a scientific planting strategy, supports automatic control of agricultural operations such as irrigation and fertilization, realizes full-cycle tracing and optimization of growth data, and aims to provide more accurate production management parameters for planters, provide decision basis for scientific and standardized production, and improve the production efficiency. Apple quality is improved, diseases are reduced, planting management is optimized, and orchard production benefits are increased.

The invention belongs to the technical field of medical information, and particularly relates to a newly diagnosed HIV infected person antiviral treatment intelligent decision support system which comprises a multi-source data integration module, a medicine knowledge module, an individualized decision support module, an interactive output module, a knowledge optimization module and a safety and privacy protection module. The system replaces doctors to query and integrate patient data across multiple systems, clinical time is saved, key information omission is avoided, comprehensive data support is provided for treatment decisions, experience decisions are converted into systematic analysis and graded risk prompts are generated by means of an individualized decision support module, scheme deviation is effectively reduced, the medication risk is reduced, and the treatment efficiency is improved. The method assists the basic level in improving the treatment normalization, allows doctors to input new experience and new achievements summarized in practice into the system after double-blind auditing, and provides continuously iterative knowledge support for the scientificity of the antiviral treatment decision of the newly diagnosed HIV infected person.

The invention provides a shuttlecock serving and picking all-in-one machine based on vision. The shuttlecock serving and picking all-in-one machine structurally comprises a moving device, a main control system, a serving device, a vision system and a picking device, the main control system and the serving device are positioned at the upper end of the moving device, the vision system is carried on the top of the serving device, and the picking device is further arranged on the back of the serving device; the limitation of traditional single-function equipment is broken through by combining a Bluetooth module and a computer underlying principle of the main control system with knowledge modules such as a sensor, an electronic element and robot control, a pitching machine and a ball picking machine are integrated into a unified system, a training-recycling closed loop is realized, and manual intervention is reduced; according to the multi-dimensional ball path control system, precise ball serving at any angle (the elevation angle of-30 degrees to 70 degrees) and in any direction (the horizontal rotation of 0-360 degrees) and a self-adaptive ball serving algorithm are achieved; according to the position of a player, the ball serving falling point is dynamically adjusted, a real confrontation scene is simulated, and the ball speed is adjusted in a stepless mode (20-120 km / h), so that different requirements from beginners to professional players are met.

The application discloses a vehicle component detection method and system based on features and priori, and a medium. The method comprises the following steps: constructing a multi-target vehicle component dataset suitable for a vehicle loss assessment scene, completing data labeling and quality checking; configuring a hardware environment and a software framework required for training, and setting training hyperparameters; based on a convolutional neural network, integrating an enhanced feature pyramid network and an embedded component knowledge module to form a detection branch and a segmentation branch; training a detection network framework by using the dataset to obtain a trained detection model, and verifying the model detection and segmentation performance; and inputting a vehicle component image to be detected into the trained detection model to perform vehicle component detection. The application focuses on the guiding effect of knowledge on a deep learning model, enhances feature knowledge of a vehicle image, and introduces priori knowledge among vehicle components to improve the detection and segmentation accuracy of the model for multi-target vehicle components, and provides reliable technical support for vehicle intelligent loss assessment.

The application relates to a complex network disease propagation tracing method based on a graph neural network. The method comprises the following steps: firstly, an individual infection sample on a contact network is obtained by using a propagation dynamics model; then, node topological features on the contact network are obtained based on a centrality method, and the individual infection sample is added; then, data preprocessing is performed on the individual infection sample, and the individual infection sample is divided into a training sample, a verification sample and a test sample; then, the training sample and the verification sample are put into a neural network model based on a message passing mechanism for training, wherein the neural network model based on the message passing mechanism comprises a feature encoding module, an MPNN network, a propagation controller module and a priori knowledge module; finally, the test sample is input into the trained neural network model based on the message passing mechanism, and a prediction result of a propagation source of the test sample is obtained. The propagation source on the complex network can be efficiently and accurately inferred.

The invention discloses an automatic bidding and tendering knowledge collection method and system based on semantic analysis and a storage medium, and relates to the technical field of digital bid evaluation. Establishing a professional database in the bidding and tendering field, and constructing a bidding and tendering knowledge graph based on association relationships of various knowledge in the professional database; obtaining a to-be-collected bidding and tendering text, preprocessing the to-be-collected bidding and tendering text, inputting the to-be-collected bidding and tendering text into the trained NLP semantic classification model, performing semantic segmentation on the bidding and tendering text, and identifying domain entities, attributes, relationships and theme information in the text; generating a corresponding semantic classification label according to the semantic analysis result; and inputting the semantic classification labels into the tendering and bidding knowledge graph, automatically extracting knowledge contents in the text in combination with the semantic classification labels, and classifying the knowledge contents into corresponding knowledge modules to complete automatic collection of knowledge. According to the method, the problems of low efficiency, poor accuracy and low standardization of current bidding and tendering knowledge collection are solved, and efficient and accurate support is provided for knowledge management.

This application provides an intelligent dialogue Q&A method and system for operation and maintenance technical services based on a large model knowledge base. It relates to the field of large model operation and maintenance dialogue technology. The method involves receiving operation and maintenance consultation requests, capturing the dynamic evolution characteristics of the problem, and associating them with the real-time updated knowledge modules of the large model knowledge base to obtain corresponding relationships. Based on these relationships and phased descriptions, dynamic scenario adaptation processing is performed. An operation and maintenance Q&A generation module with knowledge iteration functionality is invoked, combining the adaptation results and incremental knowledge to generate multiple rounds of preliminary Q&A content. The preliminary Q&A content is dynamically optimized and adjusted based on feedback information to obtain dynamically optimized Q&A content. The content is fed back to the initiator, and relevant information is associated with the knowledge iteration module to drive incremental updates, completing the intelligent dialogue Q&A process and improving the accuracy and practicality of operation and maintenance Q&A.

The invention discloses a cyclone strength prediction method and device. The method comprises the following steps: acquiring physical factor data, satellite cloud picture data and cyclone strength data in a historical period of time; preprocessing the physical factor data, the satellite cloud picture data and the cyclone intensity data to obtain a physical factor image, a satellite image and historical cyclone intensity; inputting the physical factor image, the satellite image and the historical cyclone intensity into a tropical cyclone intensity prediction model to obtain a 24-hour cyclone intensity prediction result; wherein the tropical cyclone strength prediction model comprises a hierarchical fusion sub-network, a deep convolutional neural network, a priori knowledge module and an uncertainty estimation module. And uncertainty estimation and long-term prior are fused, so that the prediction capability of the change trend of the cyclone strength and the accuracy of strength prediction are improved.

The invention discloses a generalized quantitative perception training method and system based on knowledge feature distillation, and belongs to the technical field of intelligent computing. The method comprises the following steps: constructing a full-precision floating-point teacher model and a mixed bit width quantification student model based on a neural network architecture and a target task scene; carrying out pre-training on the full-precision floating point teacher model; taking the pre-trained full-precision floating-point teacher model as a knowledge transmission subject and the mixed bit width quantification student model as a knowledge receiving subject to carry out quantification perception training; the quantitative perception training is integrated with an attention feature fusion module and a multi-level context knowledge module; integrating a knowledge feature distillation strategy in the quantitative perception training process to realize knowledge extraction, migration and fusion of multilayer features; and a knowledge distillation loss function is calculated, the knowledge distillation loss function is taken as an optimization target, training optimization of the mixed bit width quantification student model is completed, and universal quantification perception training is realized.

The application discloses a cyclone intensity prediction method and device, and the method comprises the following steps: acquiring physical factor data, satellite cloud image data and cyclone intensity data in a historical period; performing pretreatment on the physical factor data, the satellite cloud image data and the cyclone intensity data to obtain a physical factor image, a satellite image image and historical cyclone intensity; inputting the physical factor image, the satellite image image and the historical cyclone intensity into a tropical cyclone intensity prediction model to obtain a 24-hour cyclone intensity prediction result; wherein the tropical cyclone intensity prediction model comprises a hierarchical fusion subnetwork, a deep convolutional neural network, a prior knowledge module and an uncertainty estimation module. The fusion of the uncertainty estimation and the long-term prior knowledge improves the prediction ability of the cyclone intensity change trend and the accuracy of the intensity prediction.

The invention provides an equipment end animal voice recognition model training method and system based on knowledge distillation, and the method comprises the steps: firstly carrying out the dynamic disassembly of knowledge of a pre-trained animal voice recognition teacher model, and obtaining a hierarchical knowledge module set which comprises basic features, correlation decisions and adaptive migration knowledge modules; acquiring real-time resource state information of the equipment side, and constructing a student model basic structure adapted to the equipment side; then establishing a layered associated distillation constraint rule to form a layered distillation constraint mechanism; inputting the animal sound training data set into a teacher model and student model basic structure, and executing dynamic distillation training in combination with a hierarchical distillation constraint mechanism to obtain knowledge transmission efficiency data and a model parameter change sequence; and finally, carrying out dynamic convergence adjustment on the basic structure parameters of the student model based on the data to obtain an equipment end animal voice recognition model. According to the invention, the training quality and efficiency of the equipment-side model are improved.

The invention provides a construction method of a multi-granularity process knowledge base and a process file configuration type compilation method. The method is based on construction of a process model, a process step model and a process group model. A parameterized and structured work step model is formed through element combinations of different attributes; and then the plurality of process step models are combined to form the modular process model, the process content is complete and high flexibility is achieved through the process step selection units in the process step models, and the modularization degree is improved through the process group model formed by combining the process models. According to the process file configuration type compiling method, direct calling of knowledge modules with different granularities can be achieved, a process route can be rapidly formed, compiling of the process file can be rapidly and accurately completed by configuring different parameters, and meanwhile standardization and consistency of the process files compiled by different technicians can be guaranteed. According to the invention, systematic management and application of process knowledge are realized, and the process compiling efficiency and normalization are improved.

The invention provides an operation and maintenance technology service intelligent dialogue question answering method and system based on a large model knowledge base, and relates to the technical field of large model operation and maintenance dialogues, and the method comprises the steps: receiving an operation and maintenance consultation request, capturing the dynamic evolution characteristics of a question, and associating a real-time updating knowledge module of the large model knowledge base to obtain a corresponding relation; carrying out dynamic scene adaptation processing based on the corresponding relation and the stage expression information; an operation and maintenance question answering generation module with a knowledge iteration function is called, and multiple rounds of preliminary question answering content are generated in combination with the adaptation result and incremental knowledge; dynamically optimizing and adjusting the preliminary question answering content according to the feedback information to obtain dynamically optimized question answering content; and the content is fed back to the initiator, and related information is associated to the knowledge iteration module to drive incremental updating, so that the intelligent dialogue question answering process is completed, and the accuracy and practicability of operation and maintenance question answering are improved.

Methods of determining the species of birds in flight are provided along with corresponding systems. A method may include capturing a video stream of a bird in flight using at least one camera, generating a first species probability estimate by delivering images from the video stream to a neural network that has been trained to recognize species of birds from images, obtaining additional parameters from the video stream or from additional data, generating a second species probability estimate by delivering the additional parameters as input to a domain knowledge module with a domain knowledge statistical model, and generating a final species probability estimate by combining the first species probability estimate and the second species probability estimate. The additional parameters may include geometry features related to movement of the bird in flight, or parameters relating to the environment.