CN-122021717-A - Self-adaptive decision-making method and system for intelligent fruit industry and electronic equipment

CN122021717ACN 122021717 ACN122021717 ACN 122021717ACN-122021717-A

Abstract

The invention provides a self-adaptive decision-making method, a system and electronic equipment of an intelligent forest fruit industry, wherein the method comprises the steps of extracting various modal characteristics, then carrying out fusion processing to obtain fusion characteristics, inputting the fusion characteristics and corresponding service requests into a large model intelligent center, carrying out collaborative scheduling on built-in retrieval enhancement generation modules and/or multi-modal intelligent agent modules based on modal composition of the service requests and the fusion characteristics, processing the fusion characteristics to obtain decision-making results, executing service instructions based on the decision-making results, and collecting execution feedback data of the service instructions so as to carry out iterative optimization on the large model intelligent center. The intelligent system and the intelligent method can carry out comprehensive analysis based on comprehensive information by fusing modal data of multiple sources, and can realize intelligent response to complex scenes of the forest and fruit industry and improve the accuracy of self-adaptive decision of the intelligent forest and fruit industry by setting up the intelligent center of the large model, which can cooperatively schedule different professional large model capacity modules according to service requests and fused features.

Inventors

LIU FUYONG
QIU DECHUAN
QIAO JING
WANG LINAN
LIU SHENGZHI
Heng Siyu
ZHANG XI
ZHANG XIYU
Liu Panhui
LI ZEXU

Assignees

新疆科技学院

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (10)

1. An adaptive decision making method for the intelligent fruit industry, comprising: acquiring multi-mode original data related to the forest and fruit industry, and extracting various mode characteristics from the multi-mode original data; carrying out fusion processing on the multi-mode characteristics to obtain fusion characteristics; Inputting the fusion characteristics and the service requests corresponding to the multi-mode original data into a large-model intelligent center, and processing the fusion characteristics by the large-model intelligent center based on the service requests and the modes of the fusion characteristics to obtain a decision result by cooperatively scheduling a built-in retrieval enhancement generation module and/or a multi-mode intelligent agent module; Executing a service instruction based on the decision result, and acquiring execution feedback data of the service instruction; and carrying out iterative optimization on the intelligent center of the large model based on the execution feedback data.
2. The adaptive decision-making method of the intelligent forest fruit industry according to claim 1, wherein the processing the fusion feature to obtain a decision result by the large model intelligent hub based on the service request and the mode of the fusion feature by cooperatively scheduling a built-in search enhancement generation module and/or a multi-mode agent module includes: When the service request is a knowledge question and answer and the mode of the fusion feature is a pure text feature, the retrieval enhancement generation module retrieves a knowledge segment corresponding to the fusion feature from a forest and fruit industry professional knowledge base, generates a knowledge question and answer text based on the knowledge segment and the service request, and takes the knowledge question and answer text as the decision result; When the service request is identification diagnosis and the modal composition of the fusion feature contains image features, the multi-modal intelligent agent module carries out classification reasoning and/or identification reasoning on the fusion feature so as to output a target category and/or diagnosis conclusion associated with the fusion feature as the decision result; When the service request is a complex decision and the modal composition of the fusion feature comprises a plurality of modal features, the multi-modal intelligent agent module processes the fusion feature to obtain a preliminary diagnosis conclusion, the retrieval enhancement generation module uses the preliminary diagnosis conclusion as a retrieval context to carry out semantic retrieval and text generation to obtain a comprehensive decision report, and the comprehensive decision report is used as the decision result.
3. The method for adaptively deciding on a smart fruit industry as defined in claim 2, wherein the step of constructing a professional knowledge base of the fruit industry comprises: obtaining structured agricultural technology data, unstructured paper data and localized experience data; Carrying out document dicing and vectorization processing on the structured agricultural technology data, the unstructured paper data and the localized experience data to obtain document vectors; And constructing a hierarchical navigable graph index based on the document vector, and constructing the tree fruit industry expertise knowledge base based on the hierarchical navigable graph index and the document vector.
4. The method according to claim 2, wherein said performing, by said search enhancement generation module, semantic search and text generation of said preliminary diagnostic decision as a search context to obtain a comprehensive decision report, comprises: the preliminary diagnosis conclusion is vectorized by the retrieval enhancement generation module to obtain a current retrieval vector; obtaining a target knowledge segment most relevant to the current search vector in a vector space from a graph index of the tree fruit industry professional knowledge base; and inputting the target knowledge segments and the preliminary diagnosis conclusion to a preset generation type model for text generation to obtain the comprehensive decision report output by the generation type model.
5. The adaptive decision-making method of the intelligent forest and fruit industry according to claim 4, wherein the obtaining the target knowledge segment most relevant to the current search vector in the vector space from the map index of the forest and fruit industry expertise base comprises: Selecting a preset entry point node set from the graph index as a current node set; Searching neighbor nodes of each node in the current node set, and selecting a node which is closer to the current search vector in a vector space from the neighbor nodes to obtain a next round of node set; updating the next round of node set to the current node set, repeatedly executing the steps of searching for neighbor nodes of each node in the current node set, selecting nodes which are closer to the current search vector in a vector space from the neighbor nodes, and obtaining the next round of node set until the current node set is not changed; and determining the document fragment corresponding to the current node set which is not changed any more as the target knowledge fragment.
6. The method for adaptive decision making in the smart fruit industry according to any one of claims 1 to 5, wherein the fusing the multi-modal features to obtain fused features includes: taking a first modal feature of the plurality of modal features as a query and a second modal feature of the plurality of modal features as a key and a value; Calculating an attention weight between the query and the key; and carrying out weighted aggregation on the values based on the attention weight to obtain the fusion characteristic.
7. The method of adaptive decision making for the intelligent forest fruit industry according to any one of claims 1 to 5, wherein said iteratively optimizing the large model intelligent hub based on the execution feedback data comprises: Calculating importance weights of all model parameters in the intelligent center of the large model for the learned tasks; When training is carried out by utilizing the execution feedback data, constructing a total loss function comprising new task loss and a consolidation punishment item, wherein the consolidation punishment item is used for punishment on the deviation of corresponding values of each model parameter in a learned task based on the importance weight of each model parameter; And updating the model parameters aiming at minimizing the total loss function.
8. The method of adaptive decision making for the smart fruit industry according to any one of claims 1 to 5, wherein the training step of the multi-modal agent module comprises: acquiring training samples related to the forest and fruit industry and an initial multi-mode agent module, wherein the training samples comprise matched image text sample pairs and unmatched image text sample pairs; inputting the training sample into the initial multi-mode intelligent agent module to obtain a first similarity of the matched image text sample pair output by the initial multi-mode intelligent agent module, a second similarity score of the non-matched image text sample pair, a prediction result of masking content and a matched prediction result of image-text relevance of the training sample, wherein the masking content is obtained by masking images or texts in the training sample; Determining a contrast loss based on the first similarity score and the second similarity score; Determining a reconstruction loss based on the prediction result of the masked content and the tag of the masked content; determining a matching loss based on the matching prediction result and a label of the matching prediction result; determining a target loss based on the contrast loss, the reconstruction loss, and the match loss; And carrying out parameter iteration on the initial multi-mode intelligent agent module based on the target loss to obtain the multi-mode intelligent agent module.
9. An adaptive decision making system for the smart fruit industry, comprising: the acquisition unit is used for acquiring multi-mode original data related to the forest and fruit industry and extracting various mode characteristics from the multi-mode original data; the fusion processing unit is used for carrying out fusion processing on the multiple modal characteristics to obtain fusion characteristics; the decision unit is used for inputting the fusion characteristics and the service requests corresponding to the multi-mode original data into a large-model intelligent center, and the large-model intelligent center is formed by the service requests and the modes of the fusion characteristics, and cooperatively schedules a built-in retrieval enhancement generation module and/or a multi-mode intelligent agent module to process the fusion characteristics to obtain a decision result; the feedback unit is used for executing the business instruction based on the decision result and acquiring execution feedback data of the business instruction; And the optimization unit is used for carrying out iterative optimization on the intelligent center of the large model based on the execution feedback data.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, characterized in that the processor implements the adaptive decision method of the smart-fruit industry according to any one of claims 1 to 8 when executing the computer program.

Description

Self-adaptive decision-making method and system for intelligent fruit industry and electronic equipment Technical Field The invention relates to the technical field of artificial intelligence, in particular to a self-adaptive decision-making method, a self-adaptive decision-making system and electronic equipment in the intelligent fruit industry. Background With the continuous progress of big data, internet of things and artificial intelligence technology, intelligent agriculture has become a key technical direction for improving agricultural production efficiency and guaranteeing agricultural product quality. In particular, in the fruit industry, daily management faces a series of complex problems such as pest and disease monitoring, crop growth state evaluation, precise water and fertilizer management and the like. To effectively solve these problems, comprehensive analysis and judgment of information from different sources, such as field images, environmental sensor data, and text descriptions of farmers, is often required. To meet this need, some intelligent agricultural solutions have emerged in the prior art. For example, some schemes collect data by deploying environmental monitoring equipment in the farmland, then analyze the collected data using conventional machine learning algorithms to make decisions and then control related performing equipment to perform operations such as irrigation or fertilization. However, the prior art solutions described above still have significant drawbacks in practical applications. Firstly, the data processing and analyzing model of the system is mainly oriented to structured sensor data design, and when the data processing and analyzing model is oriented to unstructured data such as images, natural language and the like, the data processing and analyzing model has limited deep understanding and effective information utilization capabilities, so that valuable associated information among data from different sources cannot be fully mined and fused, and decision basis is not comprehensive enough. Secondly, the architecture of the existing system is usually solidified, the coordination capability among the functional modules is weak, and the intelligent level and the decision accuracy of the system are difficult to meet the requirements of specialized production when the complex decision scene which needs to combine professional knowledge reasoning and multi-source information perception is handled. Disclosure of Invention The invention provides a self-adaptive decision-making method, a self-adaptive decision-making system and electronic equipment in the intelligent fruit industry, which are used for solving the defects that the prior art has defects in the aspects of fusion and utilization of unstructured data processing and is difficult to support an intelligent decision-making scene requiring complex professional reasoning due to weak framework solidification and module cooperativity. The invention provides a self-adaptive decision making method for the intelligent forest fruit industry, which comprises the following steps. Acquiring multi-mode original data related to the forest and fruit industry, and extracting various mode characteristics from the multi-mode original data; carrying out fusion processing on the multi-mode characteristics to obtain fusion characteristics; Inputting the fusion characteristics and the service requests corresponding to the multi-mode original data into a large-model intelligent center, and processing the fusion characteristics by the large-model intelligent center based on the service requests and the modes of the fusion characteristics to obtain a decision result by cooperatively scheduling a built-in retrieval enhancement generation module and/or a multi-mode intelligent agent module; Executing a service instruction based on the decision result, and acquiring execution feedback data of the service instruction; and carrying out iterative optimization on the intelligent center of the large model based on the execution feedback data. According to the self-adaptive decision-making method of the intelligent forest fruit industry provided by the invention, the intelligent center of the large model is formed based on the service request and the mode of the fusion feature, and the built-in retrieval enhancement generation module and/or multi-mode agent module are cooperatively scheduled to process the fusion feature to obtain a decision-making result, and the method comprises the following steps: When the service request is a knowledge question and answer and the mode of the fusion feature is a pure text feature, the retrieval enhancement generation module retrieves a knowledge segment corresponding to the fusion feature from a forest and fruit industry professional knowledge base, generates a knowledge question and answer text based on the knowledge segment and the service request, and takes the knowledge question and answer text as the de