Agricultural remote sensing large model construction method and device based on multi-modal information
By constructing a large-scale agricultural remote sensing model based on multimodal information, and combining it with a route planning library, a self-built knowledge base, and spatiotemporal data, the problems of insufficient data security and multimodal analysis capabilities in existing technologies are solved, and highly accurate real-time agricultural decision support is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-03
- Publication Date
- 2026-03-13
AI Technical Summary
Existing large-scale agricultural remote sensing models rely on cloud-based question-and-answer platforms, which cannot guarantee the security of private data, lack the ability to comprehensively analyze multimodal data, and lack the ability to make real-time dynamic decisions and optimizations.
We constructed a large-scale agricultural remote sensing model based on multimodal information, combined with a route planning library, a self-built knowledge base, and spatiotemporal data, and designed interactive question-and-answer and operation command functions to achieve multimodal data processing and real-time decision support.
It improves the analytical accuracy and practicality of agricultural remote sensing models, ensures the security of private data, and supports real-time dynamic decision optimization.
Smart Images
Figure CN121660084A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of image information processing and farm data analysis, and specifically relates to a method and apparatus for constructing a large-scale agricultural remote sensing model based on multimodal information. Background Technology
[0002] The agricultural remote sensing big data model is a product of the deep integration of artificial intelligence and remote sensing technology in recent years. It integrates relevant knowledge in agricultural remote sensing with the semantic understanding and analytical capabilities of a language big data model to construct an intelligent decision-making and analysis system for the agricultural field. This system uses multi-source agricultural remote sensing data as its input and leverages the semantic understanding, knowledge reasoning, and data analysis capabilities of the underlying language big data model to achieve intelligent monitoring of the entire agricultural production cycle.
[0003] Existing large-scale agricultural remote sensing models primarily use extensive agricultural-related text data to train open-source large language models like DeepSeek, or directly connect to open interfaces of cloud platforms like ChatGPT to understand and apply knowledge in the agricultural remote sensing field. However, existing methods mainly rely on cloud-based question-answering platforms for interactive functions, which cannot guarantee the security of private data; current methods are generally limited to processing single-modal text or single-modal images, lacking the ability to comprehensively analyze multimodal inputs such as structured data, textual knowledge, and image information, thus limiting the accuracy and practicality of the models; current methods typically employ static knowledge base matching mechanisms, which can only generate general answers, failing to incorporate spatiotemporal contextual information such as crop planting areas and growth cycles, and lacking the ability to dynamically optimize decisions based on real-time agricultural data. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention provides a method and apparatus for constructing large-scale agricultural remote sensing models based on multimodal information. Targeting the construction needs of large-scale models in the field of agricultural remote sensing, this invention innovatively proposes a personnel guidance function design framework based on route planning, using a route planning library. By combining a basic question-and-answer model with an agricultural knowledge base, it proposes an interactive question-and-answer function design framework based on a self-built knowledge base. By binding multi-source spatiotemporal data with text tags, it designs a spatiotemporal agricultural remote sensing database construction scheme and proposes an information acquisition function design framework based on spatiotemporal data. Finally, by classifying operational problems into general operational problems and spatiotemporal operational problems, and based on a pre-designed agricultural knowledge base and a self-built spatiotemporal agricultural remote sensing database, it proposes an operational command function design framework based on comprehensive decision-making.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] A personnel guidance module based on a route planning library is constructed. The route planning library plans and stores daily operation routes according to the distribution of farm crop planting areas, farm building areas, and farm functional modules, and updates the routes when temporary operations or renovation and maintenance occur.
[0007] An interactive question-and-answer module based on a self-built knowledge base is constructed to conduct interactive question-and-answer sessions; the self-built knowledge base consists of agricultural knowledge, farm knowledge, and system knowledge.
[0008] An information acquisition module based on spatiotemporal data is constructed. The spatiotemporal data is obtained by monitoring the planting area at fixed time intervals. The monitoring data is labeled according to drought and flood conditions, pest and disease conditions, and maturity conditions. All labels of the same area are integrated into regional information labels, and a spatiotemporal agricultural remote sensing information database is established accordingly.
[0009] A comprehensive decision-making-based operation command module is constructed to distinguish between general operation problems and spatiotemporal operation problems when an operation problem is received, and to generate corresponding operation plans based on a self-built knowledge base or in combination with a spatiotemporal agricultural remote sensing information database.
[0010] The present invention also provides an agricultural remote sensing large model construction device based on multimodal information, for implementing the above method, comprising the following modules:
[0011] The system guides the construction of a personnel guidance module based on a route planning library. The route planning library plans and stores daily operation routes according to the distribution of farm crop planting areas, farm building areas, and farm functional modules, and updates the routes when temporary operations or renovation and maintenance occur.
[0012] An interactive question-and-answer module is constructed based on a self-built knowledge base to conduct interactive question-and-answer sessions; the self-built knowledge base consists of agricultural knowledge, farm knowledge, and system knowledge.
[0013] The module for acquiring spatiotemporal data is constructed. The spatiotemporal data is obtained by monitoring the planting area at fixed time intervals. The monitoring data is labeled according to drought and flood conditions, pest and disease conditions, and maturity conditions. All labels of the same area are integrated into regional information labels, and a spatiotemporal agricultural remote sensing information database is established accordingly.
[0014] The task generation module constructs a task command module based on comprehensive decision-making. When a task problem is received, it distinguishes between general task problems and spatiotemporal task problems, and generates corresponding task plans based on a self-built knowledge base or in combination with a spatiotemporal agricultural remote sensing information database.
[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the above-described method for constructing a large agricultural remote sensing model based on multimodal information.
[0016] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of the above-described method for constructing a large agricultural remote sensing model based on multimodal information.
[0017] Beneficial effects:
[0018] 1. This invention adopts a self-built knowledge base combined with a question-answering model as its overall architecture. While improving the accuracy of question answering, it eliminates the dependence on cloud platforms, ensuring that agricultural private data is processed locally throughout the process and avoiding the risk of sensitive data leakage.
[0019] 2. This invention innovatively integrates multimodal inputs such as structured data, textual knowledge, and images. By establishing a unified representation space for multimodal data and textual features and performing collaborative reasoning, it constructs a comprehensive decision-making mechanism for agricultural scenarios, significantly improving the analytical accuracy of large models in complex agricultural scenarios.
[0020] 3. This invention supports the updating of spatiotemporal data, enabling the knowledge base to be real-time and surpassing the limitations of traditional static matching in providing general answers. Attached Figure Description
[0021] Figure 1 This is a diagram illustrating the overall architecture of a large-scale agricultural remote sensing model based on multimodal information.
[0022] Figure 2 A flowchart illustrating the function of personnel guidance based on a route planning library;
[0023] Figure 3 A flowchart illustrating the functionalities of an interactive question-and-answer system based on a self-built knowledge base;
[0024] Figure 4 A flowchart illustrating the functional process of information acquisition based on spatiotemporal data;
[0025] Figure 5 A functional flowchart for operation command based on comprehensive decision-making;
[0026] Figure 6 This is a schematic diagram of an agricultural remote sensing large model construction device based on multimodal information according to the present invention. Detailed Implementation
[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0028] like Figure 1 As shown, the large-scale agricultural remote sensing model constructed in this invention utilizes multimodal information such as text, data, and images. By deeply coupling relevant locally built knowledge bases and databases with the underlying question-answering model, it achieves intelligent decision support in the field of agricultural remote sensing. When receiving a question from a client, the large-scale agricultural remote sensing model first extracts keywords from the question, determines which module it belongs to, and then calls the corresponding module to process the question, ultimately generating the processing result (i.e., ...). Figure 1 The relevant results are generated and fed back to the client.
[0029] The present invention provides a method for constructing a large-scale agricultural remote sensing model based on multimodal information, which specifically includes the following steps:
[0030] A personnel guidance module based on a route planning library is constructed. The route planning library plans and stores daily operation routes according to the distribution of farm crop planting areas, farm building areas, and farm functional modules, and updates the routes when temporary operations or renovation and maintenance occur.
[0031] An interactive question-and-answer module based on a self-built knowledge base is constructed to conduct interactive question-and-answer sessions; the self-built knowledge base consists of agricultural knowledge, farm knowledge, and system knowledge.
[0032] An information acquisition module based on spatiotemporal data is constructed. The spatiotemporal data is obtained by monitoring the planting area at fixed time intervals. The monitoring data is labeled according to drought and flood conditions, pest and disease conditions, and maturity conditions. All labels of the same area are integrated into regional information labels, and a spatiotemporal agricultural remote sensing information database is established accordingly.
[0033] A comprehensive decision-making-based operation command module is constructed to distinguish between general operation problems and spatiotemporal operation problems when an operation problem is received, and to generate corresponding operation plans based on a self-built knowledge base or in combination with a spatiotemporal agricultural remote sensing information database.
[0034] like Figure 2 As shown, the personnel guidance module based on the route planning library guides relevant personnel engaged in agricultural work to move around the farm according to the planned routes, and can introduce relevant information on the routes through voice.
[0035] First, based on the distribution of crop planting areas, farm building areas, and farm functional modules (including production modules: completing all agricultural activities from sowing to harvest; infrastructure modules: providing electricity, water, and communication; and management modules: responsible for farm operation, transportation, and personnel accommodation), daily operation routes are planned and stored in a basic route database. When temporary operations or farm renovations and maintenance occur, the basic route database is updated to form a real-time route database. The basic and real-time route databases are then integrated into a route planning database. Upon receiving a question about personnel guidance, the agricultural remote sensing model first uses a question-and-answer model to search the route planning database and generates a corresponding guidance plan based on existing routes. Simultaneously, the route in the route planning database that best matches the input question is provided as feedback.
[0036] like Figure 3 As shown, the interactive question-and-answer module based on a self-built knowledge base answers knowledge questions raised by personnel engaged in agricultural work, mainly including:
[0037] 1) Agricultural knowledge, including professional agricultural knowledge about various crops, agricultural conditions, and farming practices;
[0038] 2) Farm knowledge, including information and knowledge about crop planting, equipment operation, processing and storage on self-built farms;
[0039] 3) System knowledge, including personnel management, business strategies, safety risks and other process knowledge of self-built farms.
[0040] First, an agricultural knowledge base is constructed based on existing agricultural, farm, and system knowledge. When a question about farm-related knowledge is received, the agricultural remote sensing big data model first calls the basic question-answering model to search the agricultural knowledge base and generates a corresponding answer by combining it with existing relevant knowledge in the database. Simultaneously, the most relevant knowledge from the agricultural knowledge base that best matches the input question is fed back.
[0041] like Figure 4 As shown, the spatiotemporal data-based information acquisition module autonomously acquires various data from different times and spaces in the farm according to the voice commands of agricultural workers, and provides feedback on relevant situations or information summaries.
[0042] First, remote sensing monitoring equipment such as drones and soil monitoring devices are used to monitor various planting areas of the self-built farm at fixed time intervals. The obtained spatiotemporal data is then analyzed based on pre-defined tags such as drought / flood conditions, pest and disease status, and maturity status. After analysis, all information contained in a given area is integrated and used as its information tag. A spatiotemporal agricultural remote sensing information database is constructed based on all the obtained information tags. When a question about farm information is received (e.g., "What is the drought / flood situation in area A of the farm today?"), the agricultural remote sensing big data model first calls the basic question-answering model to search the spatiotemporal agricultural remote sensing information database and uses the information tags corresponding to the relevant area (area A) in the database as keywords to generate results. At the same time, the information tags in the spatiotemporal agricultural remote sensing information database that best match the input question are fed back.
[0043] like Figure 5 As shown, the operation command module based on comprehensive decision-making makes decisions based on the instructions of agricultural workers, combined with the current data information in the farm, and relevant agricultural knowledge, and forms specific operation plans to help direct agricultural equipment and workers to carry out relevant agricultural operations.
[0044] First, for the input task questions, the agricultural remote sensing big data model extracts keywords and categorizes them into general task questions and spatiotemporal task questions. For general task questions (e.g., how to deal with low seed germination rate, how to deal with soil compaction, etc.), the question-answering model is directly invoked to generate corresponding task plans based on relevant information in the agricultural knowledge base. For spatiotemporal task questions (e.g., how to maintain area A today), the agricultural remote sensing big data model first queries the spatiotemporal agricultural remote sensing information database, and inputs the obtained information tags as keywords into the agricultural knowledge base for retrieval, combining relevant information in the database to generate corresponding task plans. Simultaneously, the most relevant knowledge in the agricultural knowledge base that best matches the input question is fed back, as well as relevant knowledge obtained from targeted retrieval in the agricultural knowledge base based on the input question and associated with information tags in the spatiotemporal agricultural remote sensing information database is also fed back.
[0045] like Figure 6 As shown, the present invention also provides an agricultural remote sensing large model construction device based on multimodal information, used to implement the above method, comprising the following modules:
[0046] The system guides the construction of a personnel guidance module based on a route planning library. The route planning library plans and stores daily operation routes according to the distribution of farm crop planting areas, farm building areas, and farm functional modules, and updates the routes when temporary operations or renovation and maintenance occur.
[0047] An interactive question-and-answer module is constructed based on a self-built knowledge base to conduct interactive question-and-answer sessions; the self-built knowledge base consists of agricultural knowledge, farm knowledge, and system knowledge.
[0048] The module for acquiring spatiotemporal data is constructed. The spatiotemporal data is obtained by monitoring the planting area at fixed time intervals. The monitoring data is labeled according to drought and flood conditions, pest and disease conditions, and maturity conditions. All labels of the same area are integrated into regional information labels, and a spatiotemporal agricultural remote sensing information database is established accordingly.
[0049] The task generation module constructs a task command module based on comprehensive decision-making. When a task problem is received, it distinguishes between general task problems and spatiotemporal task problems, and generates corresponding task plans based on a self-built knowledge base or in combination with a spatiotemporal agricultural remote sensing information database.
[0050] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the above-described method for constructing a large agricultural remote sensing model based on multimodal information.
[0051] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of the above-described method for constructing a large agricultural remote sensing model based on multimodal information.
[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.
Claims
1. A method for constructing a large-scale agricultural remote sensing model based on multimodal information, characterized in that, include: A personnel guidance module based on a route planning library is constructed. The route planning library plans and stores daily operation routes according to the distribution of farm crop planting areas, farm building areas, and farm functional modules, and updates the routes when temporary operations or renovation and maintenance occur. An interactive question-and-answer module based on a self-built knowledge base is constructed to conduct interactive question-and-answer sessions; the self-built knowledge base consists of agricultural knowledge, farm knowledge, and system knowledge. An information acquisition module based on spatiotemporal data is constructed. The spatiotemporal data is obtained by monitoring the planting area at fixed time intervals. The monitoring data is labeled according to drought and flood conditions, pest and disease conditions, and maturity conditions. All labels of the same area are integrated into regional information labels, and a spatiotemporal agricultural remote sensing information database is established accordingly. A comprehensive decision-making-based operation command module is constructed to distinguish between general operation problems and spatiotemporal operation problems when an operation problem is received, and to generate corresponding operation plans based on a self-built knowledge base or in combination with a spatiotemporal agricultural remote sensing information database.
2. The method for constructing a large-scale agricultural remote sensing model based on multimodal information according to claim 1, characterized in that, The route planning library includes: The basic route library is used to store daily operation routes planned based on the distribution of farm crop planting areas, farm building areas, and farm functional modules. A real-time route database is used to update the routes in the basic route database when temporary work or renovation and maintenance situations occur; The basic route library and the real-time route library are integrated into a route planning library, which is used by the personnel guidance module based on the route planning library.
3. The method for constructing a large-scale agricultural remote sensing model based on multimodal information according to claim 1, characterized in that, The construction of the self-built knowledge base includes: Collect and organize agricultural knowledge, farm knowledge, and system knowledge; The knowledge is then structured to form searchable knowledge entries; The structured knowledge entries are stored in a self-built knowledge base for the interactive question-and-answer module to retrieve and generate answers.
4. The method for constructing a large-scale agricultural remote sensing model based on multimodal information according to claim 1, characterized in that, Agricultural knowledge includes information on crops, agricultural conditions, and agricultural activities; farm knowledge includes information on crop planting, equipment operation, processing, and storage on the farm; and system knowledge includes information on personnel management, business strategies, and safety risks on the farm.
5. The method for constructing a large-scale agricultural remote sensing model based on multimodal information according to claim 1, characterized in that, The steps to distinguish between general task problems and time-space task problems include: Extract keywords from the input homework questions; Determine whether the keywords contain spatiotemporal information; If it does not contain spatiotemporal information, it is classified as a general task problem; if it contains spatiotemporal information, it is classified as a spatiotemporal task problem.
6. The method for constructing a large-scale agricultural remote sensing model based on multimodal information according to claim 1, characterized in that, The spatiotemporal data was obtained using drones and soil monitoring equipment.
7. The method for constructing a large-scale agricultural remote sensing model based on multimodal information according to claim 1, characterized in that, For general operation problems, an operation plan is generated based on the self-built knowledge base. For spatiotemporal operation problems, the corresponding regional information tags are first obtained by querying the spatiotemporal agricultural remote sensing information database, and then an operation plan is generated by retrieving and generating the operation plan based on the regional information tags in the self-built knowledge base.
8. A device for constructing a large-scale agricultural remote sensing model based on multimodal information, characterized in that, Includes the following modules: The system guides the construction of a personnel guidance module based on a route planning library. The route planning library plans and stores daily operation routes according to the distribution of farm crop planting areas, farm building areas, and farm functional modules, and updates the routes when temporary operations or renovation and maintenance occur. An interactive question-and-answer building module is constructed based on a self-built knowledge base to conduct interactive question-and-answer sessions; the self-built knowledge base consists of agricultural knowledge, farm knowledge, and system knowledge. The module for acquiring spatiotemporal data is constructed. The spatiotemporal data is obtained by monitoring the planting area at fixed time intervals. The monitoring data is labeled according to drought and flood conditions, pest and disease conditions, and maturity conditions. All labels of the same area are integrated into regional information labels, and a spatiotemporal agricultural remote sensing information database is established accordingly. The task generation module constructs a task command module based on comprehensive decision-making. When a task problem is received, it distinguishes between general task problems and spatiotemporal task problems, and generates corresponding task plans based on a self-built knowledge base or in combination with a spatiotemporal agricultural remote sensing information database.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method for constructing a large agricultural remote sensing model based on multimodal information as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of a method for constructing a large agricultural remote sensing model based on multimodal information as described in any one of claims 1 to 7.