Agricultural field knowledge question-answering system and method based on large model technology

Abstract

The invention discloses an agricultural field knowledge question-answering system and method based on a large model technology, and relates to the technical field of large model application, and the system comprises a data processing module which is used for collecting agricultural knowledge, carrying out the classification according to different dimensions through a machine learning algorithm, and constructing a knowledge base; the user interface module is used for providing a friendly interface, receiving a question input by a user in a natural language form and displaying a corresponding search result; the analysis processing module is used for analyzing the problem intention of the user, calculating the semantic similarity, positioning knowledge base entries and searching text segments with the relevancy exceeding a set threshold value; the image label retrieval module is used for mining image associated information in the text fragment and marking positions for subsequent display; and the comprehensive processing module is used for receiving the retrieved text segments and the image marking information and generating comprehensive answers with text description and visual picture labels.

Description

Technical Field

[0001] The present invention relates to the field of large model application technology, and in particular to an agricultural knowledge question-answering system and method based on large model technology. Background Art

[0002] With the rapid development of modern agriculture, the deep integration of information technology with agricultural production, management, sales and other aspects has become a key force in promoting agricultural transformation and upgrading. With the continuous advancement of big data, cloud computing, the Internet of Things and artificial intelligence technologies, the ability to collect, process, analyze and apply agricultural information has been unprecedentedly improved. However, in the face of massive agricultural knowledge resources and the rapid development of agricultural science and technology, how to efficiently and accurately obtain and use this information has become a major challenge facing agricultural practitioners and researchers.

[0003] Traditional agricultural knowledge question-answering systems mostly rely on preset rule bases and keyword matching. This approach is often incapable of handling complex, ambiguous or specific situational questions, and it is difficult to meet users' needs for in-depth analysis and personalized answers. Especially when it comes to visual information such as specific crop disease and pest picture diagnosis and agricultural machinery operation demonstration, pure text descriptions often appear pale and powerless, affecting the communication effect and application efficiency of information.

[0004] In recent years, large model technologies based on deep learning, such as BERT and GPT, have made significant progress in natural language understanding and generation. They can master language rules through self-learning based on large amounts of text data, thereby achieving communication and interaction closer to that of humans. At the same time, advances in computer vision technology, especially the application of deep learning in image recognition and classification, have made it possible to automatically annotate and retrieve images, which provides a new path for the multimodal dissemination of agricultural knowledge.

[0005] Nevertheless, there are still many challenges in effectively integrating and applying these cutting-edge technologies to agricultural knowledge question-answering systems. For example, how to build a knowledge base that contains rich agricultural expertise and is updated synchronously with the latest scientific research results; how to design a model so that the large model can accurately understand the professional terms and complex problems in the agricultural field are all urgent problems to be solved. Summary of the invention

[0006] In view of the needs and deficiencies of current technological development, the present invention provides an agricultural knowledge question-answering system and method based on large model technology to solve the practical problems of agricultural knowledge acquisition and application, and promote the development of agricultural informatization in a more intelligent and efficient direction. It can effectively integrate the intelligent processing capabilities of large models with the rich information resources of the mounted knowledge base, realize accurate text answers and instant return of image labels, thereby improving the dissemination efficiency and application value of agricultural information.

[0007] In a first aspect, the present invention provides an agricultural domain knowledge Q&A system based on large model technology. The technical solutions adopted to solve the above technical problems are as follows:

[0008] An agricultural domain knowledge Q&A system based on large model technology, which includes:

[0009] A data processing module, configured to collect agricultural knowledge in various forms, classify it according to three dimensions of crops, pests and diseases, and planting techniques by using machine learning algorithms, and construct and form an ordered knowledge base;

[0010] A user interface module, configured to provide multiple access methods and a friendly interface, receive questions input by users in natural language form, and display corresponding agricultural knowledge search results;

[0011] An analysis and processing module, configured to analyze the intention of the user's question, calculate semantic similarity, locate knowledge base entries, and retrieve text fragments with a relevance exceeding a set threshold;

[0012] An image tag retrieval module, configured to utilize visual question answering and regular expressions to mine image association information in text fragments and mark positions for subsequent display;

[0013] A comprehensive processing module, configured to receive the retrieved text fragments and image marking information, and generate a comprehensive answer with both text descriptions and intuitive image tags by using a multimodal fusion algorithm.

[0014] Optionally, the data processing module specifically includes:

[0015] A data collection unit, configured to collect professional agricultural knowledge in various forms including text materials, image tag information, video tutorials, and audio commentaries;

[0016] A data classification unit, configured to finely classify the collected data according to three dimensions of crop types, pest and disease types, and planting techniques by using machine learning algorithms, and form an ordered knowledge base;

[0017] A data storage unit, configured to store the knowledge base in a data processing server.

[0018] Optionally, the analysis and processing module specifically includes:

[0019] An intention analysis unit, configured to analyze the intention of the user's input question through an integrated natural language processing model;

[0020] The semantic intelligent matching unit is used to calculate the semantic similarity between the question intention and the knowledge base entries through an integrated deep learning model, quickly locate the knowledge base entry with the highest semantic similarity to the question intention, and perform text retrieval in the located knowledge base entries based on the question intention to obtain a knowledge document that matches the question intention, and select text fragments with a relevance to the question exceeding a set threshold from the knowledge document.

[0021] Optionally, the involved image tag retrieval module uses an integrated visual question answering system and combines regular expressions to check whether the extracted text fragments mention information about a specified image or image tag. If so, it automatically marks the positions in the text fragments so that corresponding image tags can be accurately attached or relevant pictures can be directly displayed when generating the final answer.

[0022] Optionally, after the involved comprehensive processing module generates a comprehensive answer, it dynamically adjusts the format and detail level of the answer according to user preferences and device characteristics, and then presents the final answer to the user through a friendly interface, enabling the user to clearly see the text description and the corresponding image tags.

[0023] In a second aspect, the present invention provides a method for answering agricultural knowledge questions based on large model technology. The technical solution adopted to solve the above technical problems is as follows:

[0024] A method for answering agricultural knowledge questions based on large model technology includes the following steps:

[0025] S1. Collect agricultural knowledge in various forms, classify it according to three dimensions of crops, pests and diseases, and planting techniques using machine learning algorithms, and construct and form an ordered knowledge base;

[0026] S2. Provide a friendly interface to receive questions input by the user in natural language form and display the corresponding agricultural knowledge search results;

[0027] S3. Analyze the user's question intention, calculate the semantic similarity, locate the knowledge base entry and retrieve text fragments with a relevance exceeding a set threshold;

[0028] S4. Use visual question answering and regular expressions to mine image association information in the text fragments and mark the positions for subsequent display;

[0029] S5. Receive the retrieved text fragments and image marking information, use a multimodal fusion algorithm to generate a comprehensive answer with both text description and intuitive image tags, and display it through a friendly interface.

[0030] Optionally, the involved step S1 specifically includes:

[0031] Collect professional knowledge in the agricultural field in various forms, including text materials, image tag information, video tutorials, and audio commentaries;

[0032] Use machine learning algorithms to finely classify the collected data according to three dimensions: crop types, pest and disease types, and planting techniques, and form an ordered knowledge base;

[0033] Store the knowledge base in a data processing server.

[0034] Optionally, the specific steps involved in step S3 include:

[0035] Parse the problem intention with the help of a natural language processing model;

[0036] Use a deep learning model to calculate the semantic similarity between the problem and the knowledge base entries, and quickly locate the knowledge base entry with the highest semantic similarity to the problem;

[0037] Perform text retrieval in the located knowledge base entries based on the parsed problem intention, and match relevant knowledge documents that match the problem;

[0038] Search for and select text fragments in the knowledge documents whose relevance to the problem exceeds a threshold, and return the selected text fragments.

[0039] Optionally, the specific steps involved in step S4 include:

[0040] Use a visual question answering system, combined with regular expressions, to check whether the extracted text fragments mention information about specified images or image tags. If so, automatically mark the positions in the text fragments so that corresponding image tags can be accurately attached or relevant pictures can be directly displayed when generating the final answer.

[0041] Optionally, after performing step S5 to generate a comprehensive answer, according to user preferences and device characteristics, dynamically adjust the format and detail level of the answer, and then display the final answer to the user through a friendly interface, enabling the user to clearly see the text description and the corresponding image tags.

[0042] A knowledge question answering system and method in the agricultural field based on large model technology according to the present invention has the following beneficial effects compared with the prior art:

[0043] 1. The present invention can achieve accurate text answers and instant return of image tags, improving the dissemination efficiency and application value of agricultural information;

[0044] 2. The present invention not only enhances the depth and personalization of problem-solving, but also enhances the intuitiveness of problem-solving by intercepting image tags, filling the gap in the application of visual information in agricultural knowledge dissemination, providing an efficient, safe, and easy-to-use information acquisition tool for agricultural practitioners and researchers, and accelerating the application and popularization of agricultural science and technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Attached Figure 1 is a block diagram of module connections in the first embodiment of the present invention;

[0046] Attached Figure 2 is a flowchart of the method in the second embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] In order to make the technical solutions, technical problems to be solved, and technical effects of the present invention clearer and more understandable, the following describes the technical solutions of the present invention clearly and completely in conjunction with specific embodiments.

[0048] Embodiment 1:

[0049] Combined with attached Figure 1 , this embodiment proposes a knowledge Q&A system in the agricultural field based on large model technology, which includes:

[0050] A data processing module for collecting agricultural knowledge in various forms, classifying it according to three dimensions of crops, pests and diseases, and planting techniques by using machine learning algorithms, and constructing and forming an ordered knowledge base;

[0051] A user interface module for providing multiple access methods and a friendly interface, receiving questions input by users in natural language form, and displaying corresponding agricultural knowledge search results;

[0052] An analysis and processing module for analyzing the intention of the user's question, calculating the semantic similarity, locating the knowledge base entries and retrieving text fragments with a relevance exceeding a set threshold;

[0053] An image tag retrieval module for using visual question answering and regular expressions to mine the image association information in the text fragment and mark the position for subsequent display;

[0054] A comprehensive processing module for receiving the retrieved text fragments and image marking information, and generating a comprehensive answer with both text descriptions and intuitive image tags by using multi-modal fusion algorithms.

[0055] In this embodiment, the data processing module specifically includes:

[0056] A data collection unit for collecting professional knowledge in the agricultural field in various forms including text materials, image tag information, video tutorials, and audio commentaries;

[0057] A data classification unit for using machine learning algorithms to finely classify the collected data according to three dimensions: crop types, pest and disease types, and planting techniques, and form an ordered knowledge base;

[0058] A data storage unit for storing the knowledge base in a data processing server. The data processing server, with its equipped high-computing-power GPU server, NVMe SSD high-speed solid-state drive array, and combined with the latest RAID technology, ensures that the agricultural knowledge base can achieve fast read and write, and can reach second-level response even in the face of massive data, laying a data foundation for subsequent operations, and at the same time ensuring that the knowledge base can be updated in real time and new agricultural knowledge content can be incorporated at any time.

[0059] In this embodiment, the parsing and processing module specifically includes:

[0060] An intent parsing unit for parsing the intent of the user input question through an integrated natural language processing model;

[0061] A semantic intelligent matching unit for calculating the semantic similarity between the question intent and the knowledge base entries through an integrated deep learning model, quickly locating the knowledge base entry with the highest semantic similarity to the question intent, and performing text retrieval in the located knowledge base entries based on the question intent to obtain a knowledge document that matches the question intent, and selecting text fragments with a relevance to the question exceeding a set threshold from the knowledge document.

[0062] In this embodiment, the involved image tag retrieval module uses an integrated visual question answering system, and combines regular expressions to check whether there is information mentioning a specified image or image tag in the extracted text fragment. If so, it automatically marks the position in the text fragment so that the corresponding image tag can be accurately attached or the relevant picture can be directly displayed when generating the final answer.

[0063] In this embodiment, after the comprehensive processing module generates a comprehensive answer, it dynamically adjusts the format and detail level of the answer according to user preferences and device characteristics, and then displays the final answer to the user through a friendly interface, enabling the user to clearly see the text description and the corresponding picture tags.

[0064] Embodiment 2:

[0065] Combined with the attached Figure 2 , this embodiment proposes a knowledge question and answer method in the agricultural field based on large model technology, which includes the following steps:

[0066] S1. Collect agricultural knowledge in various forms, classify it according to three dimensions of crops, pests and diseases, and planting techniques using machine learning algorithms, and construct and form an ordered knowledge base. This process specifically includes:

[0067] S1.1. Collect professional knowledge in the agricultural field in various forms, including text materials, image tag information, video tutorials, and audio commentaries.

[0068] S1.2. Use machine learning algorithms to finely classify the collected data according to three dimensions: crop types, pest and disease types, and planting techniques, and form an ordered knowledge base.

[0069] S1.3. Store the knowledge base in a data processing server.

[0070] S2. Provide a friendly interface to receive questions input by users in natural language form and display corresponding agricultural knowledge search results. For example, a user inputs "How to prevent and control leaf mold disease during the tomato planting process?" through the friendly interface.

[0071] S3. Analyze the user's question intention, calculate the semantic similarity, locate the knowledge base entries, and retrieve text fragments with a relevance exceeding the set threshold. This process specifically includes:

[0072] S3.1. Analyze the question intention with the help of a natural language processing model. For example, for the above question about the prevention and control of tomato leaf mold disease, the natural language processing model can analyze the core intention that the crop involved is "tomato", the disease type is "leaf mold disease", and what the user expects to obtain is "prevention and control" related information, preparing for the subsequent accurate retrieval of the knowledge base content.

[0073] S3.2. Use a deep learning model to calculate the semantic similarity between the question and the knowledge base entries, and quickly locate the knowledge base entry with the highest semantic similarity to the question.

[0074] S3.3. Based on the parsed question intention, perform text retrieval in the located knowledge base entries to match relevant knowledge documents that match the question.

[0075] S3.4. Search and select text fragments in the knowledge documents with a relevance exceeding the threshold to the question, and return the selected text fragments.

[0076] S4. Use visual question answering and regular expressions to mine the image association information in the text fragments and mark the positions for subsequent display. This process specifically includes: using a visual question answering system, combined with regular expressions, to check whether the extracted text fragments mention information about specified images or image tags. If so, automatically mark the positions in the text fragments so that corresponding image tags can be accurately attached or relevant pictures can be directly displayed when generating the final answer.

[0077] S5. Receive the retrieved text fragments and image tagging information, and use a multi-modal fusion algorithm to generate a comprehensive answer that combines textual explanations and intuitive picture labels, and display it through a friendly interface. Specifically, the comprehensive answer not only includes a direct textual answer to the question, but also provides content such as detailed guidance on prevention and control steps, relevant recommended measures, and further explanations according to the specific requirements of the question, making the answer more comprehensive and practical, and facilitating users to understand and apply it in actual agricultural production activities.

[0078] After executing step S5 to generate a comprehensive answer, dynamically adjust the format and level of detail of the answer according to user preferences and device characteristics, and then display the final answer to the user through a friendly interface, enabling the user to clearly see the textual explanation and the corresponding picture label.

[0079] In summary, by adopting the agricultural domain knowledge Q&A system and method based on large model technology of the present invention, accurate text answers and instant return of image labels can be achieved. It not only improves the depth and personalization of question answering, but also enhances the intuitiveness of question answering by intercepting image labels, filling the gap in the application of visual information in agricultural knowledge dissemination.

[0080] The above specific application examples have elaborated in detail the principle and implementation manner of the present invention. These embodiments are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by those skilled in the art of this technology without departing from the principle of the present invention shall fall within the scope of patent protection of the present invention.

Claims

1. An agricultural knowledge question-answering system based on large model technology, characterized in that: It includes: The data processing module is used to collect various forms of agricultural knowledge, classify them according to the three dimensions of crops, pests and diseases, and planting techniques using machine learning algorithms, and construct and form an orderly knowledge base; A user interface module is used to provide multiple access methods and a friendly interface, receive questions input by users in natural language, and display corresponding agricultural knowledge search results; The parsing processing module is used to parse the user's question intention, calculate the semantic similarity, locate the knowledge base entries and retrieve the text fragments whose relevance exceeds the set threshold; Image tag retrieval module, which uses visual question answering and regular expressions to mine image-related information in text fragments and mark the location for subsequent display; The comprehensive processing module is used to receive the retrieved text fragments and image tag information, and use the multimodal fusion algorithm to generate a comprehensive answer that combines text descriptions and intuitive image labels.

2. According to claim 1, the agricultural knowledge question-answering system based on large model technology is characterized in that: The data processing module specifically includes: Data collection unit, which is used to collect agricultural domain expertise in the form of text materials, image label information, video tutorials and audio narrations; The data classification unit is used to use machine learning algorithms to finely classify the collected data according to three dimensions: crop type, pest and disease type, and planting technology, to form an orderly knowledge base; The data storage unit is used to store the knowledge base in the data processing server.

3. The agricultural knowledge question-answering system based on large model technology according to claim 1 is characterized in that: The analysis processing module specifically includes: An intention parsing unit, used to parse the intention of the user input question through an integrated natural language processing model; The semantic intelligent matching unit is used to calculate the semantic similarity between the question intent and the knowledge base entries through an integrated deep learning model, quickly locate the knowledge base entries with the highest semantic similarity to the question intent, and perform text retrieval in the located knowledge base entries based on the question intent to obtain knowledge documents that match the question intent, and select text fragments from the knowledge documents whose relevance to the question exceeds a set threshold.

4. The agricultural knowledge question-answering system based on large model technology according to claim 3 is characterized in that: The image tag retrieval module uses an integrated visual question-answering system combined with regular expressions to check whether the extracted text fragment contains information mentioning a specified image or image tag. If so, the location in the text fragment is automatically marked so that the corresponding image tag can be accurately attached or the relevant image can be directly displayed when the final answer is generated.

5. The agricultural knowledge question-answering system based on large model technology according to claim 1 is characterized in that: After the comprehensive processing module generates the comprehensive answer, it dynamically adjusts the format and level of detail of the answer according to user preferences and device characteristics, and then displays the final answer to the user through a friendly interface, so that the user can clearly see the text description and the corresponding image label.

6. A knowledge question-answering method in the agricultural field based on large model technology, characterized in that: It includes the following steps: S1. Collect various forms of agricultural knowledge, classify them according to the three dimensions of crops, pests and diseases, and planting techniques using machine learning algorithms, and construct and form an orderly knowledge base; S2. Provide a friendly interface to receive questions input by users in natural language and display corresponding agricultural knowledge search results; S3, parse the user's question intention, calculate semantic similarity, locate knowledge base entries and retrieve text fragments whose relevance exceeds the set threshold; S4. Use visual question answering and regular expressions to mine image-related information in text fragments and mark the locations for subsequent display; S5. Receive the retrieved text fragments and image tag information, use the multimodal fusion algorithm to generate a comprehensive answer that combines text descriptions and intuitive image labels, and display it through a friendly interface.

7. The agricultural knowledge question-answering method based on large model technology according to claim 6 is characterized in that: The step S1 specifically includes: Collecting agricultural expertise in the form of text materials, image labeling information, video tutorials and audio narrations; Use machine learning algorithms to finely classify the collected data according to three dimensions: crop type, pest and disease type, and planting technology, to form an orderly knowledge base; The knowledge base is stored in the data processing server.

8. The agricultural knowledge question-answering method based on large model technology according to claim 6 is characterized in that: The step S3 specifically includes: Analyze the question intent with the help of natural language processing models; Using deep learning models, we calculate the semantic similarity between questions and knowledge base entries, and quickly locate the knowledge base entry with the highest semantic similarity to the question. Perform text retrieval in the located knowledge base entries based on the parsed question intent, and match relevant knowledge documents that match the question; Search and select text fragments whose relevance to the question exceeds a threshold in the knowledge document, and return the selected text fragments.

9. The agricultural knowledge question-answering method based on large model technology according to claim 8 is characterized in that: The step S4 specifically includes: The visual question answering system is used in combination with regular expressions to check whether the extracted text fragment contains information mentioning the specified image or image tag. If so, the location in the text fragment is automatically marked so that the corresponding image tag can be accurately attached or the relevant image can be directly displayed when the final answer is generated.

10. The agricultural knowledge question-answering method based on large model technology according to claim 6 is characterized in that: After executing step S5 and generating a comprehensive answer, the format and level of detail of the answer are dynamically adjusted according to user preferences and device characteristics, and the final answer is then displayed to the user through a friendly interface so that the user can clearly see the text description and the corresponding image label.

Images