Model cooperative processing method and device, electronic equipment and computer storage medium

Abstract

The present disclosure provides a model cooperative processing method and device, electronic equipment and computer storage medium, relating to technical fields such as artificial intelligence, natural language processing, computer vision and the like. The specific implementation scheme is: obtaining to-be-processed information in an agricultural scene; sending the to-be-processed information and decision information prompt words to a large model to obtain small model information and decision parameter information output by the large model; determining a target small model set based on the small model information, and sending the decision parameter information to a related target small model in the target small model set; sending the to-be-processed information to the target small model set; receiving a small model processing result set output by the target small model set; and obtaining a processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information and the large model.

Description

TECHNICAL FIELD

[0001] The present disclosure belongs to the technical field of computers, and particularly relates to the technical fields of natural language processing, artificial intelligence, computer vision, etc. In particular, a model cooperative processing method and device, electronic equipment, and computer readable storage medium. BACKGROUND

[0002] Large models perform well in general knowledge question answering, but there are still defects in vertical fields such as agricultural planting production processes. Large models lack knowledge of vertical industries and are prone to hallucinations. Current large models are more general and more divergent in thinking and planning, and cannot perform professional thinking and planning in agricultural specific scenarios. Agricultural small models require many input parameters and are easily limited by regional conditions, and have weak generalization ability. Existing combination methods mainly use API to directly call small models through workflows or agents, and large models cannot know the internal running logic of the tools. Small models cannot be optimized, and the results of small models cannot be guaranteed. SUMMARY

[0003] The present disclosure provides a model cooperative processing method and device, electronic equipment, and computer readable storage medium.

[0004] According to a first aspect, a model cooperative processing method is provided, which includes: obtaining to-be-processed information in an agricultural scenario; sending the to-be-processed information and decision information cues to a large model to obtain small model information and decision parameter information output by the large model; determining a target small model set based on the small model information, and sending the decision parameter information to a related target small model in the target small model set; sending the to-be-processed information to the target small model set; receiving a small model processing result set output by the target small model set; and obtaining a processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model.

[0005] According to a second aspect, a model cooperative processing device is provided, which includes: an obtaining unit configured to obtain to-be-processed information in an agricultural scenario; a decision unit configured to send the to-be-processed information and decision information cues to a large model to obtain small model information and decision parameter information output by the large model; a determining unit configured to determine a target small model set based on the small model information, and send the decision parameter information to a related target small model in the target small model set; a sending unit configured to send the to-be-processed information to the target small model set; a receiving unit configured to receive a small model processing result set output by the target small model set; and an obtaining unit configured to obtain a processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model.

[0006] According to a third aspect, an electronic device is provided, comprising at least one processor; and a memory connected with the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any implementation of the first aspect.

[0007] According to a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, and the computer instructions are used to enable a computer to perform the method according to any implementation of the first aspect.

[0008] The model collaborative processing method and device provided by the embodiments of the present disclosure first acquire the to-be-processed information in the agricultural scene; secondly, the to-be-processed information and the decision information prompt word are sent to a large model to obtain small model information and decision parameter information output by the large model; thirdly, based on the small model information, a target small model set is determined, and the decision parameter information is sent to a target small model related to the target small model set; fourthly, the to-be-processed information is sent to the target small model set; then, a small model processing result set output by the target small model set is received; finally, based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information and the large model, a processing result of the to-be-processed information is obtained. The to-be-processed information is collaboratively processed by the large model and the target small model, which fully utilizes the global decision-making ability of the large model and the fine processing ability of the small model, realizes the efficiency, accuracy and flexibility of information processing in the agricultural scene, and effectively improves the decision-making efficiency and management level of agricultural production, thereby providing strong support for the intelligent development of agriculture.

[0009] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0010] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0011] Figure 1 is a flowchart of an embodiment of the model collaborative processing method according to the present disclosure;

[0012] Figure 2 is a structural schematic diagram of a system corresponding to the model collaborative processing method of the present disclosure;

[0013] Figure 3 is a structural schematic diagram of an embodiment of the model collaborative processing device of the present disclosure;

[0014] Figure 4is a block diagram of an electronic device for implementing a model collaborative processing method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0015] Unless otherwise clearly indicated, throughout the specification and claims, the term "comprise" or variations such as "comprises" or "comprising" will be understood to imply the inclusion of a stated element or group of elements but not the exclusion of any other element or group of elements.

[0016] The technical solutions of the present disclosure are described below through specific embodiments. It should be understood that one or more steps mentioned in the present disclosure do not exclude other methods and steps before and after the combination steps, or other methods and steps can be inserted between these explicitly mentioned steps. It should also be understood that these examples are only used to illustrate the present disclosure and not to limit the scope of the present disclosure. Unless otherwise specified, the numbering of each method step is only for the purpose of identifying each method step, and is not intended to limit the arrangement order of each method or to limit the implementation scope of the present disclosure. Changes or adjustments of the relative relationship can also be considered as the implementation scope of the present disclosure without substantial technical content changes.

[0017] The raw materials and instruments used in the embodiments are not specifically limited in source and can be purchased on the market or prepared according to conventional methods well known to those skilled in the art.

[0018] In the prior art, the models in the field of agriculture face two core challenges:

[0019] Limitations of large models: Although general large models have natural language interaction and common sense reasoning capabilities, the general knowledge base has the problems of insufficient coverage of agricultural professional field knowledge and poor data timeliness, which easily produces logical discontinuity or knowledge illusion in professional scenarios such as crop growth models and pest diagnosis. There is a data acquisition bottleneck when a large number of input parameters are required.

[0020] Pain points of small agricultural models: Crop growth models based on expert experience (such as DSSAT, WOFOST, APSIM) require input of high-precision soil parameters, meteorological data, etc. Model parameters have regional specificity, and the accuracy decreases when applied across regions (such as northeast corn models cannot be directly applied to southwest mountainous areas)

[0021] Combination of large models and small models: Large models have large model thinking and reasoning logic, and small models have small model theoretical rules. The current combination mainly uses workflow or agent to directly call model running result data, and the comprehensive analysis of the result data cannot know the internal running logic of the small model, and cannot fundamentally solve the contradiction between the data dependence of the small model and the professional credibility of the large model.

[0022] In view of the defects in the prior art, the present disclosure proposes a model collaborative processing method. This collaborative processing method makes full use of the global decision-making ability of large models and the fine processing ability of small models, realizes the efficiency, accuracy and flexibility of information processing, and can effectively improve the decision-making efficiency and management level of production, providing strong support for intelligent development. Figure 1 A flow 100 of one embodiment of the model collaborative processing method according to the present disclosure is shown, which comprises the following steps:

[0023] Step 101, obtaining information to be processed in an agricultural scene.

[0024] In this embodiment, the information to be processed refers to problem data to be solved related to the agricultural scene collected from the terminal and / or scene information collected from agricultural Internet of Things sensor networks, satellite remote sensing, weather stations and other data sources. The scene information includes soil moisture, temperature, light intensity, crop growth status, pest and disease conditions, and weather data, etc. After the problem data to be solved and the scene information are preliminarily processed, structured information to be processed is formed, which can be multi-modal data such as images and texts. Through the preliminary processing of the information to be processed, information that can be used for further analysis can be formed, such as soil moisture, temperature, crop growth status, pest and disease problems, etc.

[0025] In this embodiment, the implementation process of obtaining information to be processed in an agricultural scene can be described as follows: first, deploy a sensor network in farmland using Internet of Things technology. These sensors can collect real-time data such as soil moisture, temperature, light intensity, and crop growth status. At the same time, satellite remote sensing technology is used to obtain large-scale farmland image information for analyzing the distribution, growth and pest and disease conditions of crops. In addition, weather change information such as temperature, precipitation, and wind force can also be obtained by combining weather station data. These data will be transmitted to the agricultural information management system for storage and preliminary screening. Then, the system processes the collected data through data analysis algorithms to identify abnormal data points, such as soil moisture below the appropriate range or crop growth indicators deviating from the normal value. These abnormal data will be marked as information to be processed, and combined with historical data and expert knowledge base, corresponding warning information and recommended measures such as irrigation reminders, fertilization suggestions or pest and disease control solutions will be generated. Finally, these information to be processed will be pushed to the farm managers through mobile applications or SMS, etc. so that they can take timely action to optimize the agricultural production process and improve crop yield and quality.

[0026] Optionally, the above step 101 comprises: receiving agricultural scene problem or agricultural scene data processing demand information sent by a user, and taking the agricultural scene problem or agricultural scene data processing demand information as the information to be processed.

[0027] Step 102, send the to-be-processed information and decision information prompt words to the large model to obtain small model information and decision parameter information output by the large model.

[0028] In this embodiment, the small model information is information related to the small model, including small model functions, small model types, small model parameters, etc., and the decision parameter information is parameters closely related to decision-making in the decision-making process of the small model, such as cross-regional high-precision soil parameters, meteorological data, etc., wherein cross-regional refers to a region different from the region data currently analyzed by the small model, such as a northeast corn analysis model, and the decision parameter information includes corn analysis data in the southwest region.

[0029] In this embodiment, when the large model executes logic according to the rules of the agricultural small model, if it encounters information or parameters that cannot be obtained by calling existing resources through the small model, it can query the decision parameter information through Internet knowledge (such as a crop database), such as querying local historical meteorological disasters and pest information. If the execution result of this step is significantly different from the actual situation, the parameters need to be fine-tuned to achieve reasonable results. Among them, the above-mentioned existing resources can be a previously maintained variety database, a meteorological query tool interface, or a tool interface for querying soil nutrient and texture information. Alternatively, when the decision parameter information cannot be obtained from the Internet knowledge, the decision parameter information can also be information generated directly by the large model.

[0030] In this embodiment, the decision information prompt word is used to prompt the large model to determine the small model information of the small model related to the to-be-processed information, and search for decision parameter information for the determined small model. First, extract the to-be-processed information from the agricultural Internet of Things system, including soil moisture, temperature, crop growth status, pest situation, and meteorological data, etc., and generate complete input content in combination with the preset decision information prompt word (for example, “determine whether irrigation is needed according to the current soil moisture and crop growth stage” or “recommend appropriate prevention and control measures according to the pest situation”). Then, send these information to the large language model. The large language model analyzes and reasons the input to-be-processed information through its powerful natural language processing ability and knowledge base, generates targeted small model information (such as a simplified irrigation decision model or a pest control model) and specific decision parameter information (such as irrigation amount, irrigation time, pesticide type and usage dose, etc.). Finally, store the small model information and decision parameter information output by the large language model in the database of the agricultural management system for reference and execution by the farm manager or user, thereby realizing precise and efficient agricultural decision support.

[0031] Step 103, based on the small model information, determine a target small model set, and send the decision parameter information to the related target small model in the target small model set.

[0032] In this embodiment, the target small model set refers to a set of small models specifically trained for a particular agricultural task (such as irrigation, pest control, fertilizer management, etc.), wherein the target small model set includes at least one target small model, and each target small model has a corresponding specific agricultural task. These target small models can handle specific agricultural problems and generate targeted decision recommendations. For example, if the information to be processed is the suitable variety of spring corn in the northeast in 2025, the target small model set includes a weather analysis small model and a soil analysis small model. The weather analysis small model is used to analyze the local weather conditions and determine the accumulated temperature of the planting season. The soil analysis small model is used to analyze the local soil conditions and determine the soil nutrient conditions. Figure 2 As shown in the table, each target small model in the target small model set has its own theoretical rules. After determining the target small model set, the theoretical rules of all target small models in the target small model set are combined in the order of reasoning to obtain a standardized rule chain. Figure 2 In the standardized rule chain, there are three steps: the first step, the second step, and the third step. The large model can obtain the standardized rule chain of the target small model set, or process the information to be processed according to the theoretical rules corresponding to the standardized rule chain.

[0033] In this embodiment, the decision parameter information is sent to the relevant target small model in the target small model set, which includes determining the target small model that needs the decision parameter information and sending the decision parameter information to the target small model.

[0034] In this embodiment, in the agricultural intelligent system, based on the small model information output from the large language model, the system further analyzes and filters the target small model set that best matches the current information to be processed and decision requirements. These target small models are usually pre-trained special models for specific agricultural scenarios (such as irrigation optimization, pest control, fertilizer management, etc.), which can more accurately handle specific tasks. The specific implementation process is as follows: the system first analyzes the small model information output by the large language model, identifies the model features related to the current agricultural task (such as model type, applicable scenario, input and output parameters, etc.). Then, from the pre-set small model library, retrieve the target small model set that meets these characteristics. Subsequently, send the decision parameter information output by the large language model (such as irrigation amount, fertilizer ratio, control measures, etc.) to these target small models. The target small models will further optimize the decision scheme and generate the final execution instructions according to the received decision parameter information, such as specific irrigation schedules, pesticide spraying paths, etc., thereby realizing precise and efficient agricultural decision-making and automatic execution.

[0035] Step 104: Send the information to be processed to the target small model set.

[0036] In this embodiment, the execution subject on which the model collaborative processing method runs can send the to-be-processed information to each target small model in the target small model set according to the functions of the target small models (each target small model is in parallel processing of the task), or send the to-be-processed information to a target small model in the target small model set (the target small model in the target small model set is in series processing of the task). The target small model in the target small model set analyzes and processes the received to-be-processed information and specific decision parameter information such as irrigation amount, irrigation time, fertilizer type and dosage, pest control measures, and the like according to its own algorithm and training data, to generate a small model processing result set, thereby providing precise decision support for the agricultural manager.

[0037] As shown in Figure 2 After the to-be-processed information is sent to the target small model set, a small model processing result set output by the target small model set can be obtained, and the small model processing result set includes small model processing results of each target small model in the target small model set, such as Figure 2

[0038] Step 105: Receiving the small model processing result set output by the target small model set.

[0039] In this embodiment, the small model processing result set is the result set output by each target small model after processing the to-be-processed information, and the small model processing result set includes at least one small model processing result, each small model processing result corresponds to a target small model, and each small model processing result can include a specific processing result, an optimization suggestion, or early warning information, and the like.

[0040] In this embodiment, the to-be-solved problems corresponding to the processing results are different, and their specific contents are different, such as the start time and water amount of the irrigation system, the type and dosage of the fertilizer, the pest control measures, and the like. The execution subject on which the model collaborative processing method runs integrates and verifies the received processing results to ensure their accuracy and consistency, and then presents these processing results to the agricultural manager in the form of a visual interface, an SMS notification, or an automatic instruction, or directly sends them to the related agricultural equipment, thereby realizing precise agricultural decision support and automatic operation.

[0041] Step 106: Obtaining the processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model.

[0042] ​In this embodiment, the execution subject on which the model collaborative processing method runs first receives a small model processing result set output by the target small model set. These results are specific decision suggestions or optimization schemes generated by the small models based on the to-be-processed information. At the same time, the execution subject combines the scene information (such as sensor data, weather data, etc.) in the to-be-processed information, and the small model information (such as the applicable scenarios and function descriptions of the small models) and decision parameter information (such as specific irrigation amount, fertilizer amount, etc.) output by the large language model, compares and verifies these information, and uses the comprehensive analysis capability of the large model to further optimize and calibrate the processing results of the small models. Finally, the processing result of the to-be-processed information, such as a precise irrigation plan, a pest control measure, or a crop growth optimization scheme, is generated and fed back to the agricultural manager or directly sent to the relevant agricultural equipment to realize intelligent agricultural decision and automatic operation.

[0043] Optionally, the step 106 further includes determining local data of the target region based on the to-be-processed information, combining the local data with the small model processing result set to obtain the processing result of the to-be-processed information. For example, the to-be-processed information is whether northeast spring corn is suitable for planting in the local area. The determination of the local data of the target region based on the to-be-processed information includes: obtaining the local target yield analysis by searching the Internet information to obtain the yield information of the region. Other planting management, irrigation type, sowing date, and harvesting date are obtained from the Internet information data to obtain the planting information of the region. Based on the Internet information, the variety data of the region in recent years is obtained to obtain the local variety data. When the large model optimizes and calibrates the processing of the small models, the yield information, the planting information, and the local variety data are sent to the large model, so that the large model can fully refer to the local information to give accurate optimization and calibration results, and the accuracy of the processing result of the to-be-processed information is improved.

[0044] Optionally, the inference result of the large model can also be verified by the small model. The inference result of the large model needs to be verified by the small model. If it is out of limit, iteration correction (such as adjusting parameters or re-searching knowledge) is triggered. The step 106 includes determining a standardization rule chain of the target small model set based on the small model information, executing the standardization rule chain by the large model to obtain a plurality of execution results, sending the decision parameter information and the to-be-processed information to the target small model set, and collecting the processing results of each target small model in the target small model set. By comparing the small model processing result set with the collected processing results of all target small models, it is determined whether the small model processing result set is qualified. When the small model processing result set is qualified, the execution results output by the large model are verified by the small model processing result set.

[0045] When the large model executes the standardized rule chain (i.e., according to the small model rule), each step has a result output, and the result of each step needs to be verified for accuracy. For example, a crop model simulates the growth and development process of crops; soil moisture simulation, according to meteorological, initialized soil information, randomization of some initial parameters, dynamic simulation of soil water balance, which can be verified by real field moisture instrument measured real data. If there is a deviation, the initialization parameters are corrected; growth period simulation, according to accumulated temperature, field management information, initialization of small model other parameters, calculation of crop growth period, and verification of the authenticity of the growth period simulation by using field real observation records; leaf area, biomass simulation, also combined with field sampling data for verification.

[0046] The model cooperative processing method provided by the embodiment realizes rule driving and data completion, specifically, agricultural expert rules are disassembled into rule chains understandable by the large model, small model parameters are dynamically completed in combination with Internet knowledge, and data acquisition cost is reduced; dual trusted verification is realized, specifically, through cross verification of large and small model processes, large model illusion is reduced, and small model regional adaptation accuracy is improved. Generalization and application, virtual data is generated by using the large model, and the small model is applicable to other regions within a reasonable range and is applicable to multi-level scenes from family farms to large planting bases.

[0047] The model cooperative processing method provided by the embodiment of the disclosure first acquires the to-be-processed information in the agricultural scene; secondly, sends the to-be-processed information and the decision information prompt word to the large model to obtain the small model information and the decision parameter information output by the large model; thirdly, determines a target small model set based on the small model information, and sends the decision parameter information to a related target small model in the target small model set; fourthly, sends the to-be-processed information to the target small model set; fifthly, receives a small model processing result set output by the target small model set; and finally, based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model, obtains a processing result of the to-be-processed information. This cooperative processing method fully utilizes the global decision-making ability of the large model and the fine processing ability of the small model, realizes the efficiency, accuracy, and flexibility of information processing in the agricultural scene, can effectively improve the decision-making efficiency and management level of agricultural production, and provides strong support for the intelligent development of agriculture.

[0048] In some optional implementations of the disclosure, the above obtaining the processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model includes: verifying the small model processing result set by the large model based on the to-be-processed information, the small model information, and the decision parameter information to obtain a verification result; and in response to detecting that the verification result represents that the verification is qualified, extracting the processing result of the to-be-processed information in the small model processing result set.

[0049] In the optional implementation, the processing result of the to-be-processed information in the small model processing result set includes: in response to the target small model in the target small model set being a model that sequentially executes the task corresponding to the to-be-processed information, determining the last target small model in the target small model set, and extracting the processing result of the last target small model in the small model processing result set as the processing result of the to-be-processed information.

[0050] In the optional implementation, the small model processing result set has a corresponding processing specification or a result threshold. When the small model processing result set meets the processing specification or does not exceed the result threshold, the verification result represents that the verification is qualified. When the small model processing result set does not meet the processing specification or exceeds the result threshold, the verification result represents that the verification is unqualified.

[0051] In the optional implementation, the to-be-processed information, the small model information, and the decision parameter information are input into the large model. The large model verifies the small model processing result set by using its powerful comprehensive analysis capability and knowledge base. The verification process includes checking whether the output of the small model conforms to the expected decision logic, is consistent with the to-be-processed information, and meets the preset decision parameter requirement. If the verification result shows that the processing result of the small model is qualified, that is, the result is accurate and meets the expectation, the execution subject on which the model collaborative processing method runs will extract the specific processing result for the to-be-processed information from the small model processing result set that passes the verification, such as an irrigation plan, a fertilization suggestion, or a pest control measure. These verified processing results can be directly used for decision-making and execution of agricultural production, thereby improving the efficiency and accuracy of agricultural management.

[0052] Optionally, the obtaining of the processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model further includes: in response to detecting that the verification result represents that the verification is unqualified, outputting new small model information and new decision parameter information by the large model, determining a new target small model set based on the new small model information, and sending the new decision parameter information to a target small model in the new target small model set; sending the to-be-processed information to the new target small model set; receiving a new small model processing result set output by the new target small model set; verifying the new small model processing result set based on the to-be-processed information, the new small model information, and the new decision parameter information by the large model to obtain a verification result; and continuing to detect whether the verification result represents that the verification is qualified until the verification result represents that the verification is qualified.

[0053] In some optional implementations of the present disclosure, the above verification of the small model processing result set by the large model based on the to-be-processed information, the small model information, and the decision parameter information includes: determining a standardized rule chain of the target small model based on the small model information; filling the standardized rule chain based on the decision parameter information to obtain a processing logic chain; controlling the large model to process the to-be-processed information using the processing logic chain to obtain a large model processing result set; detecting whether the small model processing result set and the large model processing result set are consistent; in response to detecting that the small model processing result set and the large model processing result set are consistent, obtaining a qualified verification result; and in response to detecting that the small model processing result set and the large model processing result set are inconsistent, obtaining an unqualified verification result.

[0054] In the optional implementation, the decision flow of the target small model is decomposed into a plurality of executable steps, such as the pest control small model, which includes the following executable steps: soil moisture monitoring, water requirement calculation, irrigation scheme generation, disease symptom identification, environmental factor analysis, and prevention scheme recommendation. The executable steps are converted into a standardized rule chain by a semantic parser. The standardized rule chain includes at least two standardized rules, each of which corresponds to an execution step (the parameters in the execution step can be fixed values or values that need to be calculated by the small model). The at least two standardized rules form a chain-shaped small model rule. For example, the standardized rule chain of the plant protection small model is: 1. Determine the soil condition, variety condition, and types of diseases and pests prone to occur according to the climate condition, 2. Determine the regular occurrence of diseases and pests, 3. Analyze the meteorological conditions this year, 4. Use the disease and pest control means, 5. Use the pesticide composition, and 6. Use the product in the database.

[0055] In the optional implementation, the standardized rule chain of the variety recommendation model includes: 1) regional adaptability: screening the varieties approved by the region to ensure that the varieties adapt to the climate and soil conditions of the region. According to the historical meteorological data of the region, the accumulated temperature range of the planting season in the region is calculated. 2) Growth period and planting mode: according to the local planting mode (one season, two seasons), the number of days of crop growth cycle is determined, and suitable varieties are screened. 3) Disease resistance and stress resistance: according to the local historical disease or meteorological disaster occurrence, the varieties with disease resistance and stress resistance are screened. 4) Seed quality and field performance: screening varieties with germination rate ≥ 90%, and screening varieties that meet the local target yield. 5) Soil and fertility adaptation: for heavy clay soil, select varieties with developed root system and tolerance to barren soil; for sandy loam soil, select varieties with tolerance to dense planting and preference for water and fertilizer.

[0056] In this optional implementation, the processing logic chain includes at least two processing logics, each of which corresponds to an execution step, and each of which is an actual processing means adopted in combination with the decision parameter information when the execution step is performed. As the first standardization rule above, if the decision parameter information is the current X-place climate, the processing logic corresponding to the standardization rule is to determine the soil condition, the variety condition, and the type of diseases and pests prone to occur in the X-place under the current X-place climate.

[0057] In this optional implementation, based on the small model information (including the function, applicable scenario, and processing logic of the small model, etc.), the execution subject on which the model collaborative processing method runs determines the standardization rule chain of the target small model. The standardization rule chain is a series of predefined execution steps, each of which is used to standardize the processing process of the small model. Then, the execution subject fills in the standardization rule chain according to the decision parameter information (such as the specific irrigation amount, fertilizer amount, etc.) to generate a complete processing logic chain. The processing logic chain is a specific decision logic, which is used to guide the large model to process the to-be-processed information.

[0058] The execution subject controls the large model to process the to-be-processed information by using the processing logic chain to generate a large model processing result set. The large model processing result set is a decision suggestion or optimization scheme generated by the large model according to the processing logic chain.

[0059] In this optional implementation, the execution subject detects whether the small model processing result set and the large model processing result set are consistent. If they are consistent, it means that the processing result of the small model is reliable, and the system obtains a verified verification result; otherwise, if they are inconsistent, it means that the processing result of the small model may be biased, and the system obtains an unverified verification result.

[0060] The method for obtaining a verification result provided in this optional implementation can ensure that the processing result of the small model is consistent with the large model in logic and decision by controlling the large model to use the processing logic chain filled with the decision parameter information, thereby improving the accuracy and reliability of the agricultural intelligent system.

[0061] In some optional implementations of the present disclosure, the method further includes: in response to detecting that the verification result represents a failed verification, determining a replacement rule chain by the large model, and generating a new processing result of the to-be-processed information using the replacement rule chain; or, in response to detecting that the verification result represents a failed verification, detecting whether there is a small model processing result in the set of small model processing results that is out of the range of the corresponding large model processing result, in response to detecting that there is a small model processing result that is out of the range of the corresponding large model processing result, limiting the range of the small model processing result to obtain a new processing result of the to-be-processed information and a range limiting strategy of the target small model corresponding to the small model processing result; or, in response to detecting that the verification result represents a failed verification and the to-be-processed information represents new variety information, obtaining a processing result of the to-be-processed information from the set of large model processing results, extracting a similar region feature corresponding to the to-be-processed information by the large model to generate a virtual training data set, and training a target small model in the set of target small models using the virtual training data set.

[0062] In this optional implementation, if the verification result shows that the small model processing result is unqualified, the execution subject determines a replacement rule chain by the large model and regenerates a new processing result of the to-be-processed information using the replacement rule chain. The replacement rule chain is a decision rule re-established based on the knowledge and logic of the large model, which is used to replace the original processing logic to ensure the accuracy and reliability of the processing result.

[0063] In this optional implementation, if the verification result shows that the small model processing result is unqualified and it is detected that some results in the set of small model processing results are out of the reasonable range of the large model processing result, the execution subject will limit the range of these out-of-range small model processing results. By adjusting these results to meet the range requirements of the large model, a new processing result of the to-be-processed information is obtained, and a range limiting strategy of the corresponding target small model is generated to optimize the subsequent performance of the small model.

[0064] In this optional implementation, for the processing of new variety information, if the to-be-processed information represents new variety information (such as a new crop variety or a new agricultural scene) and the verification result is unqualified, the execution subject will directly obtain a processing result of the to-be-processed information from the set of large model processing results. At the same time, the large model extracts the similar region feature corresponding to the to-be-processed information to generate a virtual training data set. Then, the target small model in the set of target small models is retrained using this virtual training data set to improve the processing capability of the small model for new variety information.

[0065] The model collaborative processing method provided in this embodiment can flexibly adjust the processing logic according to different failed verification conditions, optimize the performance of the small model, and generate more accurate processing results, thereby improving the overall efficiency and reliability of agricultural intelligence.

[0066] In some optional implementations of the present disclosure, the sending of the to-be-processed information and the decision information prompt word to the large model to obtain the small model information and the decision parameter information output by the large model comprises: sending the to-be-processed information to the large model to obtain a standardized rule chain set and an initial small model set output by the large model; determining a target small model set and small model information of the target small model set based on the standardized rule chain set and the initial small model set; determining a target standardized rule chain based on the target small model set; and sending the target standardized rule chain and the decision information prompt word to the large model to obtain the decision parameter information output by the large model.

[0067] In the optional implementation, the to-be-processed information is sent to the large model. The large model outputs a standardized rule chain set and an initial small model set according to the information. The standardized rule chain set is a set of standardized rule chains, and the standardized rule chain set includes at least two standardized rule chains. Each standardized rule chain includes at least two standardized rules, and each standardized rule corresponds to an execution step. The execution step is used to standardize the processing process of the small model. The initial small model set is a small model that may be applicable and is preliminarily screened out by the large model according to the to-be-processed information. The execution subject further screens out a target small model set based on the standardized rule chain set and the initial small model set, and extracts related information (i.e., small model information) of the target small models. The target small model set refers to small models that are most suitable for processing the current to-be-processed information. The execution subject determines a target standardized rule chain from the target small model set. The target standardized rule chain is a rule chain selected or customized from the standardized rule chain set according to the characteristics of the target small model set, and is used to guide the subsequent processing logic.

[0068] In the optional implementation, the execution subject sends the target standardized rule chain and the decision information prompt word (for example, a specific decision requirement or a problem description) to the large model. The large model generates decision parameter information for the current to-be-processed information according to the target standardized rule chain and the decision information prompt word, such as a specific irrigation amount, a fertilization amount, a disease and pest control measure, and the like.

[0069] The method for obtaining the small model information and the decision parameter information provided in the optional implementation first determines, by the large model, a standardized rule chain set and an initial small model set through to-be-processed information; determines a target small model set and small model information based on the standardized rule chain set and the initial small model set; determines a target standardized rule chain based on the target small model set; and sends the target standardized rule chain and a decision information prompt word to the large model to obtain decision parameter information output by the large model. The execution subject can obtain the small model information and the decision parameter information suitable for the current agricultural scene from the large model, thereby providing support for subsequent processing and decision-making.

[0070] Optionally, the sending the to-be-processed information and the decision information prompt word to the large model to obtain the small model information and the decision parameter information output by the large model comprises: sending the to-be-processed information to the large model to obtain a target small model set output by the large model; determining a target standardization rule chain based on the target small model set; and sending the target standardization rule chain and the decision information prompt word to the large model to obtain the decision parameter information. The target standardization rule chain is a set of standardization rule chains applicable to the target small model set.

[0071] In some optional implementations of the present disclosure, the sending the to-be-processed information to the target small model set comprises: in response to detecting that the target small models in the target small model set have execution orders, determining the arrangement order of each target small model in the target small model set to obtain a target small model sequence arranged in sequence; and sending the to-be-processed information to the first target small model in the target small model sequence and controlling each target small model in the target small model sequence to process data according to the respective execution order.

[0072] In this optional implementation, when the execution subject needs to send the to-be-processed information to the target small model set, it will first check whether these small models have a specific execution order. If it is detected that there is an execution order, the execution subject will determine the arrangement order of each target small model in the target small model set according to the preset rules or logic, thereby generating a target small model sequence arranged in sequence. For example, some small models may need to process basic data first, while other small models rely on the output results of the previous model for further analysis or optimization.

[0073] In this optional implementation, the execution subject sends the to-be-processed information to the first target small model in the target small model sequence. This small model will process the received to-be-processed information and generate a preliminary processing result. Then, the execution subject will pass this preliminary processing result to the next target small model in the sequence for further processing. This process will be performed in sequence until all the small models in the target small model sequence complete the processing task according to their respective execution orders.

[0074] This optional implementation provides a method for sending to-be-processed information to a target small model set. When the target small models in the target small model set have execution orders, the execution subject sends the to-be-processed information to the first target small model in the target small model sequence and controls each target small model in the target small model sequence to process data according to the respective execution order, which can ensure that each target small model processes data in the correct order, thereby realizing step-by-step processing and collaborative work of complex tasks and improving the overall processing efficiency and the accuracy of the results.

[0075] In some optional implementations of the present disclosure, the to-be-processed information includes pest diagnosis scene information, and the processing result of the to-be-processed information includes a recognized pest type and a pesticide corresponding to the pest type. The model collaborative processing method further includes: sending the pest type and the pesticide to the large model to detect whether the local inventory has a pesticide of the same type as the pesticide by using the large model; in response to that no pesticide of the same type as the pesticide is detected in the local inventory, searching the Internet to check whether there is a pesticide of the same type as the pesticide and treating the pest type of the pest; and in response to that a pesticide of the same type as the pesticide and treating the pest type of the pest is searched on the Internet, giving a source of the pesticide on the Internet.

[0076] In the present embodiment, the standardized rule chain for the pest diagnosis and control scene is: calling a crop classification model tool to determine the crop type; calling a crop pest recognition model to determine what pest the crop in the image has; recommending a pesticide to be used according to the recognized pest type; finding a pesticide of the same type according to the pesticide in the local inventory; and searching the Internet for a suitable pesticide on the market if there is no inventory in the local inventory.

[0077] In the present embodiment, the execution subject on which the model collaborative processing method runs can perform the following steps 1-4:

[0078] Step 1: maintaining a set of pest databases and local pesticide databases.

[0079] Step 2: receiving an image uploaded by a user, and generating to-be-processed information for the image uploaded by the user; wherein the decision information prompt word includes: identifying the crop in the image based on the image, and then determining the pest on the crop; sending the to-be-processed information and the decision information prompt word to the large model, and the large model determining a target small model set including: a visual small model and a pest recognition small model; the system calling the visual small model (such as YOLO disease detection) to identify the crop in the image, and calling the pest small model to identify the pest of the crop as "late blight". Here, the system has two visual small models for identifying crops and identifying pests in advance; the calling mode of the two small models is included in the rules in the description of the standardized rule chain; and the late blight is a prerequisite for selecting a pesticide.

[0080] Step 3: The large model retrieves the plant protection knowledge base and combines the small model processing results given by the target small model set to generate a processing result, which includes a control scheme ("spray 25% azoxystrobin, dosage 50 ml / acre") and verification of whether it meets the pesticide safety specifications (small model verification dosage threshold). The large model calls corresponding tools or databases according to the scene rules to gradually achieve the expected results. The control scheme is based on the identified disease such as "large spot disease" and the information of the pesticides and use scheme needed to treat the large spot disease.

[0081] Step 4: If there is no azoxystrobin inventory locally, a substitute scheme is recommended based on the alternative rule chain (such as "same type fungicide -> effective component matching") and the procurement channel information is supplemented.

[0082] In this embodiment, the information to be processed relates to the disease and pest diagnosis scene, specifically including the characteristic information of the disease and pest (such as symptoms, disease site, etc.). The processing result is the identified disease and pest type and the corresponding recommended pesticide. After obtaining these processing results, the subject will send the disease and pest type and pesticide information to the large model, which first checks the local inventory to determine whether there is a pesticide of the same type as the recommended pesticide. If no matching pesticide is found in the local inventory, the subject will search the Internet to determine whether there is a pesticide of the same type that can treat the disease and pest; if a pesticide that meets the conditions is found, the subject will provide the source information of the pesticide on the Internet to help the user obtain the required pesticide.

[0083] The model cooperative processing method provided in this embodiment can quickly and effectively determine the pesticide for treating the disease and pest type of the disease and pest, ensuring that the user can quickly find a substitute pesticide in the case of insufficient local resources, and improving the efficiency of disease and pest control.

[0084] Further reference Figure 3 , as an implementation of the method shown in the above figures, the present disclosure provides an embodiment of a model cooperative processing device, which corresponds to the method embodiment shown in Figure 1 , and the device can be specifically applied in various electronic devices.

[0085] As Figure 3As shown, the model collaborative processing apparatus 300 provided by the embodiment includes an obtaining unit 301, a decision unit 302, a determining unit 303, a sending unit 304, a receiving unit 305, and an obtaining unit 306. The obtaining unit 301 can be configured to obtain to-be-processed information in an agricultural scene. The decision unit 302 can be configured to send the to-be-processed information and decision information prompt words to a large model to obtain small model information and decision parameter information output by the large model. The determining unit 303 can be configured to determine a target small model set based on the small model information, and send the decision parameter information to a target small model related to the target small model set. The sending unit 304 can be configured to send the to-be-processed information to the target small model set. The receiving unit 305 can be configured to receive a small model processing result set output by the target small model set. The obtaining unit 306 can be configured to obtain a processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model.

[0086] In the embodiment, the specific processing of the obtaining unit 301, the decision unit 302, the determining unit 303, the sending unit 304, the receiving unit 305, and the obtaining unit 306 in the model collaborative processing apparatus 300 and the technical effects brought by the specific processing can be respectively referred to Figure 1 The related description of the steps 101, 102, 103, 104, 105, and 106 in the corresponding embodiment will not be repeated here.

[0087] In some embodiments of the present disclosure, the obtaining unit 306 is further configured to: verify the small model processing result set based on the to-be-processed information, the small model information, and the decision parameter information through the large model to obtain a verification result; and in response to detecting that the verification result represents that the verification is qualified, extract a processing result of the to-be-processed information in the small model processing result set.

[0088] In some embodiments of the present disclosure, the obtaining unit 306 is further configured to: determine a standardization rule chain of the target small model based on the small model information; fill the standardization rule chain based on the decision parameter information to obtain a processing logic chain; control the large model to process the to-be-processed information by using the processing logic chain to obtain a large model processing result set; detect whether the small model processing result set and the large model processing result set are consistent; in response to detecting that the small model processing result set and the large model processing result set are consistent, obtain a verification result that the verification is qualified; and in response to detecting that the small model processing result set and the large model processing result set are inconsistent, obtain a verification result that the verification is unqualified.

[0089] In some embodiments of the present disclosure, the model cooperative processing apparatus 300 further comprises a generation unit (not shown in the figure), which is configured to: in response to detecting that the verification result represents a failed verification, determine a replacement rule chain through the large model, and generate a new processing result of the to-be-processed information using the replacement rule chain; or, in response to detecting that the verification result represents a failed verification, detect whether there is a small model processing result in the small model processing result set that is out of the range of the corresponding large model processing result, in response to detecting that there is a small model processing result that is out of the range of the corresponding large model processing result, range limit the small model processing result to obtain a new processing result of the to-be-processed information and a range limiting strategy of the target small model corresponding to the small model processing result; or, in response to detecting that the verification result represents a failed verification, and the to-be-processed information represents new variety information, obtaining a processing result of the to-be-processed information through the large model processing result set, and extracting a similar region feature corresponding to the to-be-processed information through the large model to generate a virtual training data set, and training a target small model in the target small model set using the virtual training data set.

[0090] In some embodiments of the present disclosure, the decision unit 302 is configured to: send the to-be-processed information to the large model to obtain a standardized rule chain set and an initial small model set output by the large model; determine a target small model set and small model information of the target small model set based on the standardized rule chain set and the initial small model set; determine a target standardized rule chain based on the target small model set; and send the target standardized rule chain and decision information prompt words to the large model to obtain decision parameter information output by the large model.

[0091] In some embodiments of the present disclosure, the sending unit 304 is configured to: in response to detecting that a target small model in the target small model set has an execution order, determine the arrangement order of each target small model in the target small model set to obtain a target small model sequence arranged in sequence; send the to-be-processed information to the first target small model in the target small model sequence, and control each target small model in the target small model sequence to process data according to the respective execution order.

[0092] In some embodiments of the present disclosure, the to-be-processed information includes pest diagnosis scene information, and a processing result of the to-be-processed information includes a recognized pest type and a pesticide corresponding to the pest type. The device further includes an extraction unit (not shown in the figure), which is configured to send the pest type and the pesticide to the large model to detect whether there is a pesticide of the same type as the pesticide in the local inventory by using the large model; in response to no pesticide of the same type as the pesticide being detected in the local inventory, search the Internet to check whether there is a pesticide of the same type as the pesticide and treating the pest of the pest type; and in response to the pesticide of the same type as the pesticide and treating the pest of the pest type being searched on the Internet, give the source of the pesticide on the Internet.

[0093] The model cooperative processing device provided by the embodiments of the present disclosure first acquires a to-be-processed information in an agricultural scene by the acquisition unit 301; secondly, the decision unit 302 sends the to-be-processed information and a decision information prompt word to a large model to obtain small model information and decision parameter information output by the large model; thirdly, the determination unit 303 determines a target small model set based on the small model information, and sends the decision parameter information to a related target small model in the target small model set; fourthly, the sending unit 304 sends the to-be-processed information to the target small model set; then, the receiving unit 305 receives a small model processing result set output by the target small model set; finally, the obtaining unit 306 obtains a processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model. In this way, the to-be-processed information is cooperatively processed by the large model and the target small model, and this cooperative processing mode fully utilizes the global decision-making capability of the large model and the fine processing capability of the small model, realizes the efficiency, accuracy, and flexibility of information processing in the agricultural scene, and can effectively improve the decision-making efficiency and management level of agricultural production, thereby providing strong support for the intelligent development of agriculture.

[0094] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium, and a computer program product.

[0095] Figure 4 A schematic block diagram of an electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their modes of operation, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0096] like Figure 4 As shown, the electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. The RAM 403 may also store various programs and data required for the operation of the electronic device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0097] Multiple components in electronic device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of displays, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows electronic device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0098] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as model co-processing methods. For example, in some embodiments, the model co-processing method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the model co-processing method described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the model co-processing method by any other suitable means (e.g., by means of firmware).

[0099] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a load programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0100] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable model processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, causes the machine to perform the methods / operations specified in the flowcharts and / or block diagrams. The program code can execute entirely on a machine, partly on a machine, as a stand-alone software package, partly on a machine and partly on a remote machine or entirely on a remote machine or server.

[0101] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0102] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0103] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0104] It should be understood that various forms of flow can be used instead of the ones shown above, reordering, adding or deleting steps. For example, the steps recited in the present disclosure can be performed in parallel, in series, in different orders, without limitation herein, as long as the desired results of the technology disclosed in the present disclosure can be achieved.

[0105] The foregoing description of specific exemplary embodiments of the disclosure has been presented for the purposes of illustration and description. It is not intended to be a limitation on the broad concepts disclosed herein. Obviously, many modifications and variations of the present disclosure are possible in light of the above teachings. It is intended that the scope of the disclosure be limited not with this detailed description, but rather by the claims appended hereto.

Claims

1. A model collaboration processing method characterized by comprising: The method comprises: acquiring information to be processed in an agricultural scene; the information to be processed refers to problem data to be solved related to the agricultural scene collected from a terminal and / or scene information collected from an agricultural Internet of Things sensor network, satellite remote sensing, a weather station and the like, the scene information including soil humidity, temperature, light intensity, crop growth state, pest and disease conditions and weather data; sending the information to be processed and decision information prompts to a large model to obtain small model information and decision parameter information output by the large model; the decision parameter information is a parameter closely related to decision-making in the decision-making process of a small model, and the decision parameter information is high-precision soil parameters and weather data across regions; based on the small model information, determining a target small model set and sending the decision parameter information to a target small model related in the target small model set; the target small model set includes at least one target small model, each target small model has a corresponding specific agricultural task, and the target small models can process specific agricultural problems and generate targeted decision suggestions, the target small model set includes a weather analysis small model and a soil analysis small model; sending the information to be processed to the target small model set; receiving a small model processing result set output by the target small model set; the processing result includes the start time and water quantity of an irrigation system, the type and dose of fertilization, and the prevention and control measures for pests and diseases; based on the small model processing result set, the information to be processed, the small model information, the decision parameter information and the large model, obtaining a processing result of the information to be processed, including: based on the information to be processed, the small model information and the decision parameter information, verifying the small model processing result set by the large model to obtain a verification result, including: based on the small model information, determining a standardization rule chain of the target small model; based on the decision parameter information, filling the standardization rule chain to obtain a processing logic chain; controlling the large model to process the information to be processed using the processing logic chain to obtain a large model processing result set; detecting whether the small model processing result set and the large model processing result set are consistent; in response to detecting that the small model processing result set and the large model processing result set are consistent, obtaining a verification result that is qualified; in response to detecting that the small model processing result set and the large model processing result set are inconsistent, obtaining a verification result that is unqualified; in response to detecting that the verification result represents qualified verification, extracting a processing result of the information to be processed in the small model processing result set; and the large model is a large language model; The method further comprises: in response to detecting that the verification result represents unqualified verification, determining an alternative rule chain by the large model and generating a new processing result of the information to be processed using the alternative rule chain.

2. The method of claim 1, wherein, The method further comprises: Or, in response to detecting that the verification result represents a failed verification, detecting whether there is a small model processing result in the set of small model processing results that is outside the range of the corresponding large model processing result, and in response to detecting that there is a small model processing result that is outside the range of the corresponding large model processing result, performing range limitation on the small model processing result to obtain a new processing result of the to-be-processed information and a range limitation strategy of a target small model corresponding to the small model processing result; Or, in response to detecting that the verification result represents a failed verification, and the to-be-processed information represents new variety information, obtaining a processing result of the to-be-processed information through a set of large model processing results, extracting a similar region feature corresponding to the to-be-processed information through the large model, generating a virtual training data set, and training a target small model in the set of target small models using the virtual training data set.

3. The method of claim 1, wherein the sending the to-be-processed information and the decision information prompt word to a large model to obtain small model information and decision parameter information output by the large model comprises: sending the to-be-processed information to a large model to obtain a set of standardized rule chains and an initial small model set output by the large model; determining a set of target small models and small model information of the set of target small models based on the set of standardized rule chains and the initial small model set; determining a target standardized rule chain based on the set of target small models; sending the target standardized rule chain and a decision information prompt word to a large model to obtain decision parameter information output by the large model.

4. The method of claim 1, wherein the sending the to-be-processed information to the set of target small models comprises: in response to detecting that a target small model in the set of target small models has an execution order, determining an arrangement order of each target small model in the set of target small models to obtain a sequence of target small models arranged in turn; sending the to-be-processed information to a first target small model in the sequence of target small models and controlling each target small model in the sequence of target small models to process data according to the respective execution order.

5. The method of claim 1, the information to be processed comprising: pest diagnosis scene information; the processing result of the to-be-processed information includes a recognized pest type and a pesticide corresponding to the pest type, and the method further comprises: sending the pest type and the pesticide to the large model to detect whether there is a pesticide of the same type as the pesticide in a local inventory using the large model; in response to not detecting a pesticide of the same type as the pesticide in the local inventory, searching the Internet for a pesticide of the same type as the pesticide and treating the pest type of the pest; in response to searching the Internet for a pesticide of the same type as the pesticide and treating the pest type of the pest, giving the source of the pesticide in the Internet.

6. A model cooperative processing apparatus, the apparatus comprising: an acquisition unit configured to acquire to-be-processed information in an agricultural scene; The to-be-processed information refers to problem data related to the agricultural scene collected from a terminal and / or scene information collected from an agricultural Internet of Things sensor network, satellite remote sensing, a weather station, etc., including soil humidity, temperature, light intensity, crop growth status, pest and disease conditions, and weather data; The decision unit is configured to send the to-be-processed information and decision information prompts to a large model to obtain small model information and decision parameter information output by the large model; the decision parameter information is a parameter closely related to decision-making in the decision process of the small model, and the decision parameter information is a high-precision soil parameter and weather data across regions; The determination unit is configured to determine a target small model set based on the small model information and send the decision parameter information to a related target small model in the target small model set; the target small model set includes at least one target small model, each target small model has a corresponding specific agricultural task, and the target small models can process specific agricultural problems and generate targeted decision suggestions, and the target small model set includes a weather analysis small model and a soil analysis small model; The sending unit is configured to send the to-be-processed information to the target small model set; The receiving unit is configured to receive a small model processing result set output by the target small model set; the processing result includes a start time and water quantity of an irrigation system, a type and dose of fertilization, and a pest and disease control measure; The obtaining unit is configured to obtain a processing result of the to-be-processed information based on the small model processing result set, the to-be-processed information, the small model information, the decision parameter information, and the large model; The obtaining unit is configured to: based on the to-be-processed information, the small model information, and the decision parameter information, verify the small model processing result set by the large model to obtain a verification result; in response to detecting that the verification result represents qualified verification, extract a processing result of the to-be-processed information in the small model processing result set; and the obtaining unit is further configured to: based on the small model information, determine a standardization rule chain of a target small model; based on the decision parameter information, fill the standardization rule chain to obtain a processing logic chain; and control the large model to process the to-be-processed information by using the processing logic chain to obtain a large model processing result set; Detect whether the small model processing result set and the large model processing result set are consistent; In response to detecting that the small model processing result set and the large model processing result set are consistent, a verification result of qualified verification is obtained; In response to detecting that the small model processing result set and the large model processing result set are inconsistent, a verification result of unqualified verification is obtained; The large model is a large language model; The generating unit is configured to, in response to detecting that the verification result represents unqualified verification, determine a replacement rule chain by the large model and generate a new processing result of the to-be-processed information by using the replacement rule chain.

7. An electronic device, comprising: Comprise: At least one processor; and a memory connected in communication with the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

8. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are for causing the computer to perform the method of any one of claims 1-5.

Images