A government-farmer coordination digital management system and method based on multi-modal fusion and smart contract

Abstract

The application discloses a kind of based on multi-modal fusion and intelligent contract's government and agriculture collaborative digital management system and method, including farmer terminal, integrated with lightweight convolutional neural network model is used to collect crop image, execute offline disease and insect pest preliminary diagnosis, sign intelligent contract and receive decision information;Government system server is configured as alliance chain node for deploying and executing policy intelligent contract, maintaining chain on account book;Expert terminal is used to access video consultation and provide professional prevention and cure scheme;Cloud server is deployed with deep learning model, time series prediction unit, knowledge graph reasoning engine;Multi-modal database is used to provide model enhancement analysis, whole cycle decision support and data storage service;Internet of Things sensing network is deployed in planting area for collecting environmental data and uploading. Realize the digital binding and automation execution between government and farmer's rights and responsibilities, so that the subsidy issuing cycle is significantly shortened, the income distribution process avoids human intervention, improves the efficiency and transparency of policy implementation.

Description

Technical Field

[0001] This invention relates to a digital management system and method for government-agriculture collaboration based on multimodal fusion and smart contracts, belonging to the field of digital management technology for agricultural production. Background Technology

[0002] Digitalization of agricultural production is an inevitable trend in modern agricultural development. In this process, building a collaborative management system involving multiple stakeholders—government, farmers, and technology—is of great significance for improving agricultural production efficiency, implementing pro-farmer policies, achieving precision management, and ensuring food security.

[0003] However, existing government-agriculture collaboration models still rely heavily on traditional paper agreements or simple information platforms, leading to unclear definitions of responsibilities and delays in the disbursement of policy subsidies. Furthermore, given the limited network coverage and computing power and storage resources of farmers' devices in rural areas, current agricultural AI diagnostic models that rely on powerful cloud computing are ill-suited for real-time diagnosis in offline environments. In addition, farmers' production decisions largely depend on personal experience, lacking comprehensive intelligent support that integrates time-series forecasting and domain knowledge. This results in information asymmetry and decision-making blind spots in pre-production variety selection, in-production disaster warning, and post-production sales.

[0004] Therefore, how to digitize and automate the execution of rights and responsibilities in the process of government-agriculture collaboration to improve efficiency, how to make artificial intelligence diagnostic models operate stably and efficiently in low-resource environments, and how to provide farmers with accurate decision support covering the entire agricultural production cycle have become urgent problems to be solved. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings and deficiencies of the existing technologies mentioned above by proposing a digital management system and method for government-agriculture collaboration based on multimodal fusion and smart contracts. This system uses a smart contract collaboration module to convert policy terms into on-chain executable code and automatically calculates income using a multi-factor weighted dynamic allocation algorithm. This achieves digital binding and automated execution of rights and responsibilities between the government and farmers, significantly shortening the subsidy disbursement cycle, avoiding human intervention in the income distribution process, and improving the efficiency and transparency of policy implementation.

[0006] The technical solution adopted by this invention to solve its technical problem is: a digital management system for government-agriculture collaboration based on multimodal fusion and smart contracts, the system comprising: The farmer terminal integrates a lightweight convolutional neural network model for collecting crop images, performing offline initial diagnosis of pests and diseases, signing smart contracts, and receiving decision-making information. The government system server is configured as a consortium blockchain node to deploy and execute policy smart contracts and maintain the on-chain ledger. Expert terminal, used to access video consultations and provide professional prevention and treatment solutions; The cloud server is equipped with deep learning models, time-series prediction units, and a knowledge graph inference engine. Multimodal databases are used to provide model enhancement analysis, full-lifecycle decision support, and data storage services. An Internet of Things (IoT) sensor network is deployed in the planting area to collect and upload environmental data. The farmer terminals, government system servers, expert terminals, and cloud servers are connected via network communication and collaboratively realize digital management of government and agriculture based on the multimodal data collected by the Internet of Things sensor network and farmer terminals.

[0007] Furthermore, the consortium blockchain deployed on the government system server of the present invention is a permissioned blockchain network built on multiple organizations, adopting the Raft consensus mechanism, and the policy smart contract is written in chaincode and deployed on the chain nodes.

[0008] Furthermore, the system described in this invention includes a smart contract collaboration module; The smart contract collaboration module is configured to execute a multi-factor weighted dynamic allocation algorithm to automatically calculate farmers' income according to a preset income calculation formula. The calculation formula is: Total income = Basic wage × Compliance coefficient + (Output × Unit price × Quality coefficient) + Performance bonus × Skill level coefficient. The compliance coefficient is calculated based on environmental and behavioral factors, and the skill level coefficient is determined based on the farmer's ability factor.

[0009] Furthermore, the system described in this invention includes a lightweight convolutional neural network model integrated into the farmer's terminal; The lightweight convolutional neural network model integrated into the farmer terminal is a MobileNetV3-Small model trained with knowledge distillation and quantization, used to perform offline disease and pest diagnosis on locally acquired crop images in the absence of a network.

[0010] Furthermore, the system of the present invention includes a ResNet50 model deployed on a cloud server; The deep learning model deployed on the cloud server is a ResNet50 model; the system is configured to: when the diagnostic confidence of the lightweight convolutional neural network model is lower than a first preset threshold, call the ResNet50 model to perform cloud-based verification of the uploaded image.

[0011] Furthermore, the system of the present invention is configured to: when the diagnostic confidence of the ResNet50 model reviewed in the cloud is lower than a second preset threshold or is determined to be a complex case, initiate a video consultation between the farmer terminal and the expert terminal via the WebRTC protocol.

[0012] Furthermore, the time-series prediction unit of the present invention includes a long short-term memory network model, which is used to predict output and price trends based on historical meteorological data, market conditions and production records, and supports dynamic incremental learning.

[0013] Furthermore, the knowledge graph reasoning engine of the present invention operates based on an agricultural knowledge graph that stores the correlation between crop growth, pest and disease control, soil fertility and meteorological disasters. The system is configured to automatically query the knowledge graph and generate a decision chain containing specific prevention and control solutions when it receives a forecast of a specific meteorological event, and push it to the farmer's terminal. The decision content includes pre-production planning, production management and post-production sales suggestions.

[0014] Furthermore, the IoT sensor network of the present invention is deployed with sensor groups at a density of at least one set per 50 acres. Each set includes a soil moisture sensor, a temperature and humidity sensor, and an image acquisition camera. After networking via the LoRa protocol, the data is uploaded through a gateway. The cloud server is equipped with a GPU to accelerate model inference.

[0015] This invention also provides a method for implementing a government-agriculture collaborative digital management system based on multimodal fusion and smart contracts, the method comprising the following steps: Step S1: Smart contract signing and storage, which transforms policy terms into executable code deployed on the consortium blockchain, and farmers complete the signing through digital signature; Step S2: Automated income calculation, which automatically calculates and distributes farmers' income based on a preset income calculation formula using a multi-factor weighted dynamic allocation algorithm. Step S3: Offline initial diagnosis, using the lightweight convolutional neural network model integrated in the farmer terminal to diagnose the locally collected crop images; Step S4: Cloud verification. When the diagnostic confidence of the lightweight convolutional neural network model is lower than the first preset threshold, the image is uploaded and verified by the ResNet50 model of the cloud server. Step S5: Expert consultation. When the diagnostic confidence of the ResNet50 model is lower than the second preset threshold or is determined to be a complex case, a video consultation is initiated. Step S6: Time series prediction, using the Long Short-Term Memory network model to predict crop yield and market price trends based on historical data; Step S7: Knowledge graph decision-making. Based on the agricultural knowledge graph, when a specific weather event forecast is received, the corresponding decision chain is automatically generated and pushed to the farmer's terminal.

[0016] Beneficial effects: 1. This invention transforms policy terms into on-chain executable code through a smart contract collaboration module, and automatically calculates income by combining a multi-factor weighted dynamic allocation algorithm. This realizes the digital binding and automated execution of rights and responsibilities between the government and farmers, which significantly shortens the subsidy disbursement cycle, avoids human intervention in the income distribution process, and improves the efficiency and transparency of policy implementation.

[0017] 2. This invention utilizes a lightweight AI diagnostic and offline-online hybrid diagnostic mechanism module, employing an optimized lightweight model deployed on farmer terminals to support rapid offline diagnosis. Furthermore, it combines a three-level response mechanism of cloud verification and expert consultation to ensure high availability and accuracy of diagnostic services in complex network environments, effectively adapting to the low-resource environment of rural areas.

[0018] 3. This invention utilizes a full-cycle intelligent decision support module, employing a long short-term memory network model to predict yield and price trends, and combines knowledge graphs to reason about agricultural knowledge. This provides farmers with data-driven decision-making support for each stage of pre-production planning, production management, and post-production sales, thereby improving the rationality of planting plans, the timeliness of disaster warnings, and ultimately helping farmers increase their income. Attached Figure Description

[0019] Figure 1 is a diagram of the overall system architecture of the present invention.

[0020] Figure 2 is a schematic diagram of the overall system of the present invention.

[0021] Figure 3 is a detailed diagram of the smart contract collaboration process of the present invention.

[0022] Figure 4 This is a detailed flowchart of the AI ​​diagnosis and hybrid diagnosis process of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Example 1 like Figure 1 As shown, the present invention provides a government-agriculture collaborative digital management system based on multimodal fusion and smart contracts. The system includes a smart contract collaboration module, a lightweight AI offline-online hybrid diagnostic module, and a full-cycle intelligent decision support module.

[0025] The hardware and network deployment of the system of this invention includes the following: IoT Sensing Layer: Within the planting area, one sensor group is deployed in units of 50 mu (approximately 3.3 hectares). This density has been verified in field practice to effectively cover the environmental spatial variations of typical farmland. Each sensor group includes: A soil moisture sensor (model: S-SMC-M005, measurement range: 0-100%VOL, accuracy: ±3%).

[0026] An air temperature and humidity sensor (model: AM2302, temperature range: -40~80°C, accuracy: ±0.5°C; humidity range: 0-100%RH, accuracy: ±3%RH).

[0027] A wide-angle HD camera (2 megapixels resolution, IP66 protection rating) is used to periodically capture images of the crop canopy.

[0028] The sensor group is connected internally via an RS-485 bus, and the groups form a self-organizing network using the LoRa wireless protocol (frequency band: 470MHz, transmission distance: 1-3km). The data is finally aggregated to the smart gateway (with built-in 4G communication module) deployed in the field and uploaded to the cloud server.

[0029] Terminal and server layers: Farmer terminal: A mid-to-low-end Android smartphone (≥4GB RAM, ≥64GB storage), pre-installed with a proprietary app. This app integrates an optimized lightweight convolutional neural network model (MobileNetV3-Small) and encapsulates a digital signature and contract interaction SDK.

[0030] Government administrative system server: physical server or cloud host (configuration: 4-core vCPU, 16GB memory, 500GB SSD), deploying consortium blockchain nodes (based on Hyperledger Fabric 2.4) and regulatory back-end management system.

[0031] Expert terminal: A high-performance laptop or workstation (configuration: ≥8-core CPU, 32GB memory), equipped with a professional consultation client, supporting high-definition video, image annotation, and treatment plan input.

[0032] Cloud server: Deployed using a cloud computing cluster (configuration example: 16-core vCPU, 128GB memory, equipped with NVIDIA V100 GPU, bandwidth ≥100Mbps). It hosts a multimodal database (using a MongoDB+PostgreSQL hybrid architecture), model inference service, knowledge graph engine (based on Neo4j4.0), and business middleware platform.

[0033] The implementation details of the smart contract collaboration module are shown in Figure 2: The consortium blockchain network comprises three organizations: government agencies, farmer cooperatives, and agricultural technology service centers. Each organization operates two Fabric peer nodes and one orderer node. The network uses the Raft consensus algorithm with a consensus period of 2 seconds to meet policy transaction throughput requirements while ensuring consistency. Smart contracts (chaincode) are developed using Go (version 1.16). For example, the core logic of a "high-quality wheat planting subsidy contract" chaincode includes: farmer identity verification, on-chain registration of planting area, periodic reading of environmental compliance data, and automatic subsidy calculation and payment triggering based on the multi-factor weighted dynamic allocation algorithm. After the chaincode is verified by the endorsement strategies of each organization, it is installed and instantiated on the channel.

[0034] Farmers complete real-name authentication via the identity verification interface provided by the National Government Service Platform on the app. The system then generates a digital certificate for them based on the Elliptic Curve Digital Signature Algorithm (ECDSA, curve: secp256k1). This certificate is used to sign contract signing transactions initiated on the blockchain. A complete "contract signing" transaction includes: contract ID, farmer's digital signature, and timestamp. After being sorted, packaged into blocks, and written to the ledger, a legally valid digital signature is completed. Subsequently, any changes to the contract status (such as "task completed," "subsidy triggered," etc.) are recorded on the blockchain as new transactions, forming a complete and tamper-proof audit trail.

[0035] The multi-factor weighted dynamic allocation algorithm, as the core of this module, is integrated into the chaincode and cloud service as executable code. The algorithm is triggered daily at 2:00 AM as a scheduled task, reading all factor data from the previous day from the multimodal database and performing the following calculations: Total Revenue = Base Salary × Compliance Coefficient + (Output × Unit Price × Quality Coefficient) + Performance Bonus × Skill Level Coefficient in: Compliance coefficient calculation: The system reads the soil moisture (target value 60%±10%), air temperature (target range 15-25°C), and operation records of the farmer's land plot over the past week. If the percentage of days with all environmental data meeting the standards is ≥90%, and the operation image is recognized as "compliant" by AI, the coefficient is 1.0; each non-compliance is deducted according to a preset weight.

[0036] Determination of quality coefficient: After the product is harvested, sample images are collected by camera, and the ResNet50 model in the cloud is used to grade the quality (outputting a score of 0-1). Combined with the expert spot check score (out of 10), the weighted average is converted into a quality coefficient of 0-1.

[0037] Skill level coefficient: Based on factors such as the farmer's historical diagnostic accuracy, online training hours attended, and frequency of adopting intelligent decisions from the system, the system dynamically assesses the skill level into L1-L5 levels, corresponding to coefficients of 0.8, 1.0, 1.2, 1.5, and 2.0, respectively.

[0038] The calculation result (e.g., total revenue of 2445 yuan) will automatically generate a payment instruction. After the payment instruction is confirmed by the consortium blockchain, it will trigger the API of the bank system or third-party payment platform, and transfer the funds to the farmer's bound account within 1-3 business days. At the same time, the account status will be updated on the blockchain and in the farmer's APP.

[0039] like Figure 3 As shown, the implementation process of the lightweight AI offline-online hybrid diagnostic module of the present invention includes the following: Edge Model Deployment: A lightweight convolutional neural network model for offline diagnosis was obtained through the following steps: First, a ResNet34 model was trained on a plant disease dataset containing 150,000 labeled images in the cloud as a "teacher model." Then, knowledge distillation was used to allow the MobileNetV3-Small "student model" to learn the soft-label output and intermediate layer features of the teacher model. Finally, the distilled model was quantized using INT8, converting the model weights from FP32 to INT8 format. The optimized model file size is approximately 48MB. On a test phone equipped with a Qualcomm Snapdragon 662 chip, the inference time per image (224x224) is approximately 350 milliseconds, and the memory usage is approximately 95MB.

[0040] The three-level diagnostic process specifically includes: Offline initial diagnosis: Farmers open the APP in the field and take pictures of leaves suspected of being diseased. The APP automatically calls the local model to perform real-time inference. If the output confidence level is higher than the threshold Th1 (set to 75%), the result is displayed directly (e.g., "Wheat stripe rust, confidence level 82%)" and the basic control plan (e.g., "15% triadimefon wettable powder, 100 grams per acre diluted with water for spraying").

[0041] Cloud-based verification: If the confidence level is lower than Th1, the app automatically compresses the image to below 500KB and transmits it to the cloud via a 4G / 5G network. A ResNet50 model deployed in the cloud (with an inference time of approximately 200 milliseconds under TensorRT acceleration) performs the verification. If the confidence level is higher than the threshold Th2 (set to 90%), the cloud result is taken as the final result and fed back to the terminal.

[0042] Expert Consultation: If the confidence level in the cloud is still lower than Th2, or the model identifies a mixture of multiple disease characteristics, the system automatically dispatches online experts through the signaling server and establishes a P2P video call between the farmer's terminal and the expert's terminal based on the WebRTC protocol. Experts observe the video, control the farmer's camera zoom, and can annotate the shared image, ultimately generating a consultation report (e.g., "Confirmed as a mixed infection of wheat powdery mildew and leaf rust; recommended to use 30% azoxystrobin·cyprodinil suspension at 40ml / acre, repeat after 7 days"). After farmer confirmation, the report is structured and stored in the cloud case library.

[0043] Model Iteration Closed Loop: Expert consultation reports stored in the case library contain high-quality image-label pairs and are automatically added to the model retraining dataset every month to start a round of incremental training, enabling the diagnostic model to continuously evolve.

[0044] The implementation of the full-cycle intelligent decision support module of this invention, based on data fusion and reasoning, includes the following: The time-series forecasting unit employs a Long Short-Term Memory (LSTM) network model. The model inputs include daily regional meteorological data (temperature, precipitation, sunshine) from the past 5 years, wholesale price indices of major agricultural products from the past 3 years, and production records (crop type, inputs, yield) of the target farmers from the past 2 years. The LSTM network structure consists of 3 hidden layers with 128 neurons per layer. The model is automatically trained weekly and uses a rolling forecasting approach, outputting the yield trend index and price fluctuation range for the next 4 weeks. The forecast results are displayed in chart form in the "Market Outlook" section of the farmer's app.

[0045] Knowledge Graph Decision Unit: The agricultural knowledge graph is built using Neo4j and contains approximately 100,000 entities (such as crops, pests and diseases, pesticides, weather events, and agricultural operations) and approximately 500,000 relationships (such as "susceptible to infection," "control required," "inducing," and "operation timing"). When the time-series prediction unit issues a warning of "continuous cooling accompanied by rainfall within the next 72 hours," the inference engine automatically executes the following Cypher query: MATCH(w:WeatherEvent{type:'Cooling Down and Rain'})-[r:TRIGGERS]->(d:Disease)-[s:HAS_PREVENTION]->(p:Plan) WHEREd.crop='Wheat' RETURN p.content, p.detail; The query results may return decision nodes such as "irrigate and regulate temperature before the cold wave" and "spray potassium dihydrogen phosphate foliar fertilizer to enhance resistance." The system combines these nodes into an actionable decision chain: "Warning: A cold wave is approaching. Recommendations: 1. Irrigate the fields 24 hours before the temperature drops, with the water level reaching half the height of the ridges; 2. Immediately spray 98% potassium dihydrogen phosphate, 100 grams per acre diluted in 30 kilograms of water." This decision chain is instantly sent to all wheat farmers in the relevant areas via APP push and SMS.

[0046] Example 2 This embodiment demonstrates the collaborative workflow of each module of the system through a complete business loop from disease occurrence to revenue settlement.

[0047] Scene: A rice farmer, Mr. Zhang, discovered brown streaks on some of the rice leaves in his contracted field.

[0048] Step 1: Real-time diagnosis and early warning (module collaboration) Mr. Zhang opened a mobile app and took a picture of the diseased leaves. The lightweight local model gave a preliminary diagnosis within 1.2 seconds: "Rice sheath blight, confidence level 70%". Since the confidence level was lower than the threshold Th1 (75%), the app automatically uploaded the image to the cloud.

[0049] The cloud-based ResNet50 model returned the verification result after 0.8 seconds: "Rice sheath blight, confidence level 88%". The system then immediately pushed a detailed treatment plan based on this result.

[0050] Almost simultaneously, the LSTM model of the full-cycle decision support module predicted, based on real-time meteorological data streams, that the region would experience hot and humid weather in the next 48 hours. This prediction immediately triggered knowledge graph inference.

[0051] The graph engine found that "high temperature and high humidity" are strong inducing factors for both "rice sheath blight" and "rice blast," and also discovered historical cases of mixed occurrences in Zhang's area. Therefore, the system generated an upgrade decision chain: "Emergency warning: High temperature and high humidity in the next two days are very likely to aggravate the disease and induce rice blast. Recommendations: 1. Immediately drain the waterlogged fields to reduce humidity. 2. Replace the original treatment plan with a compound agent that controls both sheath blight and rice blast, '75% azoxystrobin·tebuconazole water-dispersible granules', 15 grams per acre."

[0052] Step 2: Agricultural Implementation and Data Recording Mr. Zhang received a comprehensive decision-making plan that combined real-time diagnosis and early warning, and decided to adopt it. He carried out drainage and pesticide application operations according to the plan.

[0053] During pesticide application, the app guided him to record short videos of the pesticide preparation and application process (as part of the behavioral data). IoT sensors continuously recorded changes in temperature and humidity in the field after pesticide application.

[0054] Step 3: Automated Revenue Calculation After the rice harvest this season, Zhang reported a yield of 550 kg per mu via the APP. The system combined the average market price of the variety (provided by the time series prediction unit, which is 3.2 yuan / kg) and the quality test results (AI quality recognition score of 92, equivalent to a quality coefficient of 0.92).

[0055] The smart contract collaboration module initiates settlement: it retrieves Zhang's compliance coefficient for this quarter (environmental data compliance rate of 95%, work standardization score of 9 / 10, calculated as 1.0) and his skill level coefficient (L2, corresponding to 1.0), combined with his basic salary and performance bonus.

[0056] Substituting all data into the formula, the total revenue is automatically calculated and stored on the blockchain. The on-chain smart contract automatically takes effect after the two conditions of "production reporting completed" and "quality inspection completed" are met, and the subsidy is issued to Zhang's account. The entire process from harvest to subsidy receipt is completed within 5 working days, far exceeding the efficiency of the traditional model.

[0057] This scenario fully demonstrates how the system drives a closed loop of intelligent diagnosis, forward-looking early warning, precise decision-making, and automatic contract execution through the fusion of multimodal data (images, sensors, and markets), ultimately achieving the goal of improving management efficiency and ensuring farmers' income.

[0058] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural modifications made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

[0059] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms installation, connection, and link should be interpreted broadly, and can be mechanical or electrical connection, or internal connection between two components, or direct connection. The terms up, down, left, right, etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.

[0060] Secondly, the accompanying drawings of the embodiments disclosed in this invention only involve structures related to the embodiments disclosed in this invention. Other structures can refer to general designs. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0061] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A government-farmer coordination digital management system based on multi-modal fusion and smart contract, characterized in that, include: The farmer terminal integrates a lightweight convolutional neural network model for collecting crop images, performing offline initial diagnosis of pests and diseases, signing smart contracts, and receiving decision-making information. The government system server is configured as a consortium blockchain node to deploy and execute policy smart contracts and maintain the on-chain ledger. Expert terminal, used to access video consultations and provide professional prevention and treatment solutions; The cloud server is equipped with deep learning models, time-series prediction units, and a knowledge graph inference engine. Multimodal databases are used to provide model enhancement analysis, full-lifecycle decision support, and data storage services. An Internet of Things (IoT) sensor network is deployed in the planting area to collect and upload environmental data. The farmer terminals, government system servers, expert terminals, and cloud servers are connected via network communication and collaboratively realize digital management of government and agriculture based on the multimodal data collected by the Internet of Things sensor network and farmer terminals.

2. The government-agriculture collaborative digital management system based on multimodal fusion and smart contracts according to claim 1, characterized in that, The consortium blockchain deployed on the government system server is a permissioned blockchain network built on multiple organizations and adopts the Raft consensus mechanism. The policy smart contract is written in chaincode and deployed on the chain nodes.

3. The government-agriculture collaborative digital management system based on multimodal fusion and smart contracts according to claim 1, characterized in that, The system includes a smart contract collaboration module; The smart contract collaboration module is configured to execute a multi-factor weighted dynamic allocation algorithm to automatically calculate farmers' income according to a preset income calculation formula. The calculation formula is: Total income = Basic wage × Compliance coefficient + (Output × Unit price × Quality coefficient) + Performance bonus × Skill level coefficient. The compliance coefficient is calculated based on environmental and behavioral factors, and the skill level coefficient is determined based on the farmer's ability factor.

4. The government-agriculture collaborative digital management system based on multimodal fusion and smart contracts according to claim 1, characterized in that, The system includes a lightweight convolutional neural network model integrated into the farmer's terminal; The lightweight convolutional neural network model integrated into the farmer terminal is a MobileNetV3-Small model trained with knowledge distillation and quantization, used to perform offline disease and pest diagnosis on locally acquired crop images in the absence of a network.

5. The government-agriculture collaborative digital management system based on multimodal fusion and smart contracts according to claim 1, characterized in that, The system includes a ResNet50 deep learning model deployed on a cloud server; The deep learning model deployed on the cloud server is a ResNet50 model; the system is configured to: when the diagnostic confidence of the lightweight convolutional neural network model is lower than a first preset threshold, call the ResNet50 model to perform cloud-based verification of the uploaded image.

6. A government-agriculture collaborative digital management system based on multimodal fusion and smart contracts as described in claim 5, characterized in that, The system is configured to initiate a video consultation between the farmer's terminal and the expert's terminal via the WebRTC protocol when the diagnostic confidence of the ResNet50 model reviewed in the cloud is lower than a second preset threshold or is determined to be a complex case.

7. The government-agriculture collaborative digital management system based on multimodal fusion and smart contracts according to claim 1, characterized in that, The time-series prediction unit includes a long short-term memory network model, which is used to predict output and price trends based on historical meteorological data, market conditions and production records, and supports dynamic incremental learning.

8. The government-agriculture collaborative digital management system based on multimodal fusion and smart contracts according to claim 1, characterized in that, The knowledge graph reasoning engine operates based on an agricultural knowledge graph that stores the correlations between crop growth, pest and disease control, soil fertility, and meteorological disasters. The system is configured to automatically query the knowledge graph and generate a decision chain containing specific prevention and control solutions when it receives a forecast of a specific meteorological event. The decision chain is then pushed to the farmer's terminal. The decision content includes pre-production planning, production management, and post-production sales suggestions.

9. A government-agriculture collaborative digital management system based on multimodal fusion and smart contracts as described in claim 1, characterized in that, The IoT sensor network is deployed with at least one sensor group per 50 acres. Each group includes a soil moisture sensor, a temperature and humidity sensor, and an image acquisition camera. Data is uploaded through a gateway after networking via the LoRa protocol. The cloud server is equipped with a GPU to accelerate model inference.

10. A method for implementing a government-agriculture collaborative digital management system based on multimodal fusion and smart contracts, characterized in that, The method includes the following steps: Step S1: Smart contract signing and storage, which transforms policy terms into executable code deployed on the consortium blockchain, and farmers complete the signing through digital signature; Step S2: Automated income calculation, which automatically calculates and distributes farmers' income based on a preset income calculation formula using a multi-factor weighted dynamic allocation algorithm. Step S3: Offline initial diagnosis, using the lightweight convolutional neural network model integrated in the farmer terminal to diagnose the locally collected crop images; Step S4: Cloud verification. When the diagnostic confidence of the lightweight convolutional neural network model is lower than the first preset threshold, the image is uploaded and verified by the ResNet50 model of the cloud server. Step S5: Expert consultation. When the diagnostic confidence of the ResNet50 model is lower than the second preset threshold or is determined to be a complex case, a video consultation is initiated. Step S6: Time series prediction, using the Long Short-Term Memory network model to predict crop yield and market price trends based on historical data; Step S7: Knowledge graph decision-making. Based on the agricultural knowledge graph, when a specific weather event forecast is received, the corresponding decision chain is automatically generated and pushed to the farmer's terminal.

Images