Intelligent agricultural management system based on planting-harvesting-distribution full link

Abstract

The invention relates to an intelligent agricultural management system based on a planting-harvesting-distribution full link, and belongs to the field of agriculture, and the system comprises an online seed selection module, an order tracking module, an environment monitoring and early warning module, a pest and disease identification module, a market prediction module, a logistics scheduling module, a product estimation module and a growth state evaluation module. In the planting link, intelligent monitoring and abnormal early warning are carried out on the crop growth environment through the intelligent agent, accurate planting schemes for different crops can be formulated for farmers, and irrigation and fertilization schemes suitable for different regions can be formulated; in the production process, the intelligent agent can intelligently identify bad fruits and give out a treatment scheme, accurately identify diseases and pests and provide an effective prevention and treatment means; in the marketing link, the market trend can be predicted, the production scheme is reasonably arranged, and the production and marketing balance of agricultural products is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agriculture, and in particular to an intelligent agricultural management system based on a planting-harvesting-distribution full link. BACKGROUND

[0002] In the current market environment of the rapid development of digital agriculture, various agricultural-related platforms are emerging, and the competition in the field of intelligent agriculture is becoming increasingly fierce. The current existing technologies or products mainly focus on the transaction link of agricultural products. They build online sales platforms to provide a simple trading place for farmers and consumers. For example, community group-buying platforms represented by Meituan Xieyou provide an important role in connecting consumers and agricultural products. They integrate supply chain resources and build a convenient online sales channel, allowing consumers to purchase various agricultural products at relatively low prices. These platforms can quickly and efficiently deliver agricultural products to communities through their powerful logistics and distribution systems, meeting the daily needs of consumers. However, the business focus of these platforms is mainly on the sales and distribution links, and they do not deeply involve the production process of agricultural products. Their quality control of agricultural products often only stops at simple steps such as receiving and inspecting, and they have little understanding of the specific conditions of agricultural products during planting and breeding. For example, they cannot guide and supervise the planting methods of farmers, and farmers cannot obtain timely and professional help from these platforms when they encounter technical difficulties, pest control, and other problems during the planting process. They cannot ensure that agricultural products follow green, environmentally friendly, and safe standards during growth. This may lead to certain hidden dangers in the quality and safety of the agricultural products purchased by consumers. Moreover, due to the lack of in-depth understanding of the production link, community group-buying platforms also cannot reasonably plan and guide the production of agricultural products according to market demand changes, and cannot fundamentally solve the problem of mismatch between supply and demand of agricultural products.

[0003] In addition, some existing products in the current wave of development of smart agriculture focus on the management link of agricultural production, such as the smart agricultural management system of the "smart agricultural platform". They use advanced agricultural Internet of Things technology to collect real-time and accurate data in the farmland environment. For example, they can monitor the soil pH, fertility, humidity, and air temperature, humidity, and light intensity. At the same time, using basic data analysis capabilities, these collected data are preliminarily processed and interpreted to provide some reference for farmers to understand the environmental conditions for crop growth. However, such platforms have obvious limitations. They mainly focus on data monitoring and basic analysis in the agricultural production link, and lack integration of the entire chain from production to sales of agricultural products. That is, after the harvest of agricultural products, how to efficiently push them to the market to meet consumer demand, and how to establish a production plan that matches market demand, they do not provide effective solutions. Although farmers can obtain data on the growth environment of crops through these platforms, they still face many challenges in the sales link, such as difficulty in finding suitable sales channels and predicting market demand, which may result in unsold agricultural products and failure to maximize the economic benefits of agricultural production. SUMMARY

[0004] The present application aims to overcome the shortcomings of the prior art and provides a smart agricultural management system based on the whole link of planting-harvesting-distribution, which solves the problems existing in the prior art.

[0005] The present application is achieved by the following technical solution: a smart agricultural management system based on the whole link of planting-harvesting-distribution, the system comprising: an online seed selection module, an order tracking module, an environment monitoring and early warning module, a disease and pest identification module, a market prediction module, a logistics scheduling module, a product estimation module, and a growth state evaluation module. The online seed selection module is configured to allow consumers to select crops according to their preferences and needs, and to transmit the seed selection information to the corresponding farmers to start exclusive planting services. The order tracking module is configured to allow consumers to track the status of the order in real time through the system when the farmers start to grow the crops selected by the consumers. The environment monitoring and early warning module is configured to monitor and analyze the environmental data of the farmland, including soil humidity, temperature, pH, air temperature, humidity, and light intensity. Once the environmental data is abnormal, the module sends warning information to the farmers and provides suggestions for countermeasures. The disease and pest identification module is configured to use image recognition and deep learning technology to quickly and accurately identify the types of diseases and pests, and to provide prevention and control solutions. The market prediction module is configured to analyze the supply and demand situation and price fluctuation trend information of different fruit and vegetable varieties on the market through big data analysis and artificial intelligence analysis, and the agent predicts the trend of the agricultural product market, adjusts the planting plan, and selects the varieties with high market demand and high price for planting. The logistics scheduling module is configured to arrange logistics vehicles and distribution routes according to the distribution of orders and the time requirement of distribution, monitor the position and transportation state of the logistics vehicles in real time, and adjust the logistics plan in time according to external factors to ensure the smooth delivery. The product estimation module is configured to use the YOLOv8 model to analyze images during crop growth, identify the number and growth state of crops, and combine historical data and environmental factors to predict the final yield. The growth state evaluation module is configured to comprehensively analyze environmental data, pest and disease conditions, and growth cycles, and the agent quantitatively analyzes the health state and growth speed of crops, continuously evaluates the growth state of crops, and provides suggestions and solutions to farmers.

[0006] The system further comprises a knowledge popularization module, an intelligent planting guidance module, and a plant adoption module. The knowledge popularization module is configured to provide agricultural knowledge to consumers, including the nutritional value, storage method, cooking suggestions of different fruits and vegetables, the planting process of crops, and the prevention and control of pests and diseases. The intelligent planting guidance module is configured to use the agent to collect farmland-level environmental data, combine the growth rules of crops and historical planting data, and develop personalized planting schemes for farmers. The plant adoption module is configured for consumers to choose different appearances of land and plant varieties planted on the land according to personal preferences, and purchase the corresponding adoption rights.

[0007] The agent also determines whether the environment is suitable for crop growth through data analysis, customizes exclusive planting schemes for different crops by considering the characteristics of crop varieties, climate and soil conditions in planting areas, and seasonal changes, develops precise irrigation and fertilization schemes according to regional differences and water resource status and soil fertility, and identifies damaged fruits and gives treatment schemes.

[0008] The determination of whether the environment is suitable for crop growth through data analysis specifically includes the following contents: Real-time environmental data collection: Deploy a distributed IoT sensor network in farmland, including soil sensors to monitor and collect soil temperature, humidity, pH, and EC values, air sensors to monitor and collect air temperature, humidity, and CO2 concentration, light sensors to monitor and collect photosynthetically active radiation, and weather stations to monitor and collect wind speed and precipitation. At the same time, use Beidou positioning to achieve spatial calibration of field-level data, correlate data with specific planting areas, and provide spatial coordinate benchmarks for subsequent regional judgment. Crop growth suitability parameter library construction: Based on the agricultural knowledge base, crop variety characteristics database, and old farmer experience library, a three-dimensional threshold system of crop-growth stage-environment parameters is constructed. Combine historical planting data and machine learning models to dynamically modify the basic threshold values and analyze environmental trends to predict environmental risks in the future. Data analysis model judgment: Preprocess the collected raw data and introduce a multi-level suitability judgment algorithm for judgment. Automatically divide the warning levels according to the data deviation, regularly compare the actual crop growth feedback after warning with the judgment results, and update the parameter library threshold values and multi-factor weights through reinforcement learning algorithm to improve the judgment accuracy.

[0009] The multi-level suitability judgment algorithm includes: First layer, rule engine rapid screening: Directly compare real-time data with threshold values in the parameter library. If single-dimensional data exceeds the threshold value, it is immediately marked as abnormal. Second layer, multi-factor collaborative analysis model: Use gradient boosting tree algorithm to analyze the interaction of multiple parameters. If one of the parameters in the multi-parameter interaction exceeds the threshold value, the model judges it as a potential risk environment. Third layer, growth stage adaptation adjustment: Confirm the current growth stage of crops through planting cycle records and image recognition, and adjust the parameter threshold values of the current stage to prevent false positives caused by universal parameter threshold values.

[0010] The comprehensive crop variety characteristics, planting area climate and soil conditions, and seasonal change factors are used to customize specific planting schemes for different crops, which include the following: Multi-dimensional basic data collection and integration: Construct a crop variety characteristics database and collect regional climate and soil data in real time. Divide seasons into modules according to phenological periods, correlate typical climate parameters of each season, and map them to environmental requirements of key growth stages of crops. Intelligent decision-making model scheme generation: Combine crop germination temperature threshold values with regional spring temperature recovery curves to calculate the optimal sowing time window period; calculate planting density based on soil fertility grade, crop plant type, and row spacing formula; integrate old farmer experience library and knowledge base to generate basic schemes for irrigation period and fertilizer type; modify the generated basic schemes according to regional characteristics, and embed preventive measures in the schemes based on the seasonal climate risk database. The whole-cycle dynamic adjustment mechanism of the scheme: based on the recorded crop growth stage, the agent updates the scheme every set time, and when the environmental monitoring data deviates from the set range, the emergency adjustment mechanism of the scheme is automatically triggered, and the agent optimizes the scheme according to the feedback data of the farmers, and the scheme is decomposed into executable task chains and automatically pushed to the farmer end app.

[0011] The precise irrigation and fertilization scheme formulated according to regional differences, combined with water resource status and soil fertility, specifically includes the following contents: Multi-dimensional basic data collection and regional feature modeling: collect soil fertility data and combine regional soil survey data to establish a regional baseline of soil fertility; collect water resource data, divide water resource abundance levels and associate irrigation costs; establish a crop demand database; Precise irrigation scheme generation: calculate the crop water requirement in real time according to the crop coefficient, evaporation and transpiration, water resource abundance level and soil water retention capacity; select the irrigation method according to the regional water source type, and optimize and adjust the irrigation time combined with weather prediction and crop growth stage; output specific irrigation parameters, and automatically execute through the Internet of Things control irrigation equipment; Precise fertilization scheme generation: determine the target nutrient uptake according to the crop variety and growth stage, and calculate the natural supply proportion according to the current soil fertility data; select the fertilizer type according to the soil type, select the application method according to the regional agricultural conditions, and formulate environmental protection constraints; Regional adaptation and dynamic optimization mechanism: divide the country into 6 major agricultural ecological regions, dynamically correct each region according to regional differences, and the agent collects soil data every N days through the Internet of Things equipment, compares the actual value with the expected value of the scheme according to the crop growth state, and optimizes the fertilization scheme according to the comparison result.

[0012] The disease and pest identification module specifically includes the following contents: Multi-dimensional construction of disease and pest feature database: construct the feature database of crop diseases and pests through agricultural knowledge base, field image sample collection and old farmer experience data; Disease and pest identification model construction and training: add CBAM attention module to the ResNet-50 model to strengthen the extraction of key features of disease spot edges and insect outline, and use depth separable convolution to compress 40% of the disease and pest identification model parameters; perform data enhancement processing on the image, expand the sample, train according to the type of disease and pest + crop variety, increase the sample proportion of high-incidence diseases and pests, correct the misjudgment of similar diseases and pests through confusion matrix, and improve the recognition accuracy of the disease and pest identification model; deploy the trained disease and pest identification model on the farmer end edge device; Disease and pest identification and control of intelligent agents: Farmers upload images of diseased parts of fruits through the system, and the intelligent agent performs multi-dimensional identification and cross-validation after image preprocessing, and generates a control scheme according to chemical control, physical / biological control and emergency treatment. Finally, the control scheme is optimized according to the feedback data of farmers and the actual effect in the field.

[0013] The identification of damaged fruits and the generation of treatment schemes specifically include the following contents: Construction of damaged fruit feature database: Construct a multi-dimensional crop damaged fruit feature database, clearly define the types of disease damage, pest damage and physical damage, and label the collected damaged fruit images; Training and deployment of damaged fruit recognition model: Add a coordinate attention module to the Neck layer of the MobileNetV3 deep learning model to enhance feature extraction of damaged area edges, and use depth separable convolution to reduce damaged fruit recognition model parameters by 35% to adapt to farmers' mobile phones and field intelligent terminal devices; Data augmentation is performed on the images to expand the samples, and classification training is performed according to crop varieties + damage types. For the set crops, increase the sample weight, and optimize the ability to distinguish similar damage through the confusion matrix. Deploy the trained damaged fruit recognition model on the farmer's edge device; Identification of damaged fruits by intelligent agents: Farmers upload images of damaged fruits, and the intelligent agent performs multi-dimensional identification and cross-validation after image preprocessing; Treatment scheme generation: Intelligent agents automatically match treatment measures according to the degree of damage, track the progress of treatment through Internet of Things devices, and collect downstream feedback data to update treatment rule weights to optimize the scheme.

[0014] The product estimation module specifically includes the following contents: Crop growth image data collection and labeling: Multi-scene image collection is performed, and the collected images are manually labeled and automatically verified; Training and optimization of YOLOv8 model: Add SE attention to the Neck layer of the YOLOv8 model to strengthen the extraction of key features such as fruit outline and color, reduce model parameters, and adapt to farmers' mobile phones and field edge devices. Hyperparameter optimization is performed; first, pre-train the YOLOv8 model on a general object detection dataset, then use the crop fruit dataset to adjust the trained YOLOv8 model, optimize the recognition ability of overlapping fruits and occluded fruits, and train sub-models for different crop characteristics; The yield estimation: the YOLOv8 model outputs the fruit quantity, fruit size and fruit distribution density data after feature extraction on the input image, combines the crop variety characteristics and planting parameters, calculates the estimated total yield through the total yield = fruit quantity per unit area x total planting area x average weight of single fruit x mature rate correction factor, and introduces environmental factors to modify the result.

[0015] The present application has the following advantages: a smart agricultural management system based on planting-harvesting-distribution whole link, in the planting link, the intelligent agent can intelligently monitor and abnormally warn the crop growth environment, can formulate accurate planting schemes for different crops for farmers, and irrigation and fertilization schemes suitable for different regions. In the production process, the intelligent agent can also intelligently identify bad fruits and give treatment schemes, accurately identify diseases and insect pests and provide effective control means. In the market sales link, the market trend can be predicted, the production scheme can be reasonably arranged, and the production and marketing balance of agricultural products can be ensured. At the same time, combined with the experience of old farmers and various knowledge bases, knowledge popularization is carried out, and rich agricultural knowledge is provided for farmers and consumers. Moreover, the best harvesting time can be predicted, the growth state of crops can be evaluated, the production decision can be reasonably optimized, and even the yield can be estimated combined with the Yolov8 model, to provide all-round intelligent support for the whole agricultural production process. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The figure is a structural schematic diagram of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in combination with the drawings of the present application is not intended to limit the protection scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. The present application will be further described below in combination with the drawings.

[0018] The present application relates to a smart agricultural management system or platform based on planting-harvesting-distribution whole link, which accurately locates different types of consumers in cities who pursue fresh and customized fruits and vegetables, and farmers and cooperatives facing planting and sales difficulties, and can provide targeted services for them to meet their diversified needs.

[0019] As Figure 1As shown, the system comprises: an online seed selection module, an order tracking module, an environment monitoring and early warning module, a pest and disease identification module, a market prediction module, a logistics scheduling module, a product estimation module, a growth state evaluation module, a knowledge popularization module, an intelligent planting guidance module, and a plant adoption module; Among them, the online seed selection module: after the consumer enters the platform, he can browse a variety of crops. Here covers various common fruit and vegetable varieties, from traditional popular tomatoes, cucumbers, strawberries, etc. Consumers can learn about the characteristics of each variety in detail, such as growth cycle, taste characteristics, nutritional value, etc. At the same time, according to their own preferences and needs, they can choose from a variety of crop varieties, and the platform provides detailed information for each crop, including growth habits, suitable environment, planting difficulty, etc. such as preference for sweetness, acceptance of planting difficulty, etc. to choose the desired crop. After the selection is completed, the platform will convey the consumer's seed selection information to the corresponding farmer in a timely manner, and start exclusive planting services.

[0020] Order tracking module: when the farmer starts to plant the crops selected by the consumer, the platform provides users with comprehensive planting guidance. The intelligent agent formulates a personalized planting plan according to the user's selected crops and the environmental conditions in the region, covering sowing time, planting density, fertilization period, irrigation water volume, etc. During the planting process, consumers can track the status of the order in real time through the platform. From seed sowing, germination, growth, to fertilization, irrigation, pest control, etc. Each key node can be presented to consumers in the form of text, video, etc. Consumers can also see the expected harvest time and delivery time, and know when the fruit and vegetable they ordered will arrive. If any abnormal situation occurs during planting, such as sudden natural disasters affecting crop growth, the platform will also update the information in a timely manner, so that consumers can understand the latest developments of the order in the first time.

[0021] Knowledge popularization module: the platform has a special knowledge popularization board, which provides consumers with rich agricultural knowledge. The content includes the nutritional value of different fruits and vegetables, scientific storage methods, healthy cooking suggestions, etc. In addition, it also introduces the planting process of crops, the prevention and control knowledge of common pests and diseases, etc. so that consumers can better understand the whole process of the fruits and vegetables they eat from the field to the table, enhance their awareness and trust of agricultural products, and also cultivate their healthy eating concepts.

[0022] Intelligent planting guidance module: Farmers can obtain accurate intelligent planting guidance on the platform. The intelligent agent of the platform formulates personalized planting schemes for farmers by combining the collected farmland-level environmental data with the growth rules and historical planting data of crops. For example, it intelligently recommends appropriate types and amounts of fertilizers according to the soil fertility status, and determines the optimal sowing and irrigation time according to the local climate conditions. At the same time, it also provides detailed planting operation steps and precautions to help farmers scientifically plant and improve the yield and quality of crops.

[0023] Environmental monitoring and early warning module: The platform can monitor and analyze the environment of farmland, including soil moisture, temperature, pH, and key indicators such as air temperature, humidity, and light intensity. Once the environmental data is abnormal, such as soil moisture being too high which may cause root rot, or temperature being too low which may affect crop growth, the platform will immediately send warning information to the farmers and provide corresponding response measures recommendations to help farmers take timely action to avoid or reduce losses.

[0024] Disease and pest identification module: When farmers find abnormal symptoms on crops, they can use the disease and pest identification function of the platform for diagnosis. Farmers only need to upload photos or videos of the affected parts of the crops, and the intelligent agent uses image recognition and deep learning technology to quickly and accurately identify the type of disease and pest, and provides detailed prevention and control schemes, including recommended types of pesticides, application time, and methods. This greatly improves the efficiency of farmers in preventing and controlling diseases and pests, and reduces the loss of reduced yield caused by diseases and pests.

[0025] Market prediction module: The platform uses big data analysis and artificial intelligence technology to predict the trend of agricultural product markets. It analyzes information such as the supply and demand situation of different fruit and vegetable varieties on the market, and the price fluctuation trend, and provides market prediction reports for farmers. Farmers can adjust their planting plans based on this information, choose varieties with high market demand and high prices for planting, and avoid blind following of the market, which can lead to unsold agricultural products and improve planting income.

[0026] Further, the market prediction function of the intelligent agent is realized through the whole process of multi-source data fusion-intelligent model reasoning-dynamic scheme output, providing accurate market trend judgment and production adjustment suggestions for farmers, with the specific steps as follows: 1. Multi-dimensional market data collection and integration; The intelligent agent collects market-related data through multiple channels to build a comprehensive prediction data source, including: 1. External market dynamic data: interface with the public data of third-party agricultural e-commerce platforms (such as Yimutian, Meituan Xingyun) to obtain cross-platform price trends of similar crops and regional supply-demand differences (such as 30% higher demand for citrus in North China in winter than in summer); Integrate government-issued agricultural policies (such as the Green Food Subsidy Policy and the Agricultural Product Transportation Green Channel), weather warnings (such as typhoons that may lead to reduced fruit production in the south), and consumer trend reports (such as the demand for pre-ordered planting by 85 after families in first-tier cities increased by 42% per year).

[0027] 2. Crop-specific association data: Combine crop characteristics (such as growth cycle, storage tolerance) in the plant details table, regional adaptability (such as Xinjiang grapes with high sweetness have a 20% premium in the East China market), and regional sales patterns in the old farmer experience library (such as leafy vegetables are easily damaged during the Meiyu season in the south, so the planting quantity needs to be reduced in advance).

[0028] II. Data preprocessing and feature engineering: Clean and transform the collected raw data to extract key prediction features: 1. Data cleaning and normalization; Remove outliers (such as order quantity jumps due to system failures) and fill in missing values (such as using the historical average of the same crop in the same region to supplement the sales data for a certain week); Standardize different formats of data (such as converting price / kg and sales / ton to production value units of 10,000 yuan / acre, and converting subsidy ratios in policy documents to quantitative impact factors).

[0029] 2. Core feature extraction; Extract key dimensions that affect the market from the data, including: Time features: crop marketing cycle (such as strawberries from December to the following April are in the peak season), holiday effects (such as citrus sales increase by 50% before the Spring Festival); Regional features: different provinces' consumption preferences (such as the demand for apples in the north is 1.5 times that in the south), and logistics cost differences (transportation costs in remote areas may push up end prices); External influence features: the impact of climate disasters on supply (such as droughts leading to a 10% reduction in wheat production, which will push up prices), and the stimulation of policies on demand (such as the Rural Revitalization Consumption Voucher driving the sales growth of local specialty agricultural products).

[0030] III. Market prediction reasoning through multi-model fusion; Based on the preprocessed feature data, generate market trend predictions through multi-level model reasoning: 1. Basic trend prediction: Use LSTM time series neural networks to model historical price and sales data for a single crop and predict the basic trend for the next 3-6 months. For example, input the monthly sales data of Xinjiang grapes over the past 5 years, and output a preliminary prediction that sales in August-October 2025 will increase by 15% year-on-year.

[0031] 2、Multi-factor synergy analysis: Introduce models such as random forest and gradient boosting tree (GBDT) to analyze the comprehensive impact of multiple characteristics on the market: For example, use winter temperatures in North China, citrus production in major producing areas, and e-commerce platform promotions as input variables to predict the terminal price fluctuations of citrus (such as a sudden drop in temperature leading to increased transportation costs, which may cause prices to rise by 8%); adjust the prediction results by combining policy factors (such as organic certification subsidies), for example, when there is a policy benefit, the market acceptance of organic vegetables increases, and the premium space is expanded from 10% to 15%.

[0032] 3、Old farmer experience and knowledge base correction: Call regional market rules in the old farmer experience library (such as the price of leafy vegetables must rise after the Meiyu season in the south) and crop substitution effects in the agricultural knowledge base (such as when tomato prices are too high, consumers will turn to alternatives such as eggplants and cucumbers), and adjust the model prediction results. For example, the model initially predicts a 10% increase in tomato sales in the summer of 2025, combined with the old farmer experience that the rainy summer increases the tomato transportation loss rate by 20%, and finally the sales forecast is adjusted to 5%.

[0033] Four, prediction result subdivision and dynamic optimization; The intelligent agent disassembles the prediction results in multiple dimensions and dynamically updates them according to real-time data to ensure accuracy: 1、Market segmentation prediction; Split by region: For example, predict an 8% increase in apple demand in Shandong, while demand in Gansu increases by only 3% due to increased local production capacity; Split by consumer groups: For example, 85-year-old families have a 40% higher willingness to pay for fully visualized crop adoption than other groups, and it is recommended to increase the supply of such products.

[0034] 2、Real-time data dynamic adjustment; Update market data every 72 hours (such as real-time monitoring of platform order growth and third-party platform price changes), and if a prediction deviation is found (such as actual sales being more than 5% lower than predicted), automatically trigger model retraining and adjust prediction parameters. For example, when it is detected that transportation in Xinjiang is blocked due to a sudden epidemic, the sales forecast for the next two weeks is immediately reduced (from 10% to 3%).

[0035] Five, production adjustment scheme generation and output; Based on the prediction results, the intelligent agent generates specific production adjustment recommendations for farmers to ensure that production and sales are matched: 1、Plant polarization optimization; Variety adjustment: For example, predict a 25% increase in sales of specialty citrus (such as Rubi citrus) in 2025, and recommend that farmers reduce the planting area of ordinary citrus by 10% and plant Rubi citrus instead; Listing Time Adjustment: If it is predicted that cherries will reach their highest price in mid-May, and considering the growth cycle of the crop, it is recommended to plant 7 days earlier to ensure that the product is ready for the market at the peak price.

[0036] 2. Sales Strategy Suggestions; Pricing Guidance: If the supply-demand gap for strawberries in the winter season in North China reaches 30%, it is recommended that farmers increase the wholesale price by 15% while reserving 20% of their yield for adoption services (with a higher premium). Channel Adaptation: If it is predicted that community group buying platforms will see an increase in demand for small-packaged vegetables, it is recommended that farmers package their vegetables in 500g / portion and prioritize the community group buying channel.

[0037] 3. Risk Warning and Response; In response to potential risks predicted (such as typhoons that may cause a reduction in banana production in September, leading to increased price fluctuations), emergency solutions are proposed: such as recommending farmers to purchase agricultural insurance in advance, or signing a guaranteed purchase agreement with processing enterprises to avoid losses from price drops.

[0038] Through the above steps, the market prediction of the intelligent agent not only realizes the basic judgment of sales-price, but also considers multiple factors such as crop growth cycle, regional characteristics, and policy changes, to output production adjustment plans that can be directly implemented, helping farmers avoid overproduction and achieve precise agriculture production by adjusting production based on sales.

[0039] Logistics Scheduling Module: To ensure the freshness of fruits and vegetables, the platform is responsible for efficient logistics scheduling. According to the distribution of orders and the delivery time requirements, it reasonably arranges logistics vehicles and delivery routes. Using intelligent logistics systems, it monitors the location and transportation status of logistics vehicles in real time, ensuring that agricultural products can be delivered to consumers on time and safely. At the same time, it will adjust the logistics plan in a timely manner according to factors such as weather, such as arranging alternative routes or adjusting delivery times in advance when bad weather may affect transportation, to ensure the smooth progress of delivery.

[0040] Product Estimation Module: The platform uses YOLOv8 technology to estimate crop yields. By analyzing images of the growth process of crops, it identifies the number of crops, growth status, and other information, and combines historical data and environmental factors to predict the final yield. This function not only helps farmers understand their planting results in advance and arrange sales plans reasonably, but also provides important reference for the platform to match production and sales and schedule logistics, improving the overall operational efficiency of the platform.

[0041] Growth state evaluation module: the agent of the platform continuously evaluates the growth state of the crops. Considering factors such as environmental data, pest and disease conditions, and growth cycle, it quantitatively analyzes the health status and growth speed of the crops. Through the evaluation results, it discovers problems existing in the growth process of the crops in a timely manner and provides targeted suggestions and solutions for the farmers, ensuring that the crops can grow in the best state and improving the quality and yield of agricultural products.

[0042] Further, the evaluation of the growth state of the crops by the agent is realized through the whole process of multi-source data collection-growth stage positioning-multi-dimensional index quantification-dynamic result output, and the specific steps are as follows: I. Real-time collection and integration of multi-dimensional growth data; The agent collects all the data related to the growth of crops through the hardware device and the platform data interface, providing a basis for evaluation: 1. Environmental factor data collection: relying on a distributed Internet of Things sensor network, real-time acquisition of soil (humidity, pH value, nitrogen, phosphorus, and potassium content, error <1% RH), air (temperature, humidity, CO2 concentration, error <0.5℃), light (photosynthetically active radiation), and weather (wind speed, precipitation) data, updated at a frequency of 1 time / minute to ensure data timeliness. For example, the soil humidity of strawberry seedlings needs to be monitored in real time to maintain 60%-70%, and the air temperature needs to be monitored in the range of 15℃-20℃.

[0043] 2. Collection of crop morphology and physiology data: crop images are collected through high-definition field cameras, and the agent identifies and extracts key morphological indicators: plant height, stem diameter, leaf number / area, fruit number / size (such as tomato fruit diameter during the fruiting period and apple fruit volume during the fruit enlargement period); 3. Growth cycle and management data integration: call the biological characteristics data of crops in the plant details table, including growth cycle (such as 90-120 days from sowing to harvesting of strawberries) and key growth nodes (such as rice tillering period and fruit tree flowering period); at the same time, associate the field management operations recorded by the platform (such as fertilization time, irrigation amount, and pest control measures), analyze the influence of management behavior on growth (such as the change of leaf nitrogen content 7 days after fertilization).

[0044] II. Accurate positioning of growth stage and matching of reference parameters; The agent determines the current stage based on the growth cycle of the crops and real-time data, and matches the standard growth parameters of the stage: 1. Dynamic determination of growth stage; Through dual verification of planting cycle record + image recognition: Basic determination: preliminarily locate the stage according to the sowing time and the inherent growth cycle of the crops (such as entering the jointing stage 30 days after corn sowing); Precise calibration: Use morphological features identified by large models (such as wheat jointing stage plant height ≥ 30 cm, stem node obvious) to correct stage determination, avoid period deviation caused by environmental differences (such as low temperature delays growth, adjust stage positioning through leaf number).

[0045] Combine physiological data recorded in the plant state table, such as leaf color (judged by spectral analysis to determine chlorophyll content), flowering number, fruit setting rate, etc., to quantify crop growth vigor.

[0046] 2. Stage-specific reference parameter call; From the three-dimensional threshold system of crop-growth stage-environment parameters, the standard growth index of the current stage is called: Environmental reference: such as soil moisture 60%-70% and light intensity ≥30000 lux during apple enlargement period; Morphological reference: such as grape coloration period fruit sugar degree should be ≥16°Brix, and single ear fruit grain number ≥40 grains; Physiological reference: such as tomato fruiting period leaf chlorophyll SPAD value should be ≥50 (indicating adequate nutrition).

[0047] Three, multi-dimensional growth state index quantitative evaluation; Intelligent agent compares real-time data with reference parameters from three dimensions of environmental adaptability-morphological development-physiological health, and quantitatively evaluates the growth state: 1. Environmental adaptability evaluation; Compare real-time environmental data with stage reference through rule engine: Single factor evaluation: such as soil moisture less than 10% of the reference value (such as strawberry seedling stage humidity <54%), marked as mild water stress; Air temperature is higher than the upper limit of the reference for 3 hours (such as tomato fruiting period temperature >35℃), marked as high temperature stress risk; Multi-factor synergistic evaluation: use gradient boosting tree (GBDT) algorithm to analyze the interaction, such as high temperature (>30℃) + high humidity (>80%) combination may cause disease, even if single factor is not over standard, still determine as environmental unsuitable.

[0048] 2. Morphological development progress evaluation; Based on morphological indicators extracted by large models, calculate the development coefficient of actual value / reference value: Such as apple mature period actual single fruit weight 220g, reference value 250g, development coefficient =0.88 (judged as slightly lagging development); Calculate the daily average growth of dynamic indicators (such as actual enlargement speed), compare with the standard growth rate (such as grape enlargement period daily average weight gain should be ≥2g), less than 50% of the standard, then marked as growth retardation.

[0049] 3. Physiological health state evaluation; Disease and pest risk: Detecting the presence of disease spots (e.g., rice blast yellow spots) and insect holes (e.g., aphid damage) on leaves / fruits through image recognition technology, combined with environmental data (e.g., high humidity leading to gray mold), to assess the health level; Nutritional status: Determine if there is a deficiency (e.g., yellow leaves may indicate nitrogen deficiency) by analyzing soil nitrogen, phosphorus, and potassium content and leaf spectral data, and give a nutrition adequacy score (0-100 points, <60 points need to be fertilized) based on the "China Main Crop Fertilization Guide".

[0050] Four, comprehensive evaluation result output and dynamic adjustment; Intelligent agent integrates multi-dimensional evaluation results, generates visual reports, and drives management scheme optimization: 1. Quantitative evaluation report generation; Output growth status score (0-100 points + key indicator details: Score dimensions: environmental adaptability (30%), morphological development (40%), and physiological health (30%); Example: Strawberry evaluation during fruiting stage: 82 points - good environmental adaptability (soil moisture 65%), slightly lagging morphological development (average single fruit weight 8% lower than benchmark), no disease and pest risk.

[0051] 2. Abnormal warning and reason tracing; For scores <70 or individual indicators deviating from benchmark values by more than 15%, trigger warning and analyze reasons: For example, tomato plant height development coefficient 0.7, combined with management records to trace whether it is caused by insufficient fertilizer application during seedling stage (less than 30% of the benchmark) or continuous 5-day low light intensity below the requirement; Warning level classification: minor abnormalities (push attention reminders), moderate abnormalities (push adjustment suggestions such as increasing irrigation amount to 1.2 times the benchmark), and severe abnormalities (link to expert Q&A function, respond within 10 minutes).

[0052] 3. Dynamic adaptation of management scheme; Adjust planting scheme based on evaluation results: Growth retardation: increase fertilizer application (e.g., increase nitrogen and potassium fertilizer use by 10%) and extend light supplement time (from 6 hours / day to 8 hours / day); Environmental inadaptation: trigger drip irrigation system in drought, and open sunshade net in high temperature to ensure subsequent growth returns to benchmark track.

[0053] Five, continuous iteration and optimization of evaluation model; Intelligent agent regularly analyzes the correlation between evaluation results and actual growth feedback (e.g., final yield, quality), and updates benchmark parameters and evaluation algorithms through reinforcement learning: If it is found that apples in a certain region grow better at a soil pH of 6.0 than at the baseline value (pH 6.5), the environmental baseline for that area is dynamically revised; For new crops (such as loquat, a less common variety), the evaluation model for similar crops is reused through transfer learning, shortening the adaptation period.

[0054] Through the above steps, the intelligent agent realizes the real-time monitoring-accurate evaluation-dynamic intervention closed loop of the growth state of crops, which not only provides practical management recommendations for farmers, but also provides key basis for platform yield estimation and market prediction.

[0055] Plant adoption module: It builds a bridge for users to experience rural life deeply, allowing users to have their own cloud-based gardens without having to go to the fields. It includes the following: 1. Orchard type and land appearance selection: The platform provides users with a variety of orchard types, whether it's a tropical mango orchard, a litchi orchard, or an apple orchard with a strong fruit aroma, a grape orchard, users can choose at will. At the same time, users can also choose different land appearances according to their personal preferences, whether it's fertile and flat black land or slightly rolling hills, to meet users' different imaginations of the countryside.

[0056] 2. Plant variety selection: It covers various common and special crop varieties, from sweet and delicious strawberries, juicy and full watermelons, to nutrient-rich blueberries, and sweet and clear corn, users can choose according to their own taste preferences and planting interests to create their own personalized orchard.

[0057] 3. Adoption rights purchase: After users select the orchard type, land appearance, and plant variety they like, they can purchase the corresponding adoption rights. The purchase process is simple and convenient, and the platform provides multiple secure payment methods to ensure user financial security. After successful adoption, users become the owners of the cloud-based land.

[0058] 4. Regularly obtain growth status: With advanced Internet of Things technology and intelligent monitoring equipment, the platform collects plant growth data in real time, including temperature, humidity, light intensity, soil fertility, etc. Users can regularly obtain plant growth status reports through mobile or computer terminals, and can also view high-definition pictures and videos to intuitively understand the plant's every stage from seedling to maturity, as if they were in the field personally taking care of it.

[0059] Further, the intelligent agent also analyzes the data to determine whether the environment is suitable for crop growth; it customizes exclusive planting schemes for different crops based on crop variety characteristics, planting region climate and soil conditions, and seasonal changes; it develops precise irrigation and fertilization schemes based on regional differences, water resource status, and soil fertility; and it identifies damaged fruits and provides treatment solutions.

[0060] Further, determining whether the environment is suitable for crop growth through data analysis specifically includes the following: I. Real-time collection of high-precision environmental data; Multi-dimensional data collection terminal deployment: relying on the fusion application of 5G technology and Beidou positioning system, distributed Internet of Things sensor networks are deployed in farmland, including soil sensors (monitoring humidity, pH, EC value), air sensors (monitoring temperature, humidity, CO2 concentration), light sensors (monitoring photosynthetically active radiation), and weather stations (monitoring wind speed, precipitation). The sampling frequency of all sensors is 1 time / minute, and the data collection error is controlled to be <0.5℃ (temperature), <1% RH (humidity), ensuring the high precision of the original data. At the same time, through Beidou positioning, the data space is calibrated at the field level, accurately associating data with specific planting areas, providing spatial coordinate benchmarks for subsequent regional judgments.

[0061] II. Construction of crop growth suitability parameter library; 1. Basic parameter threshold setting; Based on agricultural knowledge base (such as "China's Major Crop Fertilization Guide" "Vegetable Planting Success Experience Promotion Manual"), crop variety characteristics database (covering the growth cycle parameters of 200+ common fruits and vegetables), and old farmer experience library, a three-dimensional threshold system of crop-growth stage-environment parameters is constructed. For example: strawberry seedling stage suitable soil moisture 60%-70%, air temperature 15℃-20℃, light intensity 20000lux-30000lux, lux is the unit of light intensity; tomato fruiting stage suitable soil pH value 6.0-7.0, air humidity 50%-60%.

[0062] 2. Dynamic parameter optimization; Combined with historical planting data (correlation analysis of crop yield and environmental data in different regions in the past 3 years) and machine learning models (random forest algorithm, LSTM time series model), the basic threshold is dynamically corrected, and the environmental trend (such as a 3-hour temperature drop) is analyzed to predict the environmental risk in the next 2 hours. For example, the soil moisture threshold for the same crop in the southern rainy area will be slightly lower than that in the northern arid area, and the model can automatically adapt to regional differences.

[0063] III. Core judgment logic of data analysis; 1. Real-time data preprocessing: the original data collected is cleaned (abnormal values such as jump data caused by sensor failure are removed) and normalized (different unit parameters are converted to standardized values in the range of 0-1) by edge computing nodes, and then uploaded to the cloud model through the 5G network.

[0064] 2. Multi-level suitability judgment: First layer, rule engine quick screening: directly compare real-time data with threshold range in parameter library, if single-dimensional data exceeds threshold (such as soil humidity is too low < 50%), it is immediately marked as mild abnormality.

[0065] Second layer, multi-factor collaborative analysis model: gradient boosting decision tree (GBDT) algorithm is used to analyze the interaction of multiple parameters (such as high temperature + high humidity may cause diseases), for example, when air temperature > 30℃ and humidity > 80%, even if a single parameter does not exceed the threshold, the model will determine it as "potential risk environment".

[0066] Third layer, growth stage adaptation adjustment: combined with the current growth stage of crops (confirmed by planting cycle record and image recognition, such as seedling stage / flowering stage / fruiting stage), the corresponding stage-specific parameter threshold is called to avoid misjudgment caused by general threshold (such as the demand for water in the wheat filling stage is higher than that in the seedling stage).

[0067] Four, abnormality early warning and dynamic feedback mechanism; 1. Early warning level division; According to the degree of data deviation, the early warning level is automatically divided: slight deviation (such as humidity is less than the suitable value within 5%): push attention reminder, suggest close monitoring; moderate deviation (such as temperature exceeds the suitable range 3-5℃): push adjustment suggestion, such as turning on sunshade net or irrigation system; serious deviation (such as soil pH < 4.5, which may cause root necrosis): trigger "emergency warning", contact farmers and technical experts at the same time.

[0068] 2. Model self-optimization and iteration; The system regularly compares the actual crop growth feedback after early warning (such as yield, disease incidence) with the model judgment result, updates the parameter library threshold and multi-factor weight through reinforcement learning algorithm, so that the judgment accuracy continuously improves with data accumulation.

[0069] Through the above content, the data analysis model realizes the whole process automation from data collection-intelligent judgment-precise early warning, which not only ensures the real-time judgment, but also ensures the accuracy of suitability judgment through the deep integration of crop specificity and environmental complexity, providing farmers with direct land-based environmental adjustment basis.

[0070] Further, considering the crop variety characteristics, planting area climate and soil conditions, and seasonal change factors, specific planting schemes are customized for different crops, including the following contents: One, multi-dimensional basic data collection and integration; 1. Construction of crop variety characteristic database: based on the plant details table, the core biological characteristic data of crops are integrated, including: (1) Growth cycle parameters (such as strawberry from sowing to harvesting about 90-120 days, cherry needs 3-5 years to bear fruit); (2) Water requirement parameters (such as the water requirement of rice is higher than that of wheat, and the water requirement of wheat is higher than that of corn); (2) Environmental sensitivity threshold (e.g. suitable day-night temperature difference for Hanyuan cherries is ≥10℃, Xinjiang grapes require annual sunshine ≥2800 hours); (3) Resistance characteristics (e.g. some citrus varieties are highly resistant to cold, can tolerate -5℃ low temperature); (4) Cultivation characteristics (e.g. tomatoes require trellis cultivation, planting density is related to row spacing).

[0071] 2. Real-time collection of regional climate and soil data: relying on 5G+Beidou fusion technology, collecting basic environmental data in the planting area, including the following data: (1) Soil data: obtain pH value (e.g. southern red soil pH is about 4.5-5.5), organic matter content, nitrogen, phosphorus and potassium levels through soil sensors; (2) Topographic features: combine Beidou positioning altitude and slope information (e.g. consider drainage and adjust planting density for mountain orchards).

[0072] 3. Dynamic factors of seasonal changes are included: divide the season module according to the phenological period (e.g. divide the dry season / rainy season in southern China, and divide the spring / summer / autumn / winter in northern China), correlate the typical climate parameters of each season (e.g. spring cold wave frequency, summer typhoon path), and map to the environmental requirements of key nodes of crop growth (e.g. wheat green-up period, fruit tree flowering period).

[0073] II. Scheme generation of intelligent decision-making model; 1. Basic scheme framework generation: based on rule engine and federated learning model, taking crop variety characteristics as the core, matching regional environmental baseline, including the following contents: (1) Sowing time: combine crop germination temperature threshold and regional spring temperature recovery curve (e.g. corn needs to be sown when soil temperature is stable ≥10℃), calculate the best window period (error ≤3 days); (2) Planting density: calculate according to soil fertility grade (e.g. high fertility land can be planted densely), crop plant type (e.g. dwarf crops have 20% higher density than tall crops), and row spacing formula (plant spacing = mature plant width × 0.8).

[0074] (3) Integrate old farmer experience library (e.g. shallow water at three-leaf stage of rice to promote tillering) and knowledge base (《China's Major Crop Fertilization Guide》) to generate basic processes such as irrigation period (e.g. sandy soil irrigation frequency is 1.5 times higher than clay soil), fertilizer type (e.g. increase phosphorus and potassium fertilizer during flowering period).

[0075] 2. Regional adaptability optimization: through the AI model deployed locally, the basic scheme is regionally corrected, such as: (1) In the arid areas of northwest China, automatically adjust the irrigation scheme to drip irrigation + mulching to conserve soil moisture, and increase the soil moisture monitoring frequency; (2) In the southern acid red soil area, when planting alkali-loving crops (such as cotton), the scheme automatically adds a pretreatment step of applying lime to adjust the pH value to 6.5-7.0.

[0076] 3. Seasonal risk dynamic avoidance: Combined with seasonal climate risk database (such as spring cold in North China and plum rain season in Yangtze River Basin), preventive measures are embedded in the scheme, including: (1) If there is a low temperature risk during the planting period, automatically delay the sowing time by 3-5 days, or recommend using mulch to keep warm; (2) Before the rainy season, for crops that fear flooding (such as peppers), the scheme adds field management items such as raising the ridge height to ≥30 cm and digging drainage ditches.

[0077] Three, dynamic adjustment mechanism of the whole cycle of the scheme; 1. Real-time adaptation of growth stage: Based on the plant state table record of crop growth stage (such as seedling stage / flowering stage / fruiting stage), the agent updates the management scheme every 72 hours, such as: (1) Automatically reduce the fertilizer concentration during the seedling stage (by 50% compared to the adult stage) to avoid burning seedlings; (2) According to the fruit number identified by the YOLOv8 recognition model, dynamically adjust the amount of topdressing (for every 10% increase in fruiting, increase the amount of nitrogen and potassium fertilizer by 5%).

[0078] 2. Environmental anomaly response adjustment: When the environmental monitoring data (such as soil moisture, air temperature) deviates from the suitable range, the scheme automatically triggers emergency adjustment, such as: (1) If the light is insufficient for 3 consecutive days (<60% of the crop's demand), add a light supplement duration of 8 hours / day to the scheme; (2) If the soil EC value is too high (≥2.5 mS / cm), immediately push the modification instructions to suspend fertilization and flood irrigation to wash salt.

[0079] 3. User interaction optimization channel: Farmers provide feedback on actual planting conditions (such as plot area, existing types of agricultural machinery), and the agent further optimizes the scheme based on the feedback, such as: (1) If the farmer does not have drip irrigation equipment, automatically adjust the irrigation scheme to furrow irrigation + timely soil loosening to conserve soil moisture; (2) When planting in small plots, replace the mechanized operation steps (such as unmanned aerial vehicle plant protection) with specific instructions for manual operation (such as evenly spraying the front and back of the leaves when using a manual sprayer).

[0080] 4. Field management process automation: Break down the scheme into executable task chains and automatically push them to the farmer's end app.

[0081] Through the above mechanism, the planting scheme generated by the intelligent agent not only ensures scientific and normative (based on authoritative knowledge base and data model), but also realizes the full-dimensional adaptation of variety-region-season-farmer conditions. The final output scheme contains quantifiable indicators (such as 2.5 kg of seeds per mu, 40 cm of plant spacing, and 60 cm of row spacing) and operation steps that can be directly executed, and farmers do not need professional knowledge to implement them.

[0082] Further, according to regional differences, combined with water resources status and soil fertility, precise irrigation and fertilization schemes are developed, including the following: I. Multi-dimensional basic data collection and regional characteristic modeling; 1. Regional differentiated data collection: (1) Soil fertility data: Real-time collection of soil nitrogen, phosphorus, and potassium content (accuracy ±5 mg / kg), organic matter ratio, and pH value (error ≤0.1) through soil sensors (deployment density up to 1 per 5 mu), combined with regional soil survey data (such as organic matter content generally >3% in Northeast black soil area and <1% in South red soil area), to establish regional baseline of soil fertility.

[0083] (2) Water resources data: Integration of regional water bureau data (such as annual precipitation, groundwater depth, and irrigation water source type) and real-time monitoring (such as river / reservoir water level and water pipeline flow), division of water resource abundance grades (such as moderate water shortage in North China Plain and abundant water in Jiangnan), and association of irrigation costs (such as energy consumption cost of extracting groundwater and loss rate of long-distance water transport).

[0084] (3) Crop water and fertilizer demand characteristics: Based on biological parameters of crop varieties in the plant details table, such as wheat jointing stage nitrogen demand (about 3-5 kg / mu) and grape swelling fruit stage potassium demand (about 8-10 kg / mu), as well as drought tolerance and poor tolerance characteristics (such as millet being 40% more drought tolerant than corn), a crop demand database is established.

[0085] II. Precise irrigation scheme generation; 1. Dynamic water demand calculation model: The intelligent agent calculates crop water demand in real time based on the Penman-Monteith formula (reference "Irrigation and Drainage Engineering") combined with the following parameters: (1) Basic parameters: Crop coefficient (such as 1.1-1.3 for rice and 0.7-0.9 for cotton) and evaporation and transpiration (derived from air temperature and humidity and light intensity); (2) Regional correction: Water resource abundance grade (automatic reduction of 20% irrigation amount threshold in water shortage areas) and soil water retention capacity (higher water replenishment frequency for sandy soil). For example, when planting corn in the arid regions of Northwest China, the model will lower the soil moisture lower threshold from the conventional 60% to 50%, and preferentially use drip irrigation (40% water saving compared to flood irrigation).

[0086] 2. Irrigation strategy generation: (1) Method adaptation: Select irrigation method according to regional water source type (e.g. use sprinkler irrigation in areas with abundant surface water, use micro-irrigation + mulching in areas with scarce groundwater); (2) Time scheduling: Combine weather forecast (e.g. delay irrigation if it rains in the next 3 days) and crop growth stage (e.g. shallow wet alternation is required during the grain filling stage of wheat, i.e. wet the soil to 70% moisture, then naturally dry to 50% before re-watering); (3) Quantitative execution: Output specific irrigation parameters (e.g. 2L / h・plant for drip irrigation, lasting 1.5 hours, with a 3-day interval), and automatically execute through Internet of Things control irrigation equipment.

[0087] Three, precise fertilization scheme generation; 1. Dynamic calculation of fertilizer demand: Based on the nutrient balance method (reference to "China's main crop fertilization guide"), calculate the amount of fertilizer based on the following factors: (1) Crop demand: Determine the target nutrient uptake according to crop variety (e.g. tomatoes require 1.2 times the amount of potassium as nitrogen during the fruiting stage) and growth stage (nitrogen requirement during the seedling stage accounts for 30% of the entire cycle); (2) Soil supply: Calculate the natural supply ratio using current soil fertility data (e.g. if the soil's available nitrogen content is >80mg / kg, reduce nitrogen fertilizer use by 30%); 2. Fertilization strategy generation: (1) Fertilizer type adaptation: Select according to soil type (e.g. use slow-release fertilizer to reduce leaching loss in sandy soil, use quick-acting fertilizer to improve absorption efficiency in clay soil); (2) Application method: Combine regional agricultural machinery conditions (e.g. use unmanned aerial vehicles for spreading in plain areas, use hole application in mountainous areas), output fertilizer type + amount + time + method combination scheme (e.g. northeast soybean field: diammonium phosphate 15kg / acre (strip application before sowing) + potassium chloride 8kg / acre (flowering stage topdressing)); (3) Environmental constraints: For ecologically sensitive areas (e.g. water source protection areas), automatically reduce nitrogen fertilizer use (reduce non-point source pollution), and recommend adding urease inhibitors (reduce ammonia volatilization by 30%).

[0088] Four, regional adaptation and dynamic optimization mechanism; 1. Dynamic correction of regional differences: Divide the country into 6 major agricultural ecological zones (e.g. Huang-Huai-Hai Plain, Yangtze River Basin), and deploy localized models in each region. For example: (1) Rainy areas in the south: Automatically add "rain after application" logic to the fertilization scheme (to avoid nutrient leaching loss); (2) Arid areas in the northwest: Embed "water and fertilizer integration" strategy in the irrigation scheme (fertilizer is applied with irrigation water, improving absorption efficiency by 20%).

[0089] 2. Real-time data feedback iteration: The agent collects soil data (fertility, humidity) every 3 days from IoT devices, combines it with crop growth status (e.g. leaf nitrogen content through spectral monitoring), and compares the actual values with the expected values of the plan: (1) If the soil nitrogen content is 10% lower than expected, automatically add 5% nitrogen fertilizer; (2) If the soil humidity decreases faster than the model predicts 3 days after irrigation, increase the next irrigation volume by 15% (adapt to changes in soil water retention capacity).

[0090] Through the above mechanism, the precise irrigation scheme can increase water resource utilization rate by 35%-50% (compared with traditional flooding irrigation), and the fertilization scheme can reduce fertilizer waste by more than 25%. At the same time, through regional characteristics adaptation, it avoids the mismatch of resources caused by one-size-fits-all (such as reducing irrigation energy consumption in southern regions, reducing the risk of soil compaction caused by excessive fertilization in northern regions).

[0091] Further, the pest and disease identification module specifically includes the following content: I. Multi-dimensional construction of pest and disease feature database; In order to support accurate identification, the agent first constructs a feature database covering common crop pests and diseases, with data sources including: (1) Integration of authoritative agricultural knowledge base: Collecting pest and disease characteristics from documents such as "Crop Pest and Disease Professionalized Unified Prevention and Control Manual" and "China Agricultural Encyclopedia Crop Volume", such as the typical characteristics of rice blast disease (yellow oval lesion + gray mold layer) and aphid colony damage + leaf curl, and labeling the host crops (such as wheat scab, tomato late blight), high-risk seasons, damage sites (leaves / fruits / roots), and morphological parameters (lesion size, insect size).

[0092] (2) Field image sample collection: Collect pest and disease images in different regions and different growth stages in cooperation with agricultural cooperatives, covering high-resolution close-ups (if real surface lesion texture), environment-related images (such as mold growth scenes in humid environments), and shooting conditions (lighting, crop variety, growth period).

[0093] (3) Digitalization of old farmer experience: Record farmers' identification experience of regional characteristic pests and diseases (such as water-stained spots of citrus anthracnose in the Meiyu season in southern China) through structured interviews, and convert them into supplementary feature labels in the database (such as a 2-fold increase in lesion expansion speed when humidity > 85%).

[0094] II. Training and deployment of pest and disease recognition model based on deep learning; An improved ResNet-50 model focused on pest and disease recognition is used, with the following specific implementation: 1. Model architecture customization and optimization; (1) Feature enhancement mechanism: Add CBAM attention module to ResNet-50 to strengthen the extraction of key features such as lesion edges and insect outlines, and solve the interference problem of complex background (branch and leaf shading, fruit overlap) in the field.

[0095] (2) Lightweight design: Use depth separable convolution to compress 40% of the model parameters, adapt to low-power devices such as farmers' mobile phones and field intelligent terminals, and process a single image in less than 0.8 seconds, meeting real-time recognition needs.

[0096] 2. Sample training and precision guarantee; (1) Data augmentation (rotation, brightness adjustment, local occlusion simulation) is performed on the image to expand the sample. Training is classified by disease and pest type + crop variety, and the sample proportion of high-incidence diseases (such as strawberry gray mold) is increased (to 20%).

[0097] (2) Confusion matrix is used to correct similar disease and pest misjudgments (such as distinguishing between tomato early blight and late blight), ultimately improving model recognition accuracy and reducing misidentification rate.

[0098] 3. Local deployment and privacy protection; The model is deployed on the farmer's edge device (intelligent monitoring terminal, mobile phone APP). The fruit image uploaded by the farmer is recognized locally without uploading to the cloud, reducing network dependence and protecting data privacy.

[0099] Three, intelligent agent disease and pest identification and prevention decision; 1. Image acquisition and preprocessing; Farmers upload fruit disease site images (supporting automatic focusing and light supplementing for auxiliary shooting) through the platform, and the intelligent agent performs the following preprocessing on the images: (1) Crop redundant background, keep disease area proportion ≥70%; (2) Enhance feature contrast (such as increase color difference between diseased areas and normal skin), highlight key features (such as mold layer and insect texture).

[0100] 2. Dimension identification and cross-validation; (1) Model preliminary identification: output disease and pest type and confidence (such as grape downy mildew).

[0101] (2) Intelligent agent cross-validation: confirm the identification result combined with the following three types of data: ① "Plant State Table" (current growth stage, such as grape coloring period for high incidence of downy mildew); ② Environmental monitoring data (such as air humidity > 85% in the past 3 days, meeting the disease conditions); ③ Pest and disease characteristic database (matching the grayish-white mold layer of downy mildew + lesion characteristics on the back of leaves).

[0102] 3. Generate targeted prevention and control plans; The agent outputs a quantitative solution from the pest and disease control measures association database (integrated from the authoritative knowledge base), which specifically includes the following: (1) Chemical control: Specify the type of pesticide (e.g., dimethomorph is recommended for downy mildew), concentration (1000 times diluted), application method (foliar spray, with emphasis on spraying the back of the fruit), and safe interval (14 days), referring to the "New Technologies for Pollution-Free Vegetable Cultivation and Pest Control."

[0103] (2) Physical / biological control: Provide alternative solutions to meet the needs of green planting (e.g., using yellow boards to trap aphids and releasing ladybugs as natural enemies, with 20 yellow boards placed per acre).

[0104] (3) Emergency treatment: If the damage level reaches a serious level (the proportion of diseased fruits is greater than 30%), the instructions for removing the diseased fruits and disinfecting the entire garden will be pushed simultaneously, and the expert Q&A function will be linked to respond to farmers' inquiries within 10 minutes.

[0105] 4. Dynamic optimization mechanism; The system collects farmer feedback every month (such as the effectiveness of the prevention and control plan) and optimizes the model and plan based on the actual results in the field: (1) For pests and diseases with an identification accuracy rate of less than 80%, additional samples are added to retrain the model; (2) Calculate the efficacy of the prevention plan (e.g., zinc thiazole is 86% effective in preventing ulcer disease), optimize the priority of the plan, and ensure that the recommended measures are efficient and feasible.

[0106] Through the above process, the intelligent agent realizes the closed loop of image collection-accurate recognition-solution push-effect feedback, solving the pain points of farmers' difficulty in identifying diseases and blindly using medicines. The current average recognition time is less than 12 seconds, and the diseased fruit loss rate is reduced by 32% after the prevention and control plan is adopted.

[0107] Furthermore, damaged fruits are identified and treatment plans are given, including the following: 1. Construction of damaged fruit characteristics database; 1. Multi-dimensional damage characteristics are included; Integrating relevant standards for agricultural product quality and safety and field practice data, we have built a database covering the characteristics of damaged fruits for over 60 crops: (1) Classification of injury types: Identify the following three core injury characteristics: ① Disease damage (e.g. lesion shape: round / irregular, color: brown / black, degree of decay: surface / deep); ② Insect damage (such as the number of insect holes, insect feces residue, and the area of ​​damaged flesh); ③ Physical damage (e.g. compression deformation rate, crack length, mechanical scratch depth).

[0108] Crop-specific annotations: Record the damage tolerance characteristics of different crops (e.g. strawberry skin thickness 0.1mm, slight compression leads to juice leakage; apple skin wax layer is thick and can tolerate light impact), mark the variety-specific damage threshold (e.g. citrus skin damage diameter <3mm is considered light damage).

[0109] 2. Sample data collection and annotation; Collect damaged fruit images from agricultural product cooperatives (covering different damage stages, lighting conditions, and shooting angles), and annotate the following information: (1) Damage parameters (e.g. lesion area ratio 15%, compression deformation rate 20%); (2) Related information (crop variety, harvesting time, storage environment); (3) Treatment result labels (e.g. light damage - marketable, severe damage - processing).

[0110] The dataset is divided into training and validation sets in the ratio of 8:2, providing the basis for model training.

[0111] II. Training and deployment of damaged fruit recognition model; Use an improved MobileNetV3 deep learning model that focuses on damaged fruit recognition scenarios.

[0112] 1. Model architecture optimization; (1) Lightweight design: Use depthwise separable convolution to reduce model parameters by 35%, adapt to low-power devices such as farmers' mobile phones and field intelligent terminals, single image processing time <1 second, meet real-time recognition needs.

[0113] (2) Feature enhancement mechanism: Add coordinate attention modules to the model's Neck layer (neck layer) to enhance feature extraction of damage area edges (such as crack outlines and lesion boundaries), solving the problem of recognition ambiguity caused by similar fruit and background colors (such as light scratches on yellow lemons).

[0114] 2. Sample training and precision improvement; (1) Data augmentation (brightness adjustment, rotation, local occlusion simulation) is performed on the images to expand the sample set and avoid model overfitting.

[0115] (2) Train by "crop variety + damage type" classification, increase sample weight (occupancy ratio up to 25%) for high-value crops (such as cherries and blueberries), optimize the ability to distinguish similar damage through confusion matrix (such as distinguishing between mechanical damage and anthracnose lesions on apples), and finally achieve an accuracy rate of 92% and a damage degree determination error of <5%.

[0116] 3, Local deployment; Deploy the model on the farmer's edge device (such as an intelligent monitoring terminal), and complete the identification locally after image acquisition, without the need to upload to the cloud, ensuring real-time response in weak network environments.

[0117] Three, damaged fruit recognition by intelligent agent; 1, Image acquisition and preprocessing; (1) Farmers upload fruit images through the platform APP (supporting automatic focus guidance to ensure clear imaging of damaged areas), or use field high-definition cameras to collect fruit growth images regularly.

[0118] (2) Intelligent agent pre-processes images: crop redundant background (retain fruit area ratio ≥80%), enhance damage feature contrast (such as increase the brightness difference between scratch area and normal skin).

[0119] 2, Multi-dimensional recognition and cross-validation; (1) Model preliminary identification: deep learning model outputs damage results (such as "strawberry-physical damage-crack length 3mm-damage ratio 8%").

[0120] (2) Intelligent agent cross-validation: combined with "plant state table" (such as the damage proneness of mature fruit), environmental data (such as temperature fluctuations during transportation may cause frostbite), and historical treatment records (such as common damage types of this variety of fruit), the identification results are re-verified to reduce the misjudgment rate (misrecognition rate <4%).

[0121] Four, generation and execution of treatment plan; 1, Hierarchical processing rule engine; Intelligent agent automatically matches treatment strategies based on damage degree, and rules come from platform supply chain data and "Agricultural Product Processing Technical Specifications": (1) Mild damage (such as skin scratches <5mm, disease spot area <10%): determined as edible grade, push priority local real-time sales plan, suggest labeling blemished fruit and discount 10%-15%.

[0122] (2) Moderate damage (such as local rot <30%, extrusion deformation rate 15%-30%): determined as processing grade, interface with jam and dried fruit processing plants, generate 4-hour cold chain transportation instructions, and specify processing pretreatment requirements (such as use after removing rot parts).

[0123] (3) Severe damage (such as whole fruit rot, insect damage area >50%): determined as harmless treatment grade, suggest deep burial and decomposition (depth ≥50cm), and trace the damage source (such as prompt to check if the storage environment humidity is over standard).

[0124] 2. Full-process tracking feedback; The agent tracks the processing progress (e.g., the raw materials have been delivered to the factory, and the defective fruits have been sold) through IoT devices and collects downstream feedback (e.g., the utilization rate of raw materials by the processing plant and the evaluation of defective fruits by consumers), and updates the processing rule weights monthly (e.g., increasing the priority of "lightly damaged fruits" for community group purchase adaptation), continuously optimizing the feasibility of the solution.

[0125] Through the above mechanism, the agent realizes the closed-loop management of damaged fruits from identification, grading, processing, and tracing, effectively reducing the loss rate of agricultural products, while ensuring the rational use of resources through grading and processing, ultimately achieving the dual improvement of agricultural product quality and economic benefits.

[0126] Further, the product estimation module specifically includes the following: I. Crop growth image data collection and labeling; 1. Multi-scene image collection; Relying on high-definition cameras and unmanned aerial inspection systems deployed in the field, images are collected regularly according to the crop growth cycle (seedling stage, flowering stage, fruiting stage, and mature stage), covering different light conditions (sunny, cloudy, and evening), different planting densities, and different varieties (such as strawberries, citrus, and grapes). The collection frequency is dynamically adjusted according to the growth stage: once every 3 days before the fruiting stage, and once a day from the fruiting stage to the mature stage, ensuring the capture of dynamic changes in fruit quantity and size.

[0127] 2. Detailed data labeling; The collected images are manually labeled and automatically verified, including the following: (1) Labeling content includes fruit quantity, single fruit diameter (pixel level), and maturity (distinguished by color characteristics, such as green unripe and red ripe); (2) Combined with Beidou positioning information, associate images with specific planting plots, and label plot area, crop variety (such as "Hanyuan cherries" and "Xinjiang grapes"), and other metadata; (3) Build a dataset of labeled samples, with 80% used for model training and 20% for verification.

[0128] II. Customized training and optimization of YOLOv8 model; 1. Model structure optimization; Lightweight and precision optimization of YOLOv8 for agricultural scene characteristics: (1) Backbone (backbone network) replacement: use MobileNetV3 as the feature extraction backbone network, reduce model parameters (compress by 40%), adapt to low-power terminals such as farmers' mobile phones and field edge devices, and ensure real-time processing (single image recognition time <0.3 seconds); (2) Attention mechanism introduction: Add SE (Squeeze-and-Excitation) attention channels to the Neck layer to enhance the extraction of key features such as fruit outline and color, and improve recognition accuracy in complex field environments. (3) Hyperparameter optimization: Optimize parameters such as learning rate (initially set to 0.01, decayed according to the cosine annealing strategy) and IoU threshold (set to 0.65) through grid search to reduce the miss rate of small fruits (such as cherries).

[0129] 2. Transfer learning and scene adaptation First, pre-train the model on a general object detection dataset (such as COCO), and then fine-tune it with a project-specific crop fruit dataset, focusing on optimizing the recognition ability of "overlapping fruits" and "occluded fruits" (such as fruits in grape clusters that are occluded by leaves). For different crop characteristics (such as strawberry's creeping growth vs. apple's hanging growth), train sub-models for each crop (such as strawberry-specific model and citrus-specific model) to further improve recognition accuracy.

[0130] Three, yield estimation calculation 1. Fruit feature extraction per plant / region After the model performs inference and recognition on the input image, the following key data is output: (1) Fruit number: Count the total number of fruits recognized in the image (including mature and immature, distinguished by annotation); (2) Fruit size: Convert pixel size to actual distance (based on camera focal length and shooting distance calibration) to get single fruit diameter / volume (such as "apple average diameter 8.5 cm"); (3) Fruit distribution density: Calculate the number of fruits per unit area (such as "35 strawberries per square meter") based on image coverage area.

[0131] 2. Yield conversion formula Combine crop variety characteristics and planting parameters to calculate the estimated total yield using the following formula: Estimated total yield = fruit number per unit area × total planting area × average single fruit weight × maturity rate correction factor.

[0132] (1) Average single fruit weight: Based on crop variety database (such as "Red Fuji apple average single fruit weight 250g", "Jufeng grape single fruit weight 10g"); (2) Maturity rate correction factor: Adjust dynamically according to the current growth stage (such as setting the maturity rate to 80% for the 7 days before maturity) and historical maturity data (such as the land's past year maturity rate of 92%); (3) Total planting area: the actual planting area of the plot obtained through Beidou positioning.

[0133] 3. Error correction mechanism; Introduce environmental factors (such as light intensity, soil fertility) to correct the results: for example, if it is monitored that the soil nitrogen content in a certain area is lower than the standard value of 15%, the single fruit weight is calculated at 90%, and the final estimated error is controlled within 5% (in line with the performance index of "estimated error <5%" in the document).

[0134] Through the above process, the YOLOv8 algorithm realizes the accurate conversion from "image recognition" to "yield quantification", providing scientific basis for farmers that can be directly used for production planning (such as arranging picking manpower in advance, connecting with sales channels), effectively solving the problem of large error in traditional yield estimation relying on experience.

[0135] The above is only the preferred embodiment of the present application, it should be understood that the present application is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and improvements, and can be within the scope of the concept described herein, by the above teaching or related art or knowledge to make changes. The changes and variations made by those skilled in the art do not deviate from the spirit and scope of the present application, and should be within the scope of protection of the appended claims of the present application.

Claims

1. A smart agricultural management system based on the entire chain of planting, harvesting and distribution, characterized by: The system includes: an online seed selection module, an order tracking module, an environmental monitoring and early warning module, a pest and disease identification module, a market forecast module, a logistics scheduling module, a product estimation module and a growth status assessment module; The online seed selection module is configured to allow consumers to select crops according to their preferences and needs, and transmit the seed selection information to the corresponding farmers to start exclusive planting services; The order tracking module is configured so that once the farmer starts to terminate the crops selected by the consumer, the consumer can track the status of the order in real time through the system; The environmental monitoring and early warning module is configured to monitor and analyze environmental data of farmland, including soil moisture, temperature, pH, air temperature, humidity, and light intensity. If any abnormal environmental data is detected, an early warning message will be immediately sent to farmers, and countermeasures will be provided. The pest and disease identification module is configured as an intelligent agent to quickly and accurately identify the types of pests and diseases using image recognition and deep learning technology, and provide prevention and control plans; The market forecasting module is configured to analyze the supply and demand of different fruit and vegetable varieties in the market and price fluctuation trend information through big data analysis and artificial intelligence. The intelligent agent predicts the trend of the agricultural product market, adjusts the planting plan, and selects varieties with high market demand and high prices for planting; The logistics scheduling module is configured to arrange logistics vehicles and delivery routes based on the distribution and delivery time requirements of orders, monitor the location and transportation status of logistics vehicles in real time, and adjust the logistics plan in a timely manner according to external factors to ensure the smooth progress of delivery; The product estimation module is configured with a YOLOv8 model to analyze images of crop growth, identify the quantity and growth status of crops, and predict the final yield by combining historical data and environmental factors. The growth status assessment module is configured to integrate environmental data, pest and disease conditions, and growth cycles. The intelligent agent conducts quantitative analysis of the health status and growth rate of crops, continuously assesses the growth status of crops, and provides suggestions and solutions to farmers.

2. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 1 is characterized by: The system also includes: a knowledge popularization module, an intelligent planting guidance module and a plant adoption module; The knowledge popularization module is configured to provide consumers with agricultural knowledge, including the nutritional value, storage methods, cooking suggestions of different fruits and vegetables, the planting process of crops, and the prevention and control of pests and diseases; The intelligent planting guidance module is configured as an intelligent agent that uses the collected farmland-level environmental data, combined with the growth patterns of crops and historical planting data, to develop personalized planting plans for farmers; The plant adoption module is configured to allow consumers to select land of different appearances and plant varieties grown on the land according to their personal preferences, and purchase corresponding adoption rights.

3. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 1 is characterized in that: The agent also uses data analysis to determine whether the environment is suitable for crop growth. It also customizes planting plans for different crops based on the characteristics of the crop variety, the climate and soil conditions of the planting area, and seasonal changes. Based on regional differences, combined with water resource status and soil fertility, precise irrigation and fertilization plans are formulated; damaged fruits are identified and treatment plans are given.

4. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 3 is characterized by: The determination of whether the environment is suitable for crop growth through data analysis specifically includes the following: Real-time environmental data collection: A distributed IoT sensor network is deployed in farmland, including soil sensors that monitor and collect soil temperature, humidity, pH, and EC values; air sensors that monitor and collect air temperature and CO2 concentration; light sensors that monitor and collect photosynthetically active radiation; and weather stations that monitor and collect wind speed and precipitation. At the same time, Beidou positioning is used to achieve field-level data spatial calibration, linking data with specific planting areas and providing a spatial coordinate benchmark for subsequent regional judgments. Construction of a crop growth suitability parameter library: Based on the agricultural knowledge base, crop variety characteristic database, and veteran farmer experience database, a three-dimensional threshold system for crop-growth stage-environmental parameters is constructed. Historical planting data and machine learning models are combined to dynamically adjust basic thresholds, analyze environmental trends, and predict environmental risks within a set timeframe. Judgment of the data analysis model: The collected raw data is preprocessed and a multi-level suitability judgment algorithm is introduced for judgment. The warning level is automatically divided according to the degree of data deviation. The actual crop growth feedback after the warning is regularly compared with the judgment result. The parameter library threshold and multi-factor weight are updated through the reinforcement learning algorithm to improve the judgment accuracy.

5. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 4 is characterized in that: The multi-level suitability judgment algorithm includes: First layer: rule engine rapid screening: Directly compares real-time data with the threshold range in the parameter library. If a single dimension of data exceeds the threshold, it is immediately marked as an anomaly. The second layer, multi-factor collaborative analysis model: uses the gradient boosting tree algorithm to analyze the interaction of multiple parameters. If any parameter in the multi-parameter interaction exceeds the threshold, the model will determine it as a potential risk environment; The third layer, growth stage adaptation and adjustment: Confirm the current growth stage of the crop through planting cycle records and image recognition, and adjust the parameter threshold of the current stage to prevent misjudgment caused by universal parameter thresholds.

6. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 3 is characterized by: The above-mentioned comprehensive consideration of crop variety characteristics, climate and soil conditions in the planting area, and seasonal changes in order to customize exclusive planting plans for different crops specifically includes the following: Multi-dimensional basic data collection and integration: Build a crop variety characteristics database and collect regional climate and soil data in real time. Divide seasonal modules according to phenological periods, associate typical climate parameters for each season, and map them to environmental requirements during key crop growth stages. Intelligent decision-making model solution generation: The optimal sowing window is calculated by combining the crop germination temperature threshold and the regional spring temperature recovery curve; planting density is calculated based on soil fertility level, crop type, and row spacing formula; a basic solution for irrigation cycles and fertilization types is generated by integrating the experience and knowledge base of experienced farmers; the generated basic solution is regionally modified based on regional characteristics, and preventive measures are embedded in the solution by combining it with the seasonal climate risk database; Dynamic adjustment mechanism for the entire plan cycle: Based on the recorded crop growth stages, the intelligent agent updates the plan at set intervals. When environmental monitoring data deviates from the set range, the emergency adjustment mechanism of the plan is automatically triggered. The intelligent agent optimizes the plan based on farmers' feedback data and breaks down the plan into executable task chains, which are automatically pushed to the farmer-side app.

7. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 3 is characterized by: The above-mentioned precise irrigation and fertilization plans are formulated based on regional differences, combined with water resource status and soil fertility, and specifically include the following: Multi-dimensional basic data collection and regional characteristic modeling: collect soil fertility data and combine it with regional soil survey data to establish a regional soil fertility baseline; collect water resource data, classify water resource abundance levels and associate them with irrigation costs; and establish a crop demand database; Precision irrigation plan generation: Calculate crop water requirements in real time based on crop coefficients, evapotranspiration, water resource abundance levels, and soil water retention capacity; select irrigation methods based on regional water source types, and optimize irrigation schedules based on weather forecasts and crop growth stages; output specific irrigation parameters and automatically execute them through IoT-controlled irrigation equipment; Precision fertilization plan generation: Determine target nutrient uptake based on crop variety and growth stage, and calculate the natural supply ratio based on current soil fertility data; select fertilizer type based on soil type, choose application method based on regional agricultural machinery conditions, and establish environmental constraints; Regional adaptation and dynamic optimization mechanism: The country is divided into six major agricultural ecological regions, and each region is dynamically adjusted according to regional differences. The intelligent agent collects soil data every N days based on IoT devices, compares the actual value with the expected value of the plan based on the crop growth status, and optimizes the fertilization plan based on the comparison results.

8. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 1 is characterized by: The pest identification module specifically includes the following contents: Multi-dimensional construction of pest and disease characteristic database: constructing a crop pest and disease characteristic database through agricultural knowledge base, field image sample collection and digitization of experienced farmers' experience; Pest and disease identification model construction and training: The CBAM attention module is added to the ResNet-50 model to enhance the extraction of key features of lesion edges and insect body contours. Depthwise separable convolution is used to compress the pest and disease identification model parameters by 40%. Data augmentation is performed on images to expand the sample size. Classification training is conducted based on pest and disease type and crop variety. The sample proportion for highly prevalent pests and diseases is increased. A confusion matrix is ​​used to correct misidentifications of similar pests and diseases to improve the recognition accuracy of the pest and disease identification model. The trained pest and disease identification model is deployed on edge devices at the farmer's end. Intelligent agent's identification and control of pests and diseases: Farmers upload images of diseased parts of fruits through the system. The intelligent agent pre-processes the images and performs multi-dimensional identification and cross-validation. It then generates control plans based on chemical control, physical / biological control, and emergency treatment. Finally, it optimizes the control plans based on farmer feedback data and actual field results.

9. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 3 is characterized by: The identification of damaged fruits and the provision of treatment plans specifically include the following: Construction of a damaged fruit feature database: Build a multi-dimensional crop damaged fruit feature database, clarify the classification of disease damage, insect damage, and physical damage, and annotate the collected damaged fruit images; Training and deployment of a damaged fruit recognition model: A coordinate attention module was added to the Neck layer of the MobileNetV3 deep learning model to enhance feature extraction of damaged area edges. Depthwise separable convolution was used to reduce the damaged fruit recognition model parameters by 35% to adapt it to farmers' mobile phones and smart field devices. Image data augmentation was performed to expand the sample size, and classification training was conducted based on crop variety and damage type. Sample weights were increased for specific crops, and the ability to distinguish between similar damage patterns was optimized using a confusion matrix. The trained damaged fruit recognition model was deployed on farmers' edge devices. Damaged fruit recognition by an intelligent agent: Farmers upload images of damaged fruits, and the intelligent agent pre-processes the images and then performs multi-dimensional recognition and cross-validation; Treatment solution generation: The intelligent agent automatically matches treatment measurements based on the damage level, tracks treatment progress through IoT devices, and collects downstream feedback data to update treatment rule weights to optimize the solution.

10. The smart agricultural management system based on the planting-harvesting-distribution full-link according to claim 1, characterized in that: The product estimation module specifically includes the following contents: Crop growth image data collection and annotation: collect images from multiple scenes, and perform manual annotation and automatic verification on the collected images; YOLOv8 model training and optimization: SE attention was added to the Neck layer of the YOLOv8 model to enhance the extraction of key features of fruit outline and color, reduce model parameters, and adapt to farmers' mobile phones and field edge devices. Hyperparameter tuning was also performed. The YOLOv8 model was pre-trained on a general object detection dataset and then adjusted using a crop fruit dataset to optimize the recognition of overlapping and occluded fruit. Sub-models were also trained for different crop characteristics. Yield estimation: The YOLOv8 model extracts features from the input image and outputs data on fruit number, fruit size, and fruit distribution density. Combined with crop variety characteristics and planting parameters, the estimated total yield is calculated using the formula: estimated total yield = number of fruits per unit area × total planting area × average weight of a single fruit × maturity rate correction coefficient. Environmental factors are then introduced to correct the results.

Images