Role-based digital agricultural management system enabling AI-powered plant disease diagnosis, expert-approved electronic prescription flow, and multi-source agricultural data integration.

Abstract

Technical field The invention relates to a computer-based agricultural management system and its methodology that utilizes role-based authorization for the end-to-end digitalization of agricultural production processes, particularly the integration of image-based artificial intelligence for plant disease diagnosis, agricultural engineer approval / prescription workflows, and Internet of Things (IoT) sensors, remote sensing (satellite / drone), weather, and market price data within a single server-client architecture. Technical problem In current agricultural software, disease diagnosis, expert consultation, official prescription issuance, monitoring of sensor and remote sensing data, market price tracking, income and expense planning, and integration with official agricultural systems (ÇKS, BKÜ-VT, B-Prescription, TARBİL, DİTAP, e-Government) are all carried out in separate applications. This fragmented structure leads to: (i) the risk of incorrect pesticide recommendations due to the use of preliminary diagnoses obtained through artificial intelligence without expert verification; (ii) the difficulty of farmers physically accessing experts; (iii) the issuance of prescriptions disconnected from diagnostic data; (iv) the inability to collect field-based historical data (planting, spraying, pests, soil, sensors, NDVI) in a single place; and (v) the inability to maintain traceable records compliant with regulations. Solution The invention is an agricultural management system connected to a common data layer on the server, offering separate interfaces and permission sets for at least three different roles: farmer, agricultural engineer, and manager (municipality / admin). The system solves the aforementioned problems through the integrated operation of the following elements: AI-based image diagnosis module: Analyzes plant photos uploaded by farmers and ranks potential diseases with a confidence score; correlates the result with the relevant field and plant record. Expert approval cycle: The diagnostic result is transmitted to an agricultural engineer for verification via the same record; the engineer confirms, corrects, or adds clarification to the diagnosis; however, only a confirmed diagnosis can trigger the prescription issuance process. Contextual electronic prescription generation module: Based on verified diagnoses, the plant protection product database, and past field application records, the engineer generates an electronic prescription containing information on the pesticide, dosage, number of applications, waiting period, and pre-harvest time; the prescription is then ready to be transferred to the official B-Prescription workflow. Multi-source field data layer: Planting history, spraying records, pest observation, soil analysis, IoT sensor measurements, and NDVI / satellite readings for each field are combined under a single entity; prescription and diagnostic data are linked to this layer. Decision support modules: Weather data, sensor threshold-based alerts, harvest forecasting, and market price monitoring / alerting mechanisms provide data for decisions regarding spraying, irrigation, and sales timing. Income-expense and product recommendation modules: Income estimation based on field size, cultivated crops, and current market prices, along with seed / fertilizer / pesticide expense tracking; a step-by-step product recommendation wizard considering crop rotation and soil suitability. Official integration interface: Abstracted integration layer for data exchange with ÇKS, e-Government, TARBİL, DİTAP, BKÜ-VT and B-Prescription systems; connection status and last synchronization time are monitored from the management panel. Expert-farmer matching and marketplace: Listing, assigning, and messaging engineers by area of ​​expertise and region; a direct, ad-based marketplace between producers and buyers. Thanks to this structure, AI diagnosis, expert verification, and official prescription are generated as a single traceable record chain; diagnosis, prescription, application, and harvest data are integrated under the same field object, ensuring a compliant and auditable data flow.

Description

1 TARIFF AI-powered plant disease diagnosis, expert-approved electronic prescription flow, and much more. Role-Based Digital Agricultural Management Enabling Resource-Based Agricultural Data Integration System 5 1. TECHNICAL FIELD The invention enables the end-to-end digitalization of agricultural production processes, particularly image-based artificial intelligence. intelligence in plant disease diagnosis, agricultural engineer approval / prescription workflow, and the Internet of Things. (IoT) sensors, remote sensing (satellite / drone), weather and market price. role-based 10 regarding the unification of resources into a single server-client architecture. A computer-aided agriculture using authorization and multi-tenant isolation. It is related to the management system (100) and the method of this system. 2. STATE OF KNOWLEDGE OF THE ART Known agricultural software solutions include disease diagnosis, expert consultation, and official prescriptions. regulation, monitoring of sensor / remote sensing data, market price tracking, income-expense 15 planning and official agricultural systems (ÇKS, BKÜ-VT, B-Reçete, TARBİL, DİTAP, e-Government) Integration is carried out in separate applications. Applications that diagnose plant diseases using image-based artificial intelligence are offered by producers. Making inferences from the uploaded photo with a single model results in a low confidence score. It does not include a backup model or verification mechanism for the situations and the preliminary diagnosis obtained is 20 It can directly generate drug recommendations based on the results. This approach carries the risk of misdiagnosis and This leads to the incorrect application of chemical pesticides. Prescription writing processes typically involve separate prescription writing software, disconnected from diagnostic data. or it is operated through a paper-based process. In this case, the engineer takes a photograph of the disease. 25 based on observed symptoms, image quality score, and historical field records. Time access is limited and there is a verifiable trace between the issued prescription and the diagnostic record. It cannot be abandoned. Existing agricultural dashboard applications track planting history, spraying, pest monitoring, soil analysis, and IoT. Sensor readings, satellite / NDVI, weather data, and diagnostic / prescription records are all grouped together in a common field object. It cannot be structurally combined underneath; each data source is kept isolated on its own screen, therefore 30 Decision support modules (harvest forecasting, revenue forecasting, product recommendation, alarm generation) are multi-source. It cannot be fed with data in real time. 2 In terms of integration with official systems (ÇKS, BKÜ-VT, B-Prescription, TARBİL, DİTAP, e-Government) Current solutions use point-specific interfaces for each system; synchronization queue, Common cross-sectional features include token management, retry / withdrawal, and health monitoring. The requirements are repeated in each integration. This situation ensures compliance with legislative changes. 5 delaying and providing service to multiple municipalities / tenants through a single installation It makes things more difficult. 3. PURPOSE OF THE INVENTION AND THE TECHNICAL PROBLEM SOLVED The purpose of the invention is to solve the problems mentioned above by (i) developing artificial intelligence-based pre-diagnosis. A workflow that cannot trigger prescription production without verification by an expert agricultural engineer. 10 (ii) provides automatic backup in low-security situations with a multi-model backup strategy. (iii) switching to alternative models, treating the field as a central data object and at least eight different (iv) a common integration abstraction layer that integrates the data from the source under this object (v) farmers, agricultural engineers and exchange data with all official agricultural systems through Multiple 15-person tenant structure with separate role boards for administrators (municipality / admin). a digital agricultural management system (100) that provides services to the municipality through a single installation to place. 4. EXPLANATION OF THE INVENTION The subject of the invention is the system (100); at least one farmer client (102), at least one agricultural engineer client (104), at least one administrator client (106), at least one field IoT sensor device (108) and optionally at least 20 a REST / HTTP API gateway (200) that is in bidirectional communication with a small receiving client (110) includes. API gateway (200); authentication middleware (202), multi-tenant (municipal) resolution middleware layer (204), role-based authorization layer (206) and multipart installation intermediate It is equipped with layer (208). Each request is; (i) passed through JWT verification, (ii) municipality / tenant 25 (iii) identity is determined, (iv) user is authorized according to their role, and visual data is retrieved as needed. It is loaded. Behind the API gateway (200); identity service (300), field management service (302), diagnostic service (304), prescription service (306), plant catalog service (308), disease catalog service (310), pesticide service (312), sensor service (314), weather service (316), market price service (318), marketplace service 30 (320), community service (322), notification service (324), analytics service (326), product recommendation service (328), management service (330), integration service (332) and drug verification service (334) are included. The data layer consists of a PostgreSQL relational database (600), file store (626), cache (628), and business processes. The queue consists of (630) components. Each service, within the scope of multi-tenant coverage, has an assigned component. 3 It operates on assets (602, 604, 606, 608, 610, 612, 614, 616, 618, 620, 622, 624); each The query is scoped by the municipality ID (624) determined by the middle layer (204). The artificial intelligence inference module (400) within the diagnostic service (304); local convolutional neural network model (402), image preprocessor (404), Top-K softmax classifier (406), backup major language 5 The model includes the visual API (408) and rule-based heuristic fallback (410) components. Which components to use is determined by the confidence score-sensitive model selector (412). is determined. Inference outputs are determined in the relational database via the label-disease analyzer (414). It is mapped to disease entities (612). Prior to inference, the image quality scorer (418) is used; brightness, Using contrast and Laplacian variance-based sharpness metrics, the usability score of the photograph is 10. It produces. The main innovation cycle of the invention is the workflow shown in Figure 3. By the farmer client (102) Artificial intelligence analysis (502) is performed on the uploaded photo (500); the result is a preliminary diagnosis (504). is removed. This preliminary diagnosis is made after the appointment of the agricultural engineer (506) to the engineer examination stage. (508) enters. Only diagnosis approved by the engineer (512), drafting of prescription 15 This is considered the condition that triggers step (514); otherwise, the rejection / correction loop (510) It is activated and, if desired, the photo is refreshed and the model is restarted. Prescription draft (514); pesticide catalog (312 / 614), disease catalog (310 / 612) and BKU-VT (806) Enriched with active ingredient verification (516). Engineer; drug, dose, number of applications, waiting period Fill in the fields for duration and pre-harvest last application time. In the publication step (518); prescription record 20 A QR token (signed by JWT) compatible with B-Prescription (808) is produced. The pharmacist uses this token in the field. Checks via verification service (334) (520); if verification is successful, the medicine is supplied. All The process is linked to the field record (602 / 700) and the farmer archive (522). Field data object (700); planting history (702), pesticide record (704), pest observation (706), soil analysis (708), IoT sensor reading (710), satellite / NDVI / drone image (712), weather (714) and 25 It brings together diagnostic and prescription records (716) resources under a common entity. This integration; harvest forecast (900), revenue forecast (902), product recommendation wizard (908) and alarm engine (912) modules It enables multi-source data streaming. The integration abstraction layer (800) has six internal services (304, 306, 302, 320, 314 etc.). official system adapter (ÇKS 802, e-Government 804, BKÜ-VT 806, B-Prescription 808, TARBİL 810, 30 DITAP acts as a bridge between price flow (812) and the price stream (816). The layer is the token vault (824), sync queue (822, retries + exponential rollback included) and health tracker (826) is equipped with. Integration status logs (622) visible on the admin panel; connected / connected It includes the "not working / problematic" status and the last synchronization time. 4 Product recommendation wizard (908); field selection, soil profile (708), previous season's crop (702), life cycle It consists of seven steps: preference, product type, variety preference, and scoring. Scoring; soil. It considers suitability, crop rotation, climate, and market price together and provides reasoned, sequential product recommendations. It generates suggestions. 5 Marketplace (1000) module; ad creation (1002), direct buyer-seller messaging (1004) and It includes the functions of DİTAP (812) and synchronization (1006). Income calculator (902); field expected gross based on size, cultivated product, harvest estimate (900) and current market price (318, 816) It generates income. Expense panel (904); seed, fertilizer, pesticide, fuel and labor items in separate modules. amount. Profitability indicator (906); estimate of profit / loss for the season based on the difference between income and expenses 10 It visualizes. 5. BRIEF DESCRIPTION OF THE FIGURES Figure 1. Seven-layered system architecture; client devices, API gateway, service modules, data The layer, integration abstraction layer, official adapters, and external APIs are arranged in a vertical stack. is shown. 15 Figure 2. Role-based authentication and routing flow; login → JWT → role / tenant Analysis → three separate role-specific dashboard and module sets. Figure 3. Main innovation cycle of the invention; three-lane (farmer / presenter / engineer) workflow diagram. AI diagnosis → expert approval → electronic prescription generation and pharmacist QR code. It shows the verification in chronological order. 20 Figure 4. Multi-model AI inference pipeline; image quality scoring, preprocessing, confidence. native CNN with score-sensitive model selector, external visual API, and rule-based heuristic backup. The routing logic between them. Figure 5. Hub-and-spoke (star) arrangement with the field data object at the center and eight different data sources. (Around planting, spraying, pests, soil, IoT sensors, satellite / NDVI, weather, diagnosis-prescription); 25 The bottom four decision support consumers. Figure 6. Structure of the integration abstraction layer; token vault, synchronization queue, and A bridge between health monitoring elements, internal services, and six official adapters. Figure 7. Marketplace and revenue-expense flow; two-column layout, cross-data links (current price, Expected harvest, announcement transmission) and DİTAP synchronization. 30 Figure 8. The seven-step flow of the product recommendation wizard; soil, planting history, weather, and price data. Feeding into the scoring step and outputting ranked suggestions. 6. APPLICATION OF THE INVENTION 6.1 — System architecture and multi-tenant structure. The system (100) consists of clients as shown in Figure 1. behind the REST / HTTP API gateway (200) accessed via devices (102, 104, 106, 108, 110), It includes eighteen service modules connected to a common data layer (600, 626, 628, 630). Multi-tenant. Resolution middleware (204); municipality ID in each request from JWT content or subdomain 5 By analyzing the area, it determines the relevant municipal entity (624) and attaches this identity to the request. Each The service provides data access to a scoped repository filtered by municipality ID. It is implemented through the design. This allows for isolated solutions for multiple municipalities on a single installation. Data and configuration are provided. 6.2 — Role-based access. Identity service (300); generates JWT access and renewal tokens. Token content; 10 User ID, role (farmer, engineer, manager, super manager), municipal ID, and optional. Includes SSO resource fields. Role-based authorization layer (206); access of each endpoint According to the role constraint in the matrix, the farmer (102) rejects or passes the request. He / she, according to his / her field and diagnoses; Engineer (104), to the farmers to whom he is assigned; Manager (106), to all within his municipality’s scope They can access the records. Cross-municipal transactions are also defined for super administrators (Figure 2). 15 6.3 — Artificial intelligence pre-diagnosis line. Farmer client (102), multi-part upload of plant photo. It transmits the image to the diagnostic service (304) via the intermediate layer (208). The service stores the image in the file repository (626). It writes to a municipality-based road and then triggers the AI ​​inference module (400). The module first It runs the image quality scorer (418); for photos below the threshold, it re-scores the image quality scorer to the user. A shooting suggestion is generated. Then the image preprocessor (404); 192×192 pixels, mid-cropping and 20 float32 applies normalization. 6.4 — Multi-model backup strategy. Confidence score-sensitive model selector (412); local first. It runs a convolutional neural network model (402, EfficientNetB0 based TFLite). Top-K softmax The highest probability produced by the classifier (406) is the calibrated threshold value (e.g. 0.65) Below, the selector redirects the inference to the backup visual API (408). The backup model also has sufficient confidence. 25 If it does not produce, rule-based heuristic backup (410) seed information consisting of plant name and visual cues. It performs mapping through the set. The outputs of all models are mapped with the confidence limiting module (416). Clamped to the 0.05–0.99 range; overly safe or overly low outputs in the upstream section. This prevents the navigation logic from being disrupted (Figure 4). 6.5 — Label-disease analysis. Label-disease analyzer (414); 30 generated by the local model Tags (e.g., Tomato___Early_blight) are first mapped in the static mapping table, then the disease map. direct matching of the island (310 / 612) in the catalog, finally with token-based fuzzy matching. It solves the problem. In fuzzy matching, it finds a word that corresponds to a plant context (e.g., tomato). In this case, the score is increased by +3 points. Treatment depends on the disease detected as a result of the analysis. The text is returned to the client as part of the preliminary diagnosis result (504). 35 6 6.6 — Expert approval cycle. The farmer, from the pre-diagnostic (504) screen, to the engineer assigned to him. (104) sends a request for approval (506). Engineer review (508); can confirm the diagnosis (512), can correct or reject (510). The system can only create a prescription draft in case of approval. Opens step (514). Rejected or corrected results; label-disease analysis training 5 It can be marked as data. 6.7 — Electronic prescription generation. Prescription draft (514); to the engineer; relevant disease (612), recommended It presents a list of pesticides (614) filtered via N:N target-pest relationship. BKU-VT verification (516); the active ingredient license of the selected pesticide and the legal limits of the recommended dose. Checks its compatibility. At the release step (518); a JWT-signed QR token is generated (valid for 180 days). 10 The token carries the prescription ID and the scope=prescription-verify field. 6.8 — Pharmacist verification. Drug verification service (334); QR token scanned by pharmacist It solves the problem, verifies the signature, checks that the prescription is in the active state, and The answer should include information on the applicable dose, waiting period, and the last application date before harvest. It returns. The verification event is logged to the prescription existence (606). 15 6.9 — Multi-source field integration. On the field asset (602), schematized in Figure 5. Field data object (700) and nine different sources (702–718) through a single read surface Decision support modules are provided; harvest forecast (900), income forecast (902), product recommendation wizard. (908) and alarm engine (912) take simultaneous readings from this surface. For example, the alarm engine; IoT When sensor data (710) falls below the threshold and weather conditions (714) have a low probability of precipitation, 20 It produces irrigation recommendations; harvest planning cannot be done before the pesticide waiting period (614) is over. It creates a warning about this. 6.10 — Integration abstraction layer. As shown in Figure 6, the integration service (332); Six official adapters (802, 804, 806, 808, 810, 812) with all internal services and external streaming. It provides the abstraction layer (800) which acts as a bridge between the providers (816, 818, 820). 25 The layer stores a separate credential for each tenant with a token vault (824), synchronization queue (822) retry failed calls with exponential withdrawal, each with health tracker (826) It records the status of the adapter (entity 622). The administrator can record the status of any adapter. You can monitor its status from the control panel. 6.11 — Marketplace and income / expense flow. As schematically illustrated in Figure 7; producer, advertisement creation 30 (1002) presents its product to the market. Market price service (318); from the wholesale price flow (816) It stores the data under the market price asset (620). The price alarm engine (910) is set by the user. It monitors target price thresholds and generates a notification via the notification distributor (914) when the threshold is met. Income calculator (902); estimate based on field area, harvest estimate (900) and current market price. 7 Calculates income. Expense panel (904); sums the costs entered item by item. Profitability indicator. (906); visualizes the difference between income and expenses both in absolute and percentage terms. 6.12 — Product recommendation wizard. The seven-step flow in Figure 8; field selection, soil profile, previous After selecting the plant season, life cycle preference, product type, and variety preference, the scoring (908g) step begins. 5 Scoring considers soil suitability (e.g., pH and NPK), crop rotation suitability (the effect of the previous crop), regional climate (weather source 714) and current market price (620) factors are weighted It lists the varieties in the plant catalog (610) based on the total and justifies the gross income range. They present together. 6.13 — IoT sensor data acquisition. Field IoT sensor devices (108); sensor service (314) POST 10 Sends data to the / sensors / readings endpoint with a JSON body. Sensor reading presence (616); device The municipality identity includes fields for temperature, air humidity, soil moisture, and timestamp. It is resolved from the device-municipality matching table. Sensor threshold alarm motor (912); predefined It generates notifications when thresholds are exceeded. 6.14 — Offline and low bandwidth support. Local convolutional neural network model (402); TFLite 15 It is approximately 15 MB in size and can generate preliminary diagnoses without requiring an external API. This allows... The system can also work in rural areas where mobile connections are weak; network quality becomes favorable again. When this happens, the job queue (630) transmits the pending confirmation requests and sensor data to the server. 6.15 — Method. The subject of the invention is the method; (a) a photograph of a plant from a farmer client (102) (500), (b) scoring of image quality (418), (c) confidence score sensitive 20 model selector (412) selects local model (402), backup model (408) and intuitive backup respectively. (410) making inferences from (502), (d) the preliminary diagnosis obtained (504) agricultural engineer (104) transmission to the client (506), (e) approval by the engineer (508 → 512) or (510) closing the feedback loop by rejection, (f) only with the condition of approved diagnosis Prescription draft (514) opening, (g) BKU-VT (806) active ingredient and dose verification (516), (h) signed 25 The issuance of electronic prescriptions (518) containing QR tokens, (i) pharmacist QR verification (520) and (j) This includes steps to link the entire record chain to the field asset (602 / 700) and the farmer archive (522). 7. LIST OF REFERENCE NUMBERS (100) Agricultural management system (general) (102) Farmer client (web / mobile interface) 30 (104) Agricultural engineer client (106) Administrator / municipal client (108) Field IoT sensor device (110) Buyer client (marketplace user) (200) REST / HTTP API gateway 35 8 (202) Authentication (JWT) middleware (204) Multi-tenant (municipal) resolution middleware (206) Role-based authorization layer (208) Multipart loading intermediate layer 5 (300) Identity / session service (302) Field management service (304) Diagnostic service (artificial intelligence orchestration) (306) Prescription service (308) Plant catalog service 10 (310) Disease catalog service (312) Plant protection product (pesticide) service (314) Sensor service (IoT data acquisition) (316) Weather service (318) Market price service 15 (320) Marketplace service (322) Community / forum service (324) Notification service (326) Analytical service (harvest forecast) (328) Product recommendation service (wizard) 20 (330) Management service (332) Integration service (334) Drug verification service (B-Prescription QR) (400) Artificial intelligence image extraction module (402) Local convolutional neural network model (EfficientNetB0 based TFLite) 25 (404) Image preprocessor (192×192 pixel center cropping) (406) Top-K softmax classifier (408) Backup image model (major language model visual API) (410) Rule-based heuristic backup (412) Confidence score-sensitive model selector (multiple backup) 30 (414) Label–Disease analyzer (matching + fuzzy matching + plant context) (416) Confidence limiting module (0.05–0.99 clamping) (418) Image quality scorer (brightness / contrast / Laplacian sharpness) (500) Photo upload steps (502) Multi-model artificial intelligence analysis step 35 (504) Preliminary diagnosis result (confidence score + alternatives) (506) Expert assignment / engineer referral (508) Engineer review step (510) Red / correction feedback loop (512) Confirmed diagnostic status (triggering condition) 40 (514) Creating a prescription draft (516) BKU-VT active ingredient verification (518) Electronic prescription issuance (QR token generation) (520) Pharmacist QR verification 9 (522) Field registration and linking to farmer archives (600) Relational database (PostgreSQL) (602) Field aggregate (604) Diagnostic presence (image URL, alternatives, confidence score) 5 (606) Prescription existence (608) User presence (role + municipality / tenant) (610) Existence of plant catalog (612) Existence of disease catalog (614) Pesticide presence and N:N target-pest relationship 10 (616) Sensor time series reading presence (618) Presence of soil analysis (620) Market price existence (622) Integration connection status existence (624) Municipality (tenant) assets 15 (626) File / image repository (628) Cache layer (Redis) (630) Job queue (BullMQ) (700) Field data object (central total) (702) Source of the history of October 20 (704) Source of pesticide record (706) Source of harmful observation (708) Soil analysis source (710) IoT sensor read source (712) Satellite / NDVI / unmanned aerial vehicle image source 25 (714) Weather source (716) Source of diagnosis and prescription records (800) Integration abstraction layer (adapter / connector) (802) CKS (Farmer Registration System) adapter (804) e-Government adapter 30 (806) BKU-VT adapter (808) B-Prescription adapter (810) TARBIL adapter (812) DITAP adapter (816) Market price flow (Izmir / Bursa) 35 (818) External weather API (820) External AI image API (822) Synchronization queue + retry / return (824) Token vault / identity repository (826) Integration health tracker 40 (900) Harvest estimation module (902) Revenue forecast module (904) Cost panel (seed / fertilizer / pesticides / fuel) (906) Profitability calculator (908) Product recommendation wizard (seven steps) (910) Price alarm engine (912) Sensor threshold alarm motor (914) Notification distributor 5 (916) Cooperative product pool / matching (1000) Marketplace module (1002) Ad creation step (1004) Direct buyer-seller messaging (1006) DITAP synchronization 10

Claims

11 REQUESTS 1. An agricultural management system (100); — with a farmer client (102), an agricultural engineer client (104) and an administrator client (106) A bidirectional REST / HTTP API gateway (200); 5 — an authentication middleware (202) that is operated sequentially on the API gateway (200), a multi-tenant resolution middleware (204) and a role-based authorization layer (206); — at least one diagnostic service (304) has a local convolutional neural network model (402), an image preprocessor (404), a Top-K softmax classifier (406), a replacement visual API model (408), a rule-based heuristic backup (410) and a confidence score-sensitive 10 among the mentioned models. an AI inference module (400) containing a model selector (412); — the approval of an agricultural engineer (512) on the diagnosis result is required as a triggering condition and a prescription that only opens the prescription draft creation step (514) when the aforementioned approval is provided service (306); — Performing active ingredient and dose verification (516) via BKU-VT adapter (806) and B-15 A prescription adapter (808) that generates an electronic prescription (518) containing a signed QR token. publishing module; — planting history (702), pesticide application record (704), pest observation (706), soil analysis (708), IoT sensor reading (710), satellite / NDVI / drone image (712), weather (714) and diagnosis and prescription A field management system that integrates records (716) and sources under a common field data object (700) 20 service (302); It is characterized by...

2. According to claim 1, the system is (100) and the model selector (412) is sensitive to the aforementioned confidence score; local The maximum probability produced by the convolutional neural network model (402) is a predetermined caliber. When the value falls below the established threshold, the inference is automatically transferred to the backup visual API model (408) 25 If this model also fails to generate sufficient confidence, then a rule-based heuristic backup is used. (410) It is characterized by having a priority logic that guides.

3. System (100) according to Request 1 or 2, and within the AI ​​inference module (400); input calculates the brightness, contrast, and sharpness metrics of the photograph based on Laplacian variance, and Image quality 30 that triggers a retake recommendation for the user for photos below the threshold. It is characterized by the presence of scorers (418).

4. According to any of the above requirements, the system is (100) and the artificial intelligence inference module (400) a confidence limiting module (416) that clamps its outputs to the range [0.05; 0.99]. It is characterized by its presence. 12 5. System (100) according to any of the above requirements, a label-disease analyzer (414) is characterized by containing; the said analyzer (i) a static mapping table, (ii) direct match to (612) island in the disease catalog and (iii) point increase based on plant context 5 It is structured.

6. According to any of the above requests, the system is (100) and the prescription service is (306). The generated electronic prescription (518) contains the prescription ID and verification scope fields. It includes a JWT signed QR token and this token is verified by a drug verification service (334) to the pharmacist By resolving it in the interface, the current status of the prescription (606) and applicable drug information 10 It is characterized by its reversal.

7. System (100) according to any of the above requirements, multi-tenant resolution interface layer (204); resolves the municipality identity via JWT content or subdomain and the relevant municipality Identifying its existence (624) and accessing data with this identity is a scoped repository (scoped It is characterized by being carried out via a repository. 15 8. According to any of the above requirements, the system is (100) and is an integration abstraction. It is characterized by containing layer (800); the said layer is at least one CKS adapter (802), an e-Government adapter (804), a BKU-VT adapter (806), a B- Prescription adapter (808), a TARBIL adapter (810) and a DITAP adapter (812) It connects to internal services via a common interface. 20 9. According to claim 8, the system is (100) and the integration abstraction layer is (800); tenant-based identity a token vault (824) that stores information, retrying failed calls with exponential rollback a synchronization queue (822) and the status of each adapter in an integration link It is characterized by including a health tracker (826) that updates the status in the presence of (622).

10. According to any of the above requirements, the system is (100) and a product recommendation wizard (908) 25 It is characterized by its inclusion of the wizard, field selection, soil profile (708), previous Seven steps including planted season (702), life cycle preference, crop type, variety preference and scoring. It is operated through; scoring, soil suitability, crop rotation, climate and market price components It is based on the weighted sum principle.

11. According to any of the above requests, the system is (100) and the field data object (700) is equal to 30 timed reading; a harvest forecast module (900), a revenue forecast module (902), an expense panel (904), a profitability indicator (906), a price alarm engine (910) and a sensor threshold alarm It is characterized by containing the engine (912). 13 12. According to any of the above requirements, the system is (100) and a marketplace module is (1000) It is characterized by its inclusion of; the module in question, the advertisement creation step (1002), directly a transaction carried out via buyer-seller messaging channel (1004) and DITAP adapter (812) Includes synchronization (1006) capabilities. 5 13. System (100) according to any of the above requirements, and field IoT sensor devices (108) temperature, air humidity and soil moisture measurements are taken by a sensor service (314) received, stored in the presence of sensor reading (616) and by a sensor threshold alarm motor (912) It is characterized by generating notifications by comparing them to predefined thresholds.

14. A computer-applied method; agricultural management according to any of requests 1–13. The system (100) is operated on and includes the following steps: (a) uploading a plant photograph from the farmer client (102) to the server (500); (b) scoring of the photograph with an image quality scorer (418); (c) a confidence score-sensitive model selector (412), local convolutional neural network model (402), backup 15 Choosing an inference path between the visual API model (408) and rule-based heuristic fallback (410) (502); (d) transmitting the preliminary diagnosis obtained (504) to the agricultural engineer client (104) (506); (e) approval (512) or rejection (510) by the engineer; (f) opening of prescription draft (514) only with the condition of approved diagnosis; (g) BKU-VT (806) active ingredient and dose verification (516); 20 (h) Publication of electronic prescription (518) containing signed QR token; (i) verification of the QR token with the drug verification service (334) in the pharmacist interface (520); (j) linking the entire record chain to the field asset (602 / 700) and farmer archive (522).

15. The method according to claim 14 is; the choice in step (c) is the maximum produced by the local model (402). 25 If the backup model also fails to generate sufficient confidence, switch to intuitive backup (410). It is characterized by its implementation.

16. The method is according to claim 14 or 15; the records obtained at each step are for multiple tenants. a scoped municipality identity (624) resolved with the analysis middle layer (204) It is characterized by being written to the repository. 30 17. The method is according to any of claims 14–16; the binding operation in step (j) is the field data. Adding a new record to the source of diagnostic and prescription records under object (700) (716) and this The record is matched by the harvest forecast (900), income forecast (902) and alarm engine (912) modules. It is characterized by its ability to be read in real time. 14 18. The method according to any of claims 14–17; in a condition where the network connection is interrupted. client photo capture and prediction via local convolutional neural network model (402) The pending data of the job queue (630) can be generated on the side when the network becomes accessible again. It is characterized by transmitting it to the server. 5 19. A computer-readable storage medium on which written commands are stored, in one or more locations. When run by more processors, the method steps of any of prompts 14–18 It is characterized by the commands it executes.

Images