AgriTech (Agriculture) Mobile App Development
Agronomist stands in the middle of a field in Krasnodar — phone catches one bar of signal, GPS jumps 10–15 meters. He needs to record an anomaly on a specific plot, photograph a disease spot, geotag it and send to monitoring system. In 30 minutes he'll be in a no-signal zone. Standard mobile app doesn't work here.
Offline Cartography and Coordinate Accuracy
AgriTech app works where internet is often absent for hours. Fields don't change daily — you can (and should) pre-download raster or vector tiles to the device.
Satellite imagery. For agronomist Sentinel-2 or DigitalGlobe matters more than road maps. Tile server with Earth Engine or Copernicus Open Access Hub data, tiles cached in MBTiles format and displayed via MapLibre GL Native. User downloads tiles for needed region before departure — sees NDVI changes, disease spots, overwet areas.
GPS accuracy. Built-in smartphone GPS gives 3–5 meters in open field, up to 15 meters under clouds. Not enough for precise plot mapping. Solutions:
- Bluetooth GNSS receiver (Trimble R1, Bad Elf GPS+) gives sub-meter accuracy via standard NMEA 0183 protocol
- SBAS (WAAS/EGNOS) correction — free, improves to 1–3 meters
- RTK via mobile internet (NTRIP client) — centimeters, requires constant network
Integration of external GPS on Android: UsbSerialForAndroid for USB-OTG or BluetoothSocket for BLE receivers. On iOS — External Accessory Framework for MFi-certified devices.
Field Tasks and Operations
Agronomist performs tasks in field: plot inspection, soil sampling, irrigation, fertilizer application. This is workflow of geotgged tasks created by agronomist-consultant in web interface, executed by field employee in mobile app.
Task status model: created → assigned → in_progress → completed. On transition to in_progress fix GPS point of start, on completed — point and photo. If user closed task without network — operation goes to queue (Room database, pending_operations table), synchronizes on connection via WorkManager with NetworkConstraint.
Crop and Disease Identification via ML
Plant photo and automatic disease detection — popular AgriTech feature. Implement via Core ML (iOS) or TensorFlow Lite (Android) with model fine-tuned on client's crops. Model packaged in mlmodel/.tflite and lives on device — works offline, critical for field.
Basic open-source models: PlantNet, iNaturalist API, or PlantVillage dataset models (38 disease classes for 14 crops). For production fine-tune on client's regional data via Transfer Learning over MobileNetV3.
Limitation: model recognizes leaf diseases but not micronutrient stress — requires multispectral drone imagery, not phone photo.
Integration with Smart Sensors and Equipment
AgriTech ecosystem includes soil sensors (Davis Instruments, Decagon), weather stations, smart irrigation systems (Netafim, Lindsay) and precision farm machinery (John Deere Operations Center API, AGCO Connect). Agronomist's mobile app — integration point where data from all sources converge in single field context.
For Bluetooth sensors on iOS: CoreBluetooth with CBCentralManager, manufacturer GATT profile. For WiFi/LoRaWAN sensors — data aggregated on server, mobile client requests via REST or WebSocket.
John Deere Operations Center provides OAuth 2.0 REST API with machine, field and operation data. Integration requires developer partnership account.
Field Map and Polygons
Fields drawn as polygons: agronomist walks boundary or draws on map with finger. Polygon input via taps on map — MKPolygon / GmsPolygon with live edge preview. Save as GeoJSON, calculate area via ST_Area(ST_Transform(geom, 32637)) on server (UTM projection for meter accuracy).
| Function | Technology |
|---|---|
| Offline maps | MapLibre GL + MBTiles |
| Satellite imagery | Sentinel Hub API / Earth Engine |
| ML diseases | TFLite / Core ML |
| External GPS | NMEA 0183 / BLE |
| Field tasks | Room + WorkManager |
Development Stages and Timeline
- Audit: what sensors/equipment already used, what GNSS accuracy needed
- Develop offline data model — fields, tasks, seasons
- Map and polygon work
- ML module — dataset preparation or use ready models
- Integration with external systems
- Pilot on one farm before production
MVP (field map, tasks, photo documentation): 8–12 weeks. Full AgriTech platform with ML, machinery and sensor integration: 5–8 months. Cost calculated individually.