feat: automatic field boundary detection (v0.4.0)

- FTW deep learning model for field boundary detection from Sentinel-2 imagery
- Dedicated ML processor Docker service (PyTorch + torchgeo + FTW weights)
- Detection API: trigger, list, accept, discard with org-scoped access
- Full-page detection UI (/farms/[id]/detect) with 3-phase workflow
- Interactive polygon drawing via MapLibre GL Draw
- Boundary review with confidence scores, bulk actions, zoom-to-select
- Boundary editing before accepting as field
- 7-step job progress tracking
- Alembic migrations for detected_boundaries, nullable job field_id
- i18n translations (en + es) for all detection strings
- Updated CHANGELOG, ROADMAP, README

Author: superzero11

CHANGELOG.md

@@ -6,7 +6,25 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+---
+
+## [0.4.0] - 2026-03-17
+
 ### Added
+- **Automatic field boundary detection** from Sentinel-2 satellite imagery using the FTW (Fields of The World) deep learning model.
+- Dedicated ML processor Docker service (`ml-processor`) with PyTorch, torchgeo, and FTW model weights.
+- Detection API endpoints: trigger detection, list/accept/discard detected boundaries.
+- Full-page detection UI (`/farms/[id]/detect`) with 3-phase workflow: draw area → detecting → review results.
+- Interactive polygon drawing via MapLibre GL Draw (pan, zoom, draw, move, edit vertices, delete).
+- Viewport preservation across detection phase transitions (draw → detecting → review).
+- Bbox overlay shown during detection processing phase.
+- Boundary review sidebar with confidence scores, area display, and accept/discard/edit actions.
+- Bulk accept-all and discard-all for detected boundaries.
+- Zoom-to-boundary on selection from sidebar or map click.
+- Boundary geometry editing with DrawMap before accepting as a field.
+- 7-step job progress tracking for detection pipeline (validate → STAC search → download → prepare → inference → polygonize → store).
+- Alembic migrations for detected boundaries table, nullable job field_id, and updated_at column.
+- English and Spanish translations for all detection UI strings.
 - Changelog page visible in-app under sidebar navigation.
 - SAVI L factor displayed on layer cards in field detail sidebar.
 

README.md

@@ -37,6 +37,7 @@ Open source self-hostable and reproducible Crop Intelligence Platform
 ## Why OpenFarm
 - Self-hostable stack with clear service boundaries (Next.js ↔ FastAPI ↔ TiTiler ↔ MinIO ↔ PostGIS)
 - Multi-index vegetation monitoring — NDVI, EVI, SAVI (configurable L factor), and NDWI from Sentinel-2 imagery
+- ML-powered automatic field boundary detection (FTW model) with interactive review workflow
 - Reproducible pipeline with provenance (Element84 STAC → COG → TiTiler tiles)
 - Tenant isolation via `X-Org-Id` + JWT; RBAC (`owner`/`admin`/`member`/`viewer`)
 - MapLibre + PMTiles (no Mapbox token needed), ECharts for time series
@@ -144,6 +145,7 @@ ruff format --check .
 - **Orgs & RBAC**: owner/admin/member/viewer with audit logging
 - **Farms & Fields**: draw/upload GeoJSON/KML, area calc, soft delete
 - **Vegetation Monitoring**: NDVI, EVI, SAVI (configurable L factor), NDWI — STAC search → COG → TiTiler tiles → time-series stats
+- **Boundary Detection**: automatic field boundary detection from Sentinel-2 imagery using FTW deep learning model — draw area, review results, accept as fields
 - **Per-Index Alerts**: configurable threshold and drop-percentage rules for each index
 - **Scouting**: geotagged observations with optional photo upload
 - **Sharing**: read-only field health reports via share links with multi-index toggle

ROADMAP.md

@@ -8,7 +8,7 @@ This document outlines where OpenFarm is today and where it's headed. If you'd l
 
 ## Current Status
 
-OpenFarm **Phase 2 (Multi-Index) is complete**. The platform delivers end-to-end satellite-powered crop intelligence with four vegetation indices (NDVI, EVI, SAVI, NDWI), auth, org management, farm/field CRUD, configurable monitoring pipelines, per-index alerts, scouting observations, shareable field health reports, and production-grade security hardening — all functional and deployed. The focus now shifts to testing, documentation, and expanding the platform with new data sources and intelligence capabilities. See [Future Ideas](#future-ideas-post-mvp) for what's next.
+OpenFarm **Phase 3 (Boundary Detection) is complete**. The platform delivers end-to-end satellite-powered crop intelligence with four vegetation indices (NDVI, EVI, SAVI, NDWI), ML-powered automatic field boundary detection, auth, org management, farm/field CRUD, configurable monitoring pipelines, per-index alerts, scouting observations, shareable field health reports, and production-grade security hardening — all functional and deployed. The focus now shifts to testing, documentation, and expanding the platform with new data sources and intelligence capabilities. See [Future Ideas](#future-ideas-post-mvp) for what's next.
 
 ---
 
@@ -98,6 +98,20 @@ OpenFarm **Phase 2 (Multi-Index) is complete**. The platform delivers end-to-end
 - [x] SAVI L factor stored in layer `params_json` and displayed in UI
 - [x] In-app changelog page with parsed Keep a Changelog rendering
 
+## Milestone 7 — Automatic Boundary Detection ✅
+
+- [x] FTW (Fields of The World) deep learning model integration for field boundary detection
+- [x] Dedicated ML processor Docker service with PyTorch, torchgeo, and model weights
+- [x] Detection API — trigger, list, accept, discard, with org-scoped access control
+- [x] Full-page detection UI (`/farms/[id]/detect`) with draw → detecting → review workflow
+- [x] Interactive polygon drawing with MapLibre GL Draw (draw, move, edit vertices, delete)
+- [x] Viewport preservation across phase transitions
+- [x] Boundary review with confidence scores, bulk accept/discard, zoom-to-boundary
+- [x] Boundary geometry editing before accepting as a field
+- [x] 7-step progress tracking (validate → STAC search → download → prepare → inference → polygonize → store)
+- [x] Alembic migrations for detected boundaries, nullable job field_id, updated_at
+- [x] i18n translations (English + Spanish) for all detection UI
+
 ---
 
 ## Future Ideas (Post-MVP)
@@ -107,14 +121,31 @@ These are under consideration but not yet committed. Grouped by theme and roughl
 ### Platform Foundations
 - **Email/Microsoft & Enterprise SSO** — support email/password, Microsoft OAuth, and SAML/OIDC for enterprise identity providers
 - **Direct API integration** — stable, versioned public API with API keys for integrating OpenFarm into existing farm management software
-- **Boundary detection** — automatic field boundary detection from satellite imagery
-- **Crop detection and classification** — ML-based crop type identification from spectral data
 - **Multi-satellite support** — Landsat, Planet (currently Sentinel-2 only)
+- **Higher-frequency monitoring** — support for daily revisit satellites (e.g., PlanetScope) for near real-time crop monitoring
+- **Higher-res imagery** — support for sub-meter commercial imagery for detailed crop monitoring
+- **Custom index builder** — UI for users to define custom indices from available bands with formula editor and visualization
+- **Data export** — export field data, stats, and reports in CSV, GeoJSON, PDF formats
+- **User roles and permissions** — more granular permissions (e.g., field-level access, read-only API keys) and user groups
+
 
 ### Agricultural Intelligence
+- **Weather data integration** — overlay forecasts and historical weather on field maps and incorporate into alert rules
+- **Crop detection and classification** — ML-based crop type identification from spectral data
+- **tree detection and classification** — ML-based tree crop identification and health monitoring
+- **Phenology tracking** — track crop growth stages and phenological events from satellite data
+- **tree canopy analysis** — canopy cover, leaf area index, and tree height estimation from high-res imagery
+- **drought monitoring** — integrate drought indices and soil moisture data for water stress assessment
+- **water stress monitoring** — integrate NDWI and soil moisture data for irrigation management
+- **soil moisture estimation** — integrate soil moisture data from remote sensing and in-situ sensors
 - **Disease/pest risk signals** — risk scoring framework combining vegetation anomalies, weather, and regional pest data
+- **nutrient deficiency detection** — identify spectral signatures of common nutrient deficiencies for early intervention
+- **Fertilizer and irrigation recommendations** — actionable insights based on crop health trends, weather forecasts, and agronomic models
+- **Anomaly detection** — unsupervised ML to identify unusual patterns in field health data that may indicate emerging issues 
+- **Historical data backfill** — backfill historical vegetation index data for existing fields to enable trend analysis from day one
 - **Yield analysis and forecasting** — predict yield from historical NDVI trends, weather, and field data
-- **Weather data integration** — overlay forecasts and historical weather on field maps
+- **Harvest timing recommendations** — optimal harvest windows based on crop maturity models and vegetation indices
+- **Climate impact modeling** — estimate carbon sequestration, emissions, and climate impact of farming practices using field data and agronomic models
 - **Carbon/sustainability reporting** — track and report carbon sequestration, emissions, and sustainability metrics
 
 ### Analytics & Workflows
@@ -125,11 +156,13 @@ These are under consideration but not yet committed. Grouped by theme and roughl
 - **Webhook/notification system** — email, Slack, or SMS on alerts
 
 ### Ecosystem & Integrations
-- **OpenFarm MCP server** — Model Context Protocol server for AI agent integration
+- **AI agent integration** — connect to LLMs for natural language insights, recommendations, and conversational interfaces  
+- **Model Context Protocol (MCP) server** — standardized interface for AI agents to query field data and trigger analysis
 - **Device/Sensor plugin framework** — connect soil sensors, weather stations, and IoT devices
 - **Machinery telemetry integration** — ingest GPS tracks and operational data from farm equipment
 - **Supply chain / traceability integrations** — link field data to downstream logistics and compliance systems
 - **Plugin system** — extensible processing pipelines for custom analysis
+- **Community data sharing** — opt-in anonymized data sharing for regional insights and benchmarking
 
 ### Enterprise & Scale
 - **Enterprise admin controls** — SSO enforcement, audit policies, usage quotas, multi-tenant admin

apps/web/messages/en.json

@@ -409,5 +409,50 @@
         "title": "Changelog",
         "description": "Latest updates and improvements to OpenFarm",
         "unreleased": "Unreleased"
+    },
+    "detection": {
+        "title": "Detect Boundaries",
+        "description": "Draw a rectangle on the map to detect field boundaries using satellite imagery.",
+        "drawBbox": "Draw area on map",
+        "drawBboxHint": "Click to place points, double-click to finish. You can move or edit the shape after drawing.",
+        "bboxArea": "Area: {area} km²",
+        "bboxTooLarge": "Area exceeds 50 km² limit",
+        "bboxTooSmall": "Area too small — draw at least 2 km²",
+        "startDetection": "Start Detection",
+        "detecting": "Detecting...",
+        "farmLabel": "Associate with farm",
+        "farmPlaceholder": "Select a farm",
+        "windowA": "Window A (planting)",
+        "windowB": "Window B (harvest)",
+        "optional": "Optional",
+        "jobCreated": "Detection job started",
+        "jobFailed": "Detection failed",
+        "jobComplete": "Detection complete — {count} boundaries found",
+        "noBoundaries": "No boundaries detected in this area",
+        "pending": "Pending",
+        "accepted": "Accepted",
+        "discarded": "Discarded",
+        "confidence": "Confidence: {value}%",
+        "areaHa": "{area} ha",
+        "detectedOn": "Detected {date}",
+        "accept": "Accept",
+        "acceptAs": "Accept as Field",
+        "discard": "Discard",
+        "edit": "Edit & Accept",
+        "fieldName": "Field name",
+        "fieldNamePlaceholder": "Enter field name",
+        "acceptSuccess": "Field created ({area} ha)",
+        "discardSuccess": "Boundary discarded",
+        "acceptAll": "Accept All",
+        "discardAll": "Discard All",
+        "progress": {
+            "validate": "Validating area",
+            "stac_search": "Searching satellite imagery",
+            "download_bands": "Downloading bands",
+            "prepare_input": "Preparing input",
+            "inference": "Running detection model",
+            "polygonize": "Converting to polygons",
+            "store_results": "Storing results"
+        }
     }
 }

apps/web/messages/es.json

@@ -409,5 +409,50 @@
         "title": "Registro de cambios",
         "description": "Últimas actualizaciones y mejoras de OpenFarm",
         "unreleased": "Próximamente"
+    },
+    "detection": {
+        "title": "Detectar Límites",
+        "description": "Dibuja un rectángulo en el mapa para detectar los límites de campos usando imágenes satelitales.",
+        "drawBbox": "Dibujar área en el mapa",
+        "drawBboxHint": "Haz clic para colocar puntos, doble clic para terminar. Puedes mover o editar la forma después de dibujar.",
+        "bboxArea": "Área: {area} km²",
+        "bboxTooLarge": "El área excede el límite de 50 km²",
+        "bboxTooSmall": "Área muy pequeña — dibuja al menos 2 km²",
+        "startDetection": "Iniciar Detección",
+        "detecting": "Detectando...",
+        "farmLabel": "Asociar con finca",
+        "farmPlaceholder": "Seleccionar una finca",
+        "windowA": "Ventana A (siembra)",
+        "windowB": "Ventana B (cosecha)",
+        "optional": "Opcional",
+        "jobCreated": "Trabajo de detección iniciado",
+        "jobFailed": "La detección falló",
+        "jobComplete": "Detección completa — {count} límites encontrados",
+        "noBoundaries": "No se detectaron límites en esta área",
+        "pending": "Pendiente",
+        "accepted": "Aceptado",
+        "discarded": "Descartado",
+        "confidence": "Confianza: {value}%",
+        "areaHa": "{area} ha",
+        "detectedOn": "Detectado {date}",
+        "accept": "Aceptar",
+        "acceptAs": "Aceptar como Campo",
+        "discard": "Descartar",
+        "edit": "Editar y Aceptar",
+        "fieldName": "Nombre del campo",
+        "fieldNamePlaceholder": "Ingresa el nombre del campo",
+        "acceptSuccess": "Campo creado ({area} ha)",
+        "discardSuccess": "Límite descartado",
+        "acceptAll": "Aceptar Todos",
+        "discardAll": "Descartar Todos",
+        "progress": {
+            "validate": "Validando área",
+            "stac_search": "Buscando imágenes satelitales",
+            "download_bands": "Descargando bandas",
+            "prepare_input": "Preparando entrada",
+            "inference": "Ejecutando modelo de detección",
+            "polygonize": "Convirtiendo a polígonos",
+            "store_results": "Almacenando resultados"
+        }
     }
 }