AI-Powered Precision Agriculture

Plant Intelligence
for the Future of Farming

Real-time health monitoring, yield forecasting, and closed-loop environment control for containerised controlled-environment agriculture — powered by deep learning.

Request a Demo Explore the Technology
Scroll
<15ms
End-to-end inference
8
Health classes
72h
Sensor history window
4
Simultaneous predictions
About ArcaNova

Precision intelligence
for every grow cycle

We build the AI brain behind next-generation containerised farms — fusing computer vision, environmental sensor data, and growth-stage awareness into one actionable intelligence layer.

ArcaNova's Plant Intelligence System transforms how controlled-environment agriculture (CEA) farms operate. Our multi-modal deep learning platform continuously monitors every plant, forecasts yield, detects nutrient deficiencies, and drives automated environment control — all in under 15 milliseconds.

Founded to close the gap between precision agriculture research and deployable farm technology, ArcaNova brings PhD-level plant science into a production-ready SaaS platform that connects directly to your sensors, actuators, and edge hardware.

Our flagship crop is Wasabi (Wasabia japonica) — among the most demanding and valuable crops for precision cultivation — but the platform is fully method-agnostic, supporting hydroponic, aeroponic, and soil-based systems.

# ArcaNova FusionPlantModel v2.0

inputs:
  ├─ img: (B, 3, 380, 380)  # RGB camera
  ├─ sensor_seq: (B, 288, 8) # 72h history
  └─ growth_stage: (B,)   # stage 0–7

outputs:
  ├─ health: (B, 8)       # classification
  ├─ nutrient: (B, 5)     # multi-label
  ├─ stage: (B, 8)       # classification
  └─ yield_g: (B, 1)     # regression

inference: <15ms   # TensorRT INT8
hardware:  "Jetson Orin Nano 8GB"
Technology

Three streams.
One fusion model.

FusionPlantModel simultaneously processes visual, environmental, and phenological signals — catching stress before it becomes visible.

🔍
Vision Encoder
EfficientNet-B4 → (B, 512)
Processes 380×380 RGB frames from inside the grow chamber. EfficientNet-B4 backbone via Noisy Student pretraining, projected to 512 dimensions. Detects leaf colour, lesions, morphology, and canopy density.
📡
Temporal Encoder
Transformer ×4 → (B, 256)
Encodes 72 hours of 8-channel sensor data at 15-minute intervals (288 timesteps) using a 4-layer pre-norm Transformer. Captures diurnal cycles, slow nutrient drift, and early stress before it appears visually.
🌱
Stage Embedding
Embedding(8, 64) · LayerNorm
Conditions every prediction on the current growth stage — germination through post-harvest. Stage-awareness ensures health thresholds, setpoints, and yield targets are always calibrated to the plant's actual developmental phase.
Fusion MLP · cat([512, 256, 64]) → 512
🏥
Health Class
8-class · softmax
Alert & intervention trigger
🧪
Nutrient Flags
5-label · sigmoid
Dosing system adjustment
📊
Growth Stage
8-class · softmax
Setpoint schedule advancement
📈
Yield Forecast
Regression · Softplus
Harvest planning & dashboard

Health Classification — 8 Classes

Full plant health spectrum, from silent early stress to active disease

Healthy
All sensors in spec, no visual symptoms
Early Stress
Sensor deviation only, no visible signs yet
Nitrogen Deficiency
Yellowing old leaves, low NO₃, EC falling
PK Deficiency
Purple/brown margins, necrotic tips
Mg/S Deficiency
Interveinal chlorosis, young leaves affected
Fungal Disease
Spore patches, powdery lesions
Other Disease
Bacterial, pest, viral mosaic
Env. Stress
Heat/cold scorch, light burn
SaaS Platform

Everything your farm
needs, in one place

API-first, multi-tenant architecture with real-time telemetry, automated alerting, and a full fleet management dashboard.

📡
Real-Time Monitoring
WebSocket telemetry updated every 30 seconds. Temperature, humidity, pH, EC, PAR, CO₂ and dissolved oxygen — all on one live dashboard per container.
🤖
Automated Environment Control
Closed-loop PID control for all environment variables. Stage-aware setpoints advance automatically as plants develop through the crop protocol.
🚨
Intelligent Alerts
Multi-channel notifications via SMS, email, and push. Custom rules per container and per stage. Severity scoring eliminates alert fatigue.
🏭
Fleet Management
Manage hundreds of containers from a single tenant view. Geographic overview, health summary, OTA firmware updates, and per-container drill-down.
📉
Yield Analytics
30-day yield forecasting per container with confidence intervals. Historical trends, COGS tracking, and harvest planning with supply-chain export.
🔗
Open API
Full OpenAPI 3.1 surface. Integrate with ERP, POS, and third-party systems. Webhook and WebSocket support for custom automation pipelines.
Supported Crops & Methods

One platform.
Any crop, any method.

The platform is fully method-agnostic. Swap the growing method without changing any hardware — only the crop protocol JSON and ML model update.

Adding a new crop requires 1–2 weeks of software work — a new crop protocol JSON and a fine-tuned model on the new dataset — and zero hardware changes for same-method crops.

Wasabi ★ Lettuce Tomato Strawberry Cannabis Microgreens Herbs Any Brassica
💧
Hydroponic
NFT · DWC · Ebb-Flow
🌫️
Aeroponic
High/low pressure mist
🌍
Soil-Based
Container · Raised bed
Edge AI Deployment

Intelligence at
the container edge

The model runs on-device inside every container — no cloud dependency for real-time control, full offline operation, and sub-15ms closed-loop response.

  • Target hardwareJetson Orin Nano 8 GB
  • GPU1024 CUDA cores (Ampere)
  • RuntimeTensorRT INT8
  • End-to-end latency< 15 ms
  • Control loopevery 30 s
  • Offline operationFull (7-day cache)
  • Local storage500 GB NVMe
  • ConnectivityWiFi 6 · LTE · LoRa
# Export pipeline

PyTorch (best_model.pt)
  → ONNX (opset 17, dynamic batch)
  → TensorRT INT8 engine via trtexec

# Control loop (every 30s)

1. read_sensors() → dict[str, float]
2. append to buffer  (max 288 entries)
3. resolve_setpoints() + interpolate
4. PID.step() × 6   per variable
5. send_command(var, pct) → HW
6. model.predict() → health alert
Get in Touch

Ready to bring
intelligence to your farm?

Whether you're a farm operator, investor, or research institution — we'd love to hear from you.

Based in France  ·  arcanova-france.com  ·  contact@arcanova-france.com