01
Edge-native visual intelligence
Hybrid inference with a MobileNetV2 baseline and EfficientFormerV2-S1 advanced path, optimized for Raspberry Pi 5 and offline use.
Baseline + transformer backends
Edge AI for Plant Health
CONNECTED TO NASA-ISRO/SAR
CONNECTED TO OPEN-METEO
Raspberry Pi 5 Ready
DELTA Plant
Edge AI for Plant Health Operations.
Explainable plant health intelligence, SAR-aware monitoring, weather fusion, and conversational workflows for Raspberry Pi 5 edge deployment.
From field capture to fungal-risk screening: transparent, offline-capable, and now paired with a hero demo that explains how a pasted GPS point becomes a 50 meter satellite observation zone.
New hero demo
Paste Google coordinates or a Google Maps URL, and DELTA Plant centers a circular footprint with a 50 m radius around that GPS point. The demo then shows a 7-day NISAR-style SAR view plus weather context and returns a simple fungal-risk outlook for that exact micro-area.
Top-1 accuracy
89.33%
600-image, 33-class public benchmark.
Top-3 accuracy
99.00%
High-confidence shortlisting for fast operator review.
Mean edge latency
41.36 ms
Measured public runtime profile on edge hardware.
Deployment
Raspberry Pi 5
Offline-capable execution with production-oriented TFLite export paths.
Hero demo
Paste Google coordinates. Get a weekly fungal-risk reading.
This public demo now calls the same live NASA-ISRO/SAR dashboard pipeline used in DELTAPLANO: your pasted GPS point becomes a 50 meter circular footprint, DELTA pulls the 7-day Open-Meteo weather window plus the NASA POWER GWETROOT/GWETTOP soil proxy, and returns the same daily breakdown and weekly fungal-risk label shown in Telegram.
The site keeps the same GPS entry surface as the bot, but the output is aligned to the live DELTAPLANO dashboard contract: latest day summary, seven daily rows, and one weekly LOW, MEDIUM, or HIGH risk label.
Acquisition radius
50 m
Same circular micro-footprint used in DELTAPLANO.
Dashboard output
7 daily rows
Latest day summary plus one weekly mean and verbal label, exactly like the Telegram dashboard.
Live sources
2 official feeds
Open-Meteo Archive for weather and NASA POWER GWETROOT/GWETTOP for soil moisture context.
On mobile you can paste either a full Google Maps URL or a simple latitude, longitude pair. The demo is tuned for iPhone 14 Pro, Galaxy S22+, and similar narrow touch screens.
Paste a Google Maps coordinate pair or URL. DELTA will lock the same 50 m radius used by the bot, call the live NASA/SAR dashboard endpoint, and return the same 7-day breakdown used in Telegram.
Scientific reading
Soil proxy input
The live dashboard uses the NASA POWER daily GWETROOT and GWETTOP products, normalized and averaged exactly like the DELTAPLANO Telegram flow.
Weather input
Temperature, humidity, and rainfall come from the same Open-Meteo Archive daily window used by the backend dashboard: daily max or min values are averaged into one row per day before the fungal-risk score is computed.
Output contract
The landing renders the same output contract as Telegram: latest-day header, daily breakdown ordered from newest to oldest, and one weekly LOW, MEDIUM, or HIGH label derived from the 7 daily scores.
NISAR 7-day demonstrator
Footprint 7,854 m²
Satellite focus
The highlighted ring marks the 50 m inspection radius around the selected GPS coordinate.
GPS center
41.90280, 12.49640
DELTA Plant analyzes a circular micro-area with a 50 m radius around this point.
Weekly fungal risk
MEDIUM
Weekly mean risk 52.0%. The live DELTAPLANO dashboard currently classifies this GPS micro-zone as medium fungal pressure.
Latest dashboard day
Today 2026-05-25
Soil 47.0%, temp 24.4 C, humidity 45%, rain 0.0 mm, and risk 44.0% on the latest 7-day dashboard row.
Data sources
Esri World Imagery + NASA POWER + Open-Meteo
The public landing now requests the same live NASA/SAR dashboard used by DELTAPLANO whenever the backend is reachable.
Scientific note: the fungal outlook is not a disease diagnosis. It is a 7-day micro-area favorability estimate derived from the same Open-Meteo plus NASA POWER pipeline used by the DELTAPLANO Telegram dashboard. If the live backend is temporarily unavailable, the page switches to a clearly marked browser fallback that keeps the same window and fungal-risk formula but uses a local soil proxy instead of NASA POWER.
Input contract
The pasted Google coordinate becomes the center of the same 50 m circle used by DELTAPLANO in the NASA-ISRO/SAR Telegram flow.
Data fusion
Open-Meteo Archive provides daily weather, while NASA POWER provides the GWETROOT and GWETTOP soil proxy used by the live Telegram dashboard.
Output alignment
The landing shows the same latest-day summary, daily breakdown columns, and weekly fungal-risk label expected from the DELTAPLANO Telegram dashboard.
Same 50 m GPS circle used in DELTAPLANO.
Daily Open-Meteo plus NASA POWER inputs aligned to the Telegram dashboard.
Latest-day summary and weekly risk label aligned to the bot output.
Audio
Hero soundtrack
Space Epic Cinematic Journey
Muted loop active
Core capabilities
Clear edge AI capabilities for operators, researchers, and partners.
DELTA Plant brings vision, explainability, sensor fusion, and conversational workflows into one system that is easier to deploy, audit, and scale.
Execution model
Offline-first
Raspberry Pi 5 deployment, auditable inference, and operator-friendly runtime paths.
Context channels
7 inputs
Leaf image evidence can be fused with environmental and agronomic context signals.
Operator surface
Telegram
Field workflows, follow-up, and operational handoff remain accessible in a single interface.
02
Explainable diagnosis
LayerCAM heatmaps show where the model looks on the leaf, improving operator confidence and research traceability.
Heatmaps and visual evidence trails
03
DELTAPLANO conversational operations
Run diagnostics, ask follow-up questions, and move through workflows directly in Telegram.
Operator workflow continuity
04
Sensor and rule fusion
Combine leaf images with 7 environmental channels and agronomic rules for contextual recommendations.
Agronomic context inside the model loop
05
Research-to-release pipeline
Training, TFLite export, benchmarking, and documentation stay reproducible and fast to iterate.
Train, export, benchmark, release
06
Investor-ready documentation
Model cards, manuals, benchmarks, and public artifacts support pilots, partnerships, and funding conversations.
Evidence for pilots and due diligence
Technology stack
A practical stack for explainable plant health on the edge.
Production baseline, advanced transformer path, LayerCAM explainability, DELTAPLANO operations, and Raspberry Pi 5 deployment in one auditable architecture.
Vision stack
2 paths
MobileNetV2 baseline plus EfficientFormerV2-S1 advanced backend with export-ready TFLite targets.
Explainability
LayerCAM
Leaf-level visual evidence stays attached to inference so operators can audit where the model focused.
Operator surface
API + Bot
Telegram operations, admin tooling, and API routing keep the deployment readable in the field.
Vision
MobileNetV2 production baseline plus EfficientFormerV2-S1 advanced backend with int8 and float32 TFLite export.
Explainability
LayerCAM heatmaps and overlays show where the model focused on the leaf.
Operations
DELTAPLANO on Telegram, admin tooling, and API endpoints keep the workflow operator-friendly.
Sensors and rules
Seven environmental channels and agronomic rules add context for better decisions.
How it works
Capture, diagnose, and act
Built for agritech pilots
1
Capture
Take leaf photos and optional sensor readings in the field, greenhouse, or lab.
Input
2
Diagnose
Run edge inference with LayerCAM explainability and agronomic context.
AI
3
Act
Receive clear recommendations and continue through DELTAPLANO follow-up.
Ops
Top-1
89.33%
Public benchmark reference on 600 validation images.
Top-3
99.00%
High-confidence support for fast operator review.
Edge latency
41.36 ms
Mean measured edge inference for the public runtime profile.
Resources and next steps
Launch DELTAPLANO, inspect the evidence, and move toward pilots.
Start with the Telegram workflow, inspect the public benchmark, and review the repository, manual, and model card.
Operator entry
Telegram
Start from the bot to test the field workflow exactly where operators will use it.
Evidence pack
Public docs
Benchmark, repository, model card, and manual stay aligned for review and due diligence.
Runtime path
Preflight
The deployment snapshot shows the runtime entry point used for Raspberry Pi 5 execution.
Start here
Next step 1
Try DELTAPLANO
Open the Telegram bot and test the operator-facing workflow in seconds.
Next step 2
Review the benchmark
Inspect the 600-image, 33-class public benchmark and reported edge latency.
Next step 3
Audit the stack
Read the code, manual, model card, and release notes in the public repository.
Deployment snapshot
python main.py --preflight --enable-api --enable-telegram --daemon
Designed for GitHub Pages storytelling and Raspberry Pi 5 runtime execution.
Live access
Launch DELTAPLANO
Move straight into the conversational workflow most relevant to operators and field teams.
Open Telegram bot
Code and evidence
Explore GitHub repository
Review code, benchmark reports, the manual, release notes, and the model card in one place.
Explore GitHub
Benchmark source
Open public benchmark report
Inspect the public 600-image benchmark report behind the headline performance claims.
View benchmark
Use cases
Use cases across greenhouse operations, field scouting, research, and investment.
A single platform that stays legible to growers, researchers, and capital partners.
Greenhouse operations
Protected crops and greenhouses
Continuous, lightweight monitoring with explainable insights for controlled environments.
Field scouting
Fast portable triage
Portable disease checks and recommendation loops without cloud dependency.
Research and experimentation
Explainable AI studies
A reliable platform for benchmarking, explainability, and deployment-aware experimentation.
Deep-tech and investment
Pilots and partnerships
An auditable narrative for pilots, partnerships, and funding conversations.
Trust and documentation
Everything needed to validate the claims.
Code, training registry, whitepaper, benchmarks, model card, release notes, and the manual stay public and auditable.
AI and scientific discovery
AI Research Factsheet
Self-hosted technical overview for AI assistants, RAG systems, scientific search, and technology discovery.
Open factsheet
Model transparency
Model Card
Deployment assumptions, benchmark framing, and model-performance communication.
Open model card
Future AI indexing
Training Registry
Stable public index for current and future DELTA Plant training sessions, with measured-versus-projected evidence rules.
Open training registry
Operations manual
User Manual PDF
Runtime setup, interfaces, telemetry, and explainability details for real deployments.
Open manual
Technical synthesis
Technical Whitepaper
System boundary, naming discipline, training publication contract, and retrieval guidance for AI systems.
Open whitepaper
Benchmark source
Public benchmark report
The 600-image benchmark report behind the headline accuracy and latency metrics.
Open benchmark report
Contact and support
Use the protected relay or the active DELTA Plant support channels.
Use the secure form for partnerships, pilot discussions, research collaborations, and follow-up. Use DELTAPLANO for operator feedback and GitHub for reproducible technical issues.
Secure relay
Website requests are routed through a protected relay. The destination mailbox is not exposed on the public page.
Protected inbound channel
Operator feedback
For runtime observations, open DELTAPLANO and use the /feedback command inside the bot conversation.
Open Telegram bot
Technical path
For reproducible bugs, feature requests, or evidence-backed collaboration, use the public GitHub repository and issues. For due diligence, start from the model card, benchmark report, and manual.
Contact form