AgroWatch

AgroWatch Edge AI is a solar powered smart farming system that delivers real time insights crop health detection offline resilience and data driven tools to boost yield and reduce operational costs

Comment

About the Project

Inspiration

The inspiration behind AgroWatch stemmed from a deep desire to bridge the gap between smallholder farmers and cutting-edge agricultural technology. In many rural areas, internet connectivity is unreliable, and energy access is limited. Yet, these regions are often the most dependent on consistent crop yields for livelihood. We wanted to create a low-power, offline-capable, and intelligent monitoring solution that could empower farmers to make better decisions without relying on expensive infrastructure.

What We Learned

Throughout the development journey, we gained valuable insights in multiple domains:

  • Edge AI: Learned how to quantize and optimize neural networks (INT8) to run efficiently on ARM Cortex-M4 processors with minimal power and latency.
  • Environmental Sensing: Explored calibration techniques for soil moisture, temperature, humidity, and light sensors to ensure accurate, consistent readings.
  • Power Optimization: Designed adaptive sampling strategies to maximize battery life and operate continuously for 7+ days with no solar input.
  • Data Visualization: Used charts and historical data plots to present complex environmental patterns in a farmer-friendly interface.
  • Frontend Architecture: Strengthened our experience with Vite, TypeScript, React, Tailwind CSS, and component libraries like shadcn-ui.

How It Was Built

We structured the system in two core layers: hardware and software.

Hardware

  • Powered by a 10W solar panel and 7Ah Li-ion battery
  • Sensor suite: temperature, humidity, soil moisture, light level
  • Edge AI board: Cortex-M4 running a 2.4MB quantized model for disease detection
  • 32GB SD card for local data storage
  • Adaptive sampling and power budgeting for offline operation

Software

  • Frontend: Vite + React + TypeScript + Tailwind CSS
  • UI: Built with shadcn-ui components and custom visualization charts
  • Data Feed: Simulated using setInterval with realistic patterns
  • AI Inference: Simulated results from diseases like Powdery Mildew or Leaf Blight with confidence scores and advice
  • ROI Calculator: Uses a model with fixed improvement assumptions:

$$ \text{Yield Increase} = 12\%, \quad \text{Pesticide Reduction} = 95\%, \quad \text{Labor Reduction} = 30\% $$

Challenges Faced

  • Power Efficiency: Balancing inference speed and power draw on low-power microcontrollers was complex.
  • Offline Support: Creating an architecture that worked reliably for 7+ days without internet was challenging.
  • Model Size Constraints: Reducing AI model size without sacrificing accuracy required careful quantization and pruning.
  • Simulated Data Design: Designing believable yet flexible simulation logic for demo purposes while preserving production readiness.

Problem Definition and Context

Global agriculture is under growing pressure to increase food production sustainably, while facing challenges such as climate change, resource scarcity, and labor inefficiencies. Smallholder farmers—who contribute over 70% of the world’s food supply—are especially vulnerable due to limited access to precision technologies, reliable energy, and stable internet connectivity.

Most modern agricultural monitoring systems rely heavily on cloud computing, continuous connectivity, and high-powered hardware, which are often unavailable or unaffordable in rural and underserved areas. This leads to a persistent gap in timely crop monitoring, early disease detection, and efficient resource management.

AgroWatch Edge AI was developed to close this gap by offering a solar-powered, offline-capable, AI-driven monitoring system that can operate independently of cloud infrastructure. It empowers farmers with actionable insights to optimize yields, reduce input waste, and make smarter decisions—all from a lightweight edge device.

Identified Constraints

Power Constraints

  • The device is powered by a 10W solar panel and 7Ah Li-ion battery.
  • It must support 7+ days of continuous operation without sunlight.
  • Edge AI inference is optimized to consume only 0.8 mJ per run, completing in under 120 ms.

Connectivity Limitations

  • Designed for offline-first operation, with no reliance on internet for core functionality.
  • Uses adaptive data batching and stores logs on a 32 GB SD card for later upload or review.

Compute Constraints

  • Runs on an ARM Cortex-M4 processor with limited flash and RAM, and no floating-point unit.
  • Requires neural network models to be quantized to INT8 and under 2.4 MB in size.

Data Constraints

  • Sensor input may be noisy due to environmental variance.
  • Model training must accommodate diverse crop types, soil conditions, and climates to remain generalizable.

Cost Constraints

  • Hardware and build must remain affordable for smallholder farms.
  • Components are carefully chosen to balance performance, durability, and price.

Environmental Constraints

  • The system must endure heat, rain, dust, and direct sunlight.
  • Sensor placement and casing are designed to minimize maintenance and maximize lifespan in the field.

Accomplishments That We're Proud Of

  • Successfully built a solar-powered, edge AI system capable of offline operation for over a week.
  • Deployed a fully functional simulation of real-time sensor monitoring and AI crop disease detection.
  • Designed a user-friendly dashboard that visualizes environmental metrics, power data, and historical trends.
  • Achieved significant power efficiency while maintaining inference speed on limited hardware.
  • Created a working ROI calculator that communicates the economic value of the system to end users.

What's Next for AgroWatch

  • Integrating a real AI model on-device to replace simulation logic.
  • Expanding disease detection classes using transfer learning and region-specific datasets.
  • Adding support for LoRaWAN and GSM modules for periodic remote sync when networks are available.
  • Building a mobile-friendly interface for in-field access and management.
  • Partnering with agricultural cooperatives and NGOs to field test and iterate based on real farmer feedback.
  • Exploring multilingual voice interfaces to improve accessibility for non-literate users in rural regions.

Built With

Share this project:

Updates