AAHAR
data driven cloud-based-prediction system for precision agriculture
Created on 9th February 2025
•
AAHAR
data driven cloud-based-prediction system for precision agriculture
The problem AAHAR solves
Punjab, one of India's largest agricultural states, faces significant challenges related to over-fertilization, which leads to long-term soil degradation and reduced productivity. Prolonged overuse or underuse of fertilizers disrupts soil health, affecting nutrient availability and crop yield.
Aahar is designed to address this issue by providing precise recommendations for UREA, MOP (Muriate of Potash), and DAP (Diammonium Phosphate) application based on real-time soil conditions. Our solution integrates a heterogeneous dataset derived from three primary sources:
NASA POWER LARC – for agro-climatic and meteorological data
Google Earth Engine – for satellite-based soil and vegetation indices
Locally Deployed Sensor Nodes – for real-time, high-resolution soil nutrient and moisture data
Our platform offers additional capabilities to support sustainable and data-driven farming:
CNN-Based Crop Disease Prediction: A deep learning model trained on plant pathology datasets enables early detection of diseases using image-based analysis.
Crop Rotation Optimization: Data-driven recommendations help improve soil fertility and prevent nutrient depletion.
Real-Time Crop Parameter Monitoring: Farmers can access continuous updates on soil moisture, pH, and nutrient levels to make informed decisions.
Frequent Soil Testing & Analysis: Automated alerts and insights help maintain soil health over multiple growing seasons.
AI-Powered Chatbot: Equipped with multilingual and voice support, the chatbot assists farmers in their native languages, offering real-time agronomic guidance and troubleshooting.
By leveraging advanced remote sensing, IoT-based soil monitoring, and AI-driven analytics, Aahar ensures optimal fertilizer usage, enhances soil sustainability, and maximizes yield while reducing environmental impact
Challenges we ran into
-
Data Preprocessing & Integration
Our heterogeneous dataset (NASA POWER LARC, Google Earth Engine, IoT sensors) had different time frequencies and missing values. We used interpolation and normalization to align data, ensuring accurate fertilizer recommendations. -
Backend for Real-Time Soil Data
Soil parameters arrived asynchronously, requiring an event-driven architecture. We built our backend with Flask, FastAPI, and PostgreSQL (TimescaleDB) to handle time-series data efficiently. -
CNN-Based Crop Disease Prediction
We trained a CNN on 70,000+ images sourced from multiple databases. Data augmentation, class balancing, and transfer learning (ResNet, EfficientNet) improved model accuracy. -
App Development Challenges
Our team was new to app development, making UI/UX design for farmers difficult. We optimized Flutter with Leaflet for maps and Chart.js for visualization, ensuring smooth performance on low-end devices. -
AI Chatbot with Multilingual & Voice Support
We built a chatbot with NLP (Google Dialogflow, fine-tuned BERT) and speech synthesis (TTS APIs) to support regional languages and voice interactions. -
Model Deployment Issues
Our CNN model was too large for cloud deployment. We optimized it with TensorFlow Lite (TFLite) and ONNX Runtime for on-device inference. -
Farmer-Friendly UI
We designed an offline-capable, color-coded dashboard with large icons, ensuring accessibility for farmers with limited digital literacy.
Despite challenges, Aahar integrates AI, remote sensing, and IoT to provide precise fertilizer recommendations, real-time monitoring, and disease detection, making farming smarter and more sustainable.
Tracks Applied (1)