This Final Year Project, submitted to IBA Karachi, is a Pakistan-specific initiative aimed at empowering agricultural stakeholders, researchers, and policymakers with data-driven insights. The system uses multispectral satellite imagery and comparative machine learning models to predict crop yield which addresses the gap of open-source, localized agricultural intelligence in the country.

An additional feature is its end-to-end integration of an LLM-based chatbot, allowing users to interact with the system in natural language. This makes complex yield data and predictions accessible even to non-technical users, enabling better planning, resource allocation, and agricultural decision-making in Pakistan.

Documentation

• Software Requirements Specification
• System Design

🧠 Model Overview

We developed a satellite-based crop yield prediction system using publicly available Landsat imagery, tailored specifically for Pakistan's agricultural landscape.

• Region Focus: Division-wise areas across Pakistan

• Temporal Coverage: Multi-year imagery during the crop growing season

• Data Source: Landsat bands — especially Band 4 (Red) and Band 5 (Near-Infrared) — known for their value in vegetation analysis

🔧 Technical Approach

• We extracted relevant infrared and near-infrared data to assess crop health and growth.

• Each sample combined geographic and actual yield data (in kg/ha), forming a rich dataset for training.

• A custom Convolutional Neural Network (CNN) was trained on this data to map satellite features to crop yield.

• Typical yield values ranged between 2000–2500 kg per hectare.

🏗️ Model Architecture

Our crop yield prediction model uses a VGG-inspired Convolutional Neural Network (CNN) optimized for regression tasks on satellite imagery:

• 11 convolutional layers organized into four sequential blocks

• Each block uses the structure:
  3×3 Conv → BatchNorm → ReLU → Dropout

• Downsampling is achieved via stride‑2 convolutions instead of max-pooling at the end of Blocks 1 to 3

• Block 4 increases feature depth to 1024 channels without further spatial downsampling

• Output from the final conv layer is a feature map of 1024×16×16, which is flattened

• Passed through a 2048-unit fully connected layer

• Ends with a single linear output neuron for regression (predicting continuous crop yield)

This architecture was chosen for its balance of depth, parameter efficiency, and spatial awareness, making it well-suited for geospatial imagery and yield prediction.

🎯 Performance

• Achieved Root Mean Square Error (RMSE) of just 11%, reflecting strong prediction accuracy.

• This allows for timely, scalable, and accurate yield forecasting, aiding better agricultural planning, policy-making, and resource allocation.

Deployment

First, run the development server:

npm run dev
# or
yarn dev
# or
pnpm dev
# or
bun dev

Open http://localhost:3000 with your browser to see the result.

You can start editing the page by modifying app/page.js. The page auto-updates as you edit the file.

This project uses next/font to automatically optimize and load Geist, a new font family for Vercel.

Authors

• Muddasir Rizwan
• Muhammad Hadi
• Ahmed Raza
• Moeen Haider
• Hasan Atiq