TerraHawk – AI Drone-Based Precision Agriculture System

Abstract

This work presents a farmer-oriented assistant that combines real-time crop detection with knowledge-based agronomic guidance. The system uses a YOLOv5-based computer vision module to detect crops and visually annotate images and videos with bounding boxes and confidence scores. It also features a bilingual chatbot (English/Kannada) that answers specific questions using a curated agricultural knowledge base. The backend is built with Fast API and offers unified REST endpoints for detection, chat, and history. Meanwhile, a React and Vite frontend delivers an interactive user interface for uploading media, viewing annotated results, and chatting with the assistant.

From a design perspective, the system features a modular architecture. The detection model is wrapped in a lightweight layer and works with a multi-stage retrieval pipeline. This pipeline uses exact matching, crop-intent keywords, fuzzy token overlap, and semantic embeddings with FAISS. This design ensures strong, low-latency performance on standard hardware. It also lowers the chance of incorrect answers by using template-based response generation and explicit logging. The paper details the overall architecture, reviews related work in crop detection and agricultural decision support, and compares the proposed crop detection subsystem to traditional and modern machine learning methods. The discussion covers strengths, limitations, and design trade-offs, especially regarding deployment challenges, extensibility, and suitability for smallholder farming contexts.