Skip to main content
Featured Project

AI-Powered Bird Detector & Repellent System

An IoT-based system that detects birds and repels them using AI-powered object detection and automated speaker activation.

AI & Embedded System Engineer
6 months
4 developers
AI-Powered Bird Detector & Repellent System - Landscape View
AI-Powered Bird Detector & Repellent System - Portrait View
ESP32-CAM Streamlit YOLOv5 Flask Solar Panel Python

Architecture Diagram

graph LR A[ESP32-CAM Solar] --> B[Flask API Server] B --> C[YOLOv5 Detection] B --> D[Streamlit Dashboard] C --> E[Speaker Trigger]

A smart agricultural IoT system designed to automatically detect birds in rice fields and trigger a speaker to scare them away.

Problem

Petani kehilangan hingga 30% hasil panen karena burung yang memakan padi. Solusi manual seperti orang-orangan sawah tidak efektif dan membutuhkan pengawasan terus-menerus.

Solution

Membangun IoT system dengan AI detection yang otomatis mengaktifkan speaker pengusir saat burung terdeteksi, mengurangi kebutuhan pengawasan manual dan meningkatkan efektivitas perlindungan panen.

Architecture Decisions

  • Chose ESP32-CAM over Raspberry Pi for cost efficiency ($10 vs $50) and lower power consumption — critical for solar-powered deployment
  • Used YOLOv5 for detection because of optimal balance between accuracy (85%+) and inference speed (~100ms)
  • Chose Flask + Streamlit for rapid prototyping — needed working dashboard in 2 weeks, not production-grade infrastructure
  • Offloaded AI inference to server instead of edge due to ESP32 memory limitations

Trade-offs

  • Limited camera resolution (640x480) for bandwidth efficiency on rural networks
  • Detection delay ~2 seconds is acceptable — birds don’t fly away instantly
  • Simple threshold-based speaker trigger vs complex behavioral patterns — simpler = more reliable

Lessons Learned

  • Start with minimal viable IoT setup, add complexity only when needed
  • Power consumption is the #1 constraint for solar-powered devices
  • Field testing is mandatory — lab conditions are nothing like real rice fields

Features

  • Real-time bird detection with YOLOv5
  • Speaker activation on detection
  • Solar-powered ESP32-CAM device
  • Web-based dashboard to monitor live camera feed
  • Configurable speaker and camera settings
  • Rice disease analysis from uploaded images
  • Automatic image capture and logging

Challenges Overcome

  • Deploying custom AI models on limited IoT hardware
  • Maintaining stable ESP32-CAM live stream to dashboard
  • Synchronizing detection events with hardware output (speaker)
  • Designing a power-efficient system with solar charging
  • Building responsive dashboard UI using Streamlit