micheline24/enhanced_microdsc: enhanced microDSC codes

Enhanced MicroDSC v1.0.0 — Initial Release

This is the first official release accompanying the paper: "Lightweight deep learning model for real-time acoustic bird pest detection on edge microcontrollers" Submitted to PeerJ Computer Science.

Our model enables real-time bird pest detection on resource-constrained microcontrollers, providing an affordable precision agriculture solution for African smallholder farmers.

Repository Structure

• experiments/ — Ablation studies and architecture comparison
• training/ — Complete model training pipeline and evaluation
• deployment/ — ESP32 Arduino implementation for real-time field deployment

Model Performance

• Architecture: Enhanced MicroDSC (7,483 parameters)
• Accuracy: 97.4–98.6% across 10 bird species
• Inference Time: <100ms on ESP32
• Target Device: ESP32 (~$7), solar-powered

Author

Micheline Kazeneza PhD Student, University of Rwanda — RSIF/ICIPE Fellow

Acknowledgments

Supported by ICIPE–World Bank Financing Agreement No. D347-3A and World Bank–Korea Trust Fund Agreement No. TF0A8639 (PASET-RSIF).