micposso/Farm-Pest-Detection-AI-Mobile-App

Android mobile application designed to assist West African farmers in detecting and managing crop pests. The app uses a fine-tuned MobileNet model that runs directly on the device, enabling offline pest identification and localized pest control recommendations for low-connectivity environments. # MkulimaGuard: On-Device Crop Pest Detection and Management App for West African Farmers. ## 🌍 Overview Crop pests remain one of the biggest challenges facing small-scale farmers in West Africa, causing significant yield losses and threatening food security. Many farmers lack access to reliable pest identification tools or internet connectivity needed for online solutions. **MkulimaGuard** is an Android mobile app designed to assist farmers in detecting and managing crop pests directly from their smartphones. The app uses an **on-device fine-tuned MobileNet model** for image recognition, providing **localized pest control advice** — even in **offline environments**. ### 🧰 Technologies Used - **React Native / Kotlin** for mobile development - **TensorFlow Lite / MobileNet** for on-device image recognition - **Local LLM** for generating natural-language pest control advice - **SQLite / AsyncStorage** for local data management --- ## 🚜 Features - 📸 **Real-time pest detection** using the device camera - 🧠 **On-device MobileNet model** (no internet required) - 🌐 **Localized pest information** and control methods - 💬 **Multilingual support** — English, Hausa, Yoruba, French - ☀️ **Lightweight design** optimized for low-end Android devices - 🔋 **Low power and data usage** to suit rural connectivity conditions --- ## 🧠 Model The MobileNet model was **fine-tuned** on a curated pest image dataset to recognize common crop pests found across West Africa. ### 🔍 Dataset - **Source:** PlantVillage and locally collected pest datasets - **Classes:** Fall Armyworm, Aphids, Stem Borers, Whiteflies, and others - **Format:** 224×224 pixel RGB images ### 📊 Performance - Model size: ~14 MB (TensorFlow Lite) - Accuracy: 92% on validation set - Latency: < 300ms per inference on mid-range Android devices ### 🧩 Conversion The trained TensorFlow model was converted to **TensorFlow Lite (.tflite)** for mobile inference and integrated into the Android app for **offline pest detecti …