VERUS5/palm-tree-classifier-app

Bilingual (Arabic/English) mobile app for identifying date palm varieties (Khalas, Razeez, Shishi) using a 5-fold ConvNeXt Small ensemble with Gemini AI RAG-powered agricultural chat. Built with Expo React Native + Express + PostgreSQL. # Semantic AI Framework for Palm Tree Variety Classification ## About the Project This project is a Semantic AI Framework designed to classify palm tree varieties (specifically Khalas, Razeez, and Shishi) and provide an interactive, knowledge-retrieval digital assistant for agricultural insights. By bridging the gap between traditional farming practices and modern artificial intelligence, this framework serves as a reliable, accessible tool for farmers and agricultural enthusiasts. ## Application Previews & Model Results Below is a look at the application in action and the performance of our underlying AI model: **1. Real-Time Image Classification** *The cross-platform mobile application allows users to upload or capture images of palm trees for instant analysis.* **2. RAG-Powered Agricultural Chatbot** *Beyond visual classification, the app features a Retrieval-Augmented Generation (RAG) chat assistant to address the agricultural knowledge gap.* **3. Model Performance (Confusion Matrix)** *Our custom ensemble of ConvNeXt-Small models demonstrated exceptional reliability. The confusion matrix above illustrates the high accuracy and minimal misclassification achieved across all three test classes (Khalas, Razeez, and Shishi) during our final testing phase.* ## How We Built It The system is divided into a cross-platform mobile application, a Retrieval-Augmented Generation (RAG) backend, and a dedicated AI inference server, ensuring a clean separation of operations. * **Frontend:** Built using React Native and Expo for a seamless, responsive user experience on both Android and iOS devices. * **Backend & RAG Engine:** Developed with Node.js and TypeScript. The RAG engine utilizes a hybrid retrieval system (combining keyword matching and semantic embeddings) alongside the Gemini GenAI model to parse queries and generate accurate agricultural advice. * **AI Server:** Hosted using Python, Flask, and PyTorch to efficiently process uploaded images and run the …