ZehraouiMazilda/IfruitVision

An AI-based system for juice detection and classification using YOLOv11, SAM, and ResNet50 applied to the “Ifruit 100%” Algerian brand # 🍊 iFruitVision — Intelligent Food Supply Chain Monitoring **iFruitVision** is a computer vision project developed as part of my Master's thesis in *Intelligent Information Systems* (UMMTO). The main goal is to **automatically recognize Algerian “Ifruit 100%” juice bottles from retail shelf images**, in order to analyze product visibility and availability in stores. --- ## 🧠 Project Overview In modern supply chains, shelf monitoring is often manual, time-consuming, and error-prone. **iFruitVision** aims to automate this process using **Artificial Intelligence** and **Image Processing**. The system detects and classifies Ifruit 100% juice bottles, even under challenging conditions such as blur, lighting variation, and complex backgrounds. --- ## ⚙️ System Architecture The project is divided into three main stages: ### 1️⃣ Image Segmentation - **Model used:** *Segment Anything Model (SAM)* by Meta AI - **Goal:** Extract all visible objects from shelf images (bottles, backgrounds, irrelevant objects, etc.). - The generated segments are saved with a **transparent background**, forming the base of the dataset. ### 2️⃣ Dataset Construction - Manual creation of a dataset containing over **4000 image segments**, divided into: - `clean/` → clear juice segments - `trash/` → irrelevant or noisy objects - Dataset balanced to **70% clean / 30% noisy** images. - **Data augmentation** applied: rotation, zoom, brightness/contrast/saturation adjustments, perspective transformation, and Gaussian blur. ### 3️⃣ Detection and Classification - **YOLOv11** is used for **object detection**. - Several classification models were evaluated: - **ResNet50** - **EfficientNet-B7** - **MobileNetV3-Large** - The final pipeline combines **YOLOv11** for detection and **ResNet50** for classifying the 13 variants of the *Ifruit 100%* brand. Download ResNet50_iFruitVision.pt - Everything is integrated into a **Streamlit web application** that: 1. Allows image upload 2. Performs automatic det …