FESG3002/Camer_FRUITCLASSIF_Resnet-SVM

An advanced image classification system tailored for identifying common fruits in Cameroon (Pineapple, Orange, Banana, Avocado, and Watermelon). This project leverages a hybrid methodology, combining Deep Learning (VGG16/ResNet) for high-level semantic features and Computer Vision (SIFT) for precise texture and shape descriptors. # Cameroon Fruit Recognizer A advanced image recognition application tailored for identifying the most frequent fruits in Cameroon using a hybrid deep learning and computer vision approach. ## Project Structure - **src/**: Source code for the application and GUI. - **models/**: Pre-trained models (VGG16 features, SVM classifier). - **data/**: Dataset and test images. - **documentation/**: Project reports, methodology details, and old READMEs. - **notebooks/**: Jupyter notebooks for experimentation and model training. - **requirements.txt**: List of dependencies for the project. ## Getting Started ### Prerequisites - Python 3.8+ - TensorFlow, OpenCV, joblib, rembg, customtkinter ### Installation 1. Clone the repository: ```bash git@github.com:FESG3002/Camer_FRUITCLASSIF_Resnet-SVM.git cd cameroon-fruit-recognizer ``` 2. Install dependencies: ```bash pip install -r requirements.txt ``` ### Running the Application To start the recognition interface: ```bash python src/interfacefrec.py ``` ## Methodology ### Dataset The models were trained on a dataset composed of: - **FID30 dataset**: A standard dataset for fruit identification. - **Manual Collection**: Specialized dataset of Cameroon-specific fruit varieties collected manually to improve local accuracy. ### Preprocessing & Feature Extraction The application targets **Ananas, Oranges, Bananes, Avocats, and Watermelons**. The pipeline includes: 1. **Background Removal**: Automated background subtraction to focus on the fruit item. 2. **Hybrid Feature Extraction**: - **CNN (Deep Learning)**: Using VGG16 to extract high-level semantic features. - **SIFT (Computer Vision)**: Using Scale-Invariant Feature Transform to capture local texture and shape descriptors. 3. **Fusion & Learning**: Combined features are fed into an SVM classifier for final prediction. ## Authors - Student from M1 group project from University of Yaounde 1 ## License MIT License