Intel-CNN-Image-Classification

The dataset consists of images categorized into six distinct classes: 🏢 Buildings | 🌲 Forest | 🏔 Glacier | ⛰ Mountain | 🌊 Sea | 🛣 Street

Two different approaches are implemented to assess performance and effectiveness:

1. CNN Model with Transfer Learning – A model leveraging VGG19, a pre-trained deep learning architecture, to enhance feature extraction and improve classification accuracy.
2. CNN Model Trained from Scratch – A custom-built convolutional neural network trained without any pre-existing weights.

• Deep Learning Model: CNN-based classifier implemented using TensorFlow/Keras.
• Transfer Learning: Uses VGG19 as the pre-trained model for improved accuracy.
• Performance Metrics: Evaluates accuracy, precision, recall, and confusion matrix.
• Modular Code Structure: Well-organized for easy modification and experimentation.

• IDE: Jupyter Lab
• Programming Language: Python
• Deep Learning Framework: TensorFlow/Keras
• Data Processing: OpenCV, NumPy, Pandas
• Visualization: Matplotlib, Seaborn
• Hardware Acceleration: GPU (CUDA-enabled for TensorFlow)

The Intel Image Dataset consists of images categorized into six natural and man-made scenery classes. It is a widely used benchmark dataset for scene recognition and classification tasks. The dataset is structured into training, validation, and test sets to facilitate model evaluation.

Dataset Structure

• Train Set: 14,034 images
• Test Set: 3,000 images
• Prediction Set: 7,301 images

[!TIP] You can download the dataset from here: Intel Image Classification Dataset

🔧 Prerequisites

[!IMPORTANT]

• Ensure Anaconda is installed, if not you can download from Anaconda and also Git (if not available, download from Github).
• Also download the mentioned dataset before running any of the notebooks, and change the paths in the notebooks whereever necessary.

⚙️ Installation

Once Anaconda is installed, open the Anaconda Prompt and run the following commands:

1. Clone the repository:

git clone https://github.com/NSANTRA/Intel-CNN-Image-Classification

2. Navigate to the project directory:

cd Intel-CNN-Image-Classification

3. Create a new Conda environment:

conda env create -f "Tensorflow.yml"

4. Activate the environment:

conda activate Tensorflow

▶️ Usage

After activating the environment:

• Open Jupyter Notebook or JupyterLab within the environment.
• Navigate to the project folder and open the desired notebook.
• Ensure dataset paths are correctly configured in each notebook.
• Run the cells sequentially to execute the project.

1️⃣ CNN Model with Transfer Learning (VGG19)

• Non-Trainable Parameters: 20,026,176

• Trainable Parameters: 4,361,606

• Total Parameters: 24,387,782

• Optimizer: Adam

• Loss Function: Sparse Categorical Crossentropy

2️⃣ CNN Model (Trained from Scratch)

• Non-trainable Parameters: 1,216

• Trainable Parameters: 22,701,734

• Total Parameters: 22,702,950

• Optimizer: Adam

• Loss Function: Sparse Categorical Crossentropy

Classification Report for CNN With Transfer Learning

🔹 Key Observations:

• ✅ High Overall Accuracy: 89% – The model performs well across all classes.
• ✅ Forest category has the highest accuracy (Precision: 0.96, Recall: 0.99, F1-score: 0.98) – Very few misclassifications.
• ✅ Buildings, Sea, and Street categories also perform well (F1-score: ~0.91).
• ✅ Glacier and Mountain have the lowest scores (F1-score: ~0.82) – These categories are harder to classify correctly.

Class-Wise Weights

🔹 Key Takeaways:

• 📌 Transfer Learning significantly boosts accuracy, with an overall F1-score of 0.89.
• 📌 Forest classification is near-perfect, while Glacier and Mountain have the most confusion.
• 📌 Further improvements can be made by refining the model’s ability to differentiate Glaciers and Mountains.

**Classification Repor