🐟 Fish Classification using Deep Learning

📌 Project Overview

This project focuses on classifying fish images into multiple categories using deep learning models. The workflow involves:

1. Training a custom CNN model from scratch.
2. Leveraging transfer learning with multiple state-of-the-art pre-trained models.
3. Evaluating performance and selecting the best model.
4. Saving trained models for future inference.
5. Deploying the best model (MobileNet) using Streamlit to create an interactive web app for predictions from user-uploaded images.

---

📂 Dataset

• Type: Image dataset containing multiple fish species.
• Classes:

1. fish sea_food shrimp
2. fish sea_food striped_red_mullet
3. fish sea_food sea_bass
4. animal fish bass
5. fish sea_food black_sea_sprat
6. fish sea_food red_mullet
7. fish sea_food gilt_head_bream
8. fish sea_food red_sea_bream
9. animal fish : 1096
10. fish sea_food trout
11. fish sea_food hourse_mackerel

• Image Size: Resized to a fixed dimension (224x224) before training.
• Source: Labmentix

---

🧠 Models Used

Model	Type	Notes
Custom CNN	From scratch	Baseline deep learning model.
ResNet50	Transfer Learning	Fine-tuned on fish dataset.
VGG16	Transfer Learning	Fine-tuned on fish dataset.
MobileNet	Transfer Learning	Achieved best accuracy (≈99%).
EfficientNetB0	Transfer Learning	Balanced performance and speed.
InceptionV3	Transfer Learning	High accuracy and robustness.

---

📊 Model Performance

Note: MobileNet was selected for deployment due to its lightweight architecture, high accuracy, and fast inference speed.

---

🛠️ Tech Stack

• Programming Language: Python
• Deep Learning: TensorFlow / Keras
• Data Handling: NumPy, Pandas
• Image Processing: OpenCV, Pillow
• Visualization: Matplotlib, Seaborn
• Web Deployment: Streamlit
• Model Deployment File: .h5 format

---

⚙️ Installation & Setup

1️⃣ Clone the Repository

git clone https://github.com/yourusername/fish-classification.git
cd fish-classification

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Download the Dataset

• Place the dataset in the data/ folder.
• Update the dataset path in the training script.

4️⃣ Train Models (Optional)

python train.py

5️⃣ Run the Streamlit Web App

streamlit run app.py

---

🖥️ Web App Features

✅ Upload an image of a fish. ✅ Model predicts fish species with confidence scores. ✅ Confidence scores are shown as a bar graph. ✅ Mobile-friendly interface.

---

📁 Project Structure

📦 fish-classification
│── app.py                                                    # Streamlit app script
│── Multiclass-Fish-Image-classification.ipynb                # Model training script
│── best_mobilenet_model.h5                                   # Saved best model as .h5 file
│── requirements.txt                                          # Python dependencies
│── runtime.txt                                               # Runtime environment
│── README.md                                                 # Project documentation
│── data/                                                     # Dataset folder

---

🚀 Deployment

• The best MobileNet model (best_mobilenet_model.h5) was deployed using Streamlit.
• The app is hosted and accessible via: Streamlit Cloud
• Click here to view the app

---

📌 Future Improvements

• Expand dataset for more species.
• Improve generalization with data augmentation.
• Deploy a multi-model comparison app.
• Optimize for mobile devices.

---