This repository showcases a deep learning project aimed at classifying fish species from the CIFAR-100 dataset. Leveraging the power of transfer learning with the VGG16 model, this implementation accurately categorizes images into five fine-grained fish species classes: aquarium fish, flatfish, ray, shark, and trout.
- Dataset: CIFAR-100, filtered for fish species classification.
- Model Architecture: Transfer learning using VGG16 with custom classification layers.
- Training Optimization: Dropout regularization and learning rate scheduling.
- Evaluation Tools: Confusion matrix and accuracy visualization.
- Visualization: Training history and prediction results.
Ensure you have the following installed:
- Python 3.7 or higher
- TensorFlow 2.x
- Keras
- NumPy
- Matplotlib
- OpenCV
Install all dependencies using:
pip install tensorflow keras numpy matplotlib opencv-pythonThe CIFAR-100 dataset comprises 60,000 32x32 color images across 100 classes. For this project, only the following fish species classes were selected:
- Aquarium Fish
- Flatfish
- Ray
- Shark
- Trout
The dataset was filtered to include only these classes, and the images were resized to 224x224 to match the input requirements of the VGG16 model.
├── Source Code/
│ └── train_fish_cifar100.py # Script for training and evaluating the model
├── README.md # Documentation
├── data/ # Directory for dataset (to be created by the user)
├── results/ # Directory for storing output (e.g., visualizations, metrics)The project utilizes a modified VGG16 model with the following components:
- Pre-trained Layers: The VGG16 model, pre-trained on ImageNet, is used as the feature extractor with its convolutional layers frozen to retain pre-learned weights.
- Custom Fully Connected Layers: Added on top of the pre-trained layers to handle the fish species classification task.
- Dropout Layers: Used to reduce overfitting during training by randomly disabling neurons.
- Softmax Activation: Ensures probabilistic outputs for multi-class classification.
- Loss Function: Sparse categorical cross-entropy is used to handle integer-encoded class labels.
- Optimizer: Adam optimizer with learning rate scheduling to adapt learning rates dynamically.
- Epochs: The model is trained for 20 epochs.
- Input Size: Images are resized from 32x32 to 224x224 to match the VGG16 input size requirements.
- Clone the repository:
git clone https://github.com/Gengo-bit/Image-Classification-of-Fish-Species-Using-the-CIFAR-100-Dataset
cd Image-Classification-of-Fish-Species-Using-the-CIFAR-100-Dataset- Prepare the CIFAR-100 dataset by downloading it and organizing the files into a data/ directory.
- Train the model by running the script:
cd Source Code python train_fish_cifar100.py - The results, including training metrics and visualizations, will be saved in the results/ directory.
The model demonstrated an accuracy range of 68% to 76%.
- Training and Validation Loss Trends:
- Training loss consistently decreases.
- Validation loss stagnates or increases after ~12 epochs, indicating overfitting.
- Confusion Matrix:
- Highlights the model's classification performance and areas of misclassification.
- Visualization:
- Displays predictions with true vs. predicted labels for test images.
- Overfitting:
- Challenge: The model overfits after a few epochs.
- Solution: Dropout layers and reduced model complexity were implemented to mitigate this issue.
- Long Training Times:
- Challenge: The large dataset and extended epochs increased training time.
- Solution: Learning rate scheduling was applied to speed up convergence.
- Fluctuating Accuracy:
- Challenge: Accuracy during training was inconsistent.
- Solution: Hyperparameter tuning and reduced epochs improved stability.
- Paul Emmanuel Corsino
- David Emmanuel Lacsao
- Klinnsonveins Yee
This project is licensed under the MIT License. See the LICENSE file for details.