Animal Classification with ResNet152V2

Deep learning model for classifying 22 different animal species using transfer learning with ResNet152V2.

Model Architecture

• Base Model: ResNet152V2 (pre-trained on ImageNet)
• Custom Layers: Global Average Pooling + Dense output layer
• Total Parameters: 174.8M (222.1M trainable)
• Input Size: 224x224 RGB images

Dataset

Uses the UCF Flying Wildlife Classification Through Detection (FWCTD) dataset with 104,495 images across 22 animal classes.

Data Split:

• Training: 70% (73,147 images)
• Validation: 20% (20,899 images)
• Testing: 10% (10,449 images)

Performance

• Test Accuracy: 96.02%
• Test Loss: 0.1575
• Training completed in 10 epochs

Requirements

tensorflow
numpy
opencv-python
matplotlib

Training Configuration

• Optimizer: Adam (learning rate: 0.001)
• Loss Function: Categorical Crossentropy
• Batch Size: 64
• Epochs: 10

Usage

from tensorflow import keras

# Load the trained model
model = keras.models.load_model('animal_classification.keras')

# Make predictions
predictions = model.predict(your_image_batch)

Results

The model achieves high accuracy with consistent performance across training, validation, and test sets, demonstrating effective transfer learning from ImageNet features.