A deep learning-powered image classification application that identifies four different species of parrots from uploaded images. The model is built using Transfer Learning with MobileNetV2 and achieves an impressive 96% classification accuracy.
-
Classifies 4 Parrot Species
- African Grey Parrot
- White Parrot
- Amazon Green Parrot
- Macaw
-
Transfer Learning using MobileNetV2
-
Lightweight and efficient architecture
-
User-friendly frontend built with Streamlit
-
Real-time image prediction
-
High classification accuracy of 96%
The model uses a pre-trained MobileNetV2 backbone for feature extraction, followed by custom classification layers:
MobileNetV2 (Pre-trained)
↓
Global Average Pooling
↓
Dropout Layer
↓
Dense Layer
↓
Softmax Output Layer (4 Classes)
- Lightweight and computationally efficient
- Excellent performance on image classification tasks
- Suitable for deployment on resource-constrained environments
| Metric | Value |
|---|---|
| Accuracy | 96% |
| Classes | 4 |
| Architecture | MobileNetV2 + Custom Head |
- Python
- TensorFlow / Keras
- MobileNetV2
- NumPy
- Matplotlib
- Streamlit
- Collect and preprocess parrot image dataset.
- Apply image augmentation and normalization.
- Use MobileNetV2 as the feature extractor.
- Add custom classification layers.
- Train and validate the model.
- Deploy the prediction interface using Streamlit.
| Class | Description |
|---|---|
| African Grey | Intelligent grey-colored parrot species |
| White Parrot | White-feathered parrot species |
| Amazon Green | Green-colored Amazon parrot |
| Macaw | Large colorful parrot species |
The frontend was developed using Streamlit and works correctly in a local environment. However, public deployment could not be completed due to compatibility issues between TensorFlow, Python versions, and hosting platform dependencies.
Despite the deployment limitation, the project runs successfully on a local machine with the required environment setup.
Subhadip Some
Aspiring AI/ML Engineer passionate about Deep Learning, Computer Vision, and building intelligent applications that solve real-world problems.