Bird Species Identification Using Deep Learning

Overview

This project implements a Convolutional Neural Network (CNN) for automatic bird species classification using audio recordings. By converting bird sounds into spectrograms and applying deep learning techniques, we've developed a robust model for ecological monitoring and wildlife conservation.

Features

Data Collection

• Dataset: 2161 audio files
• 114 distinct bird species
• Sourced from Kaggle: Bird Sound Dataset

Deployment

• Streamlit web interface
• Support for MP3 and WAV uploads
• Real-time bird species classification

Project Objectives

• Develop a CNN-based classification model using TensorFlow
• Improve bird species identification accuracy in noisy conditions
• Create an automated solution for ecological monitoring
• Support conservation activities through technology

Key Technologies

• Python
• TensorFlow
• Convolutional Neural Network (CNN)
• Keras
• Streamlit
• Librosa
• OpenCV
• NumPy

Methodology

Data Preprocessing

• Resampling audio files to standard frequency (16 kHz)
• Converting audio to spectrograms using Short-Time Fourier Transform (STFT)
• Applying noise reduction and normalization techniques
• Implementing data augmentation (time stretching, pitch shifting)

Model Architecture

• 3 Convolutional layers (32, 64, 128 filters)
• Max-pooling layers for dimensionality reduction
• 2 Fully connected layers
• Softmax output layer
• Adam optimizer (learning rate: 0.001)
• Dropout regularization

Performance Metrics

• Accuracy: 0.7875
• Precision: 0.7919
• Recall: 0.6875
• F1-Score: 0.7129

Challenges Addressed

• Handling similar bird call patterns
• Managing background noise in recordings
• Accounting for variations in bird vocalizations

Potential Applications

• Biodiversity monitoring
• Population tracking
• Habitat health assessment
• Conservation research

Installation

Prerequisites

pip install tensorflow
pip install scikit-learn
pip install opencv-python
pip install librosa
pip install numpy
pip install pandas
pip install streamlit
pip install streamlit_extras

Usage

1. Clone the repository

git clone https://github.com/your-repo/Bird-Sound-Classification.git

2. Install requirements

pip install -r requirements.txt

3. Run Streamlit Application

streamlit run main.py

4. Access the app at http://localhost:8501

License

This project is licensed under the MIT License. Please review the LICENSE file for more details.

Contact

• Email: villwin11@gmail.com
• LinkedIn: Dhanush Saravanan

Future Work

• Expand dataset with more bird species
• Improve model generalization
• Reduce computational complexity
• Real-time bird sound classification