Bird Species Classification

The Bird Species Classification Project, a term project for the Cs464- Introduction to Machine Learning course, is an academic initiative that explores the synergy between avian biodiversity and computational intelligence. We harness an array of sophisticated algorithms, including Multinomial Logistic Regression, Random Forest, and advanced neural networks like Convolutional Neural Network, along with state-of-the-art Transfer Learning methodologies embodied by EfficientNet-B0 and ResNet-50. This project meticulously adopts data preprocessing and augmentation strategies to ensure the robustness and reliability of our classification outcomes.

Motivation 🔥

The Bird Species Image Classification Project is driven by the vision to revolutionize ornithological studies, replacing meticulous manual identification with a swift, automated system powered by machine learning. This innovation seeks to bolster biodiversity conservation efforts through enhanced, precise species recognition in complex ecosystems.

Data Preprocessing and Augmentation 📊

• Data Balancing: Implemented equal image distribution for top ten bird species and improved test/validation sets.
• Grayscale Conversion: Reduced images from three color channels to one to emphasize shape and texture over color, enhancing computational efficiency.
• Image Resizing: Adjusted image resolution to 64x64 pixels for machine learning models and retained 224x224 for CNNs and Transfer Learning to balance detail and performance.
• Data Augmentation: Increased robustness through random flips, rotations, crops, brightness, and contrast adjustments, with PyTorch tensor conversion for model training.

Findings 🔍

• Notably, convolutional neural networks and transfer learning models demonstrated superior performance, especially in recognizing subtle differences between similar bird species.

Utilized Machine Learning Algorithms ⚙️

• Multinomial Logistic Regression
• Random Forest
• Convolutional Neural Network (CNN)
• Transfer Learning (EfficientNet-B0 & ResNet-50)

Future Work 🔭

Efforts are underway to extend the project's capabilities. Future enhancements include diversifying the dataset with more bird species and integrating cutting-edge machine learning techniques to refine classification accuracy. We also aim to develop real-time identification features to support biodiversity research and fieldwork.

Contributors 👥

• Ömer Tuğrul
• Selin Ataş
• Mennan Gök
• Gökay Balcı