Whale Tail Classification System - MMAI

Overview

The Whale Tail Classification System is designed to compare whale tail images, determining whether they belong to the same individual. The project is inspired by the Humpback Whale Identification competition on Kaggle.

Components

1. Data Loading and Preprocessing

   • Implemented functionality for fetching data from the competition website.
   • Processed and prepared data for training both the Convolutional Neural Network (CNN) model and feature point extraction. For additional information you can check Data loading README.

2. CNN Model

   • Trained a convolutional neural network model capable of high-precision classification when provided with two whale tail images. To see model training process in detail you can check the narrated CNN notebook.

3. Feature Points

   • Identified the optimal parameters for a feature point-based solution using the Scale-Invariant Feature Transform (SIFT) methodology. You can see the example of feature points application in the Feature points notebook.

4. API

   • Developed an API for image classification, serving as a crucial tool for the Telegram bot and website components.
   • Enables requests from the Telegram bot and website to access the classification functionality without the need to individually deploy and manage the CNN model.

5. Telegram Bot

   • Implemented a Telegram bot that accepts two whale tail images and performs classification.
   • Users can choose between feature points or the CNN model for classification.
   • Deployed the Telegram bot on the Heroku platform, ensuring continuous access to its functionality.
   • User requests are forwarded to the API, and the classification results are returned. To see more information about usage and running the bot you can check Telegram bot README.

6. iOS App

   • Developed an iOS application for convenient use of the system on iPhones.
   • Features an intuitive interface and aesthetically pleasing design for user-friendly interaction.
   • The model is stored locally on the user's device, allowing system usage even without an internet connection.

7. Website

   • Created a website for image classification, with user requests forwarded to the API.
   • Provides an accessible and user-friendly interface for interaction with the system. For additional information you can check Website README.

Conclusion

The Whale Tail Classification System offers a comprehensive solution for comparing whale tail images. Its modular design allows for flexibility in choosing classification methods, and the user-friendly interfaces provided by the Telegram bot, iOS app, and website enhance the overall accessibility of the system.