Real-Time Sonar Object Classification System

An end-to-end robotics and AI project designed to classify underwater objects (Mines vs. Rocks) based on sonar frequency responses. The project bridges the gap between cloud-based model training and local real-time deployment using the Robot Operating System (ROS 2).

Key Features:

Machine Learning Pipeline: Trained a Logistic Regression classifier on sonar dataset to achieve high-accuracy predictions.

ROS 2 Integration: Developed a custom ROS 2 package (sonar_predictor) that handles real-time data subscription and prediction publishing.

Optimized Deployment: Leveraged ONNX (Open Neural Network Exchange) to export the model, resolving dependency conflicts and ensuring high-speed inference without library version issues.

Real-Time Processing: Built to handle high-frequency data inputs, mimicking the behavior of Autonomous Underwater Vehicles (AUVs).