Doppler Radar Target Classification

Problem

Radar classification requires extracting useful information from noisy time-series signals. The goal was to classify target types from Doppler-pulse radar data using signal-processing and machine-learning methods.

Constraints

• Noisy radar signals
• Limited compute resources
• Need to transform raw signals into useful model inputs
• Competition environment with many participants

Approach

Built a pipeline using signal processing, filtering, FFT-based features, spectrogram representations, and deep-learning experimentation with CNN/RNN-style architectures.

The work focused on extracting micro-Doppler patterns and turning difficult radar signals into learnable representations.

Result

Produced a competitive radar-classification solution.

Placed 30th out of 1,000+ participants.

Commercial relevance

This case demonstrates ability to work with scientific signals, constrained data, noisy measurements, feature engineering, and ML experimentation.

It is relevant to medical imaging, biotech instrumentation, sensor AI, radar, robotics, and other domains where the input data is not clean web data.

Confidentiality note

This is a public project case study based on existing project material.