About Me

Overview

I am deeply interested about leveraging machine learning algorithms to address diverse and complex tasks. Specifically, I have worked on projects related to hyperspectral anomaly detection, semantic segmentation on automotive radar point cloud, and continuous learning. Moving forward, I am fervently driven by the passion to delve deeper into machine learning and to carve a meaningful niche through dedicated research. You can find my CV here.

Projects

Semantic Segmentation on Radar Point Cloud

During my master’s research, I focus on semantic segmentation in sparse and irregular radar point clouds, a challenging task dedicated to improving the ability to detect targets using radar sensors in autonomous driving. I proposed a network that utilizes spatio-temporal relationships in radar multi-frame data to enhance the semantics of the point cloud, thus achieving the state-of-the art performance. In particular, it shows a remarkable ability to segment small objects such as pedestrians.

This work has been accepted by IEEE Transactions on Intelligent Vehicles.

Hyperspectral Anomaly Detection

Hyperspectral anomaly detection is a critical facet of remote sensing and image analysis, providing unparalleled insights into Earth’s surface composition. Unlike traditional imaging, hyperspectral sensors capture a vast range of spectral bands, enabling detailed analysis of materials and substances. The goal of anomaly detection in hyperspectral imagery is to identify outliers or irregularities that deviate from expected spectral signatures.

During my undergraduate studies, I validated the impact of low-rank constraints in enhancing the ability of deep learning networks to extract background information from hyperspectral imagery, earning honors for an outstanding undergraduate thesis.

In the first year of my master’ program, I conducted ablation experiments for the paper titled “Hyperspectral Anomaly Detection Based on Adaptive Low-Rank Transformed Tensor”, published in IEEE Transactions on Neural Networks and Learning Systems. My primary focus was on reproducing non-open-source networks, contributing significantly to the research showcased in the publication.