Research
Ground penetrating radar: clutter removal and buried object detection
GPR data are dominated by ground-surface reflections and clutter that mask the responses of buried objects. We develop clutter removal methods that range from low-rank and sparse models (robust matrix factorization, learned RPCA, learned dictionaries) to deep networks (convolutional autoencoders, vision transformers, B-spline activations), and we combine them with deep detectors, up to joint networks that remove clutter and detect buried objects at the same time. Related work covers recovering missing GPR measurements, clutter removal on FPGA, and tomographic GPR imaging of trees and cylindrical objects.
Related publications:
- Joint Deep Learning for Simultaneous Clutter Removal and Buried Object Detection in GPR (2026)
- Enhancing Deep Learning Networks Performance By Using B-Spline Activation Functions for Clutter Removal in GPR (2024)
- Autoencoder Guided Low-Rank Approximation Approach for Clutter Removal in GPR Images (2024)
- Combining Image Restoration with Deep Detectors for Improved Target Detection in GPR (2024)
- A Vision-Transformer-Based Approach to Clutter Removal in GPR: DC-ViT (2024)
- Clutter Removal in Ground Penetrating Radar by Learned RPCA (2023)
- Clutter Removal in Ground Penetrating Radar with Learned Dictionaries (2023)
- A New Clutter Removal Method Based on Direct Robust Matrix Factorization for Buried Target Detection (2022)
- Extracting tomographic images of interior structures of cylindrical objects and trees using Ground Penetrating Radar method (2022)
- Clutter Removal for Ground Penetrating Radars on FPGA: Design and Implementation (2022)
- Missing Data Recovery via Deep Networks for Limited Ground Penetrating Radar Measurements (2022)
- A Novel Convolutional Autoencoder-Based Clutter Removal Method for Buried Threat Detection in Ground-Penetrating Radar (2022)
Through-obstacle imaging and short-range radar sensing
With low-cost, portable radar measurement systems we image what lies behind walls and inside structures. This includes background subtraction and wall mitigation with tensor RPCA and non-negative matrix factorization, imaging the interior of reinforced concrete, and detecting the vital signs of people in debris environments with stepped-frequency continuous-wave (SFCW) radar.
Related publications:
- Music guided SVD method for the detection of vital signs with SFCW radar (2025)
- Detection of Vital Signs in Debris Environment using SFCW Radar (2024)
- Imaging of Reinforced Concrete Structures Using Radar (2024)
- Fast Background Subtraction for Through Obstacle Imaging Using a Low Cost, Portable Measurement System (2022)
- A Novel Tensor RPCA Based Background Subtraction Method for Outdoor Imaging With a Low Cost Portable Radar (2022)
- Non-negative matrix factorization based approaches for wall mitigation in TWRI (2022)
SAR and ISAR imaging
We form synthetic aperture radar (SAR) and inverse synthetic aperture radar (ISAR) images from limited or incomplete measurements, using sparse recovery with nuclear norm minimization, plug-and-play ADMM with entropy-based thresholding, and deep image priors. We also collect SAR images with a low-cost radar mounted on a mini drone.
Related publications:
Radar target detection, classification and direction finding
We apply deep learning to radar detection and recognition: transformer networks for target detection, despeckling-based data augmentation for target classification, recurrent networks on range-Doppler maps for helicopter classification, and an encoder-decoder RetNet for interference mitigation in FMCW radar. For direction-of-arrival (DOA) estimation we study deep learning in MIMO radars and in the autocorrelation domain together with autoregressive modelling.
Related publications:
- An Encoder-Decoder Retnet Model for Fmcw Radar Interference Mitigation (2025)
- Radar Target Detection using Improved Transformer Neural Network (2025)
- Direction Of Arrival Estimation In Autocorrelation Domain With Deep Learning and Autoregressive Modelling (2024)
- 2D Overlapping Range-Doppler Map Approach for Helicopter Classification by Using GRU (2022)
- DOA Estimation in MIMO Radars via Deep Learning (2022)
- Despeckling Based Data Augmentation Approach in Deep Learning Based Radar Target Classification (2022)
Sonar imaging and underwater object detection
Below the waterline, we detect objects in forward-looking sonar images with transformer-guided image enhancement and cross-modality knowledge distillation, and we form synthetic aperture sonar (SAS) images from limited measurements with deep learning.
Related publications:
Electronic warfare and passive RF sensing
For electronic support (ES) systems we study receiver search strategies based on randomized low-rank and sparse decomposition, time delay estimation, TDOA-based target localization and specific emitter identification with deep learning. We also process ADS-B signals, with a reliability-based decoder for low-SNR conditions and a neural network that separates overlapping signals.
Related publications:
- A Scalable Dual-Representation Deep Learning Architecture for Specific Emitter Identification (2026)
- TDOA-Based Target Localization in Passive Radar Electronic Support Measures: A Hybrid Chan-PSO Approach (2026)
- A Blind, Coherent, Reliability-Based ADS-B Decoder for Low-SNR Environments (2026)
- ADS-B Signal Separation with a Dual-Path Convolutional-Recurrent Neural Network (2026)
- Non-Negative Matrix Factorization Guided FS-GCC Method For Time Delay Estimation in Electronic Warfare (2025)
- Randomized Low Rank and Sparse Decomposition Based Receiver Search Strategy For Electronic Support Receivers (2022)