STO-Activities: (no title)
Activity title:
Multimodal Learning for SAR/ISAR ATR
Activity Reference:
SET-349
Panel:
SET
Security Classification:
Other
Status:
Proposed
Activity type:
RTG
Start date:
2024-09-21T00:00:00Z
Actual End date:
2027-09-21T00:00:00Z
Keywords:
ATR, Deep Learning, generative AI, ISAR, Machine Learning, SAR, SET
Background:
The previous SET-283 group undertook a series of activities to investigate various aspects of modern Machine Learning and advanced signal processing for SAR/ISAR ATR, and the influence of their usage on recognition performances. The group co-organized the SET-273/RSM Specialists’ Meeting on Multidimensional Radar Imaging and ATR that took place in Marseille, France on 25-26 October 2021. Furthermore, the group strongly collaborated with SET-288 “Integrating Compressive Sensing and Machine Learning Techniques for Radar Applications” and held three joint meetings.
Objectives:
One of the overall goals of the proposed RTG is to foster collaboration between the machine learning and radar remote sensing communities. More specifically, this group aims to study the different areas where multimodality could improve the SAR/ISAR ATR process. Firstly by producing accessible, organized and diverse datasets for robust ATR research results via several complementary solutions: EM simulation, AI-based data generation and augmentations (e.g. generative AI), small-scale measurements and 3D printing, and real data. Then, by adapting, developing and testing various machine learning algorithms, from advanced signal processing to the latest deep learning methodologies (e.g finetuning YOLO or SAM foundation models for segmentation). Finally, by working on how multimodality could improve the interaction of the proposed ATR strategies with end users (e.g. SAR images automatic labeling). Additionally, this group plans to invest in explainable learning (e.g. XAI), collaborative and federated distributed learning (e.g. sharing neural net features or weights without sharing real data), and countermeasures to combat against emerging threats (e.g. DNN fooling versus AI camouflage).
Topics:
SAR/ISAR Automatic/Assisted Target Recognition on diverse radar target signatures databases (simulations, indoor and outdoor measurements, AI generation …). Machine/Deep Learning. Deep Neural Networks (e.g. CNN, YOLO, SAM …). Generative Deep Learning (GAN, Stable Diffusion, …)
Contact:
Open2Partners:
Title:
Created at 23/04/2024 19:01 by System Account
Last modified at 16/05/2024 13:00 by System Account
Go back to list
Home(NATO STO)