UAE-RS
收藏资源简介:
UAE-RS数据集是首个提供遥感领域黑盒对抗样本的数据集,由先进人工智能研究所的徐永浩和Pedram Ghamisi创建。该数据集包含通过Mixup-Attack和Mixcut-Attack方法生成的对抗样本,旨在帮助研究者设计对抗攻击抵抗力强的深度神经网络。UAE-RS数据集适用于场景分类和语义分割任务,通过模拟真实世界中无法获取目标模型详细信息的攻击场景,推动了遥感图像处理领域对抗防御技术的发展。
The UAE-RS dataset is the first dataset providing black-box adversarial samples in the remote sensing domain, created by Xu Yonghao and Pedram Ghamisi from the Advanced Artificial Intelligence Research Institute. This dataset includes adversarial samples generated via the Mixup-Attack and Mixcut-Attack methods, aiming to assist researchers in designing deep neural networks with robust resistance against adversarial attacks. The UAE-RS dataset is applicable to scene classification and semantic segmentation tasks. By simulating real-world attack scenarios where detailed information of the target model is inaccessible, it promotes the development of adversarial defense technologies in the field of remote sensing image processing.

- 1Universal Adversarial Examples in Remote Sensing: Methodology and Benchmark先进人工智能研究所(IARAI) · 2022年



