MT-SARD
收藏资源简介:
MT-SARD数据集是一个自构建的多任务合成孔径雷达(SAR)飞机识别数据集,包含原始图像、类别、掩码和投影等信息。该数据集通过基于预定结构模板的结构化注释方法构建,为深度学习提供了结构化和组成信息补充,有利于网络理解飞机的结构知识,提高识别的鲁棒性和解释能力。数据集旨在促进SAR图像中飞机识别任务的研究,特别是在提高泛化能力和解释性方面具有显著优势。
The MT-SARD dataset is a self-built multi-task synthetic aperture radar (SAR) aircraft recognition dataset that contains raw images, category labels, masks, projection information and other relevant metadata. It is constructed via a structured annotation method based on pre-defined structural templates, which provides supplementary structured and compositional information for deep learning models. This facilitates neural networks to grasp the structural knowledge of aircraft, and effectively enhances the robustness and interpretability of recognition tasks. The dataset is intended to advance research on aircraft recognition in SAR images, and exhibits prominent advantages particularly in improving model generalization ability and interpretability.

- 1MTSGL: Multi-Task Structure Guided Learning for Robust and Interpretable SAR Aircraft RecognitionIEEE Publication Technology Group · 2025年



