RD Image Datasets of OTHR Interference

This paper proposes a novel idea for the detection of the radio frequency interference (RFI) and the transient interference (TSI) in sky-wave over-the-horizon (OTH) radar, based on the classification of the range-Doppler (RD) images, since the RFI, the TSI and no interference (NoI) have different textures in the RD map. The RD image classifiers and the image databases are designed for the first time. The performances of image classification and interference detection are evaluated based on the image database that is constructed based on real data of OTH radar. New models are developed for the RFI and the TSI to construct the simulated database, so that the classifiers can use it as the training set, in no need of other data for its designing. Four classification algorithms are employed for the classifier design, where three textural features can be extracted. Experimental results show that the RD image classifiers obtain over $95\%$ classification accuracy and detection probability, at a cost of false alarm probability less than $2\%$.本文针对天波超视距（sky-wave over-the-horizon, OTH）雷达中的射频干扰（radio frequency interference, RFI）与瞬态干扰（transient interference, TSI）检测问题，提出了一种基于距离多普勒（range-Doppler, RD）图像分类的创新思路。由于射频干扰、瞬态干扰与无干扰（no interference, NoI）在RD图谱中具有各异的纹理特征，该思路具备坚实的应用基础。本文首次构建了RD图像分类器与专用图像数据集，并基于真实天波超视距雷达采集数据构建的图像数据集，对图像分类与干扰检测的综合性能进行了评估。针对射频干扰与瞬态干扰，本文研发了新型仿真模型以生成模拟数据集，使得分类器可直接以该数据集作为训练集，无需额外获取或设计其他训练数据。本次分类器设计共采用四种分类算法，可提取三类纹理特征用于分类任务。实验结果表明，所提出的RD图像分类器可实现超过95%的分类准确率与检测概率，且虚警概率控制在2%以内。

搜集汇总

背景与挑战

背景概述

该数据集是用于高频超视距雷达干扰检测的64*64大小灰度图像集合，包含模拟和真实两类数据库，图像分为射频干扰、瞬态干扰和无干扰三种类型。数据集主要用于干扰检测分类器的训练和测试，其中模拟数据库通常作为训练集，真实数据作为测试集。