Uncrewed Aircraft Detection from YAMNET Embedding Dataset

This dataset is the main dataset used by the UA Detection Team here at Embry-Riddle Aeronautical University. It provides the ability to conduct binary drone/no drone classification as well as specific drone sub-type classifcation. Dataset is gathered from field data from drones flown by the Unavail research team. Each clip of audio is fed through YAMNET to generate points in an embedding space. The no_drone directory contains various sounds from ambient noise to cars and jets. drone contains samples from the DJI Matrice M100, the Mavic 3, and the Mavic Mini 2. There are 9108 samples total across all the directories. Each .tfdata segment represents a single second of audio.

本数据集为安柏瑞德航空大学（Embry-Riddle Aeronautical University）无人机检测团队所使用的核心数据集，支持无人机/非无人机二元分类任务，同时可实现特定无人机子类型分类。该数据集源自Unavail研究团队开展无人机飞行作业时采集的现场数据。每一段音频片段都会通过YAMNET模型处理，以生成嵌入空间中的特征点。其中，no_drone目录包含从环境噪声到汽车、喷气式飞机等各类非无人机声音样本；drone目录则收录了DJI Matrice M100、Mavic 3以及Mavic Mini 2三款无人机的采集样本。所有目录下总计共有9108个样本，每个.tfdata片段对应一秒时长的音频。

搜集汇总

背景与挑战

背景概述

该数据集是一个基于YAMNET音频嵌入的无人机检测数据集，由Embry-Riddle Aeronautical University团队创建，包含9108个一秒音频样本，用于二进制无人机/非无人机分类和特定无人机子类型识别。数据来源于实地无人机飞行音频，并涵盖环境噪音、汽车和喷气机等非无人机声音，以及DJI Matrice M100、Mavic 3和Mavic Mini 2等无人机型号，通过密集神经网络分类可实现高达96%的测试准确率。