AMIVU Acoustic Map Imaging VUB-ULB Dataset

This dataset contains a collection of both real captured and simulated acoustic images of the UMAP acoustic camera. Real captured images contain a single sound source for different positions and frequencies. The images are generated by first capturing the microphone signals and later process them for different resolutions. The simulated acoustic images are generated by simulating two sound sources at mirrored angles for different frequencies between 2kHz and 10kHz, with steps of 500Hz. These frequencies are independent of each other. The sound sources start at a position of 60° and move in steps of 2° towards the center. The distance between the sound sources and the center of the array is always 1m. This gives a total of 16 positions for the sound sources and 289 images per position. Each image is generated for 4 different resolutions (640 × 480, 320 × 240, 160 × 120, 80 × 60). For each one dataset was generated using fractional delays and another without fractional delays with a total of 36992 images. Images are standardized using instance min-max normalization and logged with uint8 data type, then converted to grayscale in the range [0 - 255] and saved as PNG format with zero compression. File naming for the simulated images: Pos_<angle_of_the_sound_sources>_R_<frequency_soundsource_1>_L_<frequency_soundsource_2> For more Information and citation please refer to the paper below.
@Article{s21103453,
AUTHOR = {Almasri, Feras and Vandendriessche, Jurgen and Segers, Laurent and da Silva, Bruno and Braeken, An and Steenhaut, Kris and Touhafi, Abdellah and Debeir, Olivier},
TITLE = {XCycles Backprojection Acoustic Super-Resolution},
JOURNAL = {Sensors},
VOLUME = {21},
YEAR = {2021},
NUMBER = {10},
ARTICLE-NUMBER = {3453},
URL = {https://www.mdpi.com/1424-8220/21/10/3453},
ISSN = {1424-8220},
DOI = {10.3390/s21103453}
}

本数据集收录了UMAP声学相机（UMAP acoustic camera）的实拍与仿真声学图像合集。实拍声学图像涵盖不同位置与频率下的单一声源场景，其生成流程为：首先采集麦克风阵列信号，随后针对不同分辨率对采集信号进行后处理。仿真声学图像则通过模拟两个镜像布置的声源生成：声源频率范围为2kHz至10kHz，步长500Hz，且两个声源的频率相互独立。声源初始位置为60°，并以2°为步长向阵列中心移动，声源与阵列中心的距离始终为1米。该设置共计生成16个声源位置，每个位置对应289张声学图像，且每张图像均对应4种不同分辨率：640×480、320×240、160×120及80×60。针对每种分辨率，分别采用分数延迟（fractional delays）与非分数延迟两种方案生成数据集，最终总计生成36992张图像。所有图像均通过实例最小-最大归一化（instance min-max normalization）进行标准化，以uint8数据类型存储，随后转换至[0, 255]区间的灰度图像，并以无压缩PNG格式保存。仿真图像的文件命名规则为：Pos_<声源角度>_R_<声源1频率>_L_<声源2频率>。如需获取更多信息或引用该数据集，请参阅以下论文：
@Article{s21103453,
AUTHOR = {Almasri, Feras and Vandendriessche, Jurgen and Segers, Laurent and da Silva, Bruno and Braeken, An and Steenhaut, Kris and Touhafi, Abdellah and Debeir, Olivier},
TITLE = {XCycles Backprojection Acoustic Super-Resolution},
JOURNAL = {Sensors},
VOLUME = {21},
YEAR = {2021},
NUMBER = {10},
ARTICLE-NUMBER = {3453},
URL = {https://www.mdpi.com/1424-8220/21/10/3453},
ISSN = {1424-8220},
DOI = {10.3390/s21103453}