Underwater hyperspectral imaging dataset of manganese nodules

Manganese nodule samples were collected from a seamount in the western Pacific at a depth of approximately 4000 to 5000 meters. The samples are predominantly brown or black in color, with significant differences in their external appearance. One type has a fuller and larger form, coarser texture, and a surface with pronounced small protrusions. In contrast, the other type is flatter, smaller in size, smoother in texture, and has a relatively even surface.A self-developed underwater hyperspectral imaging system was used to acquire hyperspectral image data of the two manganese nodule samples in a laboratory setting. The hyperspectral imaging system has a spectral range of 390 nm to 1000 nm with a spectral resolution of less than 5 nm and operates using a push-broom imaging method.The samples were placed in a black fish tank filled with tap water, with a layer of marine sand spread evenly on the bottom as the background. The manganese nodule samples were placed on the surface of the sand. The distance between the samples and the system was 54 cm. A 250W halogen lamp was used as the light source to illuminate the samples.A standard diffuse reflectance calibration panel was used as a reference to calibrate the spectral reflectance of the target. Spectral bands with low signal-to-noise ratios were discarded, retaining data for 290 bands within the spectral range of 390 nm to 880 nm. The calibrated hyperspectral images of the manganese nodules were cropped to exclude irrelevant edge information, leaving only the region of interest containing the manganese nodule samples, with spatial dimensions of 272 × 527 pixels. Semantic segmentation labels were generated for the samples using Labelme software, creating ground truth maps for the objects of interest.

本数据集的锰结核样本采集自西太平洋某海山，采集水深约4000至5000米。样本主体呈棕褐色或黑色，外观存在显著差异：一类形态饱满硕大、质地粗糙，表面带有明显的细小凸起；另一类则相对扁平小巧、质地细腻，表面较为平整。
本研究采用自研水下高光谱成像系统（hyperspectral imaging system），在实验室环境下采集两类锰结核样本的高光谱图像数据。该系统光谱范围为390 nm至1000 nm，光谱分辨率优于5 nm，采用推扫式成像（push-broom imaging）方式。
实验中将样本置于盛有自来水的黑色鱼缸中，缸底均匀铺覆一层海沙作为背景，锰结核样本放置于沙面之上。样本与成像系统的距离为54 cm，采用250W卤钨灯作为光源对样本进行照明。
采用标准漫反射校正面板（diffuse reflectance calibration panel）作为参考，对目标的光谱反射率进行校正。剔除信噪比过低的光谱波段，最终保留390 nm至880 nm光谱范围内的290个波段数据。对校正后的锰结核高光谱图像进行裁剪，去除无关的边缘信息，仅保留包含锰结核样本的感兴趣区域，其空间尺寸为272 × 527像素。使用Labelme软件为样本生成语义分割标签，为目标对象创建真值标注图（ground truth）。