The MASSIMAL Dataset: UAV Hyperspectral Images and Ground Truth Observations for Coastal Habitat Mapping in Norway
Authors: Martin Hansen Skjelvareid, Eli Rinde, Kasper Hancke, Katalin Blix, Galice Guillaume Hoarau, Nicolo Grometto, Maia Røst Kile
Organizations: UiT The Arctic University of Norway, Hansine Hansens veg 18, 9019 Tromsø, Norway · Norwegian Institute for Water Research (NIVA), Sognsveien 72, 0855 Oslo, Norway. · Nord University, Universitetsall´een 11, 8026 Bodø, Norway
The MASSIMAL dataset was collected in an effort to develop new remote sensing methods for mapping shallow water coastal habitats, such as kelp forests, seagrass meadows and maerl beds. The main component of the dataset is a set of hyperspectral images collected using a unmanned aerial vehicle (UAV). Images were collected during 45 UAV flights performed at 4 main areas along the Norwegian coast, spanning across a wide latitude range, from 59 to 67 degrees north. The ground sampling distance of the hyperspectral images is 3.5 cm, and the spectral range is 400-1000 nm, sampled across 300 spectral bands. During acquisition, push-broom imagery was automatically split into images with a maximum along-track length of 2000 lines. Out of approximately 1000 hyperspectral images, 221 have been annotated, using segmentation masks and a hierarchical taxonomy. An open source image processing pipeline has been developed as part of the dataset preparation, enabling efficient generation of reflectance images and glint-corrected images based on published radiance images. The dataset also includes color images (RGB) and multispectral images collected using UAV, and georeferenced image mosaics constructed from these. Ground truth data supporting the remote sensing images include hours of underwater video, thousands of geotagged underwater images, and tens of kilometers of single-beam sonar data. The data provide a foundation for assessing habitat distribution, tracking changes, and evaluating ecosystem services, thus supporting evidence-based management in Norwegian coastal environments.
Figures & tables
Figure 1: The map in a) shows an overview of the four main areas in Norway where field work was performed, and the maps in b)-e) show specific locations within these areas. The axes of each map show latitude and longitude in degrees, but to save space, only the axes of map a) are labeled. Each location in b)-e) is annotated with the types of aerial images acquired (hyperspectral / multispectral / RGB), and ground truth sampling methods used (underwater imaging from boat or USV, drop-camera imaging, sonar, imaging while snorkeling, and imaging while walking). All maps were made available by the Norwegian Mapping Authority under the CC-BY-4.0 license [ 6 ] .
Figure 2: Overview of methods used in aerial imaging and ground truth sampling methods. Image a.1 and b.1 show multicopter and fixed-wing drones, and images a.2 and b.2 show a false-color representation of a hyperspectral push-broom image and an RGB image, respectively. Image c.1 shows the ”Otter” USV, and c.2 shows an example underwater image taken from the USV, showing bedrock, brown algae, and a large number of green sea urchins. Image d.1 shows a small boat used under field work, and image d.2 shows an underwater image of kelp canopy taken from the boat, using a camera mounted on a pole. Image e.1 shows the above-water monitor and electronics for the drop camera, and image e.2 shows an example image from the drop camera. Image f.1 shows a researcher snorkeling with a camera on a pole, and image f.2 is an example image of a maerl bed taken during snorkelling. Image g.1 shows the ”FiFish” ROV and image g.2 shows an example ROV image, again showing a maerl bed. Image h.1 shows a person walking with a GNSS antenna in the intertidal zone, and h.2 is an example of an image of rockweed taken during a walking transect.
Figure 3: Map a) shows a visualization of multiple methods for collecting ground truth, and map b) shows corresponding annotations made on hyperspectral images. Example from Søla, Vega. The coverage of hyperspectral images is indicated in both maps, with each polygon representing one flight (one dataset). The base map for both maps was made available by the Norwegian Mapping Authority under the CC-BY-4.0 license [ 6 ] .
Grouped classes
Original Classes
Not included
Deep water, Algae, Green algae, Turf, Animals, Mussels, Mytilus edilus, Trawl track, Human activity
Brown Algae
Brown algae, Kelp, Laminaria hyperborea, L. digitata, Sacchoriza polyides, S. latissima, Alaria esculenta, Rockweed, Ascophyllum nodosum, Fucus spp., Halidrys siliquosa, Chorda filum, Desmarestia aculeata
Maerl / Red Algae
Red algae, Coralline algae, Maerl
Seagrass
Seagrass, Zostera marina
Rock
Rock, Bedrock, Boulder, Cobble, Gravel
Sediment
Substrate, Sediment, Sand, Mud
Figure 4: Hierarchy of marine habitats and substrates used for annotations. The table summarizes the suggested class groupings for a simplified taxonomy. Classes listed under ”not included” are limited to single locations and/or a very small number of images.
Location
Date
Flights
Images
Annotations
DI
Ground truth
Bodø
Juvika
2021-04-18
4
60
0
Yes
Snorkeling
2022-06-24
1
36
3
Yes
Walking
Kvikstad
2023-10-13
2
53
0
1/2
Boat imaging
Saltstraumen
2022-03-12
3
68
0
Yes
Boat imaging
Sandsund
2021-04-18
1
19
0
Yes
Snorkeling
Table 1: Overview of all hyperspectral datasets, with areas, locations, number of flights (each flight corresponds to one dataset), the total number of images per location, the number of annotated images per location, whether downwelling irradiance (DI) was measured, and what types of ground truth were collected.
Figure 5: Spectral statistics for annotated regions of hyperspectral reflectance images. The upper left plot shows statistics for the the dataset as a whole, and the remaining plots shows statistics for five grouped classes.
Figure 6: Visualization of example spectra from 5 grouped annotation classes, reduced to 2 dimensions via PCA and UMAP.
Figure 7: Example image with reference gray card placed on a beach among clusters of macroalgae. Black spots in the image correspond to invalid or saturated pixels in the original radiance data. The image was taken in Juvika, Bodø (file massimal_bodo_juvika_202104181131-transects-15m_hsi_002 )
Figure 8: Two examples of images with glint (a,c) and the corresponding images with glint removed (b,d). All images are RGB representations of hyperspectral images, contrast stretched individually per band. Image a) was captured at Larvik, Ølbergholmen ( massimal_larvik_olbergholmen_202308300959-south1_hsi_008 ) and image c) at Smøla, Måholmen ( massimal_smola_maholmen_202306211129-2_hsi_009 ).
Benthic habitat mapping is fundamental for understanding marine ecosystems, guiding conservation efforts, and supporting sustainable resource management. Yet, the scarcity of large, annotated datasets limits the development and benchmarking of machine learning models in this domain. This paper introduces a thorough multi-modal dataset, comprising about a million side-scan sonar (SSS) tiles collected along the coast of Catalonia (Spain), complemented by bathymetric maps and a set of co-registered optical images from targeted surveys using an autonomous underwater vehicle (AUV). Approximately 36000 of the SSS tiles have been manually annotated with segmentation masks to enable supervised fine-tuning of classification models. All the raw sensor data, together with mosaics, are also released to support further exploration and algorithm development. To address challenges in multi-sensor data fusion for AUVs, we spatially associate optical images with corresponding SSS tiles, facilitating self-supervised, cross-modal representation learning. Accompanying open-source preprocessing and annotation tools are provided to enhance accessibility and encourage research. This resource aims to establish a standardized benchmark for underwater habitat mapping, promoting advancements in autonomous seafloor classification and multi-sensor integration.
Hayat Rajani, Valerio Franchi, Borja Martinez-Clavel Valles +3
Computer Vision and Robotics Research Institute, University of Girona, Spain · Tecnoambiente SLU, Spain
The proliferation of unmanned aerial vehicles (UAVs) has created urgent demand for precise UAV monitoring. Existing RGB-based systems rely on spatial cues that degrade at small scales, particularly with high inter-type similarity, target-clutter ambiguity, and low contrast. Multispectral imaging (MSI) encodes material-aware spectral signatures, yet MSI-based fine-grained small-UAV detection remains underexplored due to lack of dedicated datasets. We introduce UAVNet-MS, the first multispectral dataset for fine-grained small-UAV detection, comprising 15,618 temporally synchronized RGB-MSI data cubes (1440x1080) with bounding box annotations. The dataset features challenging small objects (93.7% <= 32^2 pixels, average 18^2 pixels, ~0.02% image area) under low contrast. We propose MFDNet, a dual-stream baseline addressing array-induced parallax and spatial-spectral fusion. Extensive evaluation under RGB-only, MSI-only, and RGB+MSI protocols against 20 detectors shows MFDNet achieves +6.2% AP50 improvement over best RGB-only methods, demonstrating spectral cues provide complementary material evidence beyond spatial cues. This work provides foundational dataset, strong baseline, and benchmark for multispectral UAV monitoring research.
Yihang Luo, Jun Chen, Chao Xiao +13
College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China · Aviation University of Air Force, China · Sun Yat-sen University, China
We present HyperImageNet, a large-scale benchmark for fine-grained hyperspectral land-cover understanding. The dataset contains 26,084 airborne hyperspectral image patches with 224 spectral bands and 138 fine-grained land-cover categories. Unlike existing datasets, HyperImageNet provides raw imagery, pixel-level semantic labels, and object-level instance masks, supporting both semantic and instance segmentation. Furthermore, we establish an open-environment benchmark with strict spatial separation to evaluate representative methods and the HyperFree foundation model. Experimental results demonstrate the effectiveness of HyperImageNet for fine-grained hyperspectral understanding and open-environment remote sensing research.